[clang] 16a385b - [OpenMP] Modernize the kernel launching interface and APIs
Johannes Doerfert via cfe-commits
cfe-commits at lists.llvm.org
Sat Jan 21 11:31:45 PST 2023
Author: Johannes Doerfert
Date: 2023-01-21T11:16:21-08:00
New Revision: 16a385ba21a921a42009ff971199bce56eb2a86e
URL: https://github.com/llvm/llvm-project/commit/16a385ba21a921a42009ff971199bce56eb2a86e
DIFF: https://github.com/llvm/llvm-project/commit/16a385ba21a921a42009ff971199bce56eb2a86e.diff
LOG: [OpenMP] Modernize the kernel launching interface and APIs
We already created a versioned `__tgt_kernel_arguments` struct but it
was only briefly used and its content was passed in isolation anyway.
This makes it hard to add more information in the future. With this
patch we fully embrace the struct as means to pass information from the
compiler to the plugin as part of a kernel launch.
The patch also extends and renames the struct, bumping the version
number to 2. Version 1 entries are auto-upgraded. This is in preparation
for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP
lowering, etc.
The `__tgt_target_kernel_nowait` interface was deprecated as it was
unused. Once we actually implement support for something like that, we
can add an appropriate API.
Note: Only plugins with the `launch_kernel` interface are now supported.
That means that a new clang won't be able to use an old runtime.
An old clang can still use the new runtime since the libomptarget
interface did not change.
Differential Revision: https://reviews.llvm.org/D141232
Added:
Modified:
clang/lib/CodeGen/CGOpenMPRuntime.cpp
clang/test/OpenMP/declare_mapper_codegen.cpp
clang/test/OpenMP/distribute_codegen.cpp
clang/test/OpenMP/distribute_firstprivate_codegen.cpp
clang/test/OpenMP/distribute_lastprivate_codegen.cpp
clang/test/OpenMP/distribute_parallel_for_codegen.cpp
clang/test/OpenMP/distribute_parallel_for_firstprivate_codegen.cpp
clang/test/OpenMP/distribute_parallel_for_if_codegen.cpp
clang/test/OpenMP/distribute_parallel_for_lastprivate_codegen.cpp
clang/test/OpenMP/distribute_parallel_for_num_threads_codegen.cpp
clang/test/OpenMP/distribute_parallel_for_private_codegen.cpp
clang/test/OpenMP/distribute_parallel_for_proc_bind_codegen.cpp
clang/test/OpenMP/distribute_parallel_for_simd_codegen.cpp
clang/test/OpenMP/distribute_parallel_for_simd_firstprivate_codegen.cpp
clang/test/OpenMP/distribute_parallel_for_simd_if_codegen.cpp
clang/test/OpenMP/distribute_parallel_for_simd_lastprivate_codegen.cpp
clang/test/OpenMP/distribute_parallel_for_simd_num_threads_codegen.cpp
clang/test/OpenMP/distribute_parallel_for_simd_private_codegen.cpp
clang/test/OpenMP/distribute_parallel_for_simd_proc_bind_codegen.cpp
clang/test/OpenMP/distribute_private_codegen.cpp
clang/test/OpenMP/distribute_simd_codegen.cpp
clang/test/OpenMP/distribute_simd_firstprivate_codegen.cpp
clang/test/OpenMP/distribute_simd_lastprivate_codegen.cpp
clang/test/OpenMP/distribute_simd_private_codegen.cpp
clang/test/OpenMP/distribute_simd_reduction_codegen.cpp
clang/test/OpenMP/interop_irbuilder.cpp
clang/test/OpenMP/nvptx_lambda_capturing.cpp
clang/test/OpenMP/reduction_implicit_map.cpp
clang/test/OpenMP/target_codegen.cpp
clang/test/OpenMP/target_codegen_global_capture.cpp
clang/test/OpenMP/target_data_map_codegen_hold.cpp
clang/test/OpenMP/target_depend_codegen.cpp
clang/test/OpenMP/target_has_device_addr_codegen.cpp
clang/test/OpenMP/target_has_device_addr_codegen_01.cpp
clang/test/OpenMP/target_map_codegen_03.cpp
clang/test/OpenMP/target_map_codegen_hold.cpp
clang/test/OpenMP/target_map_member_expr_codegen.cpp
clang/test/OpenMP/target_offload_mandatory_codegen.cpp
clang/test/OpenMP/target_parallel_codegen.cpp
clang/test/OpenMP/target_parallel_depend_codegen.cpp
clang/test/OpenMP/target_parallel_for_codegen.cpp
clang/test/OpenMP/target_parallel_for_depend_codegen.cpp
clang/test/OpenMP/target_parallel_for_simd_codegen.cpp
clang/test/OpenMP/target_parallel_for_simd_depend_codegen.cpp
clang/test/OpenMP/target_parallel_if_codegen.cpp
clang/test/OpenMP/target_parallel_num_threads_codegen.cpp
clang/test/OpenMP/target_simd_codegen.cpp
clang/test/OpenMP/target_simd_depend_codegen.cpp
clang/test/OpenMP/target_teams_codegen.cpp
clang/test/OpenMP/target_teams_depend_codegen.cpp
clang/test/OpenMP/target_teams_distribute_codegen.cpp
clang/test/OpenMP/target_teams_distribute_collapse_codegen.cpp
clang/test/OpenMP/target_teams_distribute_depend_codegen.cpp
clang/test/OpenMP/target_teams_distribute_dist_schedule_codegen.cpp
clang/test/OpenMP/target_teams_distribute_firstprivate_codegen.cpp
clang/test/OpenMP/target_teams_distribute_lastprivate_codegen.cpp
clang/test/OpenMP/target_teams_distribute_parallel_for_codegen.cpp
clang/test/OpenMP/target_teams_distribute_parallel_for_collapse_codegen.cpp
clang/test/OpenMP/target_teams_distribute_parallel_for_depend_codegen.cpp
clang/test/OpenMP/target_teams_distribute_parallel_for_dist_schedule_codegen.cpp
clang/test/OpenMP/target_teams_distribute_parallel_for_firstprivate_codegen.cpp
clang/test/OpenMP/target_teams_distribute_parallel_for_if_codegen.cpp
clang/test/OpenMP/target_teams_distribute_parallel_for_lastprivate_codegen.cpp
clang/test/OpenMP/target_teams_distribute_parallel_for_order_codegen.cpp
clang/test/OpenMP/target_teams_distribute_parallel_for_private_codegen.cpp
clang/test/OpenMP/target_teams_distribute_parallel_for_proc_bind_codegen.cpp
clang/test/OpenMP/target_teams_distribute_parallel_for_reduction_codegen.cpp
clang/test/OpenMP/target_teams_distribute_parallel_for_schedule_codegen.cpp
clang/test/OpenMP/target_teams_distribute_parallel_for_simd_codegen.cpp
clang/test/OpenMP/target_teams_distribute_parallel_for_simd_collapse_codegen.cpp
clang/test/OpenMP/target_teams_distribute_parallel_for_simd_depend_codegen.cpp
clang/test/OpenMP/target_teams_distribute_parallel_for_simd_dist_schedule_codegen.cpp
clang/test/OpenMP/target_teams_distribute_parallel_for_simd_firstprivate_codegen.cpp
clang/test/OpenMP/target_teams_distribute_parallel_for_simd_if_codegen.cpp
clang/test/OpenMP/target_teams_distribute_parallel_for_simd_lastprivate_codegen.cpp
clang/test/OpenMP/target_teams_distribute_parallel_for_simd_private_codegen.cpp
clang/test/OpenMP/target_teams_distribute_parallel_for_simd_proc_bind_codegen.cpp
clang/test/OpenMP/target_teams_distribute_parallel_for_simd_reduction_codegen.cpp
clang/test/OpenMP/target_teams_distribute_parallel_for_simd_schedule_codegen.cpp
clang/test/OpenMP/target_teams_distribute_private_codegen.cpp
clang/test/OpenMP/target_teams_distribute_reduction_codegen.cpp
clang/test/OpenMP/target_teams_distribute_simd_codegen.cpp
clang/test/OpenMP/target_teams_distribute_simd_collapse_codegen.cpp
clang/test/OpenMP/target_teams_distribute_simd_depend_codegen.cpp
clang/test/OpenMP/target_teams_distribute_simd_dist_schedule_codegen.cpp
clang/test/OpenMP/target_teams_distribute_simd_firstprivate_codegen.cpp
clang/test/OpenMP/target_teams_distribute_simd_lastprivate_codegen.cpp
clang/test/OpenMP/target_teams_distribute_simd_private_codegen.cpp
clang/test/OpenMP/target_teams_distribute_simd_reduction_codegen.cpp
clang/test/OpenMP/target_teams_map_codegen.cpp
clang/test/OpenMP/target_teams_num_teams_codegen.cpp
clang/test/OpenMP/target_teams_thread_limit_codegen.cpp
clang/test/OpenMP/teams_codegen.cpp
clang/test/OpenMP/teams_distribute_codegen.cpp
clang/test/OpenMP/teams_distribute_collapse_codegen.cpp
clang/test/OpenMP/teams_distribute_dist_schedule_codegen.cpp
clang/test/OpenMP/teams_distribute_firstprivate_codegen.cpp
clang/test/OpenMP/teams_distribute_lastprivate_codegen.cpp
clang/test/OpenMP/teams_distribute_parallel_for_codegen.cpp
clang/test/OpenMP/teams_distribute_parallel_for_collapse_codegen.cpp
clang/test/OpenMP/teams_distribute_parallel_for_copyin_codegen.cpp
clang/test/OpenMP/teams_distribute_parallel_for_dist_schedule_codegen.cpp
clang/test/OpenMP/teams_distribute_parallel_for_firstprivate_codegen.cpp
clang/test/OpenMP/teams_distribute_parallel_for_if_codegen.cpp
clang/test/OpenMP/teams_distribute_parallel_for_lastprivate_codegen.cpp
clang/test/OpenMP/teams_distribute_parallel_for_num_threads_codegen.cpp
clang/test/OpenMP/teams_distribute_parallel_for_private_codegen.cpp
clang/test/OpenMP/teams_distribute_parallel_for_proc_bind_codegen.cpp
clang/test/OpenMP/teams_distribute_parallel_for_reduction_codegen.cpp
clang/test/OpenMP/teams_distribute_parallel_for_schedule_codegen.cpp
clang/test/OpenMP/teams_distribute_parallel_for_simd_codegen.cpp
clang/test/OpenMP/teams_distribute_parallel_for_simd_collapse_codegen.cpp
clang/test/OpenMP/teams_distribute_parallel_for_simd_dist_schedule_codegen.cpp
clang/test/OpenMP/teams_distribute_parallel_for_simd_firstprivate_codegen.cpp
clang/test/OpenMP/teams_distribute_parallel_for_simd_if_codegen.cpp
clang/test/OpenMP/teams_distribute_parallel_for_simd_lastprivate_codegen.cpp
clang/test/OpenMP/teams_distribute_parallel_for_simd_num_threads_codegen.cpp
clang/test/OpenMP/teams_distribute_parallel_for_simd_private_codegen.cpp
clang/test/OpenMP/teams_distribute_parallel_for_simd_proc_bind_codegen.cpp
clang/test/OpenMP/teams_distribute_parallel_for_simd_reduction_codegen.cpp
clang/test/OpenMP/teams_distribute_parallel_for_simd_schedule_codegen.cpp
clang/test/OpenMP/teams_distribute_private_codegen.cpp
clang/test/OpenMP/teams_distribute_reduction_codegen.cpp
clang/test/OpenMP/teams_distribute_simd_codegen.cpp
clang/test/OpenMP/teams_distribute_simd_collapse_codegen.cpp
clang/test/OpenMP/teams_distribute_simd_dist_schedule_codegen.cpp
clang/test/OpenMP/teams_distribute_simd_firstprivate_codegen.cpp
clang/test/OpenMP/teams_distribute_simd_lastprivate_codegen.cpp
clang/test/OpenMP/teams_distribute_simd_private_codegen.cpp
clang/test/OpenMP/teams_distribute_simd_reduction_codegen.cpp
clang/test/OpenMP/teams_firstprivate_codegen.cpp
clang/test/OpenMP/teams_private_codegen.cpp
llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h
llvm/include/llvm/Frontend/OpenMP/OMPKinds.def
llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
openmp/libomptarget/include/device.h
openmp/libomptarget/include/omptarget.h
openmp/libomptarget/include/rtl.h
openmp/libomptarget/plugins-nextgen/amdgpu/src/rtl.cpp
openmp/libomptarget/plugins-nextgen/common/PluginInterface/PluginInterface.cpp
openmp/libomptarget/plugins-nextgen/common/PluginInterface/PluginInterface.h
openmp/libomptarget/plugins-nextgen/cuda/src/rtl.cpp
openmp/libomptarget/plugins-nextgen/generic-elf-64bit/src/rtl.cpp
openmp/libomptarget/plugins/amdgpu/src/rtl.cpp
openmp/libomptarget/plugins/generic-elf-64bit/src/rtl.cpp
openmp/libomptarget/plugins/remote/src/Client.h
openmp/libomptarget/plugins/remote/src/rtl.cpp
openmp/libomptarget/plugins/ve/src/rtl.cpp
openmp/libomptarget/src/LegacyAPI.cpp
openmp/libomptarget/src/device.cpp
openmp/libomptarget/src/interface.cpp
openmp/libomptarget/src/omptarget.cpp
openmp/libomptarget/src/private.h
openmp/libomptarget/src/rtl.cpp
Removed:
################################################################################
diff --git a/clang/lib/CodeGen/CGOpenMPRuntime.cpp b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
index a29146689d7b4..5335cc072e76b 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntime.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
@@ -9903,9 +9903,22 @@ void CGOpenMPRuntime::emitTargetCall(
llvm::Value *NumIterations =
emitTargetNumIterationsCall(CGF, D, SizeEmitter);
+ llvm::Value *DynCGroupMem = CGF.Builder.getInt32(0);
+
+ llvm::Value *ZeroArray =
+ llvm::Constant::getNullValue(llvm::ArrayType::get(CGF.CGM.Int32Ty, 3));
+
+ bool HasNoWait = D.hasClausesOfKind<OMPNowaitClause>();
+ llvm::Value *Flags = CGF.Builder.getInt64(HasNoWait);
+
+ llvm::Value *NumTeams3D =
+ CGF.Builder.CreateInsertValue(ZeroArray, NumTeams, {0});
+ llvm::Value *NumThreads3D =
+ CGF.Builder.CreateInsertValue(ZeroArray, NumThreads, {0});
+
// Arguments for the target kernel.
SmallVector<llvm::Value *> KernelArgs{
- CGF.Builder.getInt32(/* Version */ 1),
+ CGF.Builder.getInt32(/* Version */ 2),
PointerNum,
InputInfo.BasePointersArray.getPointer(),
InputInfo.PointersArray.getPointer(),
@@ -9913,18 +9926,13 @@ void CGOpenMPRuntime::emitTargetCall(
MapTypesArray,
MapNamesArray,
InputInfo.MappersArray.getPointer(),
- NumIterations};
-
- // Arguments passed to the 'nowait' variant.
- SmallVector<llvm::Value *> NoWaitKernelArgs{
- CGF.Builder.getInt32(0),
- llvm::ConstantPointerNull::get(CGM.VoidPtrTy),
- CGF.Builder.getInt32(0),
- llvm::ConstantPointerNull::get(CGM.VoidPtrTy),
+ NumIterations,
+ Flags,
+ NumTeams3D,
+ NumThreads3D,
+ DynCGroupMem,
};
- bool HasNoWait = D.hasClausesOfKind<OMPNowaitClause>();
-
// The target region is an outlined function launched by the runtime
// via calls to __tgt_target_kernel().
//
@@ -9938,13 +9946,9 @@ void CGOpenMPRuntime::emitTargetCall(
// __tgt_target_teams() launches a GPU kernel with the requested number
// of teams and threads so no additional calls to the runtime are required.
// Check the error code and execute the host version if required.
- CGF.Builder.restoreIP(
- HasNoWait ? OMPBuilder.emitTargetKernel(
- CGF.Builder, Return, RTLoc, DeviceID, NumTeams,
- NumThreads, OutlinedFnID, KernelArgs, NoWaitKernelArgs)
- : OMPBuilder.emitTargetKernel(CGF.Builder, Return, RTLoc,
- DeviceID, NumTeams, NumThreads,
- OutlinedFnID, KernelArgs));
+ CGF.Builder.restoreIP(OMPBuilder.emitTargetKernel(
+ CGF.Builder, Return, RTLoc, DeviceID, NumTeams, NumThreads,
+ OutlinedFnID, KernelArgs));
llvm::BasicBlock *OffloadFailedBlock =
CGF.createBasicBlock("omp_offload.failed");
diff --git a/clang/test/OpenMP/declare_mapper_codegen.cpp b/clang/test/OpenMP/declare_mapper_codegen.cpp
index aef2fcd7305c3..56fdd0ff6ee24 100644
--- a/clang/test/OpenMP/declare_mapper_codegen.cpp
+++ b/clang/test/OpenMP/declare_mapper_codegen.cpp
@@ -507,7 +507,7 @@ void foo(int a){
// CK0: }
// CK0: define internal void [[OUTLINED:@.+]](i32 {{.*}}{{[^,]+}}, [[ANON_T]]* noalias noundef [[CTXARG:%.+]])
-// CK0-DAG: call i32 @__tgt_target_kernel_nowait(%struct.ident_t* @{{.+}}, i64 -1, i32 -1, i32 0, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]], i32 0, i8* null, i32 0, i8* null)
+// CK0-DAG: call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 -1, i32 -1, i32 0, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]])
// CK0-DAG: [[BPARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 2
// CK0-DAG: store i8** [[BPGEP:%.+]], i8*** [[BPARG]]
// CK0-DAG: [[PARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 3
@@ -547,7 +547,7 @@ void foo(int a){
// CK0: }
// CK0: define internal void [[OUTLINE_1:@.+]](i32 {{.*}}%.global_tid.{{.+}}, [[ANON_T_0]]* noalias noundef [[CTXARG:%.+]])
-// CK0-DAG: call i32 @__tgt_target_kernel_nowait(%struct.ident_t* @{{.+}}, i64 -1, i32 0, i32 0, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]], i32 0, i8* null, i32 0, i8* null)
+// CK0-DAG: call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 -1, i32 0, i32 0, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]])
// CK0-DAG: [[BPARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 2
// CK0-DAG: store i8** [[BPGEP:%.+]], i8*** [[BPARG]]
// CK0-DAG: [[PARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 3
diff --git a/clang/test/OpenMP/distribute_codegen.cpp b/clang/test/OpenMP/distribute_codegen.cpp
index d26ed299ca380..a93d3e1385761 100644
--- a/clang/test/OpenMP/distribute_codegen.cpp
+++ b/clang/test/OpenMP/distribute_codegen.cpp
@@ -163,7 +163,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -180,9 +180,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK1-NEXT: store ptr null, ptr [[TMP25]], align 8
// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 4571424, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z23without_schedule_clausePfS_S_S__l56.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK1-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP29]], align 4
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z23without_schedule_clausePfS_S_S__l56.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK1-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z23without_schedule_clausePfS_S_S__l56(ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -346,7 +354,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -363,9 +371,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK1-NEXT: store ptr null, ptr [[TMP25]], align 8
// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 4571424, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18static_not_chunkedPfS_S_S__l68.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK1-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP29]], align 4
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18static_not_chunkedPfS_S_S__l68.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK1-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18static_not_chunkedPfS_S_S__l68(ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -529,7 +545,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -546,9 +562,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK1-NEXT: store ptr null, ptr [[TMP25]], align 8
// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 16908289, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z14static_chunkedPfS_S_S__l80.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK1-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP29]], align 4
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z14static_chunkedPfS_S_S__l80.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK1-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z14static_chunkedPfS_S_S__l80(ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -720,7 +744,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK1-NEXT: [[TMP10:%.*]] = zext i32 [[ADD4]] to i64
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP11]], align 4
// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP12]], align 4
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -737,9 +761,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 [[TMP10]], ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z12test_precondv_l92.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP21]], align 4
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP22]], align 4
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP23]], align 4
+// CHECK1-NEXT: [[TMP24:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z12test_precondv_l92.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP25:%.*]] = icmp ne i32 [[TMP24]], 0
+// CHECK1-NEXT: br i1 [[TMP25]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z12test_precondv_l92(i64 [[TMP1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -879,7 +911,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -896,9 +928,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_v_l108.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK1-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_v_l108.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_v_l108(i64 [[TMP1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1051,7 +1091,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1068,9 +1108,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK3-NEXT: store ptr null, ptr [[TMP25]], align 4
// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 4571424, ptr [[TMP26]], align 8
-// CHECK3-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z23without_schedule_clausePfS_S_S__l56.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK3-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z23without_schedule_clausePfS_S_S__l56.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK3-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z23without_schedule_clausePfS_S_S__l56(ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1230,7 +1278,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1247,9 +1295,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK3-NEXT: store ptr null, ptr [[TMP25]], align 4
// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 4571424, ptr [[TMP26]], align 8
-// CHECK3-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18static_not_chunkedPfS_S_S__l68.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK3-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18static_not_chunkedPfS_S_S__l68.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK3-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18static_not_chunkedPfS_S_S__l68(ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1409,7 +1465,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1426,9 +1482,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK3-NEXT: store ptr null, ptr [[TMP25]], align 4
// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 16908289, ptr [[TMP26]], align 8
-// CHECK3-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z14static_chunkedPfS_S_S__l80.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK3-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z14static_chunkedPfS_S_S__l80.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK3-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z14static_chunkedPfS_S_S__l80(ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1596,7 +1660,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK3-NEXT: [[TMP10:%.*]] = zext i32 [[ADD4]] to i64
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP11]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP11]], align 4
// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1613,9 +1677,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK3-NEXT: store ptr null, ptr [[TMP18]], align 4
// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 [[TMP10]], ptr [[TMP19]], align 8
-// CHECK3-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z12test_precondv_l92.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK3-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP20]], align 8
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP21]], align 4
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP22]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP23]], align 4
+// CHECK3-NEXT: [[TMP24:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z12test_precondv_l92.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP25:%.*]] = icmp ne i32 [[TMP24]], 0
+// CHECK3-NEXT: br i1 [[TMP25]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z12test_precondv_l92(i32 [[TMP1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1755,7 +1827,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1772,9 +1844,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK3-NEXT: store ptr null, ptr [[TMP14]], align 4
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 100, ptr [[TMP15]], align 8
-// CHECK3-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_v_l108.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK3-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK3-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_v_l108.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK3-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_v_l108(i32 [[TMP1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/distribute_firstprivate_codegen.cpp b/clang/test/OpenMP/distribute_firstprivate_codegen.cpp
index 90fdf57c6b39c..c385bc1209357 100644
--- a/clang/test/OpenMP/distribute_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/distribute_firstprivate_codegen.cpp
@@ -556,7 +556,7 @@ int main() {
// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP25]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP25]], align 4
// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 5, ptr [[TMP26]], align 4
// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -573,9 +573,17 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP32]], align 8
// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP33]], align 8
-// CHECK9-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l103.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
-// CHECK9-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK9-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l103.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK9-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l103(i64 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]], i64 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -583,18 +591,18 @@ int main() {
// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiET_v()
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP36]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP40]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP37:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP37]]
+// CHECK9-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP41]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -844,7 +852,7 @@ int main() {
// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -861,27 +869,35 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP27]], align 8
// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK9-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK9-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK9-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK9-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49(i64 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP32]]
+// CHECK9-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP36]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
@@ -1212,7 +1228,7 @@ int main() {
// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP25]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP25]], align 4
// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 5, ptr [[TMP26]], align 4
// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1229,9 +1245,17 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP32]], align 4
// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP33]], align 8
-// CHECK11-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l103.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
-// CHECK11-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK11-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l103.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK11-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l103(i32 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]], i32 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1239,18 +1263,18 @@ int main() {
// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiET_v()
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP36]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP40]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP37:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP37]]
+// CHECK11-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP41]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -1498,7 +1522,7 @@ int main() {
// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1515,27 +1539,35 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP27]], align 4
// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK11-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK11-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK11-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK11-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49(i32 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP32]]
+// CHECK11-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP36]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
diff --git a/clang/test/OpenMP/distribute_lastprivate_codegen.cpp b/clang/test/OpenMP/distribute_lastprivate_codegen.cpp
index 2fdc1fec57b81..8230ab672185b 100644
--- a/clang/test/OpenMP/distribute_lastprivate_codegen.cpp
+++ b/clang/test/OpenMP/distribute_lastprivate_codegen.cpp
@@ -541,7 +541,7 @@ int main() {
// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP25]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP25]], align 4
// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 5, ptr [[TMP26]], align 4
// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -558,9 +558,17 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP32]], align 8
// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP33]], align 8
-// CHECK9-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l98.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
-// CHECK9-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK9-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l98.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK9-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l98(i64 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]], i64 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -568,18 +576,18 @@ int main() {
// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiET_v()
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP36]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP40]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP37:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP37]]
+// CHECK9-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP41]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -847,7 +855,7 @@ int main() {
// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -864,27 +872,35 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP27]], align 8
// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK9-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK9-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK9-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK9-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49(i64 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP32]]
+// CHECK9-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP36]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
@@ -1234,7 +1250,7 @@ int main() {
// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP25]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP25]], align 4
// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 5, ptr [[TMP26]], align 4
// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1251,9 +1267,17 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP32]], align 4
// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP33]], align 8
-// CHECK11-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l98.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
-// CHECK11-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK11-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l98.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK11-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l98(i32 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]], i32 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1261,18 +1285,18 @@ int main() {
// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiET_v()
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP36]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP40]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP37:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP37]]
+// CHECK11-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP41]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -1538,7 +1562,7 @@ int main() {
// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1555,27 +1579,35 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP27]], align 4
// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK11-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK11-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK11-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK11-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49(i32 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP32]]
+// CHECK11-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP36]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
diff --git a/clang/test/OpenMP/distribute_parallel_for_codegen.cpp b/clang/test/OpenMP/distribute_parallel_for_codegen.cpp
index 095abb0d2a5fb..f333e83bf627c 100644
--- a/clang/test/OpenMP/distribute_parallel_for_codegen.cpp
+++ b/clang/test/OpenMP/distribute_parallel_for_codegen.cpp
@@ -4380,7 +4380,7 @@ int main() {
// CHECK9-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 4, ptr [[TMP24]], align 4
// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -4397,446 +4397,502 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK9-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l369.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK9-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l369.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK9-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l369(i64 [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]]) #[[ATTR2:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK9-NEXT: [[TMP35:%.*]] = load i64, ptr [[N_CASTED3]], align 8
-// CHECK9-NEXT: [[TMP36:%.*]] = load ptr, ptr [[A]], align 8
-// CHECK9-NEXT: [[TMP37:%.*]] = load ptr, ptr [[B]], align 8
-// CHECK9-NEXT: [[TMP38:%.*]] = load ptr, ptr [[C]], align 8
-// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP35]], ptr [[TMP39]], align 8
-// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP35]], ptr [[TMP40]], align 8
-// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP41]], align 8
-// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK9-NEXT: store ptr [[TMP36]], ptr [[TMP42]], align 8
-// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK9-NEXT: store ptr [[TMP36]], ptr [[TMP43]], align 8
-// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP44]], align 8
-// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP37]], ptr [[TMP45]], align 8
-// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP37]], ptr [[TMP46]], align 8
-// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP47]], align 8
-// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP38]], ptr [[TMP48]], align 8
-// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP38]], ptr [[TMP49]], align 8
-// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP50]], align 8
-// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP53:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP53]], ptr [[DOTCAPTURE_EXPR_8]], align 4
-// CHECK9-NEXT: [[TMP54:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_8]], align 4
-// CHECK9-NEXT: [[SUB10:%.*]] = sub nsw i32 [[TMP54]], 0
+// CHECK9-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK9-NEXT: [[TMP39:%.*]] = load i64, ptr [[N_CASTED3]], align 8
+// CHECK9-NEXT: [[TMP40:%.*]] = load ptr, ptr [[A]], align 8
+// CHECK9-NEXT: [[TMP41:%.*]] = load ptr, ptr [[B]], align 8
+// CHECK9-NEXT: [[TMP42:%.*]] = load ptr, ptr [[C]], align 8
+// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP39]], ptr [[TMP43]], align 8
+// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP39]], ptr [[TMP44]], align 8
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK9-NEXT: store ptr [[TMP40]], ptr [[TMP46]], align 8
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK9-NEXT: store ptr [[TMP40]], ptr [[TMP47]], align 8
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP48]], align 8
+// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP41]], ptr [[TMP49]], align 8
+// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP41]], ptr [[TMP50]], align 8
+// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP51]], align 8
+// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP42]], ptr [[TMP52]], align 8
+// CHECK9-NEXT: [[TMP53:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP42]], ptr [[TMP53]], align 8
+// CHECK9-NEXT: [[TMP54:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP54]], align 8
+// CHECK9-NEXT: [[TMP55:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP56:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP57:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP57]], ptr [[DOTCAPTURE_EXPR_8]], align 4
+// CHECK9-NEXT: [[TMP58:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_8]], align 4
+// CHECK9-NEXT: [[SUB10:%.*]] = sub nsw i32 [[TMP58]], 0
// CHECK9-NEXT: [[DIV11:%.*]] = sdiv i32 [[SUB10]], 1
// CHECK9-NEXT: [[SUB12:%.*]] = sub nsw i32 [[DIV11]], 1
// CHECK9-NEXT: store i32 [[SUB12]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK9-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK9-NEXT: [[ADD13:%.*]] = add nsw i32 [[TMP55]], 1
-// CHECK9-NEXT: [[TMP56:%.*]] = zext i32 [[ADD13]] to i64
+// CHECK9-NEXT: [[TMP59:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK9-NEXT: [[ADD13:%.*]] = add nsw i32 [[TMP59]], 1
+// CHECK9-NEXT: [[TMP60:%.*]] = zext i32 [[ADD13]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS14:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP57]], align 4
-// CHECK9-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
-// CHECK9-NEXT: store i32 4, ptr [[TMP58]], align 4
-// CHECK9-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP51]], ptr [[TMP59]], align 8
-// CHECK9-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP52]], ptr [[TMP60]], align 8
-// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.4, ptr [[TMP61]], align 8
-// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP62]], align 8
-// CHECK9-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP63]], align 8
-// CHECK9-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP64]], align 8
-// CHECK9-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP56]], ptr [[TMP65]], align 8
-// CHECK9-NEXT: [[TMP66:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l408.region_id, ptr [[KERNEL_ARGS14]])
-// CHECK9-NEXT: [[TMP67:%.*]] = icmp ne i32 [[TMP66]], 0
-// CHECK9-NEXT: br i1 [[TMP67]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
+// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP61]], align 4
+// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
+// CHECK9-NEXT: store i32 4, ptr [[TMP62]], align 4
+// CHECK9-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP55]], ptr [[TMP63]], align 8
+// CHECK9-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP56]], ptr [[TMP64]], align 8
+// CHECK9-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.4, ptr [[TMP65]], align 8
+// CHECK9-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP66]], align 8
+// CHECK9-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP67]], align 8
+// CHECK9-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP68]], align 8
+// CHECK9-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP60]], ptr [[TMP69]], align 8
+// CHECK9-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP70]], align 8
+// CHECK9-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP71]], align 4
+// CHECK9-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP72]], align 4
+// CHECK9-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP73]], align 4
+// CHECK9-NEXT: [[TMP74:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l408.region_id, ptr [[KERNEL_ARGS14]])
+// CHECK9-NEXT: [[TMP75:%.*]] = icmp ne i32 [[TMP74]], 0
+// CHECK9-NEXT: br i1 [[TMP75]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
// CHECK9: omp_offload.failed15:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l408(i64 [[TMP35]], ptr [[TMP36]], ptr [[TMP37]], ptr [[TMP38]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l408(i64 [[TMP39]], ptr [[TMP40]], ptr [[TMP41]], ptr [[TMP42]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT16]]
// CHECK9: omp_offload.cont16:
-// CHECK9-NEXT: [[TMP68:%.*]] = load i32, ptr [[CH]], align 4
-// CHECK9-NEXT: store i32 [[TMP68]], ptr [[CH_CASTED]], align 4
-// CHECK9-NEXT: [[TMP69:%.*]] = load i64, ptr [[CH_CASTED]], align 8
-// CHECK9-NEXT: [[TMP70:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP70]], ptr [[N_CASTED17]], align 4
-// CHECK9-NEXT: [[TMP71:%.*]] = load i64, ptr [[N_CASTED17]], align 8
-// CHECK9-NEXT: [[TMP72:%.*]] = load ptr, ptr [[A]], align 8
-// CHECK9-NEXT: [[TMP73:%.*]] = load ptr, ptr [[B]], align 8
-// CHECK9-NEXT: [[TMP74:%.*]] = load ptr, ptr [[C]], align 8
-// CHECK9-NEXT: [[TMP75:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP69]], ptr [[TMP75]], align 8
-// CHECK9-NEXT: [[TMP76:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP69]], ptr [[TMP76]], align 8
-// CHECK9-NEXT: [[TMP77:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP77]], align 8
-// CHECK9-NEXT: [[TMP78:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP71]], ptr [[TMP78]], align 8
-// CHECK9-NEXT: [[TMP79:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP71]], ptr [[TMP79]], align 8
-// CHECK9-NEXT: [[TMP80:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP80]], align 8
-// CHECK9-NEXT: [[TMP81:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP72]], ptr [[TMP81]], align 8
-// CHECK9-NEXT: [[TMP82:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP72]], ptr [[TMP82]], align 8
-// CHECK9-NEXT: [[TMP83:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP83]], align 8
-// CHECK9-NEXT: [[TMP84:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP73]], ptr [[TMP84]], align 8
-// CHECK9-NEXT: [[TMP85:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP73]], ptr [[TMP85]], align 8
-// CHECK9-NEXT: [[TMP86:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP86]], align 8
-// CHECK9-NEXT: [[TMP87:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP74]], ptr [[TMP87]], align 8
-// CHECK9-NEXT: [[TMP88:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP74]], ptr [[TMP88]], align 8
-// CHECK9-NEXT: [[TMP89:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 4
-// CHECK9-NEXT: store ptr null, ptr [[TMP89]], align 8
-// CHECK9-NEXT: [[TMP90:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP91:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP92:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP92]], ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK9-NEXT: [[TMP93:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK9-NEXT: [[SUB24:%.*]] = sub nsw i32 [[TMP93]], 0
+// CHECK9-NEXT: [[TMP76:%.*]] = load i32, ptr [[CH]], align 4
+// CHECK9-NEXT: store i32 [[TMP76]], ptr [[CH_CASTED]], align 4
+// CHECK9-NEXT: [[TMP77:%.*]] = load i64, ptr [[CH_CASTED]], align 8
+// CHECK9-NEXT: [[TMP78:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP78]], ptr [[N_CASTED17]], align 4
+// CHECK9-NEXT: [[TMP79:%.*]] = load i64, ptr [[N_CASTED17]], align 8
+// CHECK9-NEXT: [[TMP80:%.*]] = load ptr, ptr [[A]], align 8
+// CHECK9-NEXT: [[TMP81:%.*]] = load ptr, ptr [[B]], align 8
+// CHECK9-NEXT: [[TMP82:%.*]] = load ptr, ptr [[C]], align 8
+// CHECK9-NEXT: [[TMP83:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP77]], ptr [[TMP83]], align 8
+// CHECK9-NEXT: [[TMP84:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP77]], ptr [[TMP84]], align 8
+// CHECK9-NEXT: [[TMP85:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP85]], align 8
+// CHECK9-NEXT: [[TMP86:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP79]], ptr [[TMP86]], align 8
+// CHECK9-NEXT: [[TMP87:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP79]], ptr [[TMP87]], align 8
+// CHECK9-NEXT: [[TMP88:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP88]], align 8
+// CHECK9-NEXT: [[TMP89:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP80]], ptr [[TMP89]], align 8
+// CHECK9-NEXT: [[TMP90:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP80]], ptr [[TMP90]], align 8
+// CHECK9-NEXT: [[TMP91:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP91]], align 8
+// CHECK9-NEXT: [[TMP92:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP81]], ptr [[TMP92]], align 8
+// CHECK9-NEXT: [[TMP93:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP81]], ptr [[TMP93]], align 8
+// CHECK9-NEXT: [[TMP94:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP94]], align 8
+// CHECK9-NEXT: [[TMP95:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP82]], ptr [[TMP95]], align 8
+// CHECK9-NEXT: [[TMP96:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP82]], ptr [[TMP96]], align 8
+// CHECK9-NEXT: [[TMP97:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 4
+// CHECK9-NEXT: store ptr null, ptr [[TMP97]], align 8
+// CHECK9-NEXT: [[TMP98:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP99:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP100:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP100]], ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK9-NEXT: [[TMP101:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK9-NEXT: [[SUB24:%.*]] = sub nsw i32 [[TMP101]], 0
// CHECK9-NEXT: [[DIV25:%.*]] = sdiv i32 [[SUB24]], 1
// CHECK9-NEXT: [[SUB26:%.*]] = sub nsw i32 [[DIV25]], 1
// CHECK9-NEXT: store i32 [[SUB26]], ptr [[DOTCAPTURE_EXPR_23]], align 4
-// CHECK9-NEXT: [[TMP94:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_23]], align 4
-// CHECK9-NEXT: [[ADD27:%.*]] = add nsw i32 [[TMP94]], 1
-// CHECK9-NEXT: [[TMP95:%.*]] = zext i32 [[ADD27]] to i64
+// CHECK9-NEXT: [[TMP102:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_23]], align 4
+// CHECK9-NEXT: [[ADD27:%.*]] = add nsw i32 [[TMP102]], 1
+// CHECK9-NEXT: [[TMP103:%.*]] = zext i32 [[ADD27]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS28:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP96]], align 4
-// CHECK9-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
-// CHECK9-NEXT: store i32 5, ptr [[TMP97]], align 4
-// CHECK9-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP90]], ptr [[TMP98]], align 8
-// CHECK9-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP91]], ptr [[TMP99]], align 8
-// CHECK9-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.8, ptr [[TMP100]], align 8
-// CHECK9-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP101]], align 8
-// CHECK9-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP102]], align 8
-// CHECK9-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP103]], align 8
-// CHECK9-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP95]], ptr [[TMP104]], align 8
-// CHECK9-NEXT: [[TMP105:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l447.region_id, ptr [[KERNEL_ARGS28]])
-// CHECK9-NEXT: [[TMP106:%.*]] = icmp ne i32 [[TMP105]], 0
-// CHECK9-NEXT: br i1 [[TMP106]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
+// CHECK9-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP104]], align 4
+// CHECK9-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
+// CHECK9-NEXT: store i32 5, ptr [[TMP105]], align 4
+// CHECK9-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP98]], ptr [[TMP106]], align 8
+// CHECK9-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP99]], ptr [[TMP107]], align 8
+// CHECK9-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.8, ptr [[TMP108]], align 8
+// CHECK9-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP109]], align 8
+// CHECK9-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP110]], align 8
+// CHECK9-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP111]], align 8
+// CHECK9-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP103]], ptr [[TMP112]], align 8
+// CHECK9-NEXT: [[TMP113:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP113]], align 8
+// CHECK9-NEXT: [[TMP114:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP114]], align 4
+// CHECK9-NEXT: [[TMP115:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP115]], align 4
+// CHECK9-NEXT: [[TMP116:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP116]], align 4
+// CHECK9-NEXT: [[TMP117:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l447.region_id, ptr [[KERNEL_ARGS28]])
+// CHECK9-NEXT: [[TMP118:%.*]] = icmp ne i32 [[TMP117]], 0
+// CHECK9-NEXT: br i1 [[TMP118]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
// CHECK9: omp_offload.failed29:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l447(i64 [[TMP69]], i64 [[TMP71]], ptr [[TMP72]], ptr [[TMP73]], ptr [[TMP74]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l447(i64 [[TMP77]], i64 [[TMP79]], ptr [[TMP80]], ptr [[TMP81]], ptr [[TMP82]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT30]]
// CHECK9: omp_offload.cont30:
-// CHECK9-NEXT: [[TMP107:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP107]], ptr [[N_CASTED31]], align 4
-// CHECK9-NEXT: [[TMP108:%.*]] = load i64, ptr [[N_CASTED31]], align 8
-// CHECK9-NEXT: [[TMP109:%.*]] = load ptr, ptr [[A]], align 8
-// CHECK9-NEXT: [[TMP110:%.*]] = load ptr, ptr [[B]], align 8
-// CHECK9-NEXT: [[TMP111:%.*]] = load ptr, ptr [[C]], align 8
-// CHECK9-NEXT: [[TMP112:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP108]], ptr [[TMP112]], align 8
-// CHECK9-NEXT: [[TMP113:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP108]], ptr [[TMP113]], align 8
-// CHECK9-NEXT: [[TMP114:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP114]], align 8
-// CHECK9-NEXT: [[TMP115:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 1
-// CHECK9-NEXT: store ptr [[TMP109]], ptr [[TMP115]], align 8
-// CHECK9-NEXT: [[TMP116:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 1
-// CHECK9-NEXT: store ptr [[TMP109]], ptr [[TMP116]], align 8
-// CHECK9-NEXT: [[TMP117:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP117]], align 8
-// CHECK9-NEXT: [[TMP118:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP110]], ptr [[TMP118]], align 8
-// CHECK9-NEXT: [[TMP119:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP110]], ptr [[TMP119]], align 8
-// CHECK9-NEXT: [[TMP120:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP120]], align 8
-// CHECK9-NEXT: [[TMP121:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP111]], ptr [[TMP121]], align 8
-// CHECK9-NEXT: [[TMP122:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP111]], ptr [[TMP122]], align 8
-// CHECK9-NEXT: [[TMP123:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP123]], align 8
+// CHECK9-NEXT: [[TMP119:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP119]], ptr [[N_CASTED31]], align 4
+// CHECK9-NEXT: [[TMP120:%.*]] = load i64, ptr [[N_CASTED31]], align 8
+// CHECK9-NEXT: [[TMP121:%.*]] = load ptr, ptr [[A]], align 8
+// CHECK9-NEXT: [[TMP122:%.*]] = load ptr, ptr [[B]], align 8
+// CHECK9-NEXT: [[TMP123:%.*]] = load ptr, ptr [[C]], align 8
// CHECK9-NEXT: [[TMP124:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP120]], ptr [[TMP124]], align 8
// CHECK9-NEXT: [[TMP125:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP126:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP126]], ptr [[DOTCAPTURE_EXPR_36]], align 4
-// CHECK9-NEXT: [[TMP127:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_36]], align 4
-// CHECK9-NEXT: [[SUB38:%.*]] = sub nsw i32 [[TMP127]], 0
+// CHECK9-NEXT: store i64 [[TMP120]], ptr [[TMP125]], align 8
+// CHECK9-NEXT: [[TMP126:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP126]], align 8
+// CHECK9-NEXT: [[TMP127:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 1
+// CHECK9-NEXT: store ptr [[TMP121]], ptr [[TMP127]], align 8
+// CHECK9-NEXT: [[TMP128:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 1
+// CHECK9-NEXT: store ptr [[TMP121]], ptr [[TMP128]], align 8
+// CHECK9-NEXT: [[TMP129:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP129]], align 8
+// CHECK9-NEXT: [[TMP130:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP122]], ptr [[TMP130]], align 8
+// CHECK9-NEXT: [[TMP131:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP122]], ptr [[TMP131]], align 8
+// CHECK9-NEXT: [[TMP132:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP132]], align 8
+// CHECK9-NEXT: [[TMP133:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP123]], ptr [[TMP133]], align 8
+// CHECK9-NEXT: [[TMP134:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP123]], ptr [[TMP134]], align 8
+// CHECK9-NEXT: [[TMP135:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP135]], align 8
+// CHECK9-NEXT: [[TMP136:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP137:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP138:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP138]], ptr [[DOTCAPTURE_EXPR_36]], align 4
+// CHECK9-NEXT: [[TMP139:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_36]], align 4
+// CHECK9-NEXT: [[SUB38:%.*]] = sub nsw i32 [[TMP139]], 0
// CHECK9-NEXT: [[DIV39:%.*]] = sdiv i32 [[SUB38]], 1
// CHECK9-NEXT: [[SUB40:%.*]] = sub nsw i32 [[DIV39]], 1
// CHECK9-NEXT: store i32 [[SUB40]], ptr [[DOTCAPTURE_EXPR_37]], align 4
-// CHECK9-NEXT: [[TMP128:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_37]], align 4
-// CHECK9-NEXT: [[ADD41:%.*]] = add nsw i32 [[TMP128]], 1
-// CHECK9-NEXT: [[TMP129:%.*]] = zext i32 [[ADD41]] to i64
+// CHECK9-NEXT: [[TMP140:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_37]], align 4
+// CHECK9-NEXT: [[ADD41:%.*]] = add nsw i32 [[TMP140]], 1
+// CHECK9-NEXT: [[TMP141:%.*]] = zext i32 [[ADD41]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS42:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP130]], align 4
-// CHECK9-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 1
-// CHECK9-NEXT: store i32 4, ptr [[TMP131]], align 4
-// CHECK9-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP124]], ptr [[TMP132]], align 8
-// CHECK9-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP125]], ptr [[TMP133]], align 8
-// CHECK9-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.12, ptr [[TMP134]], align 8
-// CHECK9-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP135]], align 8
-// CHECK9-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP136]], align 8
-// CHECK9-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP137]], align 8
-// CHECK9-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP129]], ptr [[TMP138]], align 8
-// CHECK9-NEXT: [[TMP139:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l478.region_id, ptr [[KERNEL_ARGS42]])
-// CHECK9-NEXT: [[TMP140:%.*]] = icmp ne i32 [[TMP139]], 0
-// CHECK9-NEXT: br i1 [[TMP140]], label [[OMP_OFFLOAD_FAILED43:%.*]], label [[OMP_OFFLOAD_CONT44:%.*]]
+// CHECK9-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP142]], align 4
+// CHECK9-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 1
+// CHECK9-NEXT: store i32 4, ptr [[TMP143]], align 4
+// CHECK9-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP136]], ptr [[TMP144]], align 8
+// CHECK9-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP137]], ptr [[TMP145]], align 8
+// CHECK9-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.12, ptr [[TMP146]], align 8
+// CHECK9-NEXT: [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP147]], align 8
+// CHECK9-NEXT: [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP148]], align 8
+// CHECK9-NEXT: [[TMP149:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP149]], align 8
+// CHECK9-NEXT: [[TMP150:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP141]], ptr [[TMP150]], align 8
+// CHECK9-NEXT: [[TMP151:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP151]], align 8
+// CHECK9-NEXT: [[TMP152:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP152]], align 4
+// CHECK9-NEXT: [[TMP153:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP153]], align 4
+// CHECK9-NEXT: [[TMP154:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP154]], align 4
+// CHECK9-NEXT: [[TMP155:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l478.region_id, ptr [[KERNEL_ARGS42]])
+// CHECK9-NEXT: [[TMP156:%.*]] = icmp ne i32 [[TMP155]], 0
+// CHECK9-NEXT: br i1 [[TMP156]], label [[OMP_OFFLOAD_FAILED43:%.*]], label [[OMP_OFFLOAD_CONT44:%.*]]
// CHECK9: omp_offload.failed43:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l478(i64 [[TMP108]], ptr [[TMP109]], ptr [[TMP110]], ptr [[TMP111]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l478(i64 [[TMP120]], ptr [[TMP121]], ptr [[TMP122]], ptr [[TMP123]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT44]]
// CHECK9: omp_offload.cont44:
-// CHECK9-NEXT: [[TMP141:%.*]] = load i32, ptr [[CH]], align 4
-// CHECK9-NEXT: store i32 [[TMP141]], ptr [[CH_CASTED45]], align 4
-// CHECK9-NEXT: [[TMP142:%.*]] = load i64, ptr [[CH_CASTED45]], align 8
-// CHECK9-NEXT: [[TMP143:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP143]], ptr [[N_CASTED46]], align 4
-// CHECK9-NEXT: [[TMP144:%.*]] = load i64, ptr [[N_CASTED46]], align 8
-// CHECK9-NEXT: [[TMP145:%.*]] = load ptr, ptr [[A]], align 8
-// CHECK9-NEXT: [[TMP146:%.*]] = load ptr, ptr [[B]], align 8
-// CHECK9-NEXT: [[TMP147:%.*]] = load ptr, ptr [[C]], align 8
-// CHECK9-NEXT: [[TMP148:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP142]], ptr [[TMP148]], align 8
-// CHECK9-NEXT: [[TMP149:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP142]], ptr [[TMP149]], align 8
-// CHECK9-NEXT: [[TMP150:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP150]], align 8
-// CHECK9-NEXT: [[TMP151:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP144]], ptr [[TMP151]], align 8
-// CHECK9-NEXT: [[TMP152:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP144]], ptr [[TMP152]], align 8
-// CHECK9-NEXT: [[TMP153:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP153]], align 8
-// CHECK9-NEXT: [[TMP154:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP145]], ptr [[TMP154]], align 8
-// CHECK9-NEXT: [[TMP155:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP145]], ptr [[TMP155]], align 8
-// CHECK9-NEXT: [[TMP156:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP156]], align 8
-// CHECK9-NEXT: [[TMP157:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP146]], ptr [[TMP157]], align 8
-// CHECK9-NEXT: [[TMP158:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP146]], ptr [[TMP158]], align 8
-// CHECK9-NEXT: [[TMP159:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP159]], align 8
-// CHECK9-NEXT: [[TMP160:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP147]], ptr [[TMP160]], align 8
-// CHECK9-NEXT: [[TMP161:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP147]], ptr [[TMP161]], align 8
-// CHECK9-NEXT: [[TMP162:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 4
-// CHECK9-NEXT: store ptr null, ptr [[TMP162]], align 8
-// CHECK9-NEXT: [[TMP163:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP164:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP165:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP165]], ptr [[DOTCAPTURE_EXPR_51]], align 4
-// CHECK9-NEXT: [[TMP166:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_51]], align 4
-// CHECK9-NEXT: [[SUB53:%.*]] = sub nsw i32 [[TMP166]], 0
+// CHECK9-NEXT: [[TMP157:%.*]] = load i32, ptr [[CH]], align 4
+// CHECK9-NEXT: store i32 [[TMP157]], ptr [[CH_CASTED45]], align 4
+// CHECK9-NEXT: [[TMP158:%.*]] = load i64, ptr [[CH_CASTED45]], align 8
+// CHECK9-NEXT: [[TMP159:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP159]], ptr [[N_CASTED46]], align 4
+// CHECK9-NEXT: [[TMP160:%.*]] = load i64, ptr [[N_CASTED46]], align 8
+// CHECK9-NEXT: [[TMP161:%.*]] = load ptr, ptr [[A]], align 8
+// CHECK9-NEXT: [[TMP162:%.*]] = load ptr, ptr [[B]], align 8
+// CHECK9-NEXT: [[TMP163:%.*]] = load ptr, ptr [[C]], align 8
+// CHECK9-NEXT: [[TMP164:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP158]], ptr [[TMP164]], align 8
+// CHECK9-NEXT: [[TMP165:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP158]], ptr [[TMP165]], align 8
+// CHECK9-NEXT: [[TMP166:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP166]], align 8
+// CHECK9-NEXT: [[TMP167:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP160]], ptr [[TMP167]], align 8
+// CHECK9-NEXT: [[TMP168:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP160]], ptr [[TMP168]], align 8
+// CHECK9-NEXT: [[TMP169:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP169]], align 8
+// CHECK9-NEXT: [[TMP170:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP161]], ptr [[TMP170]], align 8
+// CHECK9-NEXT: [[TMP171:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP161]], ptr [[TMP171]], align 8
+// CHECK9-NEXT: [[TMP172:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP172]], align 8
+// CHECK9-NEXT: [[TMP173:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP162]], ptr [[TMP173]], align 8
+// CHECK9-NEXT: [[TMP174:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP162]], ptr [[TMP174]], align 8
+// CHECK9-NEXT: [[TMP175:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP175]], align 8
+// CHECK9-NEXT: [[TMP176:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP163]], ptr [[TMP176]], align 8
+// CHECK9-NEXT: [[TMP177:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP163]], ptr [[TMP177]], align 8
+// CHECK9-NEXT: [[TMP178:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 4
+// CHECK9-NEXT: store ptr null, ptr [[TMP178]], align 8
+// CHECK9-NEXT: [[TMP179:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP180:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP181:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP181]], ptr [[DOTCAPTURE_EXPR_51]], align 4
+// CHECK9-NEXT: [[TMP182:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_51]], align 4
+// CHECK9-NEXT: [[SUB53:%.*]] = sub nsw i32 [[TMP182]], 0
// CHECK9-NEXT: [[DIV54:%.*]] = sdiv i32 [[SUB53]], 1
// CHECK9-NEXT: [[SUB55:%.*]] = sub nsw i32 [[DIV54]], 1
// CHECK9-NEXT: store i32 [[SUB55]], ptr [[DOTCAPTURE_EXPR_52]], align 4
-// CHECK9-NEXT: [[TMP167:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_52]], align 4
-// CHECK9-NEXT: [[ADD56:%.*]] = add nsw i32 [[TMP167]], 1
-// CHECK9-NEXT: [[TMP168:%.*]] = zext i32 [[ADD56]] to i64
+// CHECK9-NEXT: [[TMP183:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_52]], align 4
+// CHECK9-NEXT: [[ADD56:%.*]] = add nsw i32 [[TMP183]], 1
+// CHECK9-NEXT: [[TMP184:%.*]] = zext i32 [[ADD56]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS57:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP169:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP169]], align 4
-// CHECK9-NEXT: [[TMP170:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 1
-// CHECK9-NEXT: store i32 5, ptr [[TMP170]], align 4
-// CHECK9-NEXT: [[TMP171:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP163]], ptr [[TMP171]], align 8
-// CHECK9-NEXT: [[TMP172:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP164]], ptr [[TMP172]], align 8
-// CHECK9-NEXT: [[TMP173:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.16, ptr [[TMP173]], align 8
-// CHECK9-NEXT: [[TMP174:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP174]], align 8
-// CHECK9-NEXT: [[TMP175:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP175]], align 8
-// CHECK9-NEXT: [[TMP176:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP176]], align 8
-// CHECK9-NEXT: [[TMP177:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP168]], ptr [[TMP177]], align 8
-// CHECK9-NEXT: [[TMP178:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l506.region_id, ptr [[KERNEL_ARGS57]])
-// CHECK9-NEXT: [[TMP179:%.*]] = icmp ne i32 [[TMP178]], 0
-// CHECK9-NEXT: br i1 [[TMP179]], label [[OMP_OFFLOAD_FAILED58:%.*]], label [[OMP_OFFLOAD_CONT59:%.*]]
+// CHECK9-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP185]], align 4
+// CHECK9-NEXT: [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 1
+// CHECK9-NEXT: store i32 5, ptr [[TMP186]], align 4
+// CHECK9-NEXT: [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP179]], ptr [[TMP187]], align 8
+// CHECK9-NEXT: [[TMP188:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP180]], ptr [[TMP188]], align 8
+// CHECK9-NEXT: [[TMP189:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.16, ptr [[TMP189]], align 8
+// CHECK9-NEXT: [[TMP190:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP190]], align 8
+// CHECK9-NEXT: [[TMP191:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP191]], align 8
+// CHECK9-NEXT: [[TMP192:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP192]], align 8
+// CHECK9-NEXT: [[TMP193:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP184]], ptr [[TMP193]], align 8
+// CHECK9-NEXT: [[TMP194:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP194]], align 8
+// CHECK9-NEXT: [[TMP195:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP195]], align 4
+// CHECK9-NEXT: [[TMP196:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP196]], align 4
+// CHECK9-NEXT: [[TMP197:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP197]], align 4
+// CHECK9-NEXT: [[TMP198:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l506.region_id, ptr [[KERNEL_ARGS57]])
+// CHECK9-NEXT: [[TMP199:%.*]] = icmp ne i32 [[TMP198]], 0
+// CHECK9-NEXT: br i1 [[TMP199]], label [[OMP_OFFLOAD_FAILED58:%.*]], label [[OMP_OFFLOAD_CONT59:%.*]]
// CHECK9: omp_offload.failed58:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l506(i64 [[TMP142]], i64 [[TMP144]], ptr [[TMP145]], ptr [[TMP146]], ptr [[TMP147]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l506(i64 [[TMP158]], i64 [[TMP160]], ptr [[TMP161]], ptr [[TMP162]], ptr [[TMP163]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT59]]
// CHECK9: omp_offload.cont59:
-// CHECK9-NEXT: [[TMP180:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP180]], ptr [[N_CASTED60]], align 4
-// CHECK9-NEXT: [[TMP181:%.*]] = load i64, ptr [[N_CASTED60]], align 8
-// CHECK9-NEXT: [[TMP182:%.*]] = load ptr, ptr [[A]], align 8
-// CHECK9-NEXT: [[TMP183:%.*]] = load ptr, ptr [[B]], align 8
-// CHECK9-NEXT: [[TMP184:%.*]] = load ptr, ptr [[C]], align 8
-// CHECK9-NEXT: [[TMP185:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP181]], ptr [[TMP185]], align 8
-// CHECK9-NEXT: [[TMP186:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP181]], ptr [[TMP186]], align 8
-// CHECK9-NEXT: [[TMP187:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP187]], align 8
-// CHECK9-NEXT: [[TMP188:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 1
-// CHECK9-NEXT: store ptr [[TMP182]], ptr [[TMP188]], align 8
-// CHECK9-NEXT: [[TMP189:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 1
-// CHECK9-NEXT: store ptr [[TMP182]], ptr [[TMP189]], align 8
-// CHECK9-NEXT: [[TMP190:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP190]], align 8
-// CHECK9-NEXT: [[TMP191:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP183]], ptr [[TMP191]], align 8
-// CHECK9-NEXT: [[TMP192:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP183]], ptr [[TMP192]], align 8
-// CHECK9-NEXT: [[TMP193:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP193]], align 8
-// CHECK9-NEXT: [[TMP194:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP184]], ptr [[TMP194]], align 8
-// CHECK9-NEXT: [[TMP195:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP184]], ptr [[TMP195]], align 8
-// CHECK9-NEXT: [[TMP196:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP196]], align 8
-// CHECK9-NEXT: [[TMP197:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP198:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP199:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP199]], ptr [[DOTCAPTURE_EXPR_65]], align 4
-// CHECK9-NEXT: [[TMP200:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_65]], align 4
-// CHECK9-NEXT: [[SUB67:%.*]] = sub nsw i32 [[TMP200]], 0
+// CHECK9-NEXT: [[TMP200:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP200]], ptr [[N_CASTED60]], align 4
+// CHECK9-NEXT: [[TMP201:%.*]] = load i64, ptr [[N_CASTED60]], align 8
+// CHECK9-NEXT: [[TMP202:%.*]] = load ptr, ptr [[A]], align 8
+// CHECK9-NEXT: [[TMP203:%.*]] = load ptr, ptr [[B]], align 8
+// CHECK9-NEXT: [[TMP204:%.*]] = load ptr, ptr [[C]], align 8
+// CHECK9-NEXT: [[TMP205:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP201]], ptr [[TMP205]], align 8
+// CHECK9-NEXT: [[TMP206:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP201]], ptr [[TMP206]], align 8
+// CHECK9-NEXT: [[TMP207:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP207]], align 8
+// CHECK9-NEXT: [[TMP208:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 1
+// CHECK9-NEXT: store ptr [[TMP202]], ptr [[TMP208]], align 8
+// CHECK9-NEXT: [[TMP209:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 1
+// CHECK9-NEXT: store ptr [[TMP202]], ptr [[TMP209]], align 8
+// CHECK9-NEXT: [[TMP210:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP210]], align 8
+// CHECK9-NEXT: [[TMP211:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP203]], ptr [[TMP211]], align 8
+// CHECK9-NEXT: [[TMP212:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP203]], ptr [[TMP212]], align 8
+// CHECK9-NEXT: [[TMP213:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP213]], align 8
+// CHECK9-NEXT: [[TMP214:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP204]], ptr [[TMP214]], align 8
+// CHECK9-NEXT: [[TMP215:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP204]], ptr [[TMP215]], align 8
+// CHECK9-NEXT: [[TMP216:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP216]], align 8
+// CHECK9-NEXT: [[TMP217:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP218:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP219:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP219]], ptr [[DOTCAPTURE_EXPR_65]], align 4
+// CHECK9-NEXT: [[TMP220:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_65]], align 4
+// CHECK9-NEXT: [[SUB67:%.*]] = sub nsw i32 [[TMP220]], 0
// CHECK9-NEXT: [[DIV68:%.*]] = sdiv i32 [[SUB67]], 1
// CHECK9-NEXT: [[SUB69:%.*]] = sub nsw i32 [[DIV68]], 1
// CHECK9-NEXT: store i32 [[SUB69]], ptr [[DOTCAPTURE_EXPR_66]], align 4
-// CHECK9-NEXT: [[TMP201:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_66]], align 4
-// CHECK9-NEXT: [[ADD70:%.*]] = add nsw i32 [[TMP201]], 1
-// CHECK9-NEXT: [[TMP202:%.*]] = zext i32 [[ADD70]] to i64
+// CHECK9-NEXT: [[TMP221:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_66]], align 4
+// CHECK9-NEXT: [[ADD70:%.*]] = add nsw i32 [[TMP221]], 1
+// CHECK9-NEXT: [[TMP222:%.*]] = zext i32 [[ADD70]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS71:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP203:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP203]], align 4
-// CHECK9-NEXT: [[TMP204:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 1
-// CHECK9-NEXT: store i32 4, ptr [[TMP204]], align 4
-// CHECK9-NEXT: [[TMP205:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP197]], ptr [[TMP205]], align 8
-// CHECK9-NEXT: [[TMP206:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP198]], ptr [[TMP206]], align 8
-// CHECK9-NEXT: [[TMP207:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.20, ptr [[TMP207]], align 8
-// CHECK9-NEXT: [[TMP208:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.21, ptr [[TMP208]], align 8
-// CHECK9-NEXT: [[TMP209:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP209]], align 8
-// CHECK9-NEXT: [[TMP210:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP210]], align 8
-// CHECK9-NEXT: [[TMP211:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP202]], ptr [[TMP211]], align 8
-// CHECK9-NEXT: [[TMP212:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l536.region_id, ptr [[KERNEL_ARGS71]])
-// CHECK9-NEXT: [[TMP213:%.*]] = icmp ne i32 [[TMP212]], 0
-// CHECK9-NEXT: br i1 [[TMP213]], label [[OMP_OFFLOAD_FAILED72:%.*]], label [[OMP_OFFLOAD_CONT73:%.*]]
+// CHECK9-NEXT: [[TMP223:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP223]], align 4
+// CHECK9-NEXT: [[TMP224:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 1
+// CHECK9-NEXT: store i32 4, ptr [[TMP224]], align 4
+// CHECK9-NEXT: [[TMP225:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP217]], ptr [[TMP225]], align 8
+// CHECK9-NEXT: [[TMP226:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP218]], ptr [[TMP226]], align 8
+// CHECK9-NEXT: [[TMP227:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.20, ptr [[TMP227]], align 8
+// CHECK9-NEXT: [[TMP228:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.21, ptr [[TMP228]], align 8
+// CHECK9-NEXT: [[TMP229:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP229]], align 8
+// CHECK9-NEXT: [[TMP230:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP230]], align 8
+// CHECK9-NEXT: [[TMP231:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP222]], ptr [[TMP231]], align 8
+// CHECK9-NEXT: [[TMP232:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP232]], align 8
+// CHECK9-NEXT: [[TMP233:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP233]], align 4
+// CHECK9-NEXT: [[TMP234:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP234]], align 4
+// CHECK9-NEXT: [[TMP235:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP235]], align 4
+// CHECK9-NEXT: [[TMP236:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l536.region_id, ptr [[KERNEL_ARGS71]])
+// CHECK9-NEXT: [[TMP237:%.*]] = icmp ne i32 [[TMP236]], 0
+// CHECK9-NEXT: br i1 [[TMP237]], label [[OMP_OFFLOAD_FAILED72:%.*]], label [[OMP_OFFLOAD_CONT73:%.*]]
// CHECK9: omp_offload.failed72:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l536(i64 [[TMP181]], ptr [[TMP182]], ptr [[TMP183]], ptr [[TMP184]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l536(i64 [[TMP201]], ptr [[TMP202]], ptr [[TMP203]], ptr [[TMP204]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT73]]
// CHECK9: omp_offload.cont73:
-// CHECK9-NEXT: [[TMP214:%.*]] = load i32, ptr [[CH]], align 4
-// CHECK9-NEXT: store i32 [[TMP214]], ptr [[CH_CASTED74]], align 4
-// CHECK9-NEXT: [[TMP215:%.*]] = load i64, ptr [[CH_CASTED74]], align 8
-// CHECK9-NEXT: [[TMP216:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP216]], ptr [[N_CASTED75]], align 4
-// CHECK9-NEXT: [[TMP217:%.*]] = load i64, ptr [[N_CASTED75]], align 8
-// CHECK9-NEXT: [[TMP218:%.*]] = load ptr, ptr [[A]], align 8
-// CHECK9-NEXT: [[TMP219:%.*]] = load ptr, ptr [[B]], align 8
-// CHECK9-NEXT: [[TMP220:%.*]] = load ptr, ptr [[C]], align 8
-// CHECK9-NEXT: [[TMP221:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP215]], ptr [[TMP221]], align 8
-// CHECK9-NEXT: [[TMP222:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP215]], ptr [[TMP222]], align 8
-// CHECK9-NEXT: [[TMP223:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP223]], align 8
-// CHECK9-NEXT: [[TMP224:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP217]], ptr [[TMP224]], align 8
-// CHECK9-NEXT: [[TMP225:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP217]], ptr [[TMP225]], align 8
-// CHECK9-NEXT: [[TMP226:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP226]], align 8
-// CHECK9-NEXT: [[TMP227:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP218]], ptr [[TMP227]], align 8
-// CHECK9-NEXT: [[TMP228:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP218]], ptr [[TMP228]], align 8
-// CHECK9-NEXT: [[TMP229:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP229]], align 8
-// CHECK9-NEXT: [[TMP230:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP219]], ptr [[TMP230]], align 8
-// CHECK9-NEXT: [[TMP231:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP219]], ptr [[TMP231]], align 8
-// CHECK9-NEXT: [[TMP232:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP232]], align 8
-// CHECK9-NEXT: [[TMP233:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP220]], ptr [[TMP233]], align 8
-// CHECK9-NEXT: [[TMP234:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP220]], ptr [[TMP234]], align 8
-// CHECK9-NEXT: [[TMP235:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 4
-// CHECK9-NEXT: store ptr null, ptr [[TMP235]], align 8
-// CHECK9-NEXT: [[TMP236:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP237:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP238:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP238]], ptr [[DOTCAPTURE_EXPR_80]], align 4
-// CHECK9-NEXT: [[TMP239:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_80]], align 4
-// CHECK9-NEXT: [[SUB82:%.*]] = sub nsw i32 [[TMP239]], 0
+// CHECK9-NEXT: [[TMP238:%.*]] = load i32, ptr [[CH]], align 4
+// CHECK9-NEXT: store i32 [[TMP238]], ptr [[CH_CASTED74]], align 4
+// CHECK9-NEXT: [[TMP239:%.*]] = load i64, ptr [[CH_CASTED74]], align 8
+// CHECK9-NEXT: [[TMP240:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP240]], ptr [[N_CASTED75]], align 4
+// CHECK9-NEXT: [[TMP241:%.*]] = load i64, ptr [[N_CASTED75]], align 8
+// CHECK9-NEXT: [[TMP242:%.*]] = load ptr, ptr [[A]], align 8
+// CHECK9-NEXT: [[TMP243:%.*]] = load ptr, ptr [[B]], align 8
+// CHECK9-NEXT: [[TMP244:%.*]] = load ptr, ptr [[C]], align 8
+// CHECK9-NEXT: [[TMP245:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP239]], ptr [[TMP245]], align 8
+// CHECK9-NEXT: [[TMP246:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP239]], ptr [[TMP246]], align 8
+// CHECK9-NEXT: [[TMP247:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP247]], align 8
+// CHECK9-NEXT: [[TMP248:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP241]], ptr [[TMP248]], align 8
+// CHECK9-NEXT: [[TMP249:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP241]], ptr [[TMP249]], align 8
+// CHECK9-NEXT: [[TMP250:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP250]], align 8
+// CHECK9-NEXT: [[TMP251:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP242]], ptr [[TMP251]], align 8
+// CHECK9-NEXT: [[TMP252:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP242]], ptr [[TMP252]], align 8
+// CHECK9-NEXT: [[TMP253:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP253]], align 8
+// CHECK9-NEXT: [[TMP254:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP243]], ptr [[TMP254]], align 8
+// CHECK9-NEXT: [[TMP255:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP243]], ptr [[TMP255]], align 8
+// CHECK9-NEXT: [[TMP256:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP256]], align 8
+// CHECK9-NEXT: [[TMP257:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP244]], ptr [[TMP257]], align 8
+// CHECK9-NEXT: [[TMP258:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP244]], ptr [[TMP258]], align 8
+// CHECK9-NEXT: [[TMP259:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 4
+// CHECK9-NEXT: store ptr null, ptr [[TMP259]], align 8
+// CHECK9-NEXT: [[TMP260:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP261:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP262:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP262]], ptr [[DOTCAPTURE_EXPR_80]], align 4
+// CHECK9-NEXT: [[TMP263:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_80]], align 4
+// CHECK9-NEXT: [[SUB82:%.*]] = sub nsw i32 [[TMP263]], 0
// CHECK9-NEXT: [[DIV83:%.*]] = sdiv i32 [[SUB82]], 1
// CHECK9-NEXT: [[SUB84:%.*]] = sub nsw i32 [[DIV83]], 1
// CHECK9-NEXT: store i32 [[SUB84]], ptr [[DOTCAPTURE_EXPR_81]], align 4
-// CHECK9-NEXT: [[TMP240:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_81]], align 4
-// CHECK9-NEXT: [[ADD85:%.*]] = add nsw i32 [[TMP240]], 1
-// CHECK9-NEXT: [[TMP241:%.*]] = zext i32 [[ADD85]] to i64
+// CHECK9-NEXT: [[TMP264:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_81]], align 4
+// CHECK9-NEXT: [[ADD85:%.*]] = add nsw i32 [[TMP264]], 1
+// CHECK9-NEXT: [[TMP265:%.*]] = zext i32 [[ADD85]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS86:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP242:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP242]], align 4
-// CHECK9-NEXT: [[TMP243:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 1
-// CHECK9-NEXT: store i32 5, ptr [[TMP243]], align 4
-// CHECK9-NEXT: [[TMP244:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP236]], ptr [[TMP244]], align 8
-// CHECK9-NEXT: [[TMP245:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP237]], ptr [[TMP245]], align 8
-// CHECK9-NEXT: [[TMP246:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.24, ptr [[TMP246]], align 8
-// CHECK9-NEXT: [[TMP247:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP247]], align 8
-// CHECK9-NEXT: [[TMP248:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP248]], align 8
-// CHECK9-NEXT: [[TMP249:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP249]], align 8
-// CHECK9-NEXT: [[TMP250:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP241]], ptr [[TMP250]], align 8
-// CHECK9-NEXT: [[TMP251:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l562.region_id, ptr [[KERNEL_ARGS86]])
-// CHECK9-NEXT: [[TMP252:%.*]] = icmp ne i32 [[TMP251]], 0
-// CHECK9-NEXT: br i1 [[TMP252]], label [[OMP_OFFLOAD_FAILED87:%.*]], label [[OMP_OFFLOAD_CONT88:%.*]]
+// CHECK9-NEXT: [[TMP266:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP266]], align 4
+// CHECK9-NEXT: [[TMP267:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 1
+// CHECK9-NEXT: store i32 5, ptr [[TMP267]], align 4
+// CHECK9-NEXT: [[TMP268:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP260]], ptr [[TMP268]], align 8
+// CHECK9-NEXT: [[TMP269:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP261]], ptr [[TMP269]], align 8
+// CHECK9-NEXT: [[TMP270:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.24, ptr [[TMP270]], align 8
+// CHECK9-NEXT: [[TMP271:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP271]], align 8
+// CHECK9-NEXT: [[TMP272:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP272]], align 8
+// CHECK9-NEXT: [[TMP273:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP273]], align 8
+// CHECK9-NEXT: [[TMP274:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP265]], ptr [[TMP274]], align 8
+// CHECK9-NEXT: [[TMP275:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP275]], align 8
+// CHECK9-NEXT: [[TMP276:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP276]], align 4
+// CHECK9-NEXT: [[TMP277:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP277]], align 4
+// CHECK9-NEXT: [[TMP278:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP278]], align 4
+// CHECK9-NEXT: [[TMP279:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l562.region_id, ptr [[KERNEL_ARGS86]])
+// CHECK9-NEXT: [[TMP280:%.*]] = icmp ne i32 [[TMP279]], 0
+// CHECK9-NEXT: br i1 [[TMP280]], label [[OMP_OFFLOAD_FAILED87:%.*]], label [[OMP_OFFLOAD_CONT88:%.*]]
// CHECK9: omp_offload.failed87:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l562(i64 [[TMP215]], i64 [[TMP217]], ptr [[TMP218]], ptr [[TMP219]], ptr [[TMP220]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l562(i64 [[TMP239]], i64 [[TMP241]], ptr [[TMP242]], ptr [[TMP243]], ptr [[TMP244]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT88]]
// CHECK9: omp_offload.cont88:
// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiET_v()
@@ -6623,7 +6679,7 @@ int main() {
// CHECK9-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 4, ptr [[TMP24]], align 4
// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -6640,446 +6696,502 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK9-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l42.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK9-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l42.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK9-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l42(i64 [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK9-NEXT: [[TMP35:%.*]] = load i64, ptr [[N_CASTED3]], align 8
-// CHECK9-NEXT: [[TMP36:%.*]] = load ptr, ptr [[A]], align 8
-// CHECK9-NEXT: [[TMP37:%.*]] = load ptr, ptr [[B]], align 8
-// CHECK9-NEXT: [[TMP38:%.*]] = load ptr, ptr [[C]], align 8
-// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP35]], ptr [[TMP39]], align 8
-// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP35]], ptr [[TMP40]], align 8
-// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP41]], align 8
-// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK9-NEXT: store ptr [[TMP36]], ptr [[TMP42]], align 8
-// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK9-NEXT: store ptr [[TMP36]], ptr [[TMP43]], align 8
-// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP44]], align 8
-// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP37]], ptr [[TMP45]], align 8
-// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP37]], ptr [[TMP46]], align 8
-// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP47]], align 8
-// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP38]], ptr [[TMP48]], align 8
-// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP38]], ptr [[TMP49]], align 8
-// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP50]], align 8
-// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP53:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP53]], ptr [[DOTCAPTURE_EXPR_8]], align 4
-// CHECK9-NEXT: [[TMP54:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_8]], align 4
-// CHECK9-NEXT: [[SUB10:%.*]] = sub nsw i32 [[TMP54]], 0
+// CHECK9-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK9-NEXT: [[TMP39:%.*]] = load i64, ptr [[N_CASTED3]], align 8
+// CHECK9-NEXT: [[TMP40:%.*]] = load ptr, ptr [[A]], align 8
+// CHECK9-NEXT: [[TMP41:%.*]] = load ptr, ptr [[B]], align 8
+// CHECK9-NEXT: [[TMP42:%.*]] = load ptr, ptr [[C]], align 8
+// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP39]], ptr [[TMP43]], align 8
+// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP39]], ptr [[TMP44]], align 8
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK9-NEXT: store ptr [[TMP40]], ptr [[TMP46]], align 8
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK9-NEXT: store ptr [[TMP40]], ptr [[TMP47]], align 8
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP48]], align 8
+// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP41]], ptr [[TMP49]], align 8
+// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP41]], ptr [[TMP50]], align 8
+// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP51]], align 8
+// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP42]], ptr [[TMP52]], align 8
+// CHECK9-NEXT: [[TMP53:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP42]], ptr [[TMP53]], align 8
+// CHECK9-NEXT: [[TMP54:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP54]], align 8
+// CHECK9-NEXT: [[TMP55:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP56:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP57:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP57]], ptr [[DOTCAPTURE_EXPR_8]], align 4
+// CHECK9-NEXT: [[TMP58:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_8]], align 4
+// CHECK9-NEXT: [[SUB10:%.*]] = sub nsw i32 [[TMP58]], 0
// CHECK9-NEXT: [[DIV11:%.*]] = sdiv i32 [[SUB10]], 1
// CHECK9-NEXT: [[SUB12:%.*]] = sub nsw i32 [[DIV11]], 1
// CHECK9-NEXT: store i32 [[SUB12]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK9-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK9-NEXT: [[ADD13:%.*]] = add nsw i32 [[TMP55]], 1
-// CHECK9-NEXT: [[TMP56:%.*]] = zext i32 [[ADD13]] to i64
+// CHECK9-NEXT: [[TMP59:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK9-NEXT: [[ADD13:%.*]] = add nsw i32 [[TMP59]], 1
+// CHECK9-NEXT: [[TMP60:%.*]] = zext i32 [[ADD13]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS14:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP57]], align 4
-// CHECK9-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
-// CHECK9-NEXT: store i32 4, ptr [[TMP58]], align 4
-// CHECK9-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP51]], ptr [[TMP59]], align 8
-// CHECK9-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP52]], ptr [[TMP60]], align 8
-// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.32, ptr [[TMP61]], align 8
-// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP62]], align 8
-// CHECK9-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP63]], align 8
-// CHECK9-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP64]], align 8
-// CHECK9-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP56]], ptr [[TMP65]], align 8
-// CHECK9-NEXT: [[TMP66:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l51.region_id, ptr [[KERNEL_ARGS14]])
-// CHECK9-NEXT: [[TMP67:%.*]] = icmp ne i32 [[TMP66]], 0
-// CHECK9-NEXT: br i1 [[TMP67]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
+// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP61]], align 4
+// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
+// CHECK9-NEXT: store i32 4, ptr [[TMP62]], align 4
+// CHECK9-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP55]], ptr [[TMP63]], align 8
+// CHECK9-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP56]], ptr [[TMP64]], align 8
+// CHECK9-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.32, ptr [[TMP65]], align 8
+// CHECK9-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP66]], align 8
+// CHECK9-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP67]], align 8
+// CHECK9-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP68]], align 8
+// CHECK9-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP60]], ptr [[TMP69]], align 8
+// CHECK9-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP70]], align 8
+// CHECK9-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP71]], align 4
+// CHECK9-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP72]], align 4
+// CHECK9-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP73]], align 4
+// CHECK9-NEXT: [[TMP74:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l51.region_id, ptr [[KERNEL_ARGS14]])
+// CHECK9-NEXT: [[TMP75:%.*]] = icmp ne i32 [[TMP74]], 0
+// CHECK9-NEXT: br i1 [[TMP75]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
// CHECK9: omp_offload.failed15:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l51(i64 [[TMP35]], ptr [[TMP36]], ptr [[TMP37]], ptr [[TMP38]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l51(i64 [[TMP39]], ptr [[TMP40]], ptr [[TMP41]], ptr [[TMP42]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT16]]
// CHECK9: omp_offload.cont16:
-// CHECK9-NEXT: [[TMP68:%.*]] = load i32, ptr [[CH]], align 4
-// CHECK9-NEXT: store i32 [[TMP68]], ptr [[CH_CASTED]], align 4
-// CHECK9-NEXT: [[TMP69:%.*]] = load i64, ptr [[CH_CASTED]], align 8
-// CHECK9-NEXT: [[TMP70:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP70]], ptr [[N_CASTED17]], align 4
-// CHECK9-NEXT: [[TMP71:%.*]] = load i64, ptr [[N_CASTED17]], align 8
-// CHECK9-NEXT: [[TMP72:%.*]] = load ptr, ptr [[A]], align 8
-// CHECK9-NEXT: [[TMP73:%.*]] = load ptr, ptr [[B]], align 8
-// CHECK9-NEXT: [[TMP74:%.*]] = load ptr, ptr [[C]], align 8
-// CHECK9-NEXT: [[TMP75:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP69]], ptr [[TMP75]], align 8
-// CHECK9-NEXT: [[TMP76:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP69]], ptr [[TMP76]], align 8
-// CHECK9-NEXT: [[TMP77:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP77]], align 8
-// CHECK9-NEXT: [[TMP78:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP71]], ptr [[TMP78]], align 8
-// CHECK9-NEXT: [[TMP79:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP71]], ptr [[TMP79]], align 8
-// CHECK9-NEXT: [[TMP80:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP80]], align 8
-// CHECK9-NEXT: [[TMP81:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP72]], ptr [[TMP81]], align 8
-// CHECK9-NEXT: [[TMP82:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP72]], ptr [[TMP82]], align 8
-// CHECK9-NEXT: [[TMP83:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP83]], align 8
-// CHECK9-NEXT: [[TMP84:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP73]], ptr [[TMP84]], align 8
-// CHECK9-NEXT: [[TMP85:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP73]], ptr [[TMP85]], align 8
-// CHECK9-NEXT: [[TMP86:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP86]], align 8
-// CHECK9-NEXT: [[TMP87:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP74]], ptr [[TMP87]], align 8
-// CHECK9-NEXT: [[TMP88:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP74]], ptr [[TMP88]], align 8
-// CHECK9-NEXT: [[TMP89:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 4
-// CHECK9-NEXT: store ptr null, ptr [[TMP89]], align 8
-// CHECK9-NEXT: [[TMP90:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP91:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP92:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP92]], ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK9-NEXT: [[TMP93:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK9-NEXT: [[SUB24:%.*]] = sub nsw i32 [[TMP93]], 0
+// CHECK9-NEXT: [[TMP76:%.*]] = load i32, ptr [[CH]], align 4
+// CHECK9-NEXT: store i32 [[TMP76]], ptr [[CH_CASTED]], align 4
+// CHECK9-NEXT: [[TMP77:%.*]] = load i64, ptr [[CH_CASTED]], align 8
+// CHECK9-NEXT: [[TMP78:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP78]], ptr [[N_CASTED17]], align 4
+// CHECK9-NEXT: [[TMP79:%.*]] = load i64, ptr [[N_CASTED17]], align 8
+// CHECK9-NEXT: [[TMP80:%.*]] = load ptr, ptr [[A]], align 8
+// CHECK9-NEXT: [[TMP81:%.*]] = load ptr, ptr [[B]], align 8
+// CHECK9-NEXT: [[TMP82:%.*]] = load ptr, ptr [[C]], align 8
+// CHECK9-NEXT: [[TMP83:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP77]], ptr [[TMP83]], align 8
+// CHECK9-NEXT: [[TMP84:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP77]], ptr [[TMP84]], align 8
+// CHECK9-NEXT: [[TMP85:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP85]], align 8
+// CHECK9-NEXT: [[TMP86:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP79]], ptr [[TMP86]], align 8
+// CHECK9-NEXT: [[TMP87:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP79]], ptr [[TMP87]], align 8
+// CHECK9-NEXT: [[TMP88:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP88]], align 8
+// CHECK9-NEXT: [[TMP89:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP80]], ptr [[TMP89]], align 8
+// CHECK9-NEXT: [[TMP90:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP80]], ptr [[TMP90]], align 8
+// CHECK9-NEXT: [[TMP91:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP91]], align 8
+// CHECK9-NEXT: [[TMP92:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP81]], ptr [[TMP92]], align 8
+// CHECK9-NEXT: [[TMP93:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP81]], ptr [[TMP93]], align 8
+// CHECK9-NEXT: [[TMP94:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP94]], align 8
+// CHECK9-NEXT: [[TMP95:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP82]], ptr [[TMP95]], align 8
+// CHECK9-NEXT: [[TMP96:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP82]], ptr [[TMP96]], align 8
+// CHECK9-NEXT: [[TMP97:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 4
+// CHECK9-NEXT: store ptr null, ptr [[TMP97]], align 8
+// CHECK9-NEXT: [[TMP98:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP99:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP100:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP100]], ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK9-NEXT: [[TMP101:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK9-NEXT: [[SUB24:%.*]] = sub nsw i32 [[TMP101]], 0
// CHECK9-NEXT: [[DIV25:%.*]] = sdiv i32 [[SUB24]], 1
// CHECK9-NEXT: [[SUB26:%.*]] = sub nsw i32 [[DIV25]], 1
// CHECK9-NEXT: store i32 [[SUB26]], ptr [[DOTCAPTURE_EXPR_23]], align 4
-// CHECK9-NEXT: [[TMP94:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_23]], align 4
-// CHECK9-NEXT: [[ADD27:%.*]] = add nsw i32 [[TMP94]], 1
-// CHECK9-NEXT: [[TMP95:%.*]] = zext i32 [[ADD27]] to i64
+// CHECK9-NEXT: [[TMP102:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_23]], align 4
+// CHECK9-NEXT: [[ADD27:%.*]] = add nsw i32 [[TMP102]], 1
+// CHECK9-NEXT: [[TMP103:%.*]] = zext i32 [[ADD27]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS28:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP96]], align 4
-// CHECK9-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
-// CHECK9-NEXT: store i32 5, ptr [[TMP97]], align 4
-// CHECK9-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP90]], ptr [[TMP98]], align 8
-// CHECK9-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP91]], ptr [[TMP99]], align 8
-// CHECK9-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.36, ptr [[TMP100]], align 8
-// CHECK9-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP101]], align 8
-// CHECK9-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP102]], align 8
-// CHECK9-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP103]], align 8
-// CHECK9-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP95]], ptr [[TMP104]], align 8
-// CHECK9-NEXT: [[TMP105:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l59.region_id, ptr [[KERNEL_ARGS28]])
-// CHECK9-NEXT: [[TMP106:%.*]] = icmp ne i32 [[TMP105]], 0
-// CHECK9-NEXT: br i1 [[TMP106]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
+// CHECK9-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP104]], align 4
+// CHECK9-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
+// CHECK9-NEXT: store i32 5, ptr [[TMP105]], align 4
+// CHECK9-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP98]], ptr [[TMP106]], align 8
+// CHECK9-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP99]], ptr [[TMP107]], align 8
+// CHECK9-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.36, ptr [[TMP108]], align 8
+// CHECK9-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP109]], align 8
+// CHECK9-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP110]], align 8
+// CHECK9-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP111]], align 8
+// CHECK9-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP103]], ptr [[TMP112]], align 8
+// CHECK9-NEXT: [[TMP113:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP113]], align 8
+// CHECK9-NEXT: [[TMP114:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP114]], align 4
+// CHECK9-NEXT: [[TMP115:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP115]], align 4
+// CHECK9-NEXT: [[TMP116:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP116]], align 4
+// CHECK9-NEXT: [[TMP117:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l59.region_id, ptr [[KERNEL_ARGS28]])
+// CHECK9-NEXT: [[TMP118:%.*]] = icmp ne i32 [[TMP117]], 0
+// CHECK9-NEXT: br i1 [[TMP118]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
// CHECK9: omp_offload.failed29:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l59(i64 [[TMP69]], i64 [[TMP71]], ptr [[TMP72]], ptr [[TMP73]], ptr [[TMP74]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l59(i64 [[TMP77]], i64 [[TMP79]], ptr [[TMP80]], ptr [[TMP81]], ptr [[TMP82]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT30]]
// CHECK9: omp_offload.cont30:
-// CHECK9-NEXT: [[TMP107:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP107]], ptr [[N_CASTED31]], align 4
-// CHECK9-NEXT: [[TMP108:%.*]] = load i64, ptr [[N_CASTED31]], align 8
-// CHECK9-NEXT: [[TMP109:%.*]] = load ptr, ptr [[A]], align 8
-// CHECK9-NEXT: [[TMP110:%.*]] = load ptr, ptr [[B]], align 8
-// CHECK9-NEXT: [[TMP111:%.*]] = load ptr, ptr [[C]], align 8
-// CHECK9-NEXT: [[TMP112:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP108]], ptr [[TMP112]], align 8
-// CHECK9-NEXT: [[TMP113:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP108]], ptr [[TMP113]], align 8
-// CHECK9-NEXT: [[TMP114:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP114]], align 8
-// CHECK9-NEXT: [[TMP115:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 1
-// CHECK9-NEXT: store ptr [[TMP109]], ptr [[TMP115]], align 8
-// CHECK9-NEXT: [[TMP116:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 1
-// CHECK9-NEXT: store ptr [[TMP109]], ptr [[TMP116]], align 8
-// CHECK9-NEXT: [[TMP117:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP117]], align 8
-// CHECK9-NEXT: [[TMP118:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP110]], ptr [[TMP118]], align 8
-// CHECK9-NEXT: [[TMP119:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP110]], ptr [[TMP119]], align 8
-// CHECK9-NEXT: [[TMP120:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP120]], align 8
-// CHECK9-NEXT: [[TMP121:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP111]], ptr [[TMP121]], align 8
-// CHECK9-NEXT: [[TMP122:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP111]], ptr [[TMP122]], align 8
-// CHECK9-NEXT: [[TMP123:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP123]], align 8
+// CHECK9-NEXT: [[TMP119:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP119]], ptr [[N_CASTED31]], align 4
+// CHECK9-NEXT: [[TMP120:%.*]] = load i64, ptr [[N_CASTED31]], align 8
+// CHECK9-NEXT: [[TMP121:%.*]] = load ptr, ptr [[A]], align 8
+// CHECK9-NEXT: [[TMP122:%.*]] = load ptr, ptr [[B]], align 8
+// CHECK9-NEXT: [[TMP123:%.*]] = load ptr, ptr [[C]], align 8
// CHECK9-NEXT: [[TMP124:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP120]], ptr [[TMP124]], align 8
// CHECK9-NEXT: [[TMP125:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP126:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP126]], ptr [[DOTCAPTURE_EXPR_36]], align 4
-// CHECK9-NEXT: [[TMP127:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_36]], align 4
-// CHECK9-NEXT: [[SUB38:%.*]] = sub nsw i32 [[TMP127]], 0
+// CHECK9-NEXT: store i64 [[TMP120]], ptr [[TMP125]], align 8
+// CHECK9-NEXT: [[TMP126:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP126]], align 8
+// CHECK9-NEXT: [[TMP127:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 1
+// CHECK9-NEXT: store ptr [[TMP121]], ptr [[TMP127]], align 8
+// CHECK9-NEXT: [[TMP128:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 1
+// CHECK9-NEXT: store ptr [[TMP121]], ptr [[TMP128]], align 8
+// CHECK9-NEXT: [[TMP129:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP129]], align 8
+// CHECK9-NEXT: [[TMP130:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP122]], ptr [[TMP130]], align 8
+// CHECK9-NEXT: [[TMP131:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP122]], ptr [[TMP131]], align 8
+// CHECK9-NEXT: [[TMP132:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP132]], align 8
+// CHECK9-NEXT: [[TMP133:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP123]], ptr [[TMP133]], align 8
+// CHECK9-NEXT: [[TMP134:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP123]], ptr [[TMP134]], align 8
+// CHECK9-NEXT: [[TMP135:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP135]], align 8
+// CHECK9-NEXT: [[TMP136:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP137:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP138:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP138]], ptr [[DOTCAPTURE_EXPR_36]], align 4
+// CHECK9-NEXT: [[TMP139:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_36]], align 4
+// CHECK9-NEXT: [[SUB38:%.*]] = sub nsw i32 [[TMP139]], 0
// CHECK9-NEXT: [[DIV39:%.*]] = sdiv i32 [[SUB38]], 1
// CHECK9-NEXT: [[SUB40:%.*]] = sub nsw i32 [[DIV39]], 1
// CHECK9-NEXT: store i32 [[SUB40]], ptr [[DOTCAPTURE_EXPR_37]], align 4
-// CHECK9-NEXT: [[TMP128:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_37]], align 4
-// CHECK9-NEXT: [[ADD41:%.*]] = add nsw i32 [[TMP128]], 1
-// CHECK9-NEXT: [[TMP129:%.*]] = zext i32 [[ADD41]] to i64
+// CHECK9-NEXT: [[TMP140:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_37]], align 4
+// CHECK9-NEXT: [[ADD41:%.*]] = add nsw i32 [[TMP140]], 1
+// CHECK9-NEXT: [[TMP141:%.*]] = zext i32 [[ADD41]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS42:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP130]], align 4
-// CHECK9-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 1
-// CHECK9-NEXT: store i32 4, ptr [[TMP131]], align 4
-// CHECK9-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP124]], ptr [[TMP132]], align 8
-// CHECK9-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP125]], ptr [[TMP133]], align 8
-// CHECK9-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.40, ptr [[TMP134]], align 8
-// CHECK9-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.41, ptr [[TMP135]], align 8
-// CHECK9-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP136]], align 8
-// CHECK9-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP137]], align 8
-// CHECK9-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP129]], ptr [[TMP138]], align 8
-// CHECK9-NEXT: [[TMP139:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l67.region_id, ptr [[KERNEL_ARGS42]])
-// CHECK9-NEXT: [[TMP140:%.*]] = icmp ne i32 [[TMP139]], 0
-// CHECK9-NEXT: br i1 [[TMP140]], label [[OMP_OFFLOAD_FAILED43:%.*]], label [[OMP_OFFLOAD_CONT44:%.*]]
+// CHECK9-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP142]], align 4
+// CHECK9-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 1
+// CHECK9-NEXT: store i32 4, ptr [[TMP143]], align 4
+// CHECK9-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP136]], ptr [[TMP144]], align 8
+// CHECK9-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP137]], ptr [[TMP145]], align 8
+// CHECK9-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.40, ptr [[TMP146]], align 8
+// CHECK9-NEXT: [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.41, ptr [[TMP147]], align 8
+// CHECK9-NEXT: [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP148]], align 8
+// CHECK9-NEXT: [[TMP149:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP149]], align 8
+// CHECK9-NEXT: [[TMP150:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP141]], ptr [[TMP150]], align 8
+// CHECK9-NEXT: [[TMP151:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP151]], align 8
+// CHECK9-NEXT: [[TMP152:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP152]], align 4
+// CHECK9-NEXT: [[TMP153:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP153]], align 4
+// CHECK9-NEXT: [[TMP154:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP154]], align 4
+// CHECK9-NEXT: [[TMP155:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l67.region_id, ptr [[KERNEL_ARGS42]])
+// CHECK9-NEXT: [[TMP156:%.*]] = icmp ne i32 [[TMP155]], 0
+// CHECK9-NEXT: br i1 [[TMP156]], label [[OMP_OFFLOAD_FAILED43:%.*]], label [[OMP_OFFLOAD_CONT44:%.*]]
// CHECK9: omp_offload.failed43:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l67(i64 [[TMP108]], ptr [[TMP109]], ptr [[TMP110]], ptr [[TMP111]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l67(i64 [[TMP120]], ptr [[TMP121]], ptr [[TMP122]], ptr [[TMP123]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT44]]
// CHECK9: omp_offload.cont44:
-// CHECK9-NEXT: [[TMP141:%.*]] = load i32, ptr [[CH]], align 4
-// CHECK9-NEXT: store i32 [[TMP141]], ptr [[CH_CASTED45]], align 4
-// CHECK9-NEXT: [[TMP142:%.*]] = load i64, ptr [[CH_CASTED45]], align 8
-// CHECK9-NEXT: [[TMP143:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP143]], ptr [[N_CASTED46]], align 4
-// CHECK9-NEXT: [[TMP144:%.*]] = load i64, ptr [[N_CASTED46]], align 8
-// CHECK9-NEXT: [[TMP145:%.*]] = load ptr, ptr [[A]], align 8
-// CHECK9-NEXT: [[TMP146:%.*]] = load ptr, ptr [[B]], align 8
-// CHECK9-NEXT: [[TMP147:%.*]] = load ptr, ptr [[C]], align 8
-// CHECK9-NEXT: [[TMP148:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP142]], ptr [[TMP148]], align 8
-// CHECK9-NEXT: [[TMP149:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP142]], ptr [[TMP149]], align 8
-// CHECK9-NEXT: [[TMP150:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP150]], align 8
-// CHECK9-NEXT: [[TMP151:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP144]], ptr [[TMP151]], align 8
-// CHECK9-NEXT: [[TMP152:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP144]], ptr [[TMP152]], align 8
-// CHECK9-NEXT: [[TMP153:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP153]], align 8
-// CHECK9-NEXT: [[TMP154:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP145]], ptr [[TMP154]], align 8
-// CHECK9-NEXT: [[TMP155:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP145]], ptr [[TMP155]], align 8
-// CHECK9-NEXT: [[TMP156:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP156]], align 8
-// CHECK9-NEXT: [[TMP157:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP146]], ptr [[TMP157]], align 8
-// CHECK9-NEXT: [[TMP158:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP146]], ptr [[TMP158]], align 8
-// CHECK9-NEXT: [[TMP159:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP159]], align 8
-// CHECK9-NEXT: [[TMP160:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP147]], ptr [[TMP160]], align 8
-// CHECK9-NEXT: [[TMP161:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP147]], ptr [[TMP161]], align 8
-// CHECK9-NEXT: [[TMP162:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 4
-// CHECK9-NEXT: store ptr null, ptr [[TMP162]], align 8
-// CHECK9-NEXT: [[TMP163:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP164:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP165:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP165]], ptr [[DOTCAPTURE_EXPR_51]], align 4
-// CHECK9-NEXT: [[TMP166:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_51]], align 4
-// CHECK9-NEXT: [[SUB53:%.*]] = sub nsw i32 [[TMP166]], 0
+// CHECK9-NEXT: [[TMP157:%.*]] = load i32, ptr [[CH]], align 4
+// CHECK9-NEXT: store i32 [[TMP157]], ptr [[CH_CASTED45]], align 4
+// CHECK9-NEXT: [[TMP158:%.*]] = load i64, ptr [[CH_CASTED45]], align 8
+// CHECK9-NEXT: [[TMP159:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP159]], ptr [[N_CASTED46]], align 4
+// CHECK9-NEXT: [[TMP160:%.*]] = load i64, ptr [[N_CASTED46]], align 8
+// CHECK9-NEXT: [[TMP161:%.*]] = load ptr, ptr [[A]], align 8
+// CHECK9-NEXT: [[TMP162:%.*]] = load ptr, ptr [[B]], align 8
+// CHECK9-NEXT: [[TMP163:%.*]] = load ptr, ptr [[C]], align 8
+// CHECK9-NEXT: [[TMP164:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP158]], ptr [[TMP164]], align 8
+// CHECK9-NEXT: [[TMP165:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP158]], ptr [[TMP165]], align 8
+// CHECK9-NEXT: [[TMP166:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP166]], align 8
+// CHECK9-NEXT: [[TMP167:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP160]], ptr [[TMP167]], align 8
+// CHECK9-NEXT: [[TMP168:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP160]], ptr [[TMP168]], align 8
+// CHECK9-NEXT: [[TMP169:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP169]], align 8
+// CHECK9-NEXT: [[TMP170:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP161]], ptr [[TMP170]], align 8
+// CHECK9-NEXT: [[TMP171:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP161]], ptr [[TMP171]], align 8
+// CHECK9-NEXT: [[TMP172:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP172]], align 8
+// CHECK9-NEXT: [[TMP173:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP162]], ptr [[TMP173]], align 8
+// CHECK9-NEXT: [[TMP174:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP162]], ptr [[TMP174]], align 8
+// CHECK9-NEXT: [[TMP175:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP175]], align 8
+// CHECK9-NEXT: [[TMP176:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP163]], ptr [[TMP176]], align 8
+// CHECK9-NEXT: [[TMP177:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP163]], ptr [[TMP177]], align 8
+// CHECK9-NEXT: [[TMP178:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 4
+// CHECK9-NEXT: store ptr null, ptr [[TMP178]], align 8
+// CHECK9-NEXT: [[TMP179:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP180:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP181:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP181]], ptr [[DOTCAPTURE_EXPR_51]], align 4
+// CHECK9-NEXT: [[TMP182:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_51]], align 4
+// CHECK9-NEXT: [[SUB53:%.*]] = sub nsw i32 [[TMP182]], 0
// CHECK9-NEXT: [[DIV54:%.*]] = sdiv i32 [[SUB53]], 1
// CHECK9-NEXT: [[SUB55:%.*]] = sub nsw i32 [[DIV54]], 1
// CHECK9-NEXT: store i32 [[SUB55]], ptr [[DOTCAPTURE_EXPR_52]], align 4
-// CHECK9-NEXT: [[TMP167:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_52]], align 4
-// CHECK9-NEXT: [[ADD56:%.*]] = add nsw i32 [[TMP167]], 1
-// CHECK9-NEXT: [[TMP168:%.*]] = zext i32 [[ADD56]] to i64
+// CHECK9-NEXT: [[TMP183:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_52]], align 4
+// CHECK9-NEXT: [[ADD56:%.*]] = add nsw i32 [[TMP183]], 1
+// CHECK9-NEXT: [[TMP184:%.*]] = zext i32 [[ADD56]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS57:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP169:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP169]], align 4
-// CHECK9-NEXT: [[TMP170:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 1
-// CHECK9-NEXT: store i32 5, ptr [[TMP170]], align 4
-// CHECK9-NEXT: [[TMP171:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP163]], ptr [[TMP171]], align 8
-// CHECK9-NEXT: [[TMP172:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP164]], ptr [[TMP172]], align 8
-// CHECK9-NEXT: [[TMP173:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.44, ptr [[TMP173]], align 8
-// CHECK9-NEXT: [[TMP174:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.45, ptr [[TMP174]], align 8
-// CHECK9-NEXT: [[TMP175:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP175]], align 8
-// CHECK9-NEXT: [[TMP176:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP176]], align 8
-// CHECK9-NEXT: [[TMP177:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP168]], ptr [[TMP177]], align 8
-// CHECK9-NEXT: [[TMP178:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l75.region_id, ptr [[KERNEL_ARGS57]])
-// CHECK9-NEXT: [[TMP179:%.*]] = icmp ne i32 [[TMP178]], 0
-// CHECK9-NEXT: br i1 [[TMP179]], label [[OMP_OFFLOAD_FAILED58:%.*]], label [[OMP_OFFLOAD_CONT59:%.*]]
+// CHECK9-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP185]], align 4
+// CHECK9-NEXT: [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 1
+// CHECK9-NEXT: store i32 5, ptr [[TMP186]], align 4
+// CHECK9-NEXT: [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP179]], ptr [[TMP187]], align 8
+// CHECK9-NEXT: [[TMP188:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP180]], ptr [[TMP188]], align 8
+// CHECK9-NEXT: [[TMP189:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.44, ptr [[TMP189]], align 8
+// CHECK9-NEXT: [[TMP190:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.45, ptr [[TMP190]], align 8
+// CHECK9-NEXT: [[TMP191:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP191]], align 8
+// CHECK9-NEXT: [[TMP192:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP192]], align 8
+// CHECK9-NEXT: [[TMP193:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP184]], ptr [[TMP193]], align 8
+// CHECK9-NEXT: [[TMP194:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP194]], align 8
+// CHECK9-NEXT: [[TMP195:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP195]], align 4
+// CHECK9-NEXT: [[TMP196:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP196]], align 4
+// CHECK9-NEXT: [[TMP197:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP197]], align 4
+// CHECK9-NEXT: [[TMP198:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l75.region_id, ptr [[KERNEL_ARGS57]])
+// CHECK9-NEXT: [[TMP199:%.*]] = icmp ne i32 [[TMP198]], 0
+// CHECK9-NEXT: br i1 [[TMP199]], label [[OMP_OFFLOAD_FAILED58:%.*]], label [[OMP_OFFLOAD_CONT59:%.*]]
// CHECK9: omp_offload.failed58:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l75(i64 [[TMP142]], i64 [[TMP144]], ptr [[TMP145]], ptr [[TMP146]], ptr [[TMP147]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l75(i64 [[TMP158]], i64 [[TMP160]], ptr [[TMP161]], ptr [[TMP162]], ptr [[TMP163]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT59]]
// CHECK9: omp_offload.cont59:
-// CHECK9-NEXT: [[TMP180:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP180]], ptr [[N_CASTED60]], align 4
-// CHECK9-NEXT: [[TMP181:%.*]] = load i64, ptr [[N_CASTED60]], align 8
-// CHECK9-NEXT: [[TMP182:%.*]] = load ptr, ptr [[A]], align 8
-// CHECK9-NEXT: [[TMP183:%.*]] = load ptr, ptr [[B]], align 8
-// CHECK9-NEXT: [[TMP184:%.*]] = load ptr, ptr [[C]], align 8
-// CHECK9-NEXT: [[TMP185:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP181]], ptr [[TMP185]], align 8
-// CHECK9-NEXT: [[TMP186:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP181]], ptr [[TMP186]], align 8
-// CHECK9-NEXT: [[TMP187:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP187]], align 8
-// CHECK9-NEXT: [[TMP188:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 1
-// CHECK9-NEXT: store ptr [[TMP182]], ptr [[TMP188]], align 8
-// CHECK9-NEXT: [[TMP189:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 1
-// CHECK9-NEXT: store ptr [[TMP182]], ptr [[TMP189]], align 8
-// CHECK9-NEXT: [[TMP190:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP190]], align 8
-// CHECK9-NEXT: [[TMP191:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP183]], ptr [[TMP191]], align 8
-// CHECK9-NEXT: [[TMP192:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP183]], ptr [[TMP192]], align 8
-// CHECK9-NEXT: [[TMP193:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP193]], align 8
-// CHECK9-NEXT: [[TMP194:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP184]], ptr [[TMP194]], align 8
-// CHECK9-NEXT: [[TMP195:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP184]], ptr [[TMP195]], align 8
-// CHECK9-NEXT: [[TMP196:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP196]], align 8
-// CHECK9-NEXT: [[TMP197:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP198:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP199:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP199]], ptr [[DOTCAPTURE_EXPR_65]], align 4
-// CHECK9-NEXT: [[TMP200:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_65]], align 4
-// CHECK9-NEXT: [[SUB67:%.*]] = sub nsw i32 [[TMP200]], 0
+// CHECK9-NEXT: [[TMP200:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP200]], ptr [[N_CASTED60]], align 4
+// CHECK9-NEXT: [[TMP201:%.*]] = load i64, ptr [[N_CASTED60]], align 8
+// CHECK9-NEXT: [[TMP202:%.*]] = load ptr, ptr [[A]], align 8
+// CHECK9-NEXT: [[TMP203:%.*]] = load ptr, ptr [[B]], align 8
+// CHECK9-NEXT: [[TMP204:%.*]] = load ptr, ptr [[C]], align 8
+// CHECK9-NEXT: [[TMP205:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP201]], ptr [[TMP205]], align 8
+// CHECK9-NEXT: [[TMP206:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP201]], ptr [[TMP206]], align 8
+// CHECK9-NEXT: [[TMP207:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP207]], align 8
+// CHECK9-NEXT: [[TMP208:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 1
+// CHECK9-NEXT: store ptr [[TMP202]], ptr [[TMP208]], align 8
+// CHECK9-NEXT: [[TMP209:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 1
+// CHECK9-NEXT: store ptr [[TMP202]], ptr [[TMP209]], align 8
+// CHECK9-NEXT: [[TMP210:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP210]], align 8
+// CHECK9-NEXT: [[TMP211:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP203]], ptr [[TMP211]], align 8
+// CHECK9-NEXT: [[TMP212:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP203]], ptr [[TMP212]], align 8
+// CHECK9-NEXT: [[TMP213:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP213]], align 8
+// CHECK9-NEXT: [[TMP214:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP204]], ptr [[TMP214]], align 8
+// CHECK9-NEXT: [[TMP215:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP204]], ptr [[TMP215]], align 8
+// CHECK9-NEXT: [[TMP216:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP216]], align 8
+// CHECK9-NEXT: [[TMP217:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP218:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP219:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP219]], ptr [[DOTCAPTURE_EXPR_65]], align 4
+// CHECK9-NEXT: [[TMP220:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_65]], align 4
+// CHECK9-NEXT: [[SUB67:%.*]] = sub nsw i32 [[TMP220]], 0
// CHECK9-NEXT: [[DIV68:%.*]] = sdiv i32 [[SUB67]], 1
// CHECK9-NEXT: [[SUB69:%.*]] = sub nsw i32 [[DIV68]], 1
// CHECK9-NEXT: store i32 [[SUB69]], ptr [[DOTCAPTURE_EXPR_66]], align 4
-// CHECK9-NEXT: [[TMP201:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_66]], align 4
-// CHECK9-NEXT: [[ADD70:%.*]] = add nsw i32 [[TMP201]], 1
-// CHECK9-NEXT: [[TMP202:%.*]] = zext i32 [[ADD70]] to i64
+// CHECK9-NEXT: [[TMP221:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_66]], align 4
+// CHECK9-NEXT: [[ADD70:%.*]] = add nsw i32 [[TMP221]], 1
+// CHECK9-NEXT: [[TMP222:%.*]] = zext i32 [[ADD70]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS71:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP203:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP203]], align 4
-// CHECK9-NEXT: [[TMP204:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 1
-// CHECK9-NEXT: store i32 4, ptr [[TMP204]], align 4
-// CHECK9-NEXT: [[TMP205:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP197]], ptr [[TMP205]], align 8
-// CHECK9-NEXT: [[TMP206:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP198]], ptr [[TMP206]], align 8
-// CHECK9-NEXT: [[TMP207:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.48, ptr [[TMP207]], align 8
-// CHECK9-NEXT: [[TMP208:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.49, ptr [[TMP208]], align 8
-// CHECK9-NEXT: [[TMP209:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP209]], align 8
-// CHECK9-NEXT: [[TMP210:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP210]], align 8
-// CHECK9-NEXT: [[TMP211:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP202]], ptr [[TMP211]], align 8
-// CHECK9-NEXT: [[TMP212:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l83.region_id, ptr [[KERNEL_ARGS71]])
-// CHECK9-NEXT: [[TMP213:%.*]] = icmp ne i32 [[TMP212]], 0
-// CHECK9-NEXT: br i1 [[TMP213]], label [[OMP_OFFLOAD_FAILED72:%.*]], label [[OMP_OFFLOAD_CONT73:%.*]]
+// CHECK9-NEXT: [[TMP223:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP223]], align 4
+// CHECK9-NEXT: [[TMP224:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 1
+// CHECK9-NEXT: store i32 4, ptr [[TMP224]], align 4
+// CHECK9-NEXT: [[TMP225:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP217]], ptr [[TMP225]], align 8
+// CHECK9-NEXT: [[TMP226:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP218]], ptr [[TMP226]], align 8
+// CHECK9-NEXT: [[TMP227:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.48, ptr [[TMP227]], align 8
+// CHECK9-NEXT: [[TMP228:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.49, ptr [[TMP228]], align 8
+// CHECK9-NEXT: [[TMP229:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP229]], align 8
+// CHECK9-NEXT: [[TMP230:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP230]], align 8
+// CHECK9-NEXT: [[TMP231:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP222]], ptr [[TMP231]], align 8
+// CHECK9-NEXT: [[TMP232:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP232]], align 8
+// CHECK9-NEXT: [[TMP233:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP233]], align 4
+// CHECK9-NEXT: [[TMP234:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP234]], align 4
+// CHECK9-NEXT: [[TMP235:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP235]], align 4
+// CHECK9-NEXT: [[TMP236:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l83.region_id, ptr [[KERNEL_ARGS71]])
+// CHECK9-NEXT: [[TMP237:%.*]] = icmp ne i32 [[TMP236]], 0
+// CHECK9-NEXT: br i1 [[TMP237]], label [[OMP_OFFLOAD_FAILED72:%.*]], label [[OMP_OFFLOAD_CONT73:%.*]]
// CHECK9: omp_offload.failed72:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l83(i64 [[TMP181]], ptr [[TMP182]], ptr [[TMP183]], ptr [[TMP184]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l83(i64 [[TMP201]], ptr [[TMP202]], ptr [[TMP203]], ptr [[TMP204]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT73]]
// CHECK9: omp_offload.cont73:
-// CHECK9-NEXT: [[TMP214:%.*]] = load i32, ptr [[CH]], align 4
-// CHECK9-NEXT: store i32 [[TMP214]], ptr [[CH_CASTED74]], align 4
-// CHECK9-NEXT: [[TMP215:%.*]] = load i64, ptr [[CH_CASTED74]], align 8
-// CHECK9-NEXT: [[TMP216:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP216]], ptr [[N_CASTED75]], align 4
-// CHECK9-NEXT: [[TMP217:%.*]] = load i64, ptr [[N_CASTED75]], align 8
-// CHECK9-NEXT: [[TMP218:%.*]] = load ptr, ptr [[A]], align 8
-// CHECK9-NEXT: [[TMP219:%.*]] = load ptr, ptr [[B]], align 8
-// CHECK9-NEXT: [[TMP220:%.*]] = load ptr, ptr [[C]], align 8
-// CHECK9-NEXT: [[TMP221:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP215]], ptr [[TMP221]], align 8
-// CHECK9-NEXT: [[TMP222:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP215]], ptr [[TMP222]], align 8
-// CHECK9-NEXT: [[TMP223:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP223]], align 8
-// CHECK9-NEXT: [[TMP224:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP217]], ptr [[TMP224]], align 8
-// CHECK9-NEXT: [[TMP225:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP217]], ptr [[TMP225]], align 8
-// CHECK9-NEXT: [[TMP226:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP226]], align 8
-// CHECK9-NEXT: [[TMP227:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP218]], ptr [[TMP227]], align 8
-// CHECK9-NEXT: [[TMP228:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP218]], ptr [[TMP228]], align 8
-// CHECK9-NEXT: [[TMP229:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP229]], align 8
-// CHECK9-NEXT: [[TMP230:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP219]], ptr [[TMP230]], align 8
-// CHECK9-NEXT: [[TMP231:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP219]], ptr [[TMP231]], align 8
-// CHECK9-NEXT: [[TMP232:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP232]], align 8
-// CHECK9-NEXT: [[TMP233:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP220]], ptr [[TMP233]], align 8
-// CHECK9-NEXT: [[TMP234:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP220]], ptr [[TMP234]], align 8
-// CHECK9-NEXT: [[TMP235:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 4
-// CHECK9-NEXT: store ptr null, ptr [[TMP235]], align 8
-// CHECK9-NEXT: [[TMP236:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP237:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP238:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP238]], ptr [[DOTCAPTURE_EXPR_80]], align 4
-// CHECK9-NEXT: [[TMP239:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_80]], align 4
-// CHECK9-NEXT: [[SUB82:%.*]] = sub nsw i32 [[TMP239]], 0
+// CHECK9-NEXT: [[TMP238:%.*]] = load i32, ptr [[CH]], align 4
+// CHECK9-NEXT: store i32 [[TMP238]], ptr [[CH_CASTED74]], align 4
+// CHECK9-NEXT: [[TMP239:%.*]] = load i64, ptr [[CH_CASTED74]], align 8
+// CHECK9-NEXT: [[TMP240:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP240]], ptr [[N_CASTED75]], align 4
+// CHECK9-NEXT: [[TMP241:%.*]] = load i64, ptr [[N_CASTED75]], align 8
+// CHECK9-NEXT: [[TMP242:%.*]] = load ptr, ptr [[A]], align 8
+// CHECK9-NEXT: [[TMP243:%.*]] = load ptr, ptr [[B]], align 8
+// CHECK9-NEXT: [[TMP244:%.*]] = load ptr, ptr [[C]], align 8
+// CHECK9-NEXT: [[TMP245:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP239]], ptr [[TMP245]], align 8
+// CHECK9-NEXT: [[TMP246:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP239]], ptr [[TMP246]], align 8
+// CHECK9-NEXT: [[TMP247:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP247]], align 8
+// CHECK9-NEXT: [[TMP248:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP241]], ptr [[TMP248]], align 8
+// CHECK9-NEXT: [[TMP249:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP241]], ptr [[TMP249]], align 8
+// CHECK9-NEXT: [[TMP250:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP250]], align 8
+// CHECK9-NEXT: [[TMP251:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP242]], ptr [[TMP251]], align 8
+// CHECK9-NEXT: [[TMP252:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP242]], ptr [[TMP252]], align 8
+// CHECK9-NEXT: [[TMP253:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP253]], align 8
+// CHECK9-NEXT: [[TMP254:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP243]], ptr [[TMP254]], align 8
+// CHECK9-NEXT: [[TMP255:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP243]], ptr [[TMP255]], align 8
+// CHECK9-NEXT: [[TMP256:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP256]], align 8
+// CHECK9-NEXT: [[TMP257:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP244]], ptr [[TMP257]], align 8
+// CHECK9-NEXT: [[TMP258:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP244]], ptr [[TMP258]], align 8
+// CHECK9-NEXT: [[TMP259:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 4
+// CHECK9-NEXT: store ptr null, ptr [[TMP259]], align 8
+// CHECK9-NEXT: [[TMP260:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP261:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP262:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP262]], ptr [[DOTCAPTURE_EXPR_80]], align 4
+// CHECK9-NEXT: [[TMP263:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_80]], align 4
+// CHECK9-NEXT: [[SUB82:%.*]] = sub nsw i32 [[TMP263]], 0
// CHECK9-NEXT: [[DIV83:%.*]] = sdiv i32 [[SUB82]], 1
// CHECK9-NEXT: [[SUB84:%.*]] = sub nsw i32 [[DIV83]], 1
// CHECK9-NEXT: store i32 [[SUB84]], ptr [[DOTCAPTURE_EXPR_81]], align 4
-// CHECK9-NEXT: [[TMP240:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_81]], align 4
-// CHECK9-NEXT: [[ADD85:%.*]] = add nsw i32 [[TMP240]], 1
-// CHECK9-NEXT: [[TMP241:%.*]] = zext i32 [[ADD85]] to i64
+// CHECK9-NEXT: [[TMP264:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_81]], align 4
+// CHECK9-NEXT: [[ADD85:%.*]] = add nsw i32 [[TMP264]], 1
+// CHECK9-NEXT: [[TMP265:%.*]] = zext i32 [[ADD85]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS86:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP242:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP242]], align 4
-// CHECK9-NEXT: [[TMP243:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 1
-// CHECK9-NEXT: store i32 5, ptr [[TMP243]], align 4
-// CHECK9-NEXT: [[TMP244:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP236]], ptr [[TMP244]], align 8
-// CHECK9-NEXT: [[TMP245:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP237]], ptr [[TMP245]], align 8
-// CHECK9-NEXT: [[TMP246:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.52, ptr [[TMP246]], align 8
-// CHECK9-NEXT: [[TMP247:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.53, ptr [[TMP247]], align 8
-// CHECK9-NEXT: [[TMP248:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP248]], align 8
-// CHECK9-NEXT: [[TMP249:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP249]], align 8
-// CHECK9-NEXT: [[TMP250:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP241]], ptr [[TMP250]], align 8
-// CHECK9-NEXT: [[TMP251:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l91.region_id, ptr [[KERNEL_ARGS86]])
-// CHECK9-NEXT: [[TMP252:%.*]] = icmp ne i32 [[TMP251]], 0
-// CHECK9-NEXT: br i1 [[TMP252]], label [[OMP_OFFLOAD_FAILED87:%.*]], label [[OMP_OFFLOAD_CONT88:%.*]]
+// CHECK9-NEXT: [[TMP266:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP266]], align 4
+// CHECK9-NEXT: [[TMP267:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 1
+// CHECK9-NEXT: store i32 5, ptr [[TMP267]], align 4
+// CHECK9-NEXT: [[TMP268:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP260]], ptr [[TMP268]], align 8
+// CHECK9-NEXT: [[TMP269:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP261]], ptr [[TMP269]], align 8
+// CHECK9-NEXT: [[TMP270:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.52, ptr [[TMP270]], align 8
+// CHECK9-NEXT: [[TMP271:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.53, ptr [[TMP271]], align 8
+// CHECK9-NEXT: [[TMP272:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP272]], align 8
+// CHECK9-NEXT: [[TMP273:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP273]], align 8
+// CHECK9-NEXT: [[TMP274:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP265]], ptr [[TMP274]], align 8
+// CHECK9-NEXT: [[TMP275:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP275]], align 8
+// CHECK9-NEXT: [[TMP276:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP276]], align 4
+// CHECK9-NEXT: [[TMP277:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP277]], align 4
+// CHECK9-NEXT: [[TMP278:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP278]], align 4
+// CHECK9-NEXT: [[TMP279:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l91.region_id, ptr [[KERNEL_ARGS86]])
+// CHECK9-NEXT: [[TMP280:%.*]] = icmp ne i32 [[TMP279]], 0
+// CHECK9-NEXT: br i1 [[TMP280]], label [[OMP_OFFLOAD_FAILED87:%.*]], label [[OMP_OFFLOAD_CONT88:%.*]]
// CHECK9: omp_offload.failed87:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l91(i64 [[TMP215]], i64 [[TMP217]], ptr [[TMP218]], ptr [[TMP219]], ptr [[TMP220]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l91(i64 [[TMP239]], i64 [[TMP241]], ptr [[TMP242]], ptr [[TMP243]], ptr [[TMP244]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT88]]
// CHECK9: omp_offload.cont88:
// CHECK9-NEXT: ret i32 0
@@ -8889,7 +9001,7 @@ int main() {
// CHECK11-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 4, ptr [[TMP24]], align 4
// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -8906,446 +9018,502 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK11-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l369.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK11-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l369.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK11-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l369(i32 [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]]) #[[ATTR2:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK11-NEXT: [[TMP35:%.*]] = load i32, ptr [[N_CASTED3]], align 4
-// CHECK11-NEXT: [[TMP36:%.*]] = load ptr, ptr [[A]], align 4
-// CHECK11-NEXT: [[TMP37:%.*]] = load ptr, ptr [[B]], align 4
-// CHECK11-NEXT: [[TMP38:%.*]] = load ptr, ptr [[C]], align 4
-// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP35]], ptr [[TMP39]], align 4
-// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP35]], ptr [[TMP40]], align 4
-// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP41]], align 4
-// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK11-NEXT: store ptr [[TMP36]], ptr [[TMP42]], align 4
-// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK11-NEXT: store ptr [[TMP36]], ptr [[TMP43]], align 4
-// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP44]], align 4
-// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP37]], ptr [[TMP45]], align 4
-// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP37]], ptr [[TMP46]], align 4
-// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP47]], align 4
-// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP38]], ptr [[TMP48]], align 4
-// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP38]], ptr [[TMP49]], align 4
-// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP50]], align 4
-// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP53:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP53]], ptr [[DOTCAPTURE_EXPR_8]], align 4
-// CHECK11-NEXT: [[TMP54:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_8]], align 4
-// CHECK11-NEXT: [[SUB10:%.*]] = sub nsw i32 [[TMP54]], 0
+// CHECK11-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK11-NEXT: [[TMP39:%.*]] = load i32, ptr [[N_CASTED3]], align 4
+// CHECK11-NEXT: [[TMP40:%.*]] = load ptr, ptr [[A]], align 4
+// CHECK11-NEXT: [[TMP41:%.*]] = load ptr, ptr [[B]], align 4
+// CHECK11-NEXT: [[TMP42:%.*]] = load ptr, ptr [[C]], align 4
+// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP39]], ptr [[TMP43]], align 4
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP39]], ptr [[TMP44]], align 4
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK11-NEXT: store ptr [[TMP40]], ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK11-NEXT: store ptr [[TMP40]], ptr [[TMP47]], align 4
+// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP48]], align 4
+// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP41]], ptr [[TMP49]], align 4
+// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP41]], ptr [[TMP50]], align 4
+// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP42]], ptr [[TMP52]], align 4
+// CHECK11-NEXT: [[TMP53:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP42]], ptr [[TMP53]], align 4
+// CHECK11-NEXT: [[TMP54:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP54]], align 4
+// CHECK11-NEXT: [[TMP55:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP56:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP57:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP57]], ptr [[DOTCAPTURE_EXPR_8]], align 4
+// CHECK11-NEXT: [[TMP58:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_8]], align 4
+// CHECK11-NEXT: [[SUB10:%.*]] = sub nsw i32 [[TMP58]], 0
// CHECK11-NEXT: [[DIV11:%.*]] = sdiv i32 [[SUB10]], 1
// CHECK11-NEXT: [[SUB12:%.*]] = sub nsw i32 [[DIV11]], 1
// CHECK11-NEXT: store i32 [[SUB12]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK11-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK11-NEXT: [[ADD13:%.*]] = add nsw i32 [[TMP55]], 1
-// CHECK11-NEXT: [[TMP56:%.*]] = zext i32 [[ADD13]] to i64
+// CHECK11-NEXT: [[TMP59:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK11-NEXT: [[ADD13:%.*]] = add nsw i32 [[TMP59]], 1
+// CHECK11-NEXT: [[TMP60:%.*]] = zext i32 [[ADD13]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS14:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP57]], align 4
-// CHECK11-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
-// CHECK11-NEXT: store i32 4, ptr [[TMP58]], align 4
-// CHECK11-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP51]], ptr [[TMP59]], align 4
-// CHECK11-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP52]], ptr [[TMP60]], align 4
-// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.4, ptr [[TMP61]], align 4
-// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP62]], align 4
-// CHECK11-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP63]], align 4
-// CHECK11-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP64]], align 4
-// CHECK11-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP56]], ptr [[TMP65]], align 8
-// CHECK11-NEXT: [[TMP66:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l408.region_id, ptr [[KERNEL_ARGS14]])
-// CHECK11-NEXT: [[TMP67:%.*]] = icmp ne i32 [[TMP66]], 0
-// CHECK11-NEXT: br i1 [[TMP67]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
+// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP61]], align 4
+// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
+// CHECK11-NEXT: store i32 4, ptr [[TMP62]], align 4
+// CHECK11-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP55]], ptr [[TMP63]], align 4
+// CHECK11-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP56]], ptr [[TMP64]], align 4
+// CHECK11-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.4, ptr [[TMP65]], align 4
+// CHECK11-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP66]], align 4
+// CHECK11-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP67]], align 4
+// CHECK11-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP68]], align 4
+// CHECK11-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP60]], ptr [[TMP69]], align 8
+// CHECK11-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP70]], align 8
+// CHECK11-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP71]], align 4
+// CHECK11-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP72]], align 4
+// CHECK11-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP73]], align 4
+// CHECK11-NEXT: [[TMP74:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l408.region_id, ptr [[KERNEL_ARGS14]])
+// CHECK11-NEXT: [[TMP75:%.*]] = icmp ne i32 [[TMP74]], 0
+// CHECK11-NEXT: br i1 [[TMP75]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
// CHECK11: omp_offload.failed15:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l408(i32 [[TMP35]], ptr [[TMP36]], ptr [[TMP37]], ptr [[TMP38]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l408(i32 [[TMP39]], ptr [[TMP40]], ptr [[TMP41]], ptr [[TMP42]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT16]]
// CHECK11: omp_offload.cont16:
-// CHECK11-NEXT: [[TMP68:%.*]] = load i32, ptr [[CH]], align 4
-// CHECK11-NEXT: store i32 [[TMP68]], ptr [[CH_CASTED]], align 4
-// CHECK11-NEXT: [[TMP69:%.*]] = load i32, ptr [[CH_CASTED]], align 4
-// CHECK11-NEXT: [[TMP70:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP70]], ptr [[N_CASTED17]], align 4
-// CHECK11-NEXT: [[TMP71:%.*]] = load i32, ptr [[N_CASTED17]], align 4
-// CHECK11-NEXT: [[TMP72:%.*]] = load ptr, ptr [[A]], align 4
-// CHECK11-NEXT: [[TMP73:%.*]] = load ptr, ptr [[B]], align 4
-// CHECK11-NEXT: [[TMP74:%.*]] = load ptr, ptr [[C]], align 4
-// CHECK11-NEXT: [[TMP75:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP69]], ptr [[TMP75]], align 4
-// CHECK11-NEXT: [[TMP76:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP69]], ptr [[TMP76]], align 4
-// CHECK11-NEXT: [[TMP77:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP77]], align 4
-// CHECK11-NEXT: [[TMP78:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP71]], ptr [[TMP78]], align 4
-// CHECK11-NEXT: [[TMP79:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP71]], ptr [[TMP79]], align 4
-// CHECK11-NEXT: [[TMP80:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP80]], align 4
-// CHECK11-NEXT: [[TMP81:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP72]], ptr [[TMP81]], align 4
-// CHECK11-NEXT: [[TMP82:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP72]], ptr [[TMP82]], align 4
-// CHECK11-NEXT: [[TMP83:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP83]], align 4
-// CHECK11-NEXT: [[TMP84:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP73]], ptr [[TMP84]], align 4
-// CHECK11-NEXT: [[TMP85:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP73]], ptr [[TMP85]], align 4
-// CHECK11-NEXT: [[TMP86:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP86]], align 4
-// CHECK11-NEXT: [[TMP87:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP74]], ptr [[TMP87]], align 4
-// CHECK11-NEXT: [[TMP88:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP74]], ptr [[TMP88]], align 4
-// CHECK11-NEXT: [[TMP89:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 4
-// CHECK11-NEXT: store ptr null, ptr [[TMP89]], align 4
-// CHECK11-NEXT: [[TMP90:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP91:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP92:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP92]], ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK11-NEXT: [[TMP93:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK11-NEXT: [[SUB24:%.*]] = sub nsw i32 [[TMP93]], 0
+// CHECK11-NEXT: [[TMP76:%.*]] = load i32, ptr [[CH]], align 4
+// CHECK11-NEXT: store i32 [[TMP76]], ptr [[CH_CASTED]], align 4
+// CHECK11-NEXT: [[TMP77:%.*]] = load i32, ptr [[CH_CASTED]], align 4
+// CHECK11-NEXT: [[TMP78:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP78]], ptr [[N_CASTED17]], align 4
+// CHECK11-NEXT: [[TMP79:%.*]] = load i32, ptr [[N_CASTED17]], align 4
+// CHECK11-NEXT: [[TMP80:%.*]] = load ptr, ptr [[A]], align 4
+// CHECK11-NEXT: [[TMP81:%.*]] = load ptr, ptr [[B]], align 4
+// CHECK11-NEXT: [[TMP82:%.*]] = load ptr, ptr [[C]], align 4
+// CHECK11-NEXT: [[TMP83:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP77]], ptr [[TMP83]], align 4
+// CHECK11-NEXT: [[TMP84:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP77]], ptr [[TMP84]], align 4
+// CHECK11-NEXT: [[TMP85:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP85]], align 4
+// CHECK11-NEXT: [[TMP86:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP79]], ptr [[TMP86]], align 4
+// CHECK11-NEXT: [[TMP87:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP79]], ptr [[TMP87]], align 4
+// CHECK11-NEXT: [[TMP88:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP88]], align 4
+// CHECK11-NEXT: [[TMP89:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP80]], ptr [[TMP89]], align 4
+// CHECK11-NEXT: [[TMP90:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP80]], ptr [[TMP90]], align 4
+// CHECK11-NEXT: [[TMP91:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP91]], align 4
+// CHECK11-NEXT: [[TMP92:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP81]], ptr [[TMP92]], align 4
+// CHECK11-NEXT: [[TMP93:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP81]], ptr [[TMP93]], align 4
+// CHECK11-NEXT: [[TMP94:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP94]], align 4
+// CHECK11-NEXT: [[TMP95:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP82]], ptr [[TMP95]], align 4
+// CHECK11-NEXT: [[TMP96:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP82]], ptr [[TMP96]], align 4
+// CHECK11-NEXT: [[TMP97:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 4
+// CHECK11-NEXT: store ptr null, ptr [[TMP97]], align 4
+// CHECK11-NEXT: [[TMP98:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP99:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP100:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP100]], ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK11-NEXT: [[TMP101:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK11-NEXT: [[SUB24:%.*]] = sub nsw i32 [[TMP101]], 0
// CHECK11-NEXT: [[DIV25:%.*]] = sdiv i32 [[SUB24]], 1
// CHECK11-NEXT: [[SUB26:%.*]] = sub nsw i32 [[DIV25]], 1
// CHECK11-NEXT: store i32 [[SUB26]], ptr [[DOTCAPTURE_EXPR_23]], align 4
-// CHECK11-NEXT: [[TMP94:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_23]], align 4
-// CHECK11-NEXT: [[ADD27:%.*]] = add nsw i32 [[TMP94]], 1
-// CHECK11-NEXT: [[TMP95:%.*]] = zext i32 [[ADD27]] to i64
+// CHECK11-NEXT: [[TMP102:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_23]], align 4
+// CHECK11-NEXT: [[ADD27:%.*]] = add nsw i32 [[TMP102]], 1
+// CHECK11-NEXT: [[TMP103:%.*]] = zext i32 [[ADD27]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS28:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP96]], align 4
-// CHECK11-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
-// CHECK11-NEXT: store i32 5, ptr [[TMP97]], align 4
-// CHECK11-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP90]], ptr [[TMP98]], align 4
-// CHECK11-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP91]], ptr [[TMP99]], align 4
-// CHECK11-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.8, ptr [[TMP100]], align 4
-// CHECK11-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP101]], align 4
-// CHECK11-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP102]], align 4
-// CHECK11-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP103]], align 4
-// CHECK11-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP95]], ptr [[TMP104]], align 8
-// CHECK11-NEXT: [[TMP105:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l447.region_id, ptr [[KERNEL_ARGS28]])
-// CHECK11-NEXT: [[TMP106:%.*]] = icmp ne i32 [[TMP105]], 0
-// CHECK11-NEXT: br i1 [[TMP106]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
+// CHECK11-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP104]], align 4
+// CHECK11-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
+// CHECK11-NEXT: store i32 5, ptr [[TMP105]], align 4
+// CHECK11-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP98]], ptr [[TMP106]], align 4
+// CHECK11-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP99]], ptr [[TMP107]], align 4
+// CHECK11-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.8, ptr [[TMP108]], align 4
+// CHECK11-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP109]], align 4
+// CHECK11-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP110]], align 4
+// CHECK11-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP111]], align 4
+// CHECK11-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP103]], ptr [[TMP112]], align 8
+// CHECK11-NEXT: [[TMP113:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP113]], align 8
+// CHECK11-NEXT: [[TMP114:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP114]], align 4
+// CHECK11-NEXT: [[TMP115:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP115]], align 4
+// CHECK11-NEXT: [[TMP116:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP116]], align 4
+// CHECK11-NEXT: [[TMP117:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l447.region_id, ptr [[KERNEL_ARGS28]])
+// CHECK11-NEXT: [[TMP118:%.*]] = icmp ne i32 [[TMP117]], 0
+// CHECK11-NEXT: br i1 [[TMP118]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
// CHECK11: omp_offload.failed29:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l447(i32 [[TMP69]], i32 [[TMP71]], ptr [[TMP72]], ptr [[TMP73]], ptr [[TMP74]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l447(i32 [[TMP77]], i32 [[TMP79]], ptr [[TMP80]], ptr [[TMP81]], ptr [[TMP82]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT30]]
// CHECK11: omp_offload.cont30:
-// CHECK11-NEXT: [[TMP107:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP107]], ptr [[N_CASTED31]], align 4
-// CHECK11-NEXT: [[TMP108:%.*]] = load i32, ptr [[N_CASTED31]], align 4
-// CHECK11-NEXT: [[TMP109:%.*]] = load ptr, ptr [[A]], align 4
-// CHECK11-NEXT: [[TMP110:%.*]] = load ptr, ptr [[B]], align 4
-// CHECK11-NEXT: [[TMP111:%.*]] = load ptr, ptr [[C]], align 4
-// CHECK11-NEXT: [[TMP112:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP108]], ptr [[TMP112]], align 4
-// CHECK11-NEXT: [[TMP113:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP108]], ptr [[TMP113]], align 4
-// CHECK11-NEXT: [[TMP114:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP114]], align 4
-// CHECK11-NEXT: [[TMP115:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 1
-// CHECK11-NEXT: store ptr [[TMP109]], ptr [[TMP115]], align 4
-// CHECK11-NEXT: [[TMP116:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 1
-// CHECK11-NEXT: store ptr [[TMP109]], ptr [[TMP116]], align 4
-// CHECK11-NEXT: [[TMP117:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP117]], align 4
-// CHECK11-NEXT: [[TMP118:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP110]], ptr [[TMP118]], align 4
-// CHECK11-NEXT: [[TMP119:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP110]], ptr [[TMP119]], align 4
-// CHECK11-NEXT: [[TMP120:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP120]], align 4
-// CHECK11-NEXT: [[TMP121:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP111]], ptr [[TMP121]], align 4
-// CHECK11-NEXT: [[TMP122:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP111]], ptr [[TMP122]], align 4
-// CHECK11-NEXT: [[TMP123:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP123]], align 4
+// CHECK11-NEXT: [[TMP119:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP119]], ptr [[N_CASTED31]], align 4
+// CHECK11-NEXT: [[TMP120:%.*]] = load i32, ptr [[N_CASTED31]], align 4
+// CHECK11-NEXT: [[TMP121:%.*]] = load ptr, ptr [[A]], align 4
+// CHECK11-NEXT: [[TMP122:%.*]] = load ptr, ptr [[B]], align 4
+// CHECK11-NEXT: [[TMP123:%.*]] = load ptr, ptr [[C]], align 4
// CHECK11-NEXT: [[TMP124:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP120]], ptr [[TMP124]], align 4
// CHECK11-NEXT: [[TMP125:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP126:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP126]], ptr [[DOTCAPTURE_EXPR_36]], align 4
-// CHECK11-NEXT: [[TMP127:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_36]], align 4
-// CHECK11-NEXT: [[SUB38:%.*]] = sub nsw i32 [[TMP127]], 0
+// CHECK11-NEXT: store i32 [[TMP120]], ptr [[TMP125]], align 4
+// CHECK11-NEXT: [[TMP126:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP126]], align 4
+// CHECK11-NEXT: [[TMP127:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 1
+// CHECK11-NEXT: store ptr [[TMP121]], ptr [[TMP127]], align 4
+// CHECK11-NEXT: [[TMP128:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 1
+// CHECK11-NEXT: store ptr [[TMP121]], ptr [[TMP128]], align 4
+// CHECK11-NEXT: [[TMP129:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP129]], align 4
+// CHECK11-NEXT: [[TMP130:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP122]], ptr [[TMP130]], align 4
+// CHECK11-NEXT: [[TMP131:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP122]], ptr [[TMP131]], align 4
+// CHECK11-NEXT: [[TMP132:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP132]], align 4
+// CHECK11-NEXT: [[TMP133:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP123]], ptr [[TMP133]], align 4
+// CHECK11-NEXT: [[TMP134:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP123]], ptr [[TMP134]], align 4
+// CHECK11-NEXT: [[TMP135:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP135]], align 4
+// CHECK11-NEXT: [[TMP136:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP137:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP138:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP138]], ptr [[DOTCAPTURE_EXPR_36]], align 4
+// CHECK11-NEXT: [[TMP139:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_36]], align 4
+// CHECK11-NEXT: [[SUB38:%.*]] = sub nsw i32 [[TMP139]], 0
// CHECK11-NEXT: [[DIV39:%.*]] = sdiv i32 [[SUB38]], 1
// CHECK11-NEXT: [[SUB40:%.*]] = sub nsw i32 [[DIV39]], 1
// CHECK11-NEXT: store i32 [[SUB40]], ptr [[DOTCAPTURE_EXPR_37]], align 4
-// CHECK11-NEXT: [[TMP128:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_37]], align 4
-// CHECK11-NEXT: [[ADD41:%.*]] = add nsw i32 [[TMP128]], 1
-// CHECK11-NEXT: [[TMP129:%.*]] = zext i32 [[ADD41]] to i64
+// CHECK11-NEXT: [[TMP140:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_37]], align 4
+// CHECK11-NEXT: [[ADD41:%.*]] = add nsw i32 [[TMP140]], 1
+// CHECK11-NEXT: [[TMP141:%.*]] = zext i32 [[ADD41]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS42:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP130]], align 4
-// CHECK11-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 1
-// CHECK11-NEXT: store i32 4, ptr [[TMP131]], align 4
-// CHECK11-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP124]], ptr [[TMP132]], align 4
-// CHECK11-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP125]], ptr [[TMP133]], align 4
-// CHECK11-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.12, ptr [[TMP134]], align 4
-// CHECK11-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP135]], align 4
-// CHECK11-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP136]], align 4
-// CHECK11-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP137]], align 4
-// CHECK11-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP129]], ptr [[TMP138]], align 8
-// CHECK11-NEXT: [[TMP139:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l478.region_id, ptr [[KERNEL_ARGS42]])
-// CHECK11-NEXT: [[TMP140:%.*]] = icmp ne i32 [[TMP139]], 0
-// CHECK11-NEXT: br i1 [[TMP140]], label [[OMP_OFFLOAD_FAILED43:%.*]], label [[OMP_OFFLOAD_CONT44:%.*]]
+// CHECK11-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP142]], align 4
+// CHECK11-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 1
+// CHECK11-NEXT: store i32 4, ptr [[TMP143]], align 4
+// CHECK11-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP136]], ptr [[TMP144]], align 4
+// CHECK11-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP137]], ptr [[TMP145]], align 4
+// CHECK11-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.12, ptr [[TMP146]], align 4
+// CHECK11-NEXT: [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP147]], align 4
+// CHECK11-NEXT: [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP148]], align 4
+// CHECK11-NEXT: [[TMP149:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP149]], align 4
+// CHECK11-NEXT: [[TMP150:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP141]], ptr [[TMP150]], align 8
+// CHECK11-NEXT: [[TMP151:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP151]], align 8
+// CHECK11-NEXT: [[TMP152:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP152]], align 4
+// CHECK11-NEXT: [[TMP153:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP153]], align 4
+// CHECK11-NEXT: [[TMP154:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP154]], align 4
+// CHECK11-NEXT: [[TMP155:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l478.region_id, ptr [[KERNEL_ARGS42]])
+// CHECK11-NEXT: [[TMP156:%.*]] = icmp ne i32 [[TMP155]], 0
+// CHECK11-NEXT: br i1 [[TMP156]], label [[OMP_OFFLOAD_FAILED43:%.*]], label [[OMP_OFFLOAD_CONT44:%.*]]
// CHECK11: omp_offload.failed43:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l478(i32 [[TMP108]], ptr [[TMP109]], ptr [[TMP110]], ptr [[TMP111]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l478(i32 [[TMP120]], ptr [[TMP121]], ptr [[TMP122]], ptr [[TMP123]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT44]]
// CHECK11: omp_offload.cont44:
-// CHECK11-NEXT: [[TMP141:%.*]] = load i32, ptr [[CH]], align 4
-// CHECK11-NEXT: store i32 [[TMP141]], ptr [[CH_CASTED45]], align 4
-// CHECK11-NEXT: [[TMP142:%.*]] = load i32, ptr [[CH_CASTED45]], align 4
-// CHECK11-NEXT: [[TMP143:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP143]], ptr [[N_CASTED46]], align 4
-// CHECK11-NEXT: [[TMP144:%.*]] = load i32, ptr [[N_CASTED46]], align 4
-// CHECK11-NEXT: [[TMP145:%.*]] = load ptr, ptr [[A]], align 4
-// CHECK11-NEXT: [[TMP146:%.*]] = load ptr, ptr [[B]], align 4
-// CHECK11-NEXT: [[TMP147:%.*]] = load ptr, ptr [[C]], align 4
-// CHECK11-NEXT: [[TMP148:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP142]], ptr [[TMP148]], align 4
-// CHECK11-NEXT: [[TMP149:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP142]], ptr [[TMP149]], align 4
-// CHECK11-NEXT: [[TMP150:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP150]], align 4
-// CHECK11-NEXT: [[TMP151:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP144]], ptr [[TMP151]], align 4
-// CHECK11-NEXT: [[TMP152:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP144]], ptr [[TMP152]], align 4
-// CHECK11-NEXT: [[TMP153:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP153]], align 4
-// CHECK11-NEXT: [[TMP154:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP145]], ptr [[TMP154]], align 4
-// CHECK11-NEXT: [[TMP155:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP145]], ptr [[TMP155]], align 4
-// CHECK11-NEXT: [[TMP156:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP156]], align 4
-// CHECK11-NEXT: [[TMP157:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP146]], ptr [[TMP157]], align 4
-// CHECK11-NEXT: [[TMP158:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP146]], ptr [[TMP158]], align 4
-// CHECK11-NEXT: [[TMP159:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP159]], align 4
-// CHECK11-NEXT: [[TMP160:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP147]], ptr [[TMP160]], align 4
-// CHECK11-NEXT: [[TMP161:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP147]], ptr [[TMP161]], align 4
-// CHECK11-NEXT: [[TMP162:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 4
-// CHECK11-NEXT: store ptr null, ptr [[TMP162]], align 4
-// CHECK11-NEXT: [[TMP163:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP164:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP165:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP165]], ptr [[DOTCAPTURE_EXPR_51]], align 4
-// CHECK11-NEXT: [[TMP166:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_51]], align 4
-// CHECK11-NEXT: [[SUB53:%.*]] = sub nsw i32 [[TMP166]], 0
+// CHECK11-NEXT: [[TMP157:%.*]] = load i32, ptr [[CH]], align 4
+// CHECK11-NEXT: store i32 [[TMP157]], ptr [[CH_CASTED45]], align 4
+// CHECK11-NEXT: [[TMP158:%.*]] = load i32, ptr [[CH_CASTED45]], align 4
+// CHECK11-NEXT: [[TMP159:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP159]], ptr [[N_CASTED46]], align 4
+// CHECK11-NEXT: [[TMP160:%.*]] = load i32, ptr [[N_CASTED46]], align 4
+// CHECK11-NEXT: [[TMP161:%.*]] = load ptr, ptr [[A]], align 4
+// CHECK11-NEXT: [[TMP162:%.*]] = load ptr, ptr [[B]], align 4
+// CHECK11-NEXT: [[TMP163:%.*]] = load ptr, ptr [[C]], align 4
+// CHECK11-NEXT: [[TMP164:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP158]], ptr [[TMP164]], align 4
+// CHECK11-NEXT: [[TMP165:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP158]], ptr [[TMP165]], align 4
+// CHECK11-NEXT: [[TMP166:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP166]], align 4
+// CHECK11-NEXT: [[TMP167:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP160]], ptr [[TMP167]], align 4
+// CHECK11-NEXT: [[TMP168:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP160]], ptr [[TMP168]], align 4
+// CHECK11-NEXT: [[TMP169:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP169]], align 4
+// CHECK11-NEXT: [[TMP170:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP161]], ptr [[TMP170]], align 4
+// CHECK11-NEXT: [[TMP171:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP161]], ptr [[TMP171]], align 4
+// CHECK11-NEXT: [[TMP172:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP172]], align 4
+// CHECK11-NEXT: [[TMP173:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP162]], ptr [[TMP173]], align 4
+// CHECK11-NEXT: [[TMP174:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP162]], ptr [[TMP174]], align 4
+// CHECK11-NEXT: [[TMP175:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP175]], align 4
+// CHECK11-NEXT: [[TMP176:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP163]], ptr [[TMP176]], align 4
+// CHECK11-NEXT: [[TMP177:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP163]], ptr [[TMP177]], align 4
+// CHECK11-NEXT: [[TMP178:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 4
+// CHECK11-NEXT: store ptr null, ptr [[TMP178]], align 4
+// CHECK11-NEXT: [[TMP179:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP180:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP181:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP181]], ptr [[DOTCAPTURE_EXPR_51]], align 4
+// CHECK11-NEXT: [[TMP182:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_51]], align 4
+// CHECK11-NEXT: [[SUB53:%.*]] = sub nsw i32 [[TMP182]], 0
// CHECK11-NEXT: [[DIV54:%.*]] = sdiv i32 [[SUB53]], 1
// CHECK11-NEXT: [[SUB55:%.*]] = sub nsw i32 [[DIV54]], 1
// CHECK11-NEXT: store i32 [[SUB55]], ptr [[DOTCAPTURE_EXPR_52]], align 4
-// CHECK11-NEXT: [[TMP167:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_52]], align 4
-// CHECK11-NEXT: [[ADD56:%.*]] = add nsw i32 [[TMP167]], 1
-// CHECK11-NEXT: [[TMP168:%.*]] = zext i32 [[ADD56]] to i64
+// CHECK11-NEXT: [[TMP183:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_52]], align 4
+// CHECK11-NEXT: [[ADD56:%.*]] = add nsw i32 [[TMP183]], 1
+// CHECK11-NEXT: [[TMP184:%.*]] = zext i32 [[ADD56]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS57:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP169:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP169]], align 4
-// CHECK11-NEXT: [[TMP170:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 1
-// CHECK11-NEXT: store i32 5, ptr [[TMP170]], align 4
-// CHECK11-NEXT: [[TMP171:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP163]], ptr [[TMP171]], align 4
-// CHECK11-NEXT: [[TMP172:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP164]], ptr [[TMP172]], align 4
-// CHECK11-NEXT: [[TMP173:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.16, ptr [[TMP173]], align 4
-// CHECK11-NEXT: [[TMP174:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP174]], align 4
-// CHECK11-NEXT: [[TMP175:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP175]], align 4
-// CHECK11-NEXT: [[TMP176:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP176]], align 4
-// CHECK11-NEXT: [[TMP177:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP168]], ptr [[TMP177]], align 8
-// CHECK11-NEXT: [[TMP178:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l506.region_id, ptr [[KERNEL_ARGS57]])
-// CHECK11-NEXT: [[TMP179:%.*]] = icmp ne i32 [[TMP178]], 0
-// CHECK11-NEXT: br i1 [[TMP179]], label [[OMP_OFFLOAD_FAILED58:%.*]], label [[OMP_OFFLOAD_CONT59:%.*]]
+// CHECK11-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP185]], align 4
+// CHECK11-NEXT: [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 1
+// CHECK11-NEXT: store i32 5, ptr [[TMP186]], align 4
+// CHECK11-NEXT: [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP179]], ptr [[TMP187]], align 4
+// CHECK11-NEXT: [[TMP188:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP180]], ptr [[TMP188]], align 4
+// CHECK11-NEXT: [[TMP189:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.16, ptr [[TMP189]], align 4
+// CHECK11-NEXT: [[TMP190:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP190]], align 4
+// CHECK11-NEXT: [[TMP191:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP191]], align 4
+// CHECK11-NEXT: [[TMP192:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP192]], align 4
+// CHECK11-NEXT: [[TMP193:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP184]], ptr [[TMP193]], align 8
+// CHECK11-NEXT: [[TMP194:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP194]], align 8
+// CHECK11-NEXT: [[TMP195:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP195]], align 4
+// CHECK11-NEXT: [[TMP196:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP196]], align 4
+// CHECK11-NEXT: [[TMP197:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP197]], align 4
+// CHECK11-NEXT: [[TMP198:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l506.region_id, ptr [[KERNEL_ARGS57]])
+// CHECK11-NEXT: [[TMP199:%.*]] = icmp ne i32 [[TMP198]], 0
+// CHECK11-NEXT: br i1 [[TMP199]], label [[OMP_OFFLOAD_FAILED58:%.*]], label [[OMP_OFFLOAD_CONT59:%.*]]
// CHECK11: omp_offload.failed58:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l506(i32 [[TMP142]], i32 [[TMP144]], ptr [[TMP145]], ptr [[TMP146]], ptr [[TMP147]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l506(i32 [[TMP158]], i32 [[TMP160]], ptr [[TMP161]], ptr [[TMP162]], ptr [[TMP163]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT59]]
// CHECK11: omp_offload.cont59:
-// CHECK11-NEXT: [[TMP180:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP180]], ptr [[N_CASTED60]], align 4
-// CHECK11-NEXT: [[TMP181:%.*]] = load i32, ptr [[N_CASTED60]], align 4
-// CHECK11-NEXT: [[TMP182:%.*]] = load ptr, ptr [[A]], align 4
-// CHECK11-NEXT: [[TMP183:%.*]] = load ptr, ptr [[B]], align 4
-// CHECK11-NEXT: [[TMP184:%.*]] = load ptr, ptr [[C]], align 4
-// CHECK11-NEXT: [[TMP185:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP181]], ptr [[TMP185]], align 4
-// CHECK11-NEXT: [[TMP186:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP181]], ptr [[TMP186]], align 4
-// CHECK11-NEXT: [[TMP187:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP187]], align 4
-// CHECK11-NEXT: [[TMP188:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 1
-// CHECK11-NEXT: store ptr [[TMP182]], ptr [[TMP188]], align 4
-// CHECK11-NEXT: [[TMP189:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 1
-// CHECK11-NEXT: store ptr [[TMP182]], ptr [[TMP189]], align 4
-// CHECK11-NEXT: [[TMP190:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP190]], align 4
-// CHECK11-NEXT: [[TMP191:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP183]], ptr [[TMP191]], align 4
-// CHECK11-NEXT: [[TMP192:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP183]], ptr [[TMP192]], align 4
-// CHECK11-NEXT: [[TMP193:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP193]], align 4
-// CHECK11-NEXT: [[TMP194:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP184]], ptr [[TMP194]], align 4
-// CHECK11-NEXT: [[TMP195:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP184]], ptr [[TMP195]], align 4
-// CHECK11-NEXT: [[TMP196:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP196]], align 4
-// CHECK11-NEXT: [[TMP197:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP198:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP199:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP199]], ptr [[DOTCAPTURE_EXPR_65]], align 4
-// CHECK11-NEXT: [[TMP200:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_65]], align 4
-// CHECK11-NEXT: [[SUB67:%.*]] = sub nsw i32 [[TMP200]], 0
+// CHECK11-NEXT: [[TMP200:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP200]], ptr [[N_CASTED60]], align 4
+// CHECK11-NEXT: [[TMP201:%.*]] = load i32, ptr [[N_CASTED60]], align 4
+// CHECK11-NEXT: [[TMP202:%.*]] = load ptr, ptr [[A]], align 4
+// CHECK11-NEXT: [[TMP203:%.*]] = load ptr, ptr [[B]], align 4
+// CHECK11-NEXT: [[TMP204:%.*]] = load ptr, ptr [[C]], align 4
+// CHECK11-NEXT: [[TMP205:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP201]], ptr [[TMP205]], align 4
+// CHECK11-NEXT: [[TMP206:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP201]], ptr [[TMP206]], align 4
+// CHECK11-NEXT: [[TMP207:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP207]], align 4
+// CHECK11-NEXT: [[TMP208:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 1
+// CHECK11-NEXT: store ptr [[TMP202]], ptr [[TMP208]], align 4
+// CHECK11-NEXT: [[TMP209:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 1
+// CHECK11-NEXT: store ptr [[TMP202]], ptr [[TMP209]], align 4
+// CHECK11-NEXT: [[TMP210:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP210]], align 4
+// CHECK11-NEXT: [[TMP211:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP203]], ptr [[TMP211]], align 4
+// CHECK11-NEXT: [[TMP212:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP203]], ptr [[TMP212]], align 4
+// CHECK11-NEXT: [[TMP213:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP213]], align 4
+// CHECK11-NEXT: [[TMP214:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP204]], ptr [[TMP214]], align 4
+// CHECK11-NEXT: [[TMP215:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP204]], ptr [[TMP215]], align 4
+// CHECK11-NEXT: [[TMP216:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP216]], align 4
+// CHECK11-NEXT: [[TMP217:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP218:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP219:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP219]], ptr [[DOTCAPTURE_EXPR_65]], align 4
+// CHECK11-NEXT: [[TMP220:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_65]], align 4
+// CHECK11-NEXT: [[SUB67:%.*]] = sub nsw i32 [[TMP220]], 0
// CHECK11-NEXT: [[DIV68:%.*]] = sdiv i32 [[SUB67]], 1
// CHECK11-NEXT: [[SUB69:%.*]] = sub nsw i32 [[DIV68]], 1
// CHECK11-NEXT: store i32 [[SUB69]], ptr [[DOTCAPTURE_EXPR_66]], align 4
-// CHECK11-NEXT: [[TMP201:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_66]], align 4
-// CHECK11-NEXT: [[ADD70:%.*]] = add nsw i32 [[TMP201]], 1
-// CHECK11-NEXT: [[TMP202:%.*]] = zext i32 [[ADD70]] to i64
+// CHECK11-NEXT: [[TMP221:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_66]], align 4
+// CHECK11-NEXT: [[ADD70:%.*]] = add nsw i32 [[TMP221]], 1
+// CHECK11-NEXT: [[TMP222:%.*]] = zext i32 [[ADD70]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS71:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP203:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP203]], align 4
-// CHECK11-NEXT: [[TMP204:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 1
-// CHECK11-NEXT: store i32 4, ptr [[TMP204]], align 4
-// CHECK11-NEXT: [[TMP205:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP197]], ptr [[TMP205]], align 4
-// CHECK11-NEXT: [[TMP206:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP198]], ptr [[TMP206]], align 4
-// CHECK11-NEXT: [[TMP207:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.20, ptr [[TMP207]], align 4
-// CHECK11-NEXT: [[TMP208:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.21, ptr [[TMP208]], align 4
-// CHECK11-NEXT: [[TMP209:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP209]], align 4
-// CHECK11-NEXT: [[TMP210:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP210]], align 4
-// CHECK11-NEXT: [[TMP211:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP202]], ptr [[TMP211]], align 8
-// CHECK11-NEXT: [[TMP212:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l536.region_id, ptr [[KERNEL_ARGS71]])
-// CHECK11-NEXT: [[TMP213:%.*]] = icmp ne i32 [[TMP212]], 0
-// CHECK11-NEXT: br i1 [[TMP213]], label [[OMP_OFFLOAD_FAILED72:%.*]], label [[OMP_OFFLOAD_CONT73:%.*]]
+// CHECK11-NEXT: [[TMP223:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP223]], align 4
+// CHECK11-NEXT: [[TMP224:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 1
+// CHECK11-NEXT: store i32 4, ptr [[TMP224]], align 4
+// CHECK11-NEXT: [[TMP225:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP217]], ptr [[TMP225]], align 4
+// CHECK11-NEXT: [[TMP226:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP218]], ptr [[TMP226]], align 4
+// CHECK11-NEXT: [[TMP227:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.20, ptr [[TMP227]], align 4
+// CHECK11-NEXT: [[TMP228:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.21, ptr [[TMP228]], align 4
+// CHECK11-NEXT: [[TMP229:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP229]], align 4
+// CHECK11-NEXT: [[TMP230:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP230]], align 4
+// CHECK11-NEXT: [[TMP231:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP222]], ptr [[TMP231]], align 8
+// CHECK11-NEXT: [[TMP232:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP232]], align 8
+// CHECK11-NEXT: [[TMP233:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP233]], align 4
+// CHECK11-NEXT: [[TMP234:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP234]], align 4
+// CHECK11-NEXT: [[TMP235:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP235]], align 4
+// CHECK11-NEXT: [[TMP236:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l536.region_id, ptr [[KERNEL_ARGS71]])
+// CHECK11-NEXT: [[TMP237:%.*]] = icmp ne i32 [[TMP236]], 0
+// CHECK11-NEXT: br i1 [[TMP237]], label [[OMP_OFFLOAD_FAILED72:%.*]], label [[OMP_OFFLOAD_CONT73:%.*]]
// CHECK11: omp_offload.failed72:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l536(i32 [[TMP181]], ptr [[TMP182]], ptr [[TMP183]], ptr [[TMP184]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l536(i32 [[TMP201]], ptr [[TMP202]], ptr [[TMP203]], ptr [[TMP204]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT73]]
// CHECK11: omp_offload.cont73:
-// CHECK11-NEXT: [[TMP214:%.*]] = load i32, ptr [[CH]], align 4
-// CHECK11-NEXT: store i32 [[TMP214]], ptr [[CH_CASTED74]], align 4
-// CHECK11-NEXT: [[TMP215:%.*]] = load i32, ptr [[CH_CASTED74]], align 4
-// CHECK11-NEXT: [[TMP216:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP216]], ptr [[N_CASTED75]], align 4
-// CHECK11-NEXT: [[TMP217:%.*]] = load i32, ptr [[N_CASTED75]], align 4
-// CHECK11-NEXT: [[TMP218:%.*]] = load ptr, ptr [[A]], align 4
-// CHECK11-NEXT: [[TMP219:%.*]] = load ptr, ptr [[B]], align 4
-// CHECK11-NEXT: [[TMP220:%.*]] = load ptr, ptr [[C]], align 4
-// CHECK11-NEXT: [[TMP221:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP215]], ptr [[TMP221]], align 4
-// CHECK11-NEXT: [[TMP222:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP215]], ptr [[TMP222]], align 4
-// CHECK11-NEXT: [[TMP223:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP223]], align 4
-// CHECK11-NEXT: [[TMP224:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP217]], ptr [[TMP224]], align 4
-// CHECK11-NEXT: [[TMP225:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP217]], ptr [[TMP225]], align 4
-// CHECK11-NEXT: [[TMP226:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP226]], align 4
-// CHECK11-NEXT: [[TMP227:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP218]], ptr [[TMP227]], align 4
-// CHECK11-NEXT: [[TMP228:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP218]], ptr [[TMP228]], align 4
-// CHECK11-NEXT: [[TMP229:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP229]], align 4
-// CHECK11-NEXT: [[TMP230:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP219]], ptr [[TMP230]], align 4
-// CHECK11-NEXT: [[TMP231:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP219]], ptr [[TMP231]], align 4
-// CHECK11-NEXT: [[TMP232:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP232]], align 4
-// CHECK11-NEXT: [[TMP233:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP220]], ptr [[TMP233]], align 4
-// CHECK11-NEXT: [[TMP234:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP220]], ptr [[TMP234]], align 4
-// CHECK11-NEXT: [[TMP235:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 4
-// CHECK11-NEXT: store ptr null, ptr [[TMP235]], align 4
-// CHECK11-NEXT: [[TMP236:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP237:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP238:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP238]], ptr [[DOTCAPTURE_EXPR_80]], align 4
-// CHECK11-NEXT: [[TMP239:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_80]], align 4
-// CHECK11-NEXT: [[SUB82:%.*]] = sub nsw i32 [[TMP239]], 0
+// CHECK11-NEXT: [[TMP238:%.*]] = load i32, ptr [[CH]], align 4
+// CHECK11-NEXT: store i32 [[TMP238]], ptr [[CH_CASTED74]], align 4
+// CHECK11-NEXT: [[TMP239:%.*]] = load i32, ptr [[CH_CASTED74]], align 4
+// CHECK11-NEXT: [[TMP240:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP240]], ptr [[N_CASTED75]], align 4
+// CHECK11-NEXT: [[TMP241:%.*]] = load i32, ptr [[N_CASTED75]], align 4
+// CHECK11-NEXT: [[TMP242:%.*]] = load ptr, ptr [[A]], align 4
+// CHECK11-NEXT: [[TMP243:%.*]] = load ptr, ptr [[B]], align 4
+// CHECK11-NEXT: [[TMP244:%.*]] = load ptr, ptr [[C]], align 4
+// CHECK11-NEXT: [[TMP245:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP239]], ptr [[TMP245]], align 4
+// CHECK11-NEXT: [[TMP246:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP239]], ptr [[TMP246]], align 4
+// CHECK11-NEXT: [[TMP247:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP247]], align 4
+// CHECK11-NEXT: [[TMP248:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP241]], ptr [[TMP248]], align 4
+// CHECK11-NEXT: [[TMP249:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP241]], ptr [[TMP249]], align 4
+// CHECK11-NEXT: [[TMP250:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP250]], align 4
+// CHECK11-NEXT: [[TMP251:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP242]], ptr [[TMP251]], align 4
+// CHECK11-NEXT: [[TMP252:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP242]], ptr [[TMP252]], align 4
+// CHECK11-NEXT: [[TMP253:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP253]], align 4
+// CHECK11-NEXT: [[TMP254:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP243]], ptr [[TMP254]], align 4
+// CHECK11-NEXT: [[TMP255:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP243]], ptr [[TMP255]], align 4
+// CHECK11-NEXT: [[TMP256:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP256]], align 4
+// CHECK11-NEXT: [[TMP257:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP244]], ptr [[TMP257]], align 4
+// CHECK11-NEXT: [[TMP258:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP244]], ptr [[TMP258]], align 4
+// CHECK11-NEXT: [[TMP259:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 4
+// CHECK11-NEXT: store ptr null, ptr [[TMP259]], align 4
+// CHECK11-NEXT: [[TMP260:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP261:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP262:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP262]], ptr [[DOTCAPTURE_EXPR_80]], align 4
+// CHECK11-NEXT: [[TMP263:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_80]], align 4
+// CHECK11-NEXT: [[SUB82:%.*]] = sub nsw i32 [[TMP263]], 0
// CHECK11-NEXT: [[DIV83:%.*]] = sdiv i32 [[SUB82]], 1
// CHECK11-NEXT: [[SUB84:%.*]] = sub nsw i32 [[DIV83]], 1
// CHECK11-NEXT: store i32 [[SUB84]], ptr [[DOTCAPTURE_EXPR_81]], align 4
-// CHECK11-NEXT: [[TMP240:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_81]], align 4
-// CHECK11-NEXT: [[ADD85:%.*]] = add nsw i32 [[TMP240]], 1
-// CHECK11-NEXT: [[TMP241:%.*]] = zext i32 [[ADD85]] to i64
+// CHECK11-NEXT: [[TMP264:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_81]], align 4
+// CHECK11-NEXT: [[ADD85:%.*]] = add nsw i32 [[TMP264]], 1
+// CHECK11-NEXT: [[TMP265:%.*]] = zext i32 [[ADD85]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS86:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP242:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP242]], align 4
-// CHECK11-NEXT: [[TMP243:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 1
-// CHECK11-NEXT: store i32 5, ptr [[TMP243]], align 4
-// CHECK11-NEXT: [[TMP244:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP236]], ptr [[TMP244]], align 4
-// CHECK11-NEXT: [[TMP245:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP237]], ptr [[TMP245]], align 4
-// CHECK11-NEXT: [[TMP246:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.24, ptr [[TMP246]], align 4
-// CHECK11-NEXT: [[TMP247:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP247]], align 4
-// CHECK11-NEXT: [[TMP248:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP248]], align 4
-// CHECK11-NEXT: [[TMP249:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP249]], align 4
-// CHECK11-NEXT: [[TMP250:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP241]], ptr [[TMP250]], align 8
-// CHECK11-NEXT: [[TMP251:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l562.region_id, ptr [[KERNEL_ARGS86]])
-// CHECK11-NEXT: [[TMP252:%.*]] = icmp ne i32 [[TMP251]], 0
-// CHECK11-NEXT: br i1 [[TMP252]], label [[OMP_OFFLOAD_FAILED87:%.*]], label [[OMP_OFFLOAD_CONT88:%.*]]
+// CHECK11-NEXT: [[TMP266:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP266]], align 4
+// CHECK11-NEXT: [[TMP267:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 1
+// CHECK11-NEXT: store i32 5, ptr [[TMP267]], align 4
+// CHECK11-NEXT: [[TMP268:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP260]], ptr [[TMP268]], align 4
+// CHECK11-NEXT: [[TMP269:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP261]], ptr [[TMP269]], align 4
+// CHECK11-NEXT: [[TMP270:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.24, ptr [[TMP270]], align 4
+// CHECK11-NEXT: [[TMP271:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP271]], align 4
+// CHECK11-NEXT: [[TMP272:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP272]], align 4
+// CHECK11-NEXT: [[TMP273:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP273]], align 4
+// CHECK11-NEXT: [[TMP274:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP265]], ptr [[TMP274]], align 8
+// CHECK11-NEXT: [[TMP275:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP275]], align 8
+// CHECK11-NEXT: [[TMP276:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP276]], align 4
+// CHECK11-NEXT: [[TMP277:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP277]], align 4
+// CHECK11-NEXT: [[TMP278:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP278]], align 4
+// CHECK11-NEXT: [[TMP279:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l562.region_id, ptr [[KERNEL_ARGS86]])
+// CHECK11-NEXT: [[TMP280:%.*]] = icmp ne i32 [[TMP279]], 0
+// CHECK11-NEXT: br i1 [[TMP280]], label [[OMP_OFFLOAD_FAILED87:%.*]], label [[OMP_OFFLOAD_CONT88:%.*]]
// CHECK11: omp_offload.failed87:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l562(i32 [[TMP215]], i32 [[TMP217]], ptr [[TMP218]], ptr [[TMP219]], ptr [[TMP220]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l562(i32 [[TMP239]], i32 [[TMP241]], ptr [[TMP242]], ptr [[TMP243]], ptr [[TMP244]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT88]]
// CHECK11: omp_offload.cont88:
// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiET_v()
@@ -11081,7 +11249,7 @@ int main() {
// CHECK11-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 4, ptr [[TMP24]], align 4
// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -11098,446 +11266,502 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK11-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l42.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK11-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l42.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK11-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l42(i32 [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK11-NEXT: [[TMP35:%.*]] = load i32, ptr [[N_CASTED3]], align 4
-// CHECK11-NEXT: [[TMP36:%.*]] = load ptr, ptr [[A]], align 4
-// CHECK11-NEXT: [[TMP37:%.*]] = load ptr, ptr [[B]], align 4
-// CHECK11-NEXT: [[TMP38:%.*]] = load ptr, ptr [[C]], align 4
-// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP35]], ptr [[TMP39]], align 4
-// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP35]], ptr [[TMP40]], align 4
-// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP41]], align 4
-// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK11-NEXT: store ptr [[TMP36]], ptr [[TMP42]], align 4
-// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK11-NEXT: store ptr [[TMP36]], ptr [[TMP43]], align 4
-// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP44]], align 4
-// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP37]], ptr [[TMP45]], align 4
-// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP37]], ptr [[TMP46]], align 4
-// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP47]], align 4
-// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP38]], ptr [[TMP48]], align 4
-// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP38]], ptr [[TMP49]], align 4
-// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP50]], align 4
-// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP53:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP53]], ptr [[DOTCAPTURE_EXPR_8]], align 4
-// CHECK11-NEXT: [[TMP54:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_8]], align 4
-// CHECK11-NEXT: [[SUB10:%.*]] = sub nsw i32 [[TMP54]], 0
+// CHECK11-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK11-NEXT: [[TMP39:%.*]] = load i32, ptr [[N_CASTED3]], align 4
+// CHECK11-NEXT: [[TMP40:%.*]] = load ptr, ptr [[A]], align 4
+// CHECK11-NEXT: [[TMP41:%.*]] = load ptr, ptr [[B]], align 4
+// CHECK11-NEXT: [[TMP42:%.*]] = load ptr, ptr [[C]], align 4
+// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP39]], ptr [[TMP43]], align 4
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP39]], ptr [[TMP44]], align 4
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK11-NEXT: store ptr [[TMP40]], ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK11-NEXT: store ptr [[TMP40]], ptr [[TMP47]], align 4
+// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP48]], align 4
+// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP41]], ptr [[TMP49]], align 4
+// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP41]], ptr [[TMP50]], align 4
+// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP42]], ptr [[TMP52]], align 4
+// CHECK11-NEXT: [[TMP53:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP42]], ptr [[TMP53]], align 4
+// CHECK11-NEXT: [[TMP54:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP54]], align 4
+// CHECK11-NEXT: [[TMP55:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP56:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP57:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP57]], ptr [[DOTCAPTURE_EXPR_8]], align 4
+// CHECK11-NEXT: [[TMP58:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_8]], align 4
+// CHECK11-NEXT: [[SUB10:%.*]] = sub nsw i32 [[TMP58]], 0
// CHECK11-NEXT: [[DIV11:%.*]] = sdiv i32 [[SUB10]], 1
// CHECK11-NEXT: [[SUB12:%.*]] = sub nsw i32 [[DIV11]], 1
// CHECK11-NEXT: store i32 [[SUB12]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK11-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK11-NEXT: [[ADD13:%.*]] = add nsw i32 [[TMP55]], 1
-// CHECK11-NEXT: [[TMP56:%.*]] = zext i32 [[ADD13]] to i64
+// CHECK11-NEXT: [[TMP59:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK11-NEXT: [[ADD13:%.*]] = add nsw i32 [[TMP59]], 1
+// CHECK11-NEXT: [[TMP60:%.*]] = zext i32 [[ADD13]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS14:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP57]], align 4
-// CHECK11-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
-// CHECK11-NEXT: store i32 4, ptr [[TMP58]], align 4
-// CHECK11-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP51]], ptr [[TMP59]], align 4
-// CHECK11-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP52]], ptr [[TMP60]], align 4
-// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.32, ptr [[TMP61]], align 4
-// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP62]], align 4
-// CHECK11-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP63]], align 4
-// CHECK11-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP64]], align 4
-// CHECK11-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP56]], ptr [[TMP65]], align 8
-// CHECK11-NEXT: [[TMP66:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l51.region_id, ptr [[KERNEL_ARGS14]])
-// CHECK11-NEXT: [[TMP67:%.*]] = icmp ne i32 [[TMP66]], 0
-// CHECK11-NEXT: br i1 [[TMP67]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
+// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP61]], align 4
+// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
+// CHECK11-NEXT: store i32 4, ptr [[TMP62]], align 4
+// CHECK11-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP55]], ptr [[TMP63]], align 4
+// CHECK11-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP56]], ptr [[TMP64]], align 4
+// CHECK11-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.32, ptr [[TMP65]], align 4
+// CHECK11-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP66]], align 4
+// CHECK11-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP67]], align 4
+// CHECK11-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP68]], align 4
+// CHECK11-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP60]], ptr [[TMP69]], align 8
+// CHECK11-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP70]], align 8
+// CHECK11-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP71]], align 4
+// CHECK11-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP72]], align 4
+// CHECK11-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP73]], align 4
+// CHECK11-NEXT: [[TMP74:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l51.region_id, ptr [[KERNEL_ARGS14]])
+// CHECK11-NEXT: [[TMP75:%.*]] = icmp ne i32 [[TMP74]], 0
+// CHECK11-NEXT: br i1 [[TMP75]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
// CHECK11: omp_offload.failed15:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l51(i32 [[TMP35]], ptr [[TMP36]], ptr [[TMP37]], ptr [[TMP38]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l51(i32 [[TMP39]], ptr [[TMP40]], ptr [[TMP41]], ptr [[TMP42]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT16]]
// CHECK11: omp_offload.cont16:
-// CHECK11-NEXT: [[TMP68:%.*]] = load i32, ptr [[CH]], align 4
-// CHECK11-NEXT: store i32 [[TMP68]], ptr [[CH_CASTED]], align 4
-// CHECK11-NEXT: [[TMP69:%.*]] = load i32, ptr [[CH_CASTED]], align 4
-// CHECK11-NEXT: [[TMP70:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP70]], ptr [[N_CASTED17]], align 4
-// CHECK11-NEXT: [[TMP71:%.*]] = load i32, ptr [[N_CASTED17]], align 4
-// CHECK11-NEXT: [[TMP72:%.*]] = load ptr, ptr [[A]], align 4
-// CHECK11-NEXT: [[TMP73:%.*]] = load ptr, ptr [[B]], align 4
-// CHECK11-NEXT: [[TMP74:%.*]] = load ptr, ptr [[C]], align 4
-// CHECK11-NEXT: [[TMP75:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP69]], ptr [[TMP75]], align 4
-// CHECK11-NEXT: [[TMP76:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP69]], ptr [[TMP76]], align 4
-// CHECK11-NEXT: [[TMP77:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP77]], align 4
-// CHECK11-NEXT: [[TMP78:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP71]], ptr [[TMP78]], align 4
-// CHECK11-NEXT: [[TMP79:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP71]], ptr [[TMP79]], align 4
-// CHECK11-NEXT: [[TMP80:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP80]], align 4
-// CHECK11-NEXT: [[TMP81:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP72]], ptr [[TMP81]], align 4
-// CHECK11-NEXT: [[TMP82:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP72]], ptr [[TMP82]], align 4
-// CHECK11-NEXT: [[TMP83:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP83]], align 4
-// CHECK11-NEXT: [[TMP84:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP73]], ptr [[TMP84]], align 4
-// CHECK11-NEXT: [[TMP85:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP73]], ptr [[TMP85]], align 4
-// CHECK11-NEXT: [[TMP86:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP86]], align 4
-// CHECK11-NEXT: [[TMP87:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP74]], ptr [[TMP87]], align 4
-// CHECK11-NEXT: [[TMP88:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP74]], ptr [[TMP88]], align 4
-// CHECK11-NEXT: [[TMP89:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 4
-// CHECK11-NEXT: store ptr null, ptr [[TMP89]], align 4
-// CHECK11-NEXT: [[TMP90:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP91:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP92:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP92]], ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK11-NEXT: [[TMP93:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK11-NEXT: [[SUB24:%.*]] = sub nsw i32 [[TMP93]], 0
+// CHECK11-NEXT: [[TMP76:%.*]] = load i32, ptr [[CH]], align 4
+// CHECK11-NEXT: store i32 [[TMP76]], ptr [[CH_CASTED]], align 4
+// CHECK11-NEXT: [[TMP77:%.*]] = load i32, ptr [[CH_CASTED]], align 4
+// CHECK11-NEXT: [[TMP78:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP78]], ptr [[N_CASTED17]], align 4
+// CHECK11-NEXT: [[TMP79:%.*]] = load i32, ptr [[N_CASTED17]], align 4
+// CHECK11-NEXT: [[TMP80:%.*]] = load ptr, ptr [[A]], align 4
+// CHECK11-NEXT: [[TMP81:%.*]] = load ptr, ptr [[B]], align 4
+// CHECK11-NEXT: [[TMP82:%.*]] = load ptr, ptr [[C]], align 4
+// CHECK11-NEXT: [[TMP83:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP77]], ptr [[TMP83]], align 4
+// CHECK11-NEXT: [[TMP84:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP77]], ptr [[TMP84]], align 4
+// CHECK11-NEXT: [[TMP85:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP85]], align 4
+// CHECK11-NEXT: [[TMP86:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP79]], ptr [[TMP86]], align 4
+// CHECK11-NEXT: [[TMP87:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP79]], ptr [[TMP87]], align 4
+// CHECK11-NEXT: [[TMP88:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP88]], align 4
+// CHECK11-NEXT: [[TMP89:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP80]], ptr [[TMP89]], align 4
+// CHECK11-NEXT: [[TMP90:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP80]], ptr [[TMP90]], align 4
+// CHECK11-NEXT: [[TMP91:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP91]], align 4
+// CHECK11-NEXT: [[TMP92:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP81]], ptr [[TMP92]], align 4
+// CHECK11-NEXT: [[TMP93:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP81]], ptr [[TMP93]], align 4
+// CHECK11-NEXT: [[TMP94:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP94]], align 4
+// CHECK11-NEXT: [[TMP95:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP82]], ptr [[TMP95]], align 4
+// CHECK11-NEXT: [[TMP96:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP82]], ptr [[TMP96]], align 4
+// CHECK11-NEXT: [[TMP97:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 4
+// CHECK11-NEXT: store ptr null, ptr [[TMP97]], align 4
+// CHECK11-NEXT: [[TMP98:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP99:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP100:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP100]], ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK11-NEXT: [[TMP101:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK11-NEXT: [[SUB24:%.*]] = sub nsw i32 [[TMP101]], 0
// CHECK11-NEXT: [[DIV25:%.*]] = sdiv i32 [[SUB24]], 1
// CHECK11-NEXT: [[SUB26:%.*]] = sub nsw i32 [[DIV25]], 1
// CHECK11-NEXT: store i32 [[SUB26]], ptr [[DOTCAPTURE_EXPR_23]], align 4
-// CHECK11-NEXT: [[TMP94:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_23]], align 4
-// CHECK11-NEXT: [[ADD27:%.*]] = add nsw i32 [[TMP94]], 1
-// CHECK11-NEXT: [[TMP95:%.*]] = zext i32 [[ADD27]] to i64
+// CHECK11-NEXT: [[TMP102:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_23]], align 4
+// CHECK11-NEXT: [[ADD27:%.*]] = add nsw i32 [[TMP102]], 1
+// CHECK11-NEXT: [[TMP103:%.*]] = zext i32 [[ADD27]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS28:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP96]], align 4
-// CHECK11-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
-// CHECK11-NEXT: store i32 5, ptr [[TMP97]], align 4
-// CHECK11-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP90]], ptr [[TMP98]], align 4
-// CHECK11-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP91]], ptr [[TMP99]], align 4
-// CHECK11-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.36, ptr [[TMP100]], align 4
-// CHECK11-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP101]], align 4
-// CHECK11-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP102]], align 4
-// CHECK11-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP103]], align 4
-// CHECK11-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP95]], ptr [[TMP104]], align 8
-// CHECK11-NEXT: [[TMP105:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l59.region_id, ptr [[KERNEL_ARGS28]])
-// CHECK11-NEXT: [[TMP106:%.*]] = icmp ne i32 [[TMP105]], 0
-// CHECK11-NEXT: br i1 [[TMP106]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
+// CHECK11-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP104]], align 4
+// CHECK11-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
+// CHECK11-NEXT: store i32 5, ptr [[TMP105]], align 4
+// CHECK11-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP98]], ptr [[TMP106]], align 4
+// CHECK11-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP99]], ptr [[TMP107]], align 4
+// CHECK11-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.36, ptr [[TMP108]], align 4
+// CHECK11-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP109]], align 4
+// CHECK11-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP110]], align 4
+// CHECK11-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP111]], align 4
+// CHECK11-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP103]], ptr [[TMP112]], align 8
+// CHECK11-NEXT: [[TMP113:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP113]], align 8
+// CHECK11-NEXT: [[TMP114:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP114]], align 4
+// CHECK11-NEXT: [[TMP115:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP115]], align 4
+// CHECK11-NEXT: [[TMP116:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP116]], align 4
+// CHECK11-NEXT: [[TMP117:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l59.region_id, ptr [[KERNEL_ARGS28]])
+// CHECK11-NEXT: [[TMP118:%.*]] = icmp ne i32 [[TMP117]], 0
+// CHECK11-NEXT: br i1 [[TMP118]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
// CHECK11: omp_offload.failed29:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l59(i32 [[TMP69]], i32 [[TMP71]], ptr [[TMP72]], ptr [[TMP73]], ptr [[TMP74]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l59(i32 [[TMP77]], i32 [[TMP79]], ptr [[TMP80]], ptr [[TMP81]], ptr [[TMP82]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT30]]
// CHECK11: omp_offload.cont30:
-// CHECK11-NEXT: [[TMP107:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP107]], ptr [[N_CASTED31]], align 4
-// CHECK11-NEXT: [[TMP108:%.*]] = load i32, ptr [[N_CASTED31]], align 4
-// CHECK11-NEXT: [[TMP109:%.*]] = load ptr, ptr [[A]], align 4
-// CHECK11-NEXT: [[TMP110:%.*]] = load ptr, ptr [[B]], align 4
-// CHECK11-NEXT: [[TMP111:%.*]] = load ptr, ptr [[C]], align 4
-// CHECK11-NEXT: [[TMP112:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP108]], ptr [[TMP112]], align 4
-// CHECK11-NEXT: [[TMP113:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP108]], ptr [[TMP113]], align 4
-// CHECK11-NEXT: [[TMP114:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP114]], align 4
-// CHECK11-NEXT: [[TMP115:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 1
-// CHECK11-NEXT: store ptr [[TMP109]], ptr [[TMP115]], align 4
-// CHECK11-NEXT: [[TMP116:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 1
-// CHECK11-NEXT: store ptr [[TMP109]], ptr [[TMP116]], align 4
-// CHECK11-NEXT: [[TMP117:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP117]], align 4
-// CHECK11-NEXT: [[TMP118:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP110]], ptr [[TMP118]], align 4
-// CHECK11-NEXT: [[TMP119:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP110]], ptr [[TMP119]], align 4
-// CHECK11-NEXT: [[TMP120:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP120]], align 4
-// CHECK11-NEXT: [[TMP121:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP111]], ptr [[TMP121]], align 4
-// CHECK11-NEXT: [[TMP122:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP111]], ptr [[TMP122]], align 4
-// CHECK11-NEXT: [[TMP123:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP123]], align 4
+// CHECK11-NEXT: [[TMP119:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP119]], ptr [[N_CASTED31]], align 4
+// CHECK11-NEXT: [[TMP120:%.*]] = load i32, ptr [[N_CASTED31]], align 4
+// CHECK11-NEXT: [[TMP121:%.*]] = load ptr, ptr [[A]], align 4
+// CHECK11-NEXT: [[TMP122:%.*]] = load ptr, ptr [[B]], align 4
+// CHECK11-NEXT: [[TMP123:%.*]] = load ptr, ptr [[C]], align 4
// CHECK11-NEXT: [[TMP124:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP120]], ptr [[TMP124]], align 4
// CHECK11-NEXT: [[TMP125:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP126:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP126]], ptr [[DOTCAPTURE_EXPR_36]], align 4
-// CHECK11-NEXT: [[TMP127:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_36]], align 4
-// CHECK11-NEXT: [[SUB38:%.*]] = sub nsw i32 [[TMP127]], 0
+// CHECK11-NEXT: store i32 [[TMP120]], ptr [[TMP125]], align 4
+// CHECK11-NEXT: [[TMP126:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP126]], align 4
+// CHECK11-NEXT: [[TMP127:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 1
+// CHECK11-NEXT: store ptr [[TMP121]], ptr [[TMP127]], align 4
+// CHECK11-NEXT: [[TMP128:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 1
+// CHECK11-NEXT: store ptr [[TMP121]], ptr [[TMP128]], align 4
+// CHECK11-NEXT: [[TMP129:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP129]], align 4
+// CHECK11-NEXT: [[TMP130:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP122]], ptr [[TMP130]], align 4
+// CHECK11-NEXT: [[TMP131:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP122]], ptr [[TMP131]], align 4
+// CHECK11-NEXT: [[TMP132:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP132]], align 4
+// CHECK11-NEXT: [[TMP133:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP123]], ptr [[TMP133]], align 4
+// CHECK11-NEXT: [[TMP134:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP123]], ptr [[TMP134]], align 4
+// CHECK11-NEXT: [[TMP135:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP135]], align 4
+// CHECK11-NEXT: [[TMP136:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP137:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP138:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP138]], ptr [[DOTCAPTURE_EXPR_36]], align 4
+// CHECK11-NEXT: [[TMP139:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_36]], align 4
+// CHECK11-NEXT: [[SUB38:%.*]] = sub nsw i32 [[TMP139]], 0
// CHECK11-NEXT: [[DIV39:%.*]] = sdiv i32 [[SUB38]], 1
// CHECK11-NEXT: [[SUB40:%.*]] = sub nsw i32 [[DIV39]], 1
// CHECK11-NEXT: store i32 [[SUB40]], ptr [[DOTCAPTURE_EXPR_37]], align 4
-// CHECK11-NEXT: [[TMP128:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_37]], align 4
-// CHECK11-NEXT: [[ADD41:%.*]] = add nsw i32 [[TMP128]], 1
-// CHECK11-NEXT: [[TMP129:%.*]] = zext i32 [[ADD41]] to i64
+// CHECK11-NEXT: [[TMP140:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_37]], align 4
+// CHECK11-NEXT: [[ADD41:%.*]] = add nsw i32 [[TMP140]], 1
+// CHECK11-NEXT: [[TMP141:%.*]] = zext i32 [[ADD41]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS42:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP130]], align 4
-// CHECK11-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 1
-// CHECK11-NEXT: store i32 4, ptr [[TMP131]], align 4
-// CHECK11-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP124]], ptr [[TMP132]], align 4
-// CHECK11-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP125]], ptr [[TMP133]], align 4
-// CHECK11-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.40, ptr [[TMP134]], align 4
-// CHECK11-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.41, ptr [[TMP135]], align 4
-// CHECK11-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP136]], align 4
-// CHECK11-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP137]], align 4
-// CHECK11-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP129]], ptr [[TMP138]], align 8
-// CHECK11-NEXT: [[TMP139:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l67.region_id, ptr [[KERNEL_ARGS42]])
-// CHECK11-NEXT: [[TMP140:%.*]] = icmp ne i32 [[TMP139]], 0
-// CHECK11-NEXT: br i1 [[TMP140]], label [[OMP_OFFLOAD_FAILED43:%.*]], label [[OMP_OFFLOAD_CONT44:%.*]]
+// CHECK11-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP142]], align 4
+// CHECK11-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 1
+// CHECK11-NEXT: store i32 4, ptr [[TMP143]], align 4
+// CHECK11-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP136]], ptr [[TMP144]], align 4
+// CHECK11-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP137]], ptr [[TMP145]], align 4
+// CHECK11-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.40, ptr [[TMP146]], align 4
+// CHECK11-NEXT: [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.41, ptr [[TMP147]], align 4
+// CHECK11-NEXT: [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP148]], align 4
+// CHECK11-NEXT: [[TMP149:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP149]], align 4
+// CHECK11-NEXT: [[TMP150:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP141]], ptr [[TMP150]], align 8
+// CHECK11-NEXT: [[TMP151:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP151]], align 8
+// CHECK11-NEXT: [[TMP152:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP152]], align 4
+// CHECK11-NEXT: [[TMP153:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP153]], align 4
+// CHECK11-NEXT: [[TMP154:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP154]], align 4
+// CHECK11-NEXT: [[TMP155:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l67.region_id, ptr [[KERNEL_ARGS42]])
+// CHECK11-NEXT: [[TMP156:%.*]] = icmp ne i32 [[TMP155]], 0
+// CHECK11-NEXT: br i1 [[TMP156]], label [[OMP_OFFLOAD_FAILED43:%.*]], label [[OMP_OFFLOAD_CONT44:%.*]]
// CHECK11: omp_offload.failed43:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l67(i32 [[TMP108]], ptr [[TMP109]], ptr [[TMP110]], ptr [[TMP111]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l67(i32 [[TMP120]], ptr [[TMP121]], ptr [[TMP122]], ptr [[TMP123]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT44]]
// CHECK11: omp_offload.cont44:
-// CHECK11-NEXT: [[TMP141:%.*]] = load i32, ptr [[CH]], align 4
-// CHECK11-NEXT: store i32 [[TMP141]], ptr [[CH_CASTED45]], align 4
-// CHECK11-NEXT: [[TMP142:%.*]] = load i32, ptr [[CH_CASTED45]], align 4
-// CHECK11-NEXT: [[TMP143:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP143]], ptr [[N_CASTED46]], align 4
-// CHECK11-NEXT: [[TMP144:%.*]] = load i32, ptr [[N_CASTED46]], align 4
-// CHECK11-NEXT: [[TMP145:%.*]] = load ptr, ptr [[A]], align 4
-// CHECK11-NEXT: [[TMP146:%.*]] = load ptr, ptr [[B]], align 4
-// CHECK11-NEXT: [[TMP147:%.*]] = load ptr, ptr [[C]], align 4
-// CHECK11-NEXT: [[TMP148:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP142]], ptr [[TMP148]], align 4
-// CHECK11-NEXT: [[TMP149:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP142]], ptr [[TMP149]], align 4
-// CHECK11-NEXT: [[TMP150:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP150]], align 4
-// CHECK11-NEXT: [[TMP151:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP144]], ptr [[TMP151]], align 4
-// CHECK11-NEXT: [[TMP152:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP144]], ptr [[TMP152]], align 4
-// CHECK11-NEXT: [[TMP153:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP153]], align 4
-// CHECK11-NEXT: [[TMP154:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP145]], ptr [[TMP154]], align 4
-// CHECK11-NEXT: [[TMP155:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP145]], ptr [[TMP155]], align 4
-// CHECK11-NEXT: [[TMP156:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP156]], align 4
-// CHECK11-NEXT: [[TMP157:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP146]], ptr [[TMP157]], align 4
-// CHECK11-NEXT: [[TMP158:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP146]], ptr [[TMP158]], align 4
-// CHECK11-NEXT: [[TMP159:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP159]], align 4
-// CHECK11-NEXT: [[TMP160:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP147]], ptr [[TMP160]], align 4
-// CHECK11-NEXT: [[TMP161:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP147]], ptr [[TMP161]], align 4
-// CHECK11-NEXT: [[TMP162:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 4
-// CHECK11-NEXT: store ptr null, ptr [[TMP162]], align 4
-// CHECK11-NEXT: [[TMP163:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP164:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP165:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP165]], ptr [[DOTCAPTURE_EXPR_51]], align 4
-// CHECK11-NEXT: [[TMP166:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_51]], align 4
-// CHECK11-NEXT: [[SUB53:%.*]] = sub nsw i32 [[TMP166]], 0
+// CHECK11-NEXT: [[TMP157:%.*]] = load i32, ptr [[CH]], align 4
+// CHECK11-NEXT: store i32 [[TMP157]], ptr [[CH_CASTED45]], align 4
+// CHECK11-NEXT: [[TMP158:%.*]] = load i32, ptr [[CH_CASTED45]], align 4
+// CHECK11-NEXT: [[TMP159:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP159]], ptr [[N_CASTED46]], align 4
+// CHECK11-NEXT: [[TMP160:%.*]] = load i32, ptr [[N_CASTED46]], align 4
+// CHECK11-NEXT: [[TMP161:%.*]] = load ptr, ptr [[A]], align 4
+// CHECK11-NEXT: [[TMP162:%.*]] = load ptr, ptr [[B]], align 4
+// CHECK11-NEXT: [[TMP163:%.*]] = load ptr, ptr [[C]], align 4
+// CHECK11-NEXT: [[TMP164:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP158]], ptr [[TMP164]], align 4
+// CHECK11-NEXT: [[TMP165:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP158]], ptr [[TMP165]], align 4
+// CHECK11-NEXT: [[TMP166:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP166]], align 4
+// CHECK11-NEXT: [[TMP167:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP160]], ptr [[TMP167]], align 4
+// CHECK11-NEXT: [[TMP168:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP160]], ptr [[TMP168]], align 4
+// CHECK11-NEXT: [[TMP169:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP169]], align 4
+// CHECK11-NEXT: [[TMP170:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP161]], ptr [[TMP170]], align 4
+// CHECK11-NEXT: [[TMP171:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP161]], ptr [[TMP171]], align 4
+// CHECK11-NEXT: [[TMP172:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP172]], align 4
+// CHECK11-NEXT: [[TMP173:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP162]], ptr [[TMP173]], align 4
+// CHECK11-NEXT: [[TMP174:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP162]], ptr [[TMP174]], align 4
+// CHECK11-NEXT: [[TMP175:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP175]], align 4
+// CHECK11-NEXT: [[TMP176:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP163]], ptr [[TMP176]], align 4
+// CHECK11-NEXT: [[TMP177:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP163]], ptr [[TMP177]], align 4
+// CHECK11-NEXT: [[TMP178:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 4
+// CHECK11-NEXT: store ptr null, ptr [[TMP178]], align 4
+// CHECK11-NEXT: [[TMP179:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP180:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP181:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP181]], ptr [[DOTCAPTURE_EXPR_51]], align 4
+// CHECK11-NEXT: [[TMP182:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_51]], align 4
+// CHECK11-NEXT: [[SUB53:%.*]] = sub nsw i32 [[TMP182]], 0
// CHECK11-NEXT: [[DIV54:%.*]] = sdiv i32 [[SUB53]], 1
// CHECK11-NEXT: [[SUB55:%.*]] = sub nsw i32 [[DIV54]], 1
// CHECK11-NEXT: store i32 [[SUB55]], ptr [[DOTCAPTURE_EXPR_52]], align 4
-// CHECK11-NEXT: [[TMP167:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_52]], align 4
-// CHECK11-NEXT: [[ADD56:%.*]] = add nsw i32 [[TMP167]], 1
-// CHECK11-NEXT: [[TMP168:%.*]] = zext i32 [[ADD56]] to i64
+// CHECK11-NEXT: [[TMP183:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_52]], align 4
+// CHECK11-NEXT: [[ADD56:%.*]] = add nsw i32 [[TMP183]], 1
+// CHECK11-NEXT: [[TMP184:%.*]] = zext i32 [[ADD56]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS57:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP169:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP169]], align 4
-// CHECK11-NEXT: [[TMP170:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 1
-// CHECK11-NEXT: store i32 5, ptr [[TMP170]], align 4
-// CHECK11-NEXT: [[TMP171:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP163]], ptr [[TMP171]], align 4
-// CHECK11-NEXT: [[TMP172:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP164]], ptr [[TMP172]], align 4
-// CHECK11-NEXT: [[TMP173:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.44, ptr [[TMP173]], align 4
-// CHECK11-NEXT: [[TMP174:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.45, ptr [[TMP174]], align 4
-// CHECK11-NEXT: [[TMP175:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP175]], align 4
-// CHECK11-NEXT: [[TMP176:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP176]], align 4
-// CHECK11-NEXT: [[TMP177:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP168]], ptr [[TMP177]], align 8
-// CHECK11-NEXT: [[TMP178:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l75.region_id, ptr [[KERNEL_ARGS57]])
-// CHECK11-NEXT: [[TMP179:%.*]] = icmp ne i32 [[TMP178]], 0
-// CHECK11-NEXT: br i1 [[TMP179]], label [[OMP_OFFLOAD_FAILED58:%.*]], label [[OMP_OFFLOAD_CONT59:%.*]]
+// CHECK11-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP185]], align 4
+// CHECK11-NEXT: [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 1
+// CHECK11-NEXT: store i32 5, ptr [[TMP186]], align 4
+// CHECK11-NEXT: [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP179]], ptr [[TMP187]], align 4
+// CHECK11-NEXT: [[TMP188:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP180]], ptr [[TMP188]], align 4
+// CHECK11-NEXT: [[TMP189:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.44, ptr [[TMP189]], align 4
+// CHECK11-NEXT: [[TMP190:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.45, ptr [[TMP190]], align 4
+// CHECK11-NEXT: [[TMP191:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP191]], align 4
+// CHECK11-NEXT: [[TMP192:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP192]], align 4
+// CHECK11-NEXT: [[TMP193:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP184]], ptr [[TMP193]], align 8
+// CHECK11-NEXT: [[TMP194:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP194]], align 8
+// CHECK11-NEXT: [[TMP195:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP195]], align 4
+// CHECK11-NEXT: [[TMP196:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP196]], align 4
+// CHECK11-NEXT: [[TMP197:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP197]], align 4
+// CHECK11-NEXT: [[TMP198:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l75.region_id, ptr [[KERNEL_ARGS57]])
+// CHECK11-NEXT: [[TMP199:%.*]] = icmp ne i32 [[TMP198]], 0
+// CHECK11-NEXT: br i1 [[TMP199]], label [[OMP_OFFLOAD_FAILED58:%.*]], label [[OMP_OFFLOAD_CONT59:%.*]]
// CHECK11: omp_offload.failed58:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l75(i32 [[TMP142]], i32 [[TMP144]], ptr [[TMP145]], ptr [[TMP146]], ptr [[TMP147]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l75(i32 [[TMP158]], i32 [[TMP160]], ptr [[TMP161]], ptr [[TMP162]], ptr [[TMP163]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT59]]
// CHECK11: omp_offload.cont59:
-// CHECK11-NEXT: [[TMP180:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP180]], ptr [[N_CASTED60]], align 4
-// CHECK11-NEXT: [[TMP181:%.*]] = load i32, ptr [[N_CASTED60]], align 4
-// CHECK11-NEXT: [[TMP182:%.*]] = load ptr, ptr [[A]], align 4
-// CHECK11-NEXT: [[TMP183:%.*]] = load ptr, ptr [[B]], align 4
-// CHECK11-NEXT: [[TMP184:%.*]] = load ptr, ptr [[C]], align 4
-// CHECK11-NEXT: [[TMP185:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP181]], ptr [[TMP185]], align 4
-// CHECK11-NEXT: [[TMP186:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP181]], ptr [[TMP186]], align 4
-// CHECK11-NEXT: [[TMP187:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP187]], align 4
-// CHECK11-NEXT: [[TMP188:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 1
-// CHECK11-NEXT: store ptr [[TMP182]], ptr [[TMP188]], align 4
-// CHECK11-NEXT: [[TMP189:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 1
-// CHECK11-NEXT: store ptr [[TMP182]], ptr [[TMP189]], align 4
-// CHECK11-NEXT: [[TMP190:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP190]], align 4
-// CHECK11-NEXT: [[TMP191:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP183]], ptr [[TMP191]], align 4
-// CHECK11-NEXT: [[TMP192:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP183]], ptr [[TMP192]], align 4
-// CHECK11-NEXT: [[TMP193:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP193]], align 4
-// CHECK11-NEXT: [[TMP194:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP184]], ptr [[TMP194]], align 4
-// CHECK11-NEXT: [[TMP195:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP184]], ptr [[TMP195]], align 4
-// CHECK11-NEXT: [[TMP196:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP196]], align 4
-// CHECK11-NEXT: [[TMP197:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP198:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP199:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP199]], ptr [[DOTCAPTURE_EXPR_65]], align 4
-// CHECK11-NEXT: [[TMP200:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_65]], align 4
-// CHECK11-NEXT: [[SUB67:%.*]] = sub nsw i32 [[TMP200]], 0
+// CHECK11-NEXT: [[TMP200:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP200]], ptr [[N_CASTED60]], align 4
+// CHECK11-NEXT: [[TMP201:%.*]] = load i32, ptr [[N_CASTED60]], align 4
+// CHECK11-NEXT: [[TMP202:%.*]] = load ptr, ptr [[A]], align 4
+// CHECK11-NEXT: [[TMP203:%.*]] = load ptr, ptr [[B]], align 4
+// CHECK11-NEXT: [[TMP204:%.*]] = load ptr, ptr [[C]], align 4
+// CHECK11-NEXT: [[TMP205:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP201]], ptr [[TMP205]], align 4
+// CHECK11-NEXT: [[TMP206:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP201]], ptr [[TMP206]], align 4
+// CHECK11-NEXT: [[TMP207:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP207]], align 4
+// CHECK11-NEXT: [[TMP208:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 1
+// CHECK11-NEXT: store ptr [[TMP202]], ptr [[TMP208]], align 4
+// CHECK11-NEXT: [[TMP209:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 1
+// CHECK11-NEXT: store ptr [[TMP202]], ptr [[TMP209]], align 4
+// CHECK11-NEXT: [[TMP210:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP210]], align 4
+// CHECK11-NEXT: [[TMP211:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP203]], ptr [[TMP211]], align 4
+// CHECK11-NEXT: [[TMP212:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP203]], ptr [[TMP212]], align 4
+// CHECK11-NEXT: [[TMP213:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP213]], align 4
+// CHECK11-NEXT: [[TMP214:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP204]], ptr [[TMP214]], align 4
+// CHECK11-NEXT: [[TMP215:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP204]], ptr [[TMP215]], align 4
+// CHECK11-NEXT: [[TMP216:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP216]], align 4
+// CHECK11-NEXT: [[TMP217:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP218:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP219:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP219]], ptr [[DOTCAPTURE_EXPR_65]], align 4
+// CHECK11-NEXT: [[TMP220:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_65]], align 4
+// CHECK11-NEXT: [[SUB67:%.*]] = sub nsw i32 [[TMP220]], 0
// CHECK11-NEXT: [[DIV68:%.*]] = sdiv i32 [[SUB67]], 1
// CHECK11-NEXT: [[SUB69:%.*]] = sub nsw i32 [[DIV68]], 1
// CHECK11-NEXT: store i32 [[SUB69]], ptr [[DOTCAPTURE_EXPR_66]], align 4
-// CHECK11-NEXT: [[TMP201:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_66]], align 4
-// CHECK11-NEXT: [[ADD70:%.*]] = add nsw i32 [[TMP201]], 1
-// CHECK11-NEXT: [[TMP202:%.*]] = zext i32 [[ADD70]] to i64
+// CHECK11-NEXT: [[TMP221:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_66]], align 4
+// CHECK11-NEXT: [[ADD70:%.*]] = add nsw i32 [[TMP221]], 1
+// CHECK11-NEXT: [[TMP222:%.*]] = zext i32 [[ADD70]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS71:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP203:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP203]], align 4
-// CHECK11-NEXT: [[TMP204:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 1
-// CHECK11-NEXT: store i32 4, ptr [[TMP204]], align 4
-// CHECK11-NEXT: [[TMP205:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP197]], ptr [[TMP205]], align 4
-// CHECK11-NEXT: [[TMP206:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP198]], ptr [[TMP206]], align 4
-// CHECK11-NEXT: [[TMP207:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.48, ptr [[TMP207]], align 4
-// CHECK11-NEXT: [[TMP208:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.49, ptr [[TMP208]], align 4
-// CHECK11-NEXT: [[TMP209:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP209]], align 4
-// CHECK11-NEXT: [[TMP210:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP210]], align 4
-// CHECK11-NEXT: [[TMP211:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP202]], ptr [[TMP211]], align 8
-// CHECK11-NEXT: [[TMP212:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l83.region_id, ptr [[KERNEL_ARGS71]])
-// CHECK11-NEXT: [[TMP213:%.*]] = icmp ne i32 [[TMP212]], 0
-// CHECK11-NEXT: br i1 [[TMP213]], label [[OMP_OFFLOAD_FAILED72:%.*]], label [[OMP_OFFLOAD_CONT73:%.*]]
+// CHECK11-NEXT: [[TMP223:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP223]], align 4
+// CHECK11-NEXT: [[TMP224:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 1
+// CHECK11-NEXT: store i32 4, ptr [[TMP224]], align 4
+// CHECK11-NEXT: [[TMP225:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP217]], ptr [[TMP225]], align 4
+// CHECK11-NEXT: [[TMP226:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP218]], ptr [[TMP226]], align 4
+// CHECK11-NEXT: [[TMP227:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.48, ptr [[TMP227]], align 4
+// CHECK11-NEXT: [[TMP228:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.49, ptr [[TMP228]], align 4
+// CHECK11-NEXT: [[TMP229:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP229]], align 4
+// CHECK11-NEXT: [[TMP230:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP230]], align 4
+// CHECK11-NEXT: [[TMP231:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP222]], ptr [[TMP231]], align 8
+// CHECK11-NEXT: [[TMP232:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP232]], align 8
+// CHECK11-NEXT: [[TMP233:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP233]], align 4
+// CHECK11-NEXT: [[TMP234:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP234]], align 4
+// CHECK11-NEXT: [[TMP235:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP235]], align 4
+// CHECK11-NEXT: [[TMP236:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l83.region_id, ptr [[KERNEL_ARGS71]])
+// CHECK11-NEXT: [[TMP237:%.*]] = icmp ne i32 [[TMP236]], 0
+// CHECK11-NEXT: br i1 [[TMP237]], label [[OMP_OFFLOAD_FAILED72:%.*]], label [[OMP_OFFLOAD_CONT73:%.*]]
// CHECK11: omp_offload.failed72:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l83(i32 [[TMP181]], ptr [[TMP182]], ptr [[TMP183]], ptr [[TMP184]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l83(i32 [[TMP201]], ptr [[TMP202]], ptr [[TMP203]], ptr [[TMP204]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT73]]
// CHECK11: omp_offload.cont73:
-// CHECK11-NEXT: [[TMP214:%.*]] = load i32, ptr [[CH]], align 4
-// CHECK11-NEXT: store i32 [[TMP214]], ptr [[CH_CASTED74]], align 4
-// CHECK11-NEXT: [[TMP215:%.*]] = load i32, ptr [[CH_CASTED74]], align 4
-// CHECK11-NEXT: [[TMP216:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP216]], ptr [[N_CASTED75]], align 4
-// CHECK11-NEXT: [[TMP217:%.*]] = load i32, ptr [[N_CASTED75]], align 4
-// CHECK11-NEXT: [[TMP218:%.*]] = load ptr, ptr [[A]], align 4
-// CHECK11-NEXT: [[TMP219:%.*]] = load ptr, ptr [[B]], align 4
-// CHECK11-NEXT: [[TMP220:%.*]] = load ptr, ptr [[C]], align 4
-// CHECK11-NEXT: [[TMP221:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP215]], ptr [[TMP221]], align 4
-// CHECK11-NEXT: [[TMP222:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP215]], ptr [[TMP222]], align 4
-// CHECK11-NEXT: [[TMP223:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP223]], align 4
-// CHECK11-NEXT: [[TMP224:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP217]], ptr [[TMP224]], align 4
-// CHECK11-NEXT: [[TMP225:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP217]], ptr [[TMP225]], align 4
-// CHECK11-NEXT: [[TMP226:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP226]], align 4
-// CHECK11-NEXT: [[TMP227:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP218]], ptr [[TMP227]], align 4
-// CHECK11-NEXT: [[TMP228:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP218]], ptr [[TMP228]], align 4
-// CHECK11-NEXT: [[TMP229:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP229]], align 4
-// CHECK11-NEXT: [[TMP230:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP219]], ptr [[TMP230]], align 4
-// CHECK11-NEXT: [[TMP231:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP219]], ptr [[TMP231]], align 4
-// CHECK11-NEXT: [[TMP232:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP232]], align 4
-// CHECK11-NEXT: [[TMP233:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP220]], ptr [[TMP233]], align 4
-// CHECK11-NEXT: [[TMP234:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP220]], ptr [[TMP234]], align 4
-// CHECK11-NEXT: [[TMP235:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 4
-// CHECK11-NEXT: store ptr null, ptr [[TMP235]], align 4
-// CHECK11-NEXT: [[TMP236:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP237:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP238:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP238]], ptr [[DOTCAPTURE_EXPR_80]], align 4
-// CHECK11-NEXT: [[TMP239:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_80]], align 4
-// CHECK11-NEXT: [[SUB82:%.*]] = sub nsw i32 [[TMP239]], 0
+// CHECK11-NEXT: [[TMP238:%.*]] = load i32, ptr [[CH]], align 4
+// CHECK11-NEXT: store i32 [[TMP238]], ptr [[CH_CASTED74]], align 4
+// CHECK11-NEXT: [[TMP239:%.*]] = load i32, ptr [[CH_CASTED74]], align 4
+// CHECK11-NEXT: [[TMP240:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP240]], ptr [[N_CASTED75]], align 4
+// CHECK11-NEXT: [[TMP241:%.*]] = load i32, ptr [[N_CASTED75]], align 4
+// CHECK11-NEXT: [[TMP242:%.*]] = load ptr, ptr [[A]], align 4
+// CHECK11-NEXT: [[TMP243:%.*]] = load ptr, ptr [[B]], align 4
+// CHECK11-NEXT: [[TMP244:%.*]] = load ptr, ptr [[C]], align 4
+// CHECK11-NEXT: [[TMP245:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP239]], ptr [[TMP245]], align 4
+// CHECK11-NEXT: [[TMP246:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP239]], ptr [[TMP246]], align 4
+// CHECK11-NEXT: [[TMP247:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP247]], align 4
+// CHECK11-NEXT: [[TMP248:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP241]], ptr [[TMP248]], align 4
+// CHECK11-NEXT: [[TMP249:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP241]], ptr [[TMP249]], align 4
+// CHECK11-NEXT: [[TMP250:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP250]], align 4
+// CHECK11-NEXT: [[TMP251:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP242]], ptr [[TMP251]], align 4
+// CHECK11-NEXT: [[TMP252:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP242]], ptr [[TMP252]], align 4
+// CHECK11-NEXT: [[TMP253:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP253]], align 4
+// CHECK11-NEXT: [[TMP254:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP243]], ptr [[TMP254]], align 4
+// CHECK11-NEXT: [[TMP255:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP243]], ptr [[TMP255]], align 4
+// CHECK11-NEXT: [[TMP256:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP256]], align 4
+// CHECK11-NEXT: [[TMP257:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP244]], ptr [[TMP257]], align 4
+// CHECK11-NEXT: [[TMP258:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP244]], ptr [[TMP258]], align 4
+// CHECK11-NEXT: [[TMP259:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 4
+// CHECK11-NEXT: store ptr null, ptr [[TMP259]], align 4
+// CHECK11-NEXT: [[TMP260:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP261:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP262:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP262]], ptr [[DOTCAPTURE_EXPR_80]], align 4
+// CHECK11-NEXT: [[TMP263:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_80]], align 4
+// CHECK11-NEXT: [[SUB82:%.*]] = sub nsw i32 [[TMP263]], 0
// CHECK11-NEXT: [[DIV83:%.*]] = sdiv i32 [[SUB82]], 1
// CHECK11-NEXT: [[SUB84:%.*]] = sub nsw i32 [[DIV83]], 1
// CHECK11-NEXT: store i32 [[SUB84]], ptr [[DOTCAPTURE_EXPR_81]], align 4
-// CHECK11-NEXT: [[TMP240:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_81]], align 4
-// CHECK11-NEXT: [[ADD85:%.*]] = add nsw i32 [[TMP240]], 1
-// CHECK11-NEXT: [[TMP241:%.*]] = zext i32 [[ADD85]] to i64
+// CHECK11-NEXT: [[TMP264:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_81]], align 4
+// CHECK11-NEXT: [[ADD85:%.*]] = add nsw i32 [[TMP264]], 1
+// CHECK11-NEXT: [[TMP265:%.*]] = zext i32 [[ADD85]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS86:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP242:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP242]], align 4
-// CHECK11-NEXT: [[TMP243:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 1
-// CHECK11-NEXT: store i32 5, ptr [[TMP243]], align 4
-// CHECK11-NEXT: [[TMP244:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP236]], ptr [[TMP244]], align 4
-// CHECK11-NEXT: [[TMP245:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP237]], ptr [[TMP245]], align 4
-// CHECK11-NEXT: [[TMP246:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.52, ptr [[TMP246]], align 4
-// CHECK11-NEXT: [[TMP247:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.53, ptr [[TMP247]], align 4
-// CHECK11-NEXT: [[TMP248:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP248]], align 4
-// CHECK11-NEXT: [[TMP249:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP249]], align 4
-// CHECK11-NEXT: [[TMP250:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP241]], ptr [[TMP250]], align 8
-// CHECK11-NEXT: [[TMP251:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l91.region_id, ptr [[KERNEL_ARGS86]])
-// CHECK11-NEXT: [[TMP252:%.*]] = icmp ne i32 [[TMP251]], 0
-// CHECK11-NEXT: br i1 [[TMP252]], label [[OMP_OFFLOAD_FAILED87:%.*]], label [[OMP_OFFLOAD_CONT88:%.*]]
+// CHECK11-NEXT: [[TMP266:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP266]], align 4
+// CHECK11-NEXT: [[TMP267:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 1
+// CHECK11-NEXT: store i32 5, ptr [[TMP267]], align 4
+// CHECK11-NEXT: [[TMP268:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP260]], ptr [[TMP268]], align 4
+// CHECK11-NEXT: [[TMP269:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP261]], ptr [[TMP269]], align 4
+// CHECK11-NEXT: [[TMP270:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.52, ptr [[TMP270]], align 4
+// CHECK11-NEXT: [[TMP271:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.53, ptr [[TMP271]], align 4
+// CHECK11-NEXT: [[TMP272:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP272]], align 4
+// CHECK11-NEXT: [[TMP273:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP273]], align 4
+// CHECK11-NEXT: [[TMP274:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP265]], ptr [[TMP274]], align 8
+// CHECK11-NEXT: [[TMP275:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP275]], align 8
+// CHECK11-NEXT: [[TMP276:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP276]], align 4
+// CHECK11-NEXT: [[TMP277:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP277]], align 4
+// CHECK11-NEXT: [[TMP278:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP278]], align 4
+// CHECK11-NEXT: [[TMP279:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l91.region_id, ptr [[KERNEL_ARGS86]])
+// CHECK11-NEXT: [[TMP280:%.*]] = icmp ne i32 [[TMP279]], 0
+// CHECK11-NEXT: br i1 [[TMP280]], label [[OMP_OFFLOAD_FAILED87:%.*]], label [[OMP_OFFLOAD_CONT88:%.*]]
// CHECK11: omp_offload.failed87:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l91(i32 [[TMP215]], i32 [[TMP217]], ptr [[TMP218]], ptr [[TMP219]], ptr [[TMP220]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l91(i32 [[TMP239]], i32 [[TMP241]], ptr [[TMP242]], ptr [[TMP243]], ptr [[TMP244]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT88]]
// CHECK11: omp_offload.cont88:
// CHECK11-NEXT: ret i32 0
diff --git a/clang/test/OpenMP/distribute_parallel_for_firstprivate_codegen.cpp b/clang/test/OpenMP/distribute_parallel_for_firstprivate_codegen.cpp
index 70558140985d0..dfeb1898823f5 100644
--- a/clang/test/OpenMP/distribute_parallel_for_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/distribute_parallel_for_firstprivate_codegen.cpp
@@ -839,7 +839,7 @@ int main() {
// CHECK8-NEXT: [[TMP24:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK8-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK8-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK8-NEXT: store i32 1, ptr [[TMP25]], align 4
+// CHECK8-NEXT: store i32 2, ptr [[TMP25]], align 4
// CHECK8-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK8-NEXT: store i32 5, ptr [[TMP26]], align 4
// CHECK8-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -856,9 +856,17 @@ int main() {
// CHECK8-NEXT: store ptr null, ptr [[TMP32]], align 8
// CHECK8-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK8-NEXT: store i64 2, ptr [[TMP33]], align 8
-// CHECK8-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.region_id, ptr [[KERNEL_ARGS]])
-// CHECK8-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
-// CHECK8-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK8-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK8-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK8-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK8-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK8-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK8-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK8-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK8-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK8-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.region_id, ptr [[KERNEL_ARGS]])
+// CHECK8-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK8-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK8: omp_offload.failed:
// CHECK8-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139(i64 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]], i64 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK8-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -866,18 +874,18 @@ int main() {
// CHECK8-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiET_v()
// CHECK8-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK8-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK8-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK8-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK8-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK8: arraydestroy.body:
-// CHECK8-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP36]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK8-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP40]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK8-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK8-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK8-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK8-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK8: arraydestroy.done2:
// CHECK8-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK8-NEXT: [[TMP37:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK8-NEXT: ret i32 [[TMP37]]
+// CHECK8-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK8-NEXT: ret i32 [[TMP41]]
//
//
// CHECK8-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -1257,7 +1265,7 @@ int main() {
// CHECK8-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK8-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK8-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK8-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK8-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK8-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK8-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK8-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1274,27 +1282,35 @@ int main() {
// CHECK8-NEXT: store ptr null, ptr [[TMP27]], align 8
// CHECK8-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK8-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK8-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l48.region_id, ptr [[KERNEL_ARGS]])
-// CHECK8-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK8-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK8-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK8-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK8-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK8-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK8-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK8-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK8-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK8-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK8-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l48.region_id, ptr [[KERNEL_ARGS]])
+// CHECK8-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK8-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK8: omp_offload.failed:
// CHECK8-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l48(i64 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK8-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK8: omp_offload.cont:
// CHECK8-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK8-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK8-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK8-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK8-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK8: arraydestroy.body:
-// CHECK8-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK8-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK8-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK8-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK8-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK8-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK8: arraydestroy.done2:
// CHECK8-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK8-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK8-NEXT: ret i32 [[TMP32]]
+// CHECK8-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK8-NEXT: ret i32 [[TMP36]]
//
//
// CHECK8-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
@@ -1749,7 +1765,7 @@ int main() {
// CHECK10-NEXT: [[TMP24:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK10-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK10-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK10-NEXT: store i32 1, ptr [[TMP25]], align 4
+// CHECK10-NEXT: store i32 2, ptr [[TMP25]], align 4
// CHECK10-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK10-NEXT: store i32 5, ptr [[TMP26]], align 4
// CHECK10-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1766,9 +1782,17 @@ int main() {
// CHECK10-NEXT: store ptr null, ptr [[TMP32]], align 4
// CHECK10-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK10-NEXT: store i64 2, ptr [[TMP33]], align 8
-// CHECK10-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.region_id, ptr [[KERNEL_ARGS]])
-// CHECK10-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
-// CHECK10-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK10-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK10-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK10-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK10-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK10-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK10-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK10-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK10-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK10-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.region_id, ptr [[KERNEL_ARGS]])
+// CHECK10-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK10-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK10: omp_offload.failed:
// CHECK10-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139(i32 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]], i32 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK10-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1776,18 +1800,18 @@ int main() {
// CHECK10-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiET_v()
// CHECK10-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK10-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK10-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK10-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK10-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK10: arraydestroy.body:
-// CHECK10-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP36]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK10-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP40]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK10-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK10-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK10-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK10-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK10: arraydestroy.done2:
// CHECK10-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK10-NEXT: [[TMP37:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK10-NEXT: ret i32 [[TMP37]]
+// CHECK10-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK10-NEXT: ret i32 [[TMP41]]
//
//
// CHECK10-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -2161,7 +2185,7 @@ int main() {
// CHECK10-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK10-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK10-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK10-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK10-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK10-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK10-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK10-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2178,27 +2202,35 @@ int main() {
// CHECK10-NEXT: store ptr null, ptr [[TMP27]], align 4
// CHECK10-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK10-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK10-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l48.region_id, ptr [[KERNEL_ARGS]])
-// CHECK10-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK10-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK10-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK10-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK10-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK10-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK10-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK10-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK10-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK10-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK10-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l48.region_id, ptr [[KERNEL_ARGS]])
+// CHECK10-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK10-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK10: omp_offload.failed:
// CHECK10-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l48(i32 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK10-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK10: omp_offload.cont:
// CHECK10-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK10-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK10-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK10-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK10-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK10: arraydestroy.body:
-// CHECK10-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK10-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK10-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK10-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK10-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK10-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK10: arraydestroy.done2:
// CHECK10-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK10-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK10-NEXT: ret i32 [[TMP32]]
+// CHECK10-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK10-NEXT: ret i32 [[TMP36]]
//
//
// CHECK10-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
diff --git a/clang/test/OpenMP/distribute_parallel_for_if_codegen.cpp b/clang/test/OpenMP/distribute_parallel_for_if_codegen.cpp
index 9c2d7981ecfd6..7ca6b251f3e0a 100644
--- a/clang/test/OpenMP/distribute_parallel_for_if_codegen.cpp
+++ b/clang/test/OpenMP/distribute_parallel_for_if_codegen.cpp
@@ -127,7 +127,7 @@ int main() {
// CHECK1-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -144,35 +144,51 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l47.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l47.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l47() #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l52.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l52.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK1-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l52() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -474,7 +490,7 @@ int main() {
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -491,78 +507,102 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l85.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l85.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l85() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK1-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK1: omp_offload.cont4:
-// CHECK1-NEXT: [[TMP22:%.*]] = load i32, ptr @Arg, align 4
-// CHECK1-NEXT: store i32 [[TMP22]], ptr [[ARG_CASTED]], align 4
-// CHECK1-NEXT: [[TMP23:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP23]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP23]], ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP30:%.*]] = load i32, ptr @Arg, align 4
+// CHECK1-NEXT: store i32 [[TMP30]], ptr [[ARG_CASTED]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP31]], ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP31]], ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP29]], align 4
-// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP30]], align 4
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP27]], ptr [[TMP31]], align 8
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP28]], ptr [[TMP32]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP35]], align 8
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP36]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l103.region_id, ptr [[KERNEL_ARGS6]])
-// CHECK1-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK1-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP38]], align 4
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP44]], align 8
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP45]], align 8
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP46]], align 8
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP47]], align 4
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP48]], align 4
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP49]], align 4
+// CHECK1-NEXT: [[TMP50:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l103.region_id, ptr [[KERNEL_ARGS6]])
+// CHECK1-NEXT: [[TMP51:%.*]] = icmp ne i32 [[TMP50]], 0
+// CHECK1-NEXT: br i1 [[TMP51]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK1: omp_offload.failed7:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l103(i64 [[TMP23]]) #[[ATTR2]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l103(i64 [[TMP31]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK1: omp_offload.cont8:
-// CHECK1-NEXT: [[TMP40:%.*]] = load i32, ptr @Arg, align 4
-// CHECK1-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP40]])
+// CHECK1-NEXT: [[TMP52:%.*]] = load i32, ptr @Arg, align 4
+// CHECK1-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP52]])
// CHECK1-NEXT: ret i32 [[CALL]]
//
//
@@ -1018,7 +1058,7 @@ int main() {
// CHECK1-NEXT: store i32 [[ARG]], ptr [[ARG_ADDR]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1035,74 +1075,98 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l63.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l63.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l63() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l69.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l69.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK1-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l69() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK1: omp_offload.cont4:
-// CHECK1-NEXT: [[TMP22:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
-// CHECK1-NEXT: store i32 [[TMP22]], ptr [[ARG_CASTED]], align 4
-// CHECK1-NEXT: [[TMP23:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP23]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP23]], ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP30:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
+// CHECK1-NEXT: store i32 [[TMP30]], ptr [[ARG_CASTED]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP31]], ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP31]], ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP29]], align 4
-// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP30]], align 4
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP27]], ptr [[TMP31]], align 8
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP28]], ptr [[TMP32]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.16, ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP35]], align 8
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP36]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l75.region_id, ptr [[KERNEL_ARGS6]])
-// CHECK1-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK1-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP38]], align 4
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.16, ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP44]], align 8
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP45]], align 8
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP46]], align 8
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP47]], align 4
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP48]], align 4
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP49]], align 4
+// CHECK1-NEXT: [[TMP50:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l75.region_id, ptr [[KERNEL_ARGS6]])
+// CHECK1-NEXT: [[TMP51:%.*]] = icmp ne i32 [[TMP50]], 0
+// CHECK1-NEXT: br i1 [[TMP51]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK1: omp_offload.failed7:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l75(i64 [[TMP23]]) #[[ATTR2]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l75(i64 [[TMP31]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK1: omp_offload.cont8:
// CHECK1-NEXT: ret i32 0
diff --git a/clang/test/OpenMP/distribute_parallel_for_lastprivate_codegen.cpp b/clang/test/OpenMP/distribute_parallel_for_lastprivate_codegen.cpp
index bb045a2044a0d..fdc1c68b5785c 100644
--- a/clang/test/OpenMP/distribute_parallel_for_lastprivate_codegen.cpp
+++ b/clang/test/OpenMP/distribute_parallel_for_lastprivate_codegen.cpp
@@ -815,7 +815,7 @@ int main() {
// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP25]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP25]], align 4
// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 5, ptr [[TMP26]], align 4
// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -832,9 +832,17 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP32]], align 8
// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP33]], align 8
-// CHECK9-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l123.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
-// CHECK9-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK9-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l123.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK9-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l123(i64 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]], i64 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -842,18 +850,18 @@ int main() {
// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiET_v()
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP36]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP40]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP37:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP37]]
+// CHECK9-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP41]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -1269,7 +1277,7 @@ int main() {
// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1286,27 +1294,35 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP27]], align 8
// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK9-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK9-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK9-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK9-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49(i64 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP32]]
+// CHECK9-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP36]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
@@ -1798,7 +1814,7 @@ int main() {
// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP25]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP25]], align 4
// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 5, ptr [[TMP26]], align 4
// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1815,9 +1831,17 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP32]], align 4
// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP33]], align 8
-// CHECK11-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l123.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
-// CHECK11-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK11-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l123.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK11-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l123(i32 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]], i32 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1825,18 +1849,18 @@ int main() {
// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiET_v()
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP36]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP40]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP37:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP37]]
+// CHECK11-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP41]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -2246,7 +2270,7 @@ int main() {
// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2263,27 +2287,35 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP27]], align 4
// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK11-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK11-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK11-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK11-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49(i32 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP32]]
+// CHECK11-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP36]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
diff --git a/clang/test/OpenMP/distribute_parallel_for_num_threads_codegen.cpp b/clang/test/OpenMP/distribute_parallel_for_num_threads_codegen.cpp
index 36be58ca78a9f..9c874b81f2b71 100644
--- a/clang/test/OpenMP/distribute_parallel_for_num_threads_codegen.cpp
+++ b/clang/test/OpenMP/distribute_parallel_for_num_threads_codegen.cpp
@@ -115,7 +115,7 @@ int main() {
// CHECK1-NEXT: store i8 [[CALL]], ptr [[A]], align 1
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -132,61 +132,77 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68() #[[ATTR6:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: lpad:
-// CHECK1-NEXT: [[TMP11:%.*]] = landingpad { ptr, i32 }
+// CHECK1-NEXT: [[TMP15:%.*]] = landingpad { ptr, i32 }
// CHECK1-NEXT: cleanup
-// CHECK1-NEXT: [[TMP12:%.*]] = extractvalue { ptr, i32 } [[TMP11]], 0
-// CHECK1-NEXT: store ptr [[TMP12]], ptr [[EXN_SLOT]], align 8
-// CHECK1-NEXT: [[TMP13:%.*]] = extractvalue { ptr, i32 } [[TMP11]], 1
-// CHECK1-NEXT: store i32 [[TMP13]], ptr [[EHSELECTOR_SLOT]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = extractvalue { ptr, i32 } [[TMP15]], 0
+// CHECK1-NEXT: store ptr [[TMP16]], ptr [[EXN_SLOT]], align 8
+// CHECK1-NEXT: [[TMP17:%.*]] = extractvalue { ptr, i32 } [[TMP15]], 1
+// CHECK1-NEXT: store i32 [[TMP17]], ptr [[EHSELECTOR_SLOT]], align 4
// CHECK1-NEXT: call void @_ZN1SD1Ev(ptr noundef nonnull align 8 dereferenceable(24) [[S]]) #[[ATTR6]]
// CHECK1-NEXT: br label [[EH_RESUME:%.*]]
// CHECK1: omp_offload.cont:
-// CHECK1-NEXT: [[TMP14:%.*]] = load i8, ptr [[A]], align 1
-// CHECK1-NEXT: store i8 [[TMP14]], ptr [[A_CASTED]], align 1
-// CHECK1-NEXT: [[TMP15:%.*]] = load i64, ptr [[A_CASTED]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP15]], ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP15]], ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP18:%.*]] = load i8, ptr [[A]], align 1
+// CHECK1-NEXT: store i8 [[TMP18]], ptr [[A_CASTED]], align 1
+// CHECK1-NEXT: [[TMP19:%.*]] = load i64, ptr [[A_CASTED]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP19]], ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP19]], ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK1-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74(i64 [[TMP15]]) #[[ATTR6]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74(i64 [[TMP19]]) #[[ATTR6]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK1: omp_offload.cont4:
-// CHECK1-NEXT: [[TMP32:%.*]] = load i8, ptr [[A]], align 1
-// CHECK1-NEXT: [[CONV:%.*]] = sext i8 [[TMP32]] to i32
+// CHECK1-NEXT: [[TMP40:%.*]] = load i8, ptr [[A]], align 1
+// CHECK1-NEXT: [[CONV:%.*]] = sext i8 [[TMP40]] to i32
// CHECK1-NEXT: [[CALL6:%.*]] = invoke noundef i32 @_Z5tmainIcLi5EEiv()
// CHECK1-NEXT: to label [[INVOKE_CONT5:%.*]] unwind label [[LPAD]]
// CHECK1: invoke.cont5:
@@ -197,8 +213,8 @@ int main() {
// CHECK1-NEXT: [[ADD9:%.*]] = add nsw i32 [[ADD]], [[CALL8]]
// CHECK1-NEXT: store i32 [[ADD9]], ptr [[RETVAL]], align 4
// CHECK1-NEXT: call void @_ZN1SD1Ev(ptr noundef nonnull align 8 dereferenceable(24) [[S]]) #[[ATTR6]]
-// CHECK1-NEXT: [[TMP33:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK1-NEXT: ret i32 [[TMP33]]
+// CHECK1-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK1-NEXT: ret i32 [[TMP41]]
// CHECK1: eh.resume:
// CHECK1-NEXT: [[EXN:%.*]] = load ptr, ptr [[EXN_SLOT]], align 8
// CHECK1-NEXT: [[SEL:%.*]] = load i32, ptr [[EHSELECTOR_SLOT]], align 4
@@ -548,7 +564,7 @@ int main() {
// CHECK1-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -565,35 +581,51 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l52() #[[ATTR6]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK1-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l57() #[[ATTR6]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -608,7 +640,7 @@ int main() {
// CHECK1-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -625,35 +657,51 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l52() #[[ATTR6]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK1-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l57() #[[ATTR6]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -1326,7 +1374,7 @@ int main() {
// CHECK5-NEXT: store i8 [[CALL]], ptr [[A]], align 1
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK5-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK5-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1343,61 +1391,77 @@ int main() {
// CHECK5-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK5-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK5-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK5-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK5-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK5-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK5-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68() #[[ATTR6:[0-9]+]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK5: lpad:
-// CHECK5-NEXT: [[TMP11:%.*]] = landingpad { ptr, i32 }
+// CHECK5-NEXT: [[TMP15:%.*]] = landingpad { ptr, i32 }
// CHECK5-NEXT: cleanup
-// CHECK5-NEXT: [[TMP12:%.*]] = extractvalue { ptr, i32 } [[TMP11]], 0
-// CHECK5-NEXT: store ptr [[TMP12]], ptr [[EXN_SLOT]], align 8
-// CHECK5-NEXT: [[TMP13:%.*]] = extractvalue { ptr, i32 } [[TMP11]], 1
-// CHECK5-NEXT: store i32 [[TMP13]], ptr [[EHSELECTOR_SLOT]], align 4
+// CHECK5-NEXT: [[TMP16:%.*]] = extractvalue { ptr, i32 } [[TMP15]], 0
+// CHECK5-NEXT: store ptr [[TMP16]], ptr [[EXN_SLOT]], align 8
+// CHECK5-NEXT: [[TMP17:%.*]] = extractvalue { ptr, i32 } [[TMP15]], 1
+// CHECK5-NEXT: store i32 [[TMP17]], ptr [[EHSELECTOR_SLOT]], align 4
// CHECK5-NEXT: call void @_ZN1SD1Ev(ptr noundef nonnull align 8 dereferenceable(24) [[S]]) #[[ATTR6]]
// CHECK5-NEXT: br label [[EH_RESUME:%.*]]
// CHECK5: omp_offload.cont:
-// CHECK5-NEXT: [[TMP14:%.*]] = load i8, ptr [[A]], align 1
-// CHECK5-NEXT: store i8 [[TMP14]], ptr [[A_CASTED]], align 1
-// CHECK5-NEXT: [[TMP15:%.*]] = load i64, ptr [[A_CASTED]], align 8
-// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK5-NEXT: store i64 [[TMP15]], ptr [[TMP16]], align 8
-// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK5-NEXT: store i64 [[TMP15]], ptr [[TMP17]], align 8
-// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK5-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP18:%.*]] = load i8, ptr [[A]], align 1
+// CHECK5-NEXT: store i8 [[TMP18]], ptr [[A_CASTED]], align 1
+// CHECK5-NEXT: [[TMP19:%.*]] = load i64, ptr [[A_CASTED]], align 8
+// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK5-NEXT: store i64 [[TMP19]], ptr [[TMP20]], align 8
+// CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK5-NEXT: store i64 [[TMP19]], ptr [[TMP21]], align 8
+// CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK5-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK5-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK5-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK5-NEXT: store ptr @.offload_sizes, ptr [[TMP25]], align 8
-// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK5-NEXT: store ptr @.offload_maptypes, ptr [[TMP26]], align 8
-// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK5-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
-// CHECK5-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK5-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK5-NEXT: store i64 100, ptr [[TMP29]], align 8
-// CHECK5-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK5-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK5-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK5-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
+// CHECK5-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK5-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK5-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK5-NEXT: store ptr @.offload_sizes, ptr [[TMP29]], align 8
+// CHECK5-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK5-NEXT: store ptr @.offload_maptypes, ptr [[TMP30]], align 8
+// CHECK5-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK5-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK5-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK5-NEXT: store i64 100, ptr [[TMP33]], align 8
+// CHECK5-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK5-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK5-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK5-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK5-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK5-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK5-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK5: omp_offload.failed3:
-// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74(i64 [[TMP15]]) #[[ATTR6]]
+// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74(i64 [[TMP19]]) #[[ATTR6]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK5: omp_offload.cont4:
-// CHECK5-NEXT: [[TMP32:%.*]] = load i8, ptr [[A]], align 1
-// CHECK5-NEXT: [[CONV:%.*]] = sext i8 [[TMP32]] to i32
+// CHECK5-NEXT: [[TMP40:%.*]] = load i8, ptr [[A]], align 1
+// CHECK5-NEXT: [[CONV:%.*]] = sext i8 [[TMP40]] to i32
// CHECK5-NEXT: [[CALL6:%.*]] = invoke noundef i32 @_Z5tmainIcLi5EEiv()
// CHECK5-NEXT: to label [[INVOKE_CONT5:%.*]] unwind label [[LPAD]]
// CHECK5: invoke.cont5:
@@ -1408,8 +1472,8 @@ int main() {
// CHECK5-NEXT: [[ADD9:%.*]] = add nsw i32 [[ADD]], [[CALL8]]
// CHECK5-NEXT: store i32 [[ADD9]], ptr [[RETVAL]], align 4
// CHECK5-NEXT: call void @_ZN1SD1Ev(ptr noundef nonnull align 8 dereferenceable(24) [[S]]) #[[ATTR6]]
-// CHECK5-NEXT: [[TMP33:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK5-NEXT: ret i32 [[TMP33]]
+// CHECK5-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK5-NEXT: ret i32 [[TMP41]]
// CHECK5: eh.resume:
// CHECK5-NEXT: [[EXN:%.*]] = load ptr, ptr [[EXN_SLOT]], align 8
// CHECK5-NEXT: [[SEL:%.*]] = load i32, ptr [[EHSELECTOR_SLOT]], align 4
@@ -1759,7 +1823,7 @@ int main() {
// CHECK5-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK5-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK5-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1776,35 +1840,51 @@ int main() {
// CHECK5-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK5-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK5-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK5-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK5-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK5-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK5-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l52() #[[ATTR6]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK5: omp_offload.cont:
// CHECK5-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK5-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK5-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK5-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK5-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK5-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK5-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK5-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK5-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK5-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK5-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK5-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK5-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK5-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK5-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK5-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK5-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK5-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK5-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK5-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK5-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK5-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK5-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK5-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK5: omp_offload.failed3:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l57() #[[ATTR6]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -1819,7 +1899,7 @@ int main() {
// CHECK5-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK5-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK5-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1836,35 +1916,51 @@ int main() {
// CHECK5-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK5-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK5-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK5-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK5-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK5-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK5-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l52() #[[ATTR6]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK5: omp_offload.cont:
// CHECK5-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK5-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK5-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK5-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK5-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK5-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK5-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK5-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK5-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK5-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK5-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK5-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK5-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK5-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK5-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK5-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK5-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK5-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK5-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK5-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK5-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK5-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK5-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK5-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK5: omp_offload.failed3:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l57() #[[ATTR6]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -2537,7 +2633,7 @@ int main() {
// CHECK9-NEXT: store i8 [[CALL]], ptr [[A]], align 1
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK9-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK9-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2554,61 +2650,77 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK9-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK9-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK9-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK9-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK9-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK9-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68() #[[ATTR6:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: lpad:
-// CHECK9-NEXT: [[TMP11:%.*]] = landingpad { ptr, i32 }
+// CHECK9-NEXT: [[TMP15:%.*]] = landingpad { ptr, i32 }
// CHECK9-NEXT: cleanup
-// CHECK9-NEXT: [[TMP12:%.*]] = extractvalue { ptr, i32 } [[TMP11]], 0
-// CHECK9-NEXT: store ptr [[TMP12]], ptr [[EXN_SLOT]], align 8
-// CHECK9-NEXT: [[TMP13:%.*]] = extractvalue { ptr, i32 } [[TMP11]], 1
-// CHECK9-NEXT: store i32 [[TMP13]], ptr [[EHSELECTOR_SLOT]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = extractvalue { ptr, i32 } [[TMP15]], 0
+// CHECK9-NEXT: store ptr [[TMP16]], ptr [[EXN_SLOT]], align 8
+// CHECK9-NEXT: [[TMP17:%.*]] = extractvalue { ptr, i32 } [[TMP15]], 1
+// CHECK9-NEXT: store i32 [[TMP17]], ptr [[EHSELECTOR_SLOT]], align 4
// CHECK9-NEXT: call void @_ZN1SD1Ev(ptr noundef nonnull align 8 dereferenceable(24) [[S]]) #[[ATTR6]]
// CHECK9-NEXT: br label [[EH_RESUME:%.*]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP14:%.*]] = load i8, ptr [[A]], align 1
-// CHECK9-NEXT: store i8 [[TMP14]], ptr [[A_CASTED]], align 1
-// CHECK9-NEXT: [[TMP15:%.*]] = load i64, ptr [[A_CASTED]], align 8
-// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP15]], ptr [[TMP16]], align 8
-// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP15]], ptr [[TMP17]], align 8
-// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP18:%.*]] = load i8, ptr [[A]], align 1
+// CHECK9-NEXT: store i8 [[TMP18]], ptr [[A_CASTED]], align 1
+// CHECK9-NEXT: [[TMP19:%.*]] = load i64, ptr [[A_CASTED]], align 8
+// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP19]], ptr [[TMP20]], align 8
+// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP19]], ptr [[TMP21]], align 8
+// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK9-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes, ptr [[TMP25]], align 8
-// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes, ptr [[TMP26]], align 8
-// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK9-NEXT: store i64 100, ptr [[TMP29]], align 8
-// CHECK9-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK9-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK9-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK9-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes, ptr [[TMP29]], align 8
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes, ptr [[TMP30]], align 8
+// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK9-NEXT: store i64 100, ptr [[TMP33]], align 8
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK9-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK9-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK9-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK9: omp_offload.failed3:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74(i64 [[TMP15]]) #[[ATTR6]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74(i64 [[TMP19]]) #[[ATTR6]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK9: omp_offload.cont4:
-// CHECK9-NEXT: [[TMP32:%.*]] = load i8, ptr [[A]], align 1
-// CHECK9-NEXT: [[CONV:%.*]] = sext i8 [[TMP32]] to i32
+// CHECK9-NEXT: [[TMP40:%.*]] = load i8, ptr [[A]], align 1
+// CHECK9-NEXT: [[CONV:%.*]] = sext i8 [[TMP40]] to i32
// CHECK9-NEXT: [[CALL6:%.*]] = invoke noundef i32 @_Z5tmainIcLi5EEiv()
// CHECK9-NEXT: to label [[INVOKE_CONT5:%.*]] unwind label [[LPAD]]
// CHECK9: invoke.cont5:
@@ -2619,8 +2731,8 @@ int main() {
// CHECK9-NEXT: [[ADD9:%.*]] = add nsw i32 [[ADD]], [[CALL8]]
// CHECK9-NEXT: store i32 [[ADD9]], ptr [[RETVAL]], align 4
// CHECK9-NEXT: call void @_ZN1SD1Ev(ptr noundef nonnull align 8 dereferenceable(24) [[S]]) #[[ATTR6]]
-// CHECK9-NEXT: [[TMP33:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP33]]
+// CHECK9-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP41]]
// CHECK9: eh.resume:
// CHECK9-NEXT: [[EXN:%.*]] = load ptr, ptr [[EXN_SLOT]], align 8
// CHECK9-NEXT: [[SEL:%.*]] = load i32, ptr [[EHSELECTOR_SLOT]], align 4
@@ -2970,7 +3082,7 @@ int main() {
// CHECK9-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK9-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK9-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2987,35 +3099,51 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK9-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK9-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK9-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK9-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK9-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK9-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l52() #[[ATTR6]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK9-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK9-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK9-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK9-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK9-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK9-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK9-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK9-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK9-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK9-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK9-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK9-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK9-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK9-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK9-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK9: omp_offload.failed3:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l57() #[[ATTR6]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -3030,7 +3158,7 @@ int main() {
// CHECK9-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK9-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK9-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3047,35 +3175,51 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK9-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK9-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK9-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK9-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK9-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK9-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l52() #[[ATTR6]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK9-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK9-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK9-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK9-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK9-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK9-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK9-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK9-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK9-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK9-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK9-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK9-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK9-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK9-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK9-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK9: omp_offload.failed3:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l57() #[[ATTR6]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -3748,7 +3892,7 @@ int main() {
// CHECK13-NEXT: store i8 [[CALL]], ptr [[A]], align 1
// CHECK13-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK13-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK13-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK13-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK13-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK13-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3765,61 +3909,77 @@ int main() {
// CHECK13-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK13-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK13-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK13-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68.region_id, ptr [[KERNEL_ARGS]])
-// CHECK13-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK13-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK13-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK13-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK13-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK13-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK13-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68.region_id, ptr [[KERNEL_ARGS]])
+// CHECK13-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK13-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK13: omp_offload.failed:
// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68() #[[ATTR6:[0-9]+]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK13: lpad:
-// CHECK13-NEXT: [[TMP11:%.*]] = landingpad { ptr, i32 }
+// CHECK13-NEXT: [[TMP15:%.*]] = landingpad { ptr, i32 }
// CHECK13-NEXT: cleanup
-// CHECK13-NEXT: [[TMP12:%.*]] = extractvalue { ptr, i32 } [[TMP11]], 0
-// CHECK13-NEXT: store ptr [[TMP12]], ptr [[EXN_SLOT]], align 8
-// CHECK13-NEXT: [[TMP13:%.*]] = extractvalue { ptr, i32 } [[TMP11]], 1
-// CHECK13-NEXT: store i32 [[TMP13]], ptr [[EHSELECTOR_SLOT]], align 4
+// CHECK13-NEXT: [[TMP16:%.*]] = extractvalue { ptr, i32 } [[TMP15]], 0
+// CHECK13-NEXT: store ptr [[TMP16]], ptr [[EXN_SLOT]], align 8
+// CHECK13-NEXT: [[TMP17:%.*]] = extractvalue { ptr, i32 } [[TMP15]], 1
+// CHECK13-NEXT: store i32 [[TMP17]], ptr [[EHSELECTOR_SLOT]], align 4
// CHECK13-NEXT: call void @_ZN1SD1Ev(ptr noundef nonnull align 8 dereferenceable(24) [[S]]) #[[ATTR6]]
// CHECK13-NEXT: br label [[EH_RESUME:%.*]]
// CHECK13: omp_offload.cont:
-// CHECK13-NEXT: [[TMP14:%.*]] = load i8, ptr [[A]], align 1
-// CHECK13-NEXT: store i8 [[TMP14]], ptr [[A_CASTED]], align 1
-// CHECK13-NEXT: [[TMP15:%.*]] = load i64, ptr [[A_CASTED]], align 8
-// CHECK13-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP15]], ptr [[TMP16]], align 8
-// CHECK13-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP15]], ptr [[TMP17]], align 8
-// CHECK13-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK13-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP18:%.*]] = load i8, ptr [[A]], align 1
+// CHECK13-NEXT: store i8 [[TMP18]], ptr [[A_CASTED]], align 1
+// CHECK13-NEXT: [[TMP19:%.*]] = load i64, ptr [[A_CASTED]], align 8
+// CHECK13-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP19]], ptr [[TMP20]], align 8
+// CHECK13-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP19]], ptr [[TMP21]], align 8
+// CHECK13-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK13-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK13-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK13-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK13-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK13-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK13-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK13-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK13-NEXT: store ptr @.offload_sizes, ptr [[TMP25]], align 8
-// CHECK13-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes, ptr [[TMP26]], align 8
-// CHECK13-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK13-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK13-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK13-NEXT: store i64 100, ptr [[TMP29]], align 8
-// CHECK13-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK13-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK13-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK13-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK13-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK13-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK13-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK13-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK13-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK13-NEXT: store ptr @.offload_sizes, ptr [[TMP29]], align 8
+// CHECK13-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes, ptr [[TMP30]], align 8
+// CHECK13-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK13-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK13-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK13-NEXT: store i64 100, ptr [[TMP33]], align 8
+// CHECK13-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK13-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK13-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK13-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK13-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK13-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK13-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK13: omp_offload.failed3:
-// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74(i64 [[TMP15]]) #[[ATTR6]]
+// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74(i64 [[TMP19]]) #[[ATTR6]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK13: omp_offload.cont4:
-// CHECK13-NEXT: [[TMP32:%.*]] = load i8, ptr [[A]], align 1
-// CHECK13-NEXT: [[CONV:%.*]] = sext i8 [[TMP32]] to i32
+// CHECK13-NEXT: [[TMP40:%.*]] = load i8, ptr [[A]], align 1
+// CHECK13-NEXT: [[CONV:%.*]] = sext i8 [[TMP40]] to i32
// CHECK13-NEXT: [[CALL6:%.*]] = invoke noundef i32 @_Z5tmainIcLi5EEiv()
// CHECK13-NEXT: to label [[INVOKE_CONT5:%.*]] unwind label [[LPAD]]
// CHECK13: invoke.cont5:
@@ -3830,8 +3990,8 @@ int main() {
// CHECK13-NEXT: [[ADD9:%.*]] = add nsw i32 [[ADD]], [[CALL8]]
// CHECK13-NEXT: store i32 [[ADD9]], ptr [[RETVAL]], align 4
// CHECK13-NEXT: call void @_ZN1SD1Ev(ptr noundef nonnull align 8 dereferenceable(24) [[S]]) #[[ATTR6]]
-// CHECK13-NEXT: [[TMP33:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK13-NEXT: ret i32 [[TMP33]]
+// CHECK13-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK13-NEXT: ret i32 [[TMP41]]
// CHECK13: eh.resume:
// CHECK13-NEXT: [[EXN:%.*]] = load ptr, ptr [[EXN_SLOT]], align 8
// CHECK13-NEXT: [[SEL:%.*]] = load i32, ptr [[EHSELECTOR_SLOT]], align 4
@@ -4181,7 +4341,7 @@ int main() {
// CHECK13-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK13-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK13-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK13-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK13-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK13-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK13-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -4198,35 +4358,51 @@ int main() {
// CHECK13-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK13-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK13-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK13-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
-// CHECK13-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK13-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK13-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK13-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK13-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK13-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK13-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
+// CHECK13-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK13-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK13: omp_offload.failed:
// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l52() #[[ATTR6]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK13: omp_offload.cont:
// CHECK13-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK13-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK13-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK13-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK13-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK13-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK13-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK13-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK13-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK13-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK13-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK13-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK13-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK13-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK13-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK13-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK13-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK13-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK13-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK13-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK13-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK13-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK13-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK13-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK13-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK13-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK13-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK13-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK13-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK13-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK13-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK13-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK13-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK13-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK13-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK13-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK13-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK13-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK13-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK13-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK13: omp_offload.failed3:
// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l57() #[[ATTR6]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -4241,7 +4417,7 @@ int main() {
// CHECK13-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK13-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK13-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK13-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK13-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK13-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK13-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -4258,35 +4434,51 @@ int main() {
// CHECK13-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK13-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK13-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK13-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
-// CHECK13-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK13-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK13-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK13-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK13-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK13-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK13-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
+// CHECK13-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK13-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK13: omp_offload.failed:
// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l52() #[[ATTR6]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK13: omp_offload.cont:
// CHECK13-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK13-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK13-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK13-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK13-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK13-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK13-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK13-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK13-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK13-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK13-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK13-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK13-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK13-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK13-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK13-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK13-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK13-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK13-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK13-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK13-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK13-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK13-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK13-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK13-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK13-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK13-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK13-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK13-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK13-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK13-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK13-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK13-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK13-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK13-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK13-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK13-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK13-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK13-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK13-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK13: omp_offload.failed3:
// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l57() #[[ATTR6]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT4]]
diff --git a/clang/test/OpenMP/distribute_parallel_for_private_codegen.cpp b/clang/test/OpenMP/distribute_parallel_for_private_codegen.cpp
index bc5a84ae88dcc..9a606bc2f4581 100644
--- a/clang/test/OpenMP/distribute_parallel_for_private_codegen.cpp
+++ b/clang/test/OpenMP/distribute_parallel_for_private_codegen.cpp
@@ -535,7 +535,7 @@ int main() {
// CHECK9-NEXT: store ptr undef, ptr [[_TMP1]], align 8
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK9-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK9-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -552,9 +552,17 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK9-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK9-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l95.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK9-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK9-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK9-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l95.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK9-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l95() #[[ATTR4:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -562,18 +570,18 @@ int main() {
// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiET_v()
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP12]]
+// CHECK9-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP16]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -848,7 +856,7 @@ int main() {
// CHECK9-NEXT: store ptr undef, ptr [[_TMP1]], align 8
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK9-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK9-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -865,27 +873,35 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK9-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK9-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK9-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK9-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK9-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK9-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49() #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP12]]
+// CHECK9-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP16]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
@@ -1237,7 +1253,7 @@ int main() {
// CHECK11-NEXT: store ptr undef, ptr [[_TMP1]], align 4
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK11-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK11-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1254,9 +1270,17 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK11-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK11-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l95.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK11-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK11-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK11-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK11-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l95.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK11-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l95() #[[ATTR4:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1264,18 +1288,18 @@ int main() {
// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiET_v()
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP12]]
+// CHECK11-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP16]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -1544,7 +1568,7 @@ int main() {
// CHECK11-NEXT: store ptr undef, ptr [[_TMP1]], align 4
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK11-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK11-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1561,27 +1585,35 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK11-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK11-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK11-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK11-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK11-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK11-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK11-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49() #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP12]]
+// CHECK11-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP16]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
diff --git a/clang/test/OpenMP/distribute_parallel_for_proc_bind_codegen.cpp b/clang/test/OpenMP/distribute_parallel_for_proc_bind_codegen.cpp
index 5935879968be2..903dbb6de2880 100644
--- a/clang/test/OpenMP/distribute_parallel_for_proc_bind_codegen.cpp
+++ b/clang/test/OpenMP/distribute_parallel_for_proc_bind_codegen.cpp
@@ -62,7 +62,7 @@ int main() {
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -79,35 +79,51 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 1000, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l37.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l37.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l37() #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 1000, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l41.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 1000, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l41.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK1-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l41() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -398,7 +414,7 @@ int main() {
// CHECK1-NEXT: [[TMP:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -415,9 +431,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 1000, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l29.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l29.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l29() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/distribute_parallel_for_simd_codegen.cpp b/clang/test/OpenMP/distribute_parallel_for_simd_codegen.cpp
index cb9a0e12f9c99..a0819493d2cee 100644
--- a/clang/test/OpenMP/distribute_parallel_for_simd_codegen.cpp
+++ b/clang/test/OpenMP/distribute_parallel_for_simd_codegen.cpp
@@ -4770,7 +4770,7 @@ int main() {
// CHECK9-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 4, ptr [[TMP24]], align 4
// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -4787,446 +4787,502 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK9-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l368.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK9-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP34]], align 4
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l368.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK9-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l368(i64 [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]]) #[[ATTR2:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK9-NEXT: [[TMP35:%.*]] = load i64, ptr [[N_CASTED3]], align 8
-// CHECK9-NEXT: [[TMP36:%.*]] = load ptr, ptr [[A]], align 8
-// CHECK9-NEXT: [[TMP37:%.*]] = load ptr, ptr [[B]], align 8
-// CHECK9-NEXT: [[TMP38:%.*]] = load ptr, ptr [[C]], align 8
-// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP35]], ptr [[TMP39]], align 8
-// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP35]], ptr [[TMP40]], align 8
-// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP41]], align 8
-// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK9-NEXT: store ptr [[TMP36]], ptr [[TMP42]], align 8
-// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK9-NEXT: store ptr [[TMP36]], ptr [[TMP43]], align 8
-// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP44]], align 8
-// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP37]], ptr [[TMP45]], align 8
-// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP37]], ptr [[TMP46]], align 8
-// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP47]], align 8
-// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP38]], ptr [[TMP48]], align 8
-// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP38]], ptr [[TMP49]], align 8
-// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP50]], align 8
-// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP53:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP53]], ptr [[DOTCAPTURE_EXPR_8]], align 4
-// CHECK9-NEXT: [[TMP54:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_8]], align 4
-// CHECK9-NEXT: [[SUB10:%.*]] = sub nsw i32 [[TMP54]], 0
+// CHECK9-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK9-NEXT: [[TMP39:%.*]] = load i64, ptr [[N_CASTED3]], align 8
+// CHECK9-NEXT: [[TMP40:%.*]] = load ptr, ptr [[A]], align 8
+// CHECK9-NEXT: [[TMP41:%.*]] = load ptr, ptr [[B]], align 8
+// CHECK9-NEXT: [[TMP42:%.*]] = load ptr, ptr [[C]], align 8
+// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP39]], ptr [[TMP43]], align 8
+// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP39]], ptr [[TMP44]], align 8
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK9-NEXT: store ptr [[TMP40]], ptr [[TMP46]], align 8
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK9-NEXT: store ptr [[TMP40]], ptr [[TMP47]], align 8
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP48]], align 8
+// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP41]], ptr [[TMP49]], align 8
+// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP41]], ptr [[TMP50]], align 8
+// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP51]], align 8
+// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP42]], ptr [[TMP52]], align 8
+// CHECK9-NEXT: [[TMP53:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP42]], ptr [[TMP53]], align 8
+// CHECK9-NEXT: [[TMP54:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP54]], align 8
+// CHECK9-NEXT: [[TMP55:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP56:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP57:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP57]], ptr [[DOTCAPTURE_EXPR_8]], align 4
+// CHECK9-NEXT: [[TMP58:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_8]], align 4
+// CHECK9-NEXT: [[SUB10:%.*]] = sub nsw i32 [[TMP58]], 0
// CHECK9-NEXT: [[DIV11:%.*]] = sdiv i32 [[SUB10]], 1
// CHECK9-NEXT: [[SUB12:%.*]] = sub nsw i32 [[DIV11]], 1
// CHECK9-NEXT: store i32 [[SUB12]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK9-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK9-NEXT: [[ADD13:%.*]] = add nsw i32 [[TMP55]], 1
-// CHECK9-NEXT: [[TMP56:%.*]] = zext i32 [[ADD13]] to i64
+// CHECK9-NEXT: [[TMP59:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK9-NEXT: [[ADD13:%.*]] = add nsw i32 [[TMP59]], 1
+// CHECK9-NEXT: [[TMP60:%.*]] = zext i32 [[ADD13]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS14:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP57]], align 4
-// CHECK9-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
-// CHECK9-NEXT: store i32 4, ptr [[TMP58]], align 4
-// CHECK9-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP51]], ptr [[TMP59]], align 8
-// CHECK9-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP52]], ptr [[TMP60]], align 8
-// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.4, ptr [[TMP61]], align 8
-// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP62]], align 8
-// CHECK9-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP63]], align 8
-// CHECK9-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP64]], align 8
-// CHECK9-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP56]], ptr [[TMP65]], align 8
-// CHECK9-NEXT: [[TMP66:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l407.region_id, ptr [[KERNEL_ARGS14]])
-// CHECK9-NEXT: [[TMP67:%.*]] = icmp ne i32 [[TMP66]], 0
-// CHECK9-NEXT: br i1 [[TMP67]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
+// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP61]], align 4
+// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
+// CHECK9-NEXT: store i32 4, ptr [[TMP62]], align 4
+// CHECK9-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP55]], ptr [[TMP63]], align 8
+// CHECK9-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP56]], ptr [[TMP64]], align 8
+// CHECK9-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.4, ptr [[TMP65]], align 8
+// CHECK9-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP66]], align 8
+// CHECK9-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP67]], align 8
+// CHECK9-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP68]], align 8
+// CHECK9-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP60]], ptr [[TMP69]], align 8
+// CHECK9-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP70]], align 8
+// CHECK9-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP71]], align 4
+// CHECK9-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP72]], align 4
+// CHECK9-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP73]], align 4
+// CHECK9-NEXT: [[TMP74:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l407.region_id, ptr [[KERNEL_ARGS14]])
+// CHECK9-NEXT: [[TMP75:%.*]] = icmp ne i32 [[TMP74]], 0
+// CHECK9-NEXT: br i1 [[TMP75]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
// CHECK9: omp_offload.failed15:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l407(i64 [[TMP35]], ptr [[TMP36]], ptr [[TMP37]], ptr [[TMP38]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l407(i64 [[TMP39]], ptr [[TMP40]], ptr [[TMP41]], ptr [[TMP42]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT16]]
// CHECK9: omp_offload.cont16:
-// CHECK9-NEXT: [[TMP68:%.*]] = load i32, ptr [[CH]], align 4
-// CHECK9-NEXT: store i32 [[TMP68]], ptr [[CH_CASTED]], align 4
-// CHECK9-NEXT: [[TMP69:%.*]] = load i64, ptr [[CH_CASTED]], align 8
-// CHECK9-NEXT: [[TMP70:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP70]], ptr [[N_CASTED17]], align 4
-// CHECK9-NEXT: [[TMP71:%.*]] = load i64, ptr [[N_CASTED17]], align 8
-// CHECK9-NEXT: [[TMP72:%.*]] = load ptr, ptr [[A]], align 8
-// CHECK9-NEXT: [[TMP73:%.*]] = load ptr, ptr [[B]], align 8
-// CHECK9-NEXT: [[TMP74:%.*]] = load ptr, ptr [[C]], align 8
-// CHECK9-NEXT: [[TMP75:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP69]], ptr [[TMP75]], align 8
-// CHECK9-NEXT: [[TMP76:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP69]], ptr [[TMP76]], align 8
-// CHECK9-NEXT: [[TMP77:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP77]], align 8
-// CHECK9-NEXT: [[TMP78:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP71]], ptr [[TMP78]], align 8
-// CHECK9-NEXT: [[TMP79:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP71]], ptr [[TMP79]], align 8
-// CHECK9-NEXT: [[TMP80:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP80]], align 8
-// CHECK9-NEXT: [[TMP81:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP72]], ptr [[TMP81]], align 8
-// CHECK9-NEXT: [[TMP82:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP72]], ptr [[TMP82]], align 8
-// CHECK9-NEXT: [[TMP83:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP83]], align 8
-// CHECK9-NEXT: [[TMP84:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP73]], ptr [[TMP84]], align 8
-// CHECK9-NEXT: [[TMP85:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP73]], ptr [[TMP85]], align 8
-// CHECK9-NEXT: [[TMP86:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP86]], align 8
-// CHECK9-NEXT: [[TMP87:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP74]], ptr [[TMP87]], align 8
-// CHECK9-NEXT: [[TMP88:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP74]], ptr [[TMP88]], align 8
-// CHECK9-NEXT: [[TMP89:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 4
-// CHECK9-NEXT: store ptr null, ptr [[TMP89]], align 8
-// CHECK9-NEXT: [[TMP90:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP91:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP92:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP92]], ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK9-NEXT: [[TMP93:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK9-NEXT: [[SUB24:%.*]] = sub nsw i32 [[TMP93]], 0
+// CHECK9-NEXT: [[TMP76:%.*]] = load i32, ptr [[CH]], align 4
+// CHECK9-NEXT: store i32 [[TMP76]], ptr [[CH_CASTED]], align 4
+// CHECK9-NEXT: [[TMP77:%.*]] = load i64, ptr [[CH_CASTED]], align 8
+// CHECK9-NEXT: [[TMP78:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP78]], ptr [[N_CASTED17]], align 4
+// CHECK9-NEXT: [[TMP79:%.*]] = load i64, ptr [[N_CASTED17]], align 8
+// CHECK9-NEXT: [[TMP80:%.*]] = load ptr, ptr [[A]], align 8
+// CHECK9-NEXT: [[TMP81:%.*]] = load ptr, ptr [[B]], align 8
+// CHECK9-NEXT: [[TMP82:%.*]] = load ptr, ptr [[C]], align 8
+// CHECK9-NEXT: [[TMP83:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP77]], ptr [[TMP83]], align 8
+// CHECK9-NEXT: [[TMP84:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP77]], ptr [[TMP84]], align 8
+// CHECK9-NEXT: [[TMP85:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP85]], align 8
+// CHECK9-NEXT: [[TMP86:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP79]], ptr [[TMP86]], align 8
+// CHECK9-NEXT: [[TMP87:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP79]], ptr [[TMP87]], align 8
+// CHECK9-NEXT: [[TMP88:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP88]], align 8
+// CHECK9-NEXT: [[TMP89:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP80]], ptr [[TMP89]], align 8
+// CHECK9-NEXT: [[TMP90:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP80]], ptr [[TMP90]], align 8
+// CHECK9-NEXT: [[TMP91:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP91]], align 8
+// CHECK9-NEXT: [[TMP92:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP81]], ptr [[TMP92]], align 8
+// CHECK9-NEXT: [[TMP93:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP81]], ptr [[TMP93]], align 8
+// CHECK9-NEXT: [[TMP94:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP94]], align 8
+// CHECK9-NEXT: [[TMP95:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP82]], ptr [[TMP95]], align 8
+// CHECK9-NEXT: [[TMP96:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP82]], ptr [[TMP96]], align 8
+// CHECK9-NEXT: [[TMP97:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 4
+// CHECK9-NEXT: store ptr null, ptr [[TMP97]], align 8
+// CHECK9-NEXT: [[TMP98:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP99:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP100:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP100]], ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK9-NEXT: [[TMP101:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK9-NEXT: [[SUB24:%.*]] = sub nsw i32 [[TMP101]], 0
// CHECK9-NEXT: [[DIV25:%.*]] = sdiv i32 [[SUB24]], 1
// CHECK9-NEXT: [[SUB26:%.*]] = sub nsw i32 [[DIV25]], 1
// CHECK9-NEXT: store i32 [[SUB26]], ptr [[DOTCAPTURE_EXPR_23]], align 4
-// CHECK9-NEXT: [[TMP94:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_23]], align 4
-// CHECK9-NEXT: [[ADD27:%.*]] = add nsw i32 [[TMP94]], 1
-// CHECK9-NEXT: [[TMP95:%.*]] = zext i32 [[ADD27]] to i64
+// CHECK9-NEXT: [[TMP102:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_23]], align 4
+// CHECK9-NEXT: [[ADD27:%.*]] = add nsw i32 [[TMP102]], 1
+// CHECK9-NEXT: [[TMP103:%.*]] = zext i32 [[ADD27]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS28:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP96]], align 4
-// CHECK9-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
-// CHECK9-NEXT: store i32 5, ptr [[TMP97]], align 4
-// CHECK9-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP90]], ptr [[TMP98]], align 8
-// CHECK9-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP91]], ptr [[TMP99]], align 8
-// CHECK9-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.8, ptr [[TMP100]], align 8
-// CHECK9-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP101]], align 8
-// CHECK9-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP102]], align 8
-// CHECK9-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP103]], align 8
-// CHECK9-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP95]], ptr [[TMP104]], align 8
-// CHECK9-NEXT: [[TMP105:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l446.region_id, ptr [[KERNEL_ARGS28]])
-// CHECK9-NEXT: [[TMP106:%.*]] = icmp ne i32 [[TMP105]], 0
-// CHECK9-NEXT: br i1 [[TMP106]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
+// CHECK9-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP104]], align 4
+// CHECK9-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
+// CHECK9-NEXT: store i32 5, ptr [[TMP105]], align 4
+// CHECK9-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP98]], ptr [[TMP106]], align 8
+// CHECK9-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP99]], ptr [[TMP107]], align 8
+// CHECK9-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.8, ptr [[TMP108]], align 8
+// CHECK9-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP109]], align 8
+// CHECK9-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP110]], align 8
+// CHECK9-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP111]], align 8
+// CHECK9-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP103]], ptr [[TMP112]], align 8
+// CHECK9-NEXT: [[TMP113:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP113]], align 8
+// CHECK9-NEXT: [[TMP114:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP114]], align 4
+// CHECK9-NEXT: [[TMP115:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP115]], align 4
+// CHECK9-NEXT: [[TMP116:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP116]], align 4
+// CHECK9-NEXT: [[TMP117:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l446.region_id, ptr [[KERNEL_ARGS28]])
+// CHECK9-NEXT: [[TMP118:%.*]] = icmp ne i32 [[TMP117]], 0
+// CHECK9-NEXT: br i1 [[TMP118]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
// CHECK9: omp_offload.failed29:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l446(i64 [[TMP69]], i64 [[TMP71]], ptr [[TMP72]], ptr [[TMP73]], ptr [[TMP74]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l446(i64 [[TMP77]], i64 [[TMP79]], ptr [[TMP80]], ptr [[TMP81]], ptr [[TMP82]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT30]]
// CHECK9: omp_offload.cont30:
-// CHECK9-NEXT: [[TMP107:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP107]], ptr [[N_CASTED31]], align 4
-// CHECK9-NEXT: [[TMP108:%.*]] = load i64, ptr [[N_CASTED31]], align 8
-// CHECK9-NEXT: [[TMP109:%.*]] = load ptr, ptr [[A]], align 8
-// CHECK9-NEXT: [[TMP110:%.*]] = load ptr, ptr [[B]], align 8
-// CHECK9-NEXT: [[TMP111:%.*]] = load ptr, ptr [[C]], align 8
-// CHECK9-NEXT: [[TMP112:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP108]], ptr [[TMP112]], align 8
-// CHECK9-NEXT: [[TMP113:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP108]], ptr [[TMP113]], align 8
-// CHECK9-NEXT: [[TMP114:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP114]], align 8
-// CHECK9-NEXT: [[TMP115:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 1
-// CHECK9-NEXT: store ptr [[TMP109]], ptr [[TMP115]], align 8
-// CHECK9-NEXT: [[TMP116:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 1
-// CHECK9-NEXT: store ptr [[TMP109]], ptr [[TMP116]], align 8
-// CHECK9-NEXT: [[TMP117:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP117]], align 8
-// CHECK9-NEXT: [[TMP118:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP110]], ptr [[TMP118]], align 8
-// CHECK9-NEXT: [[TMP119:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP110]], ptr [[TMP119]], align 8
-// CHECK9-NEXT: [[TMP120:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP120]], align 8
-// CHECK9-NEXT: [[TMP121:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP111]], ptr [[TMP121]], align 8
-// CHECK9-NEXT: [[TMP122:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP111]], ptr [[TMP122]], align 8
-// CHECK9-NEXT: [[TMP123:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP123]], align 8
+// CHECK9-NEXT: [[TMP119:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP119]], ptr [[N_CASTED31]], align 4
+// CHECK9-NEXT: [[TMP120:%.*]] = load i64, ptr [[N_CASTED31]], align 8
+// CHECK9-NEXT: [[TMP121:%.*]] = load ptr, ptr [[A]], align 8
+// CHECK9-NEXT: [[TMP122:%.*]] = load ptr, ptr [[B]], align 8
+// CHECK9-NEXT: [[TMP123:%.*]] = load ptr, ptr [[C]], align 8
// CHECK9-NEXT: [[TMP124:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP120]], ptr [[TMP124]], align 8
// CHECK9-NEXT: [[TMP125:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP126:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP126]], ptr [[DOTCAPTURE_EXPR_36]], align 4
-// CHECK9-NEXT: [[TMP127:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_36]], align 4
-// CHECK9-NEXT: [[SUB38:%.*]] = sub nsw i32 [[TMP127]], 0
+// CHECK9-NEXT: store i64 [[TMP120]], ptr [[TMP125]], align 8
+// CHECK9-NEXT: [[TMP126:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP126]], align 8
+// CHECK9-NEXT: [[TMP127:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 1
+// CHECK9-NEXT: store ptr [[TMP121]], ptr [[TMP127]], align 8
+// CHECK9-NEXT: [[TMP128:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 1
+// CHECK9-NEXT: store ptr [[TMP121]], ptr [[TMP128]], align 8
+// CHECK9-NEXT: [[TMP129:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP129]], align 8
+// CHECK9-NEXT: [[TMP130:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP122]], ptr [[TMP130]], align 8
+// CHECK9-NEXT: [[TMP131:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP122]], ptr [[TMP131]], align 8
+// CHECK9-NEXT: [[TMP132:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP132]], align 8
+// CHECK9-NEXT: [[TMP133:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP123]], ptr [[TMP133]], align 8
+// CHECK9-NEXT: [[TMP134:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP123]], ptr [[TMP134]], align 8
+// CHECK9-NEXT: [[TMP135:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP135]], align 8
+// CHECK9-NEXT: [[TMP136:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP137:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP138:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP138]], ptr [[DOTCAPTURE_EXPR_36]], align 4
+// CHECK9-NEXT: [[TMP139:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_36]], align 4
+// CHECK9-NEXT: [[SUB38:%.*]] = sub nsw i32 [[TMP139]], 0
// CHECK9-NEXT: [[DIV39:%.*]] = sdiv i32 [[SUB38]], 1
// CHECK9-NEXT: [[SUB40:%.*]] = sub nsw i32 [[DIV39]], 1
// CHECK9-NEXT: store i32 [[SUB40]], ptr [[DOTCAPTURE_EXPR_37]], align 4
-// CHECK9-NEXT: [[TMP128:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_37]], align 4
-// CHECK9-NEXT: [[ADD41:%.*]] = add nsw i32 [[TMP128]], 1
-// CHECK9-NEXT: [[TMP129:%.*]] = zext i32 [[ADD41]] to i64
+// CHECK9-NEXT: [[TMP140:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_37]], align 4
+// CHECK9-NEXT: [[ADD41:%.*]] = add nsw i32 [[TMP140]], 1
+// CHECK9-NEXT: [[TMP141:%.*]] = zext i32 [[ADD41]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS42:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP130]], align 4
-// CHECK9-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 1
-// CHECK9-NEXT: store i32 4, ptr [[TMP131]], align 4
-// CHECK9-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP124]], ptr [[TMP132]], align 8
-// CHECK9-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP125]], ptr [[TMP133]], align 8
-// CHECK9-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.12, ptr [[TMP134]], align 8
-// CHECK9-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP135]], align 8
-// CHECK9-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP136]], align 8
-// CHECK9-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP137]], align 8
-// CHECK9-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP129]], ptr [[TMP138]], align 8
-// CHECK9-NEXT: [[TMP139:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l477.region_id, ptr [[KERNEL_ARGS42]])
-// CHECK9-NEXT: [[TMP140:%.*]] = icmp ne i32 [[TMP139]], 0
-// CHECK9-NEXT: br i1 [[TMP140]], label [[OMP_OFFLOAD_FAILED43:%.*]], label [[OMP_OFFLOAD_CONT44:%.*]]
+// CHECK9-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP142]], align 4
+// CHECK9-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 1
+// CHECK9-NEXT: store i32 4, ptr [[TMP143]], align 4
+// CHECK9-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP136]], ptr [[TMP144]], align 8
+// CHECK9-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP137]], ptr [[TMP145]], align 8
+// CHECK9-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.12, ptr [[TMP146]], align 8
+// CHECK9-NEXT: [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP147]], align 8
+// CHECK9-NEXT: [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP148]], align 8
+// CHECK9-NEXT: [[TMP149:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP149]], align 8
+// CHECK9-NEXT: [[TMP150:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP141]], ptr [[TMP150]], align 8
+// CHECK9-NEXT: [[TMP151:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP151]], align 8
+// CHECK9-NEXT: [[TMP152:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP152]], align 4
+// CHECK9-NEXT: [[TMP153:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP153]], align 4
+// CHECK9-NEXT: [[TMP154:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP154]], align 4
+// CHECK9-NEXT: [[TMP155:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l477.region_id, ptr [[KERNEL_ARGS42]])
+// CHECK9-NEXT: [[TMP156:%.*]] = icmp ne i32 [[TMP155]], 0
+// CHECK9-NEXT: br i1 [[TMP156]], label [[OMP_OFFLOAD_FAILED43:%.*]], label [[OMP_OFFLOAD_CONT44:%.*]]
// CHECK9: omp_offload.failed43:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l477(i64 [[TMP108]], ptr [[TMP109]], ptr [[TMP110]], ptr [[TMP111]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l477(i64 [[TMP120]], ptr [[TMP121]], ptr [[TMP122]], ptr [[TMP123]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT44]]
// CHECK9: omp_offload.cont44:
-// CHECK9-NEXT: [[TMP141:%.*]] = load i32, ptr [[CH]], align 4
-// CHECK9-NEXT: store i32 [[TMP141]], ptr [[CH_CASTED45]], align 4
-// CHECK9-NEXT: [[TMP142:%.*]] = load i64, ptr [[CH_CASTED45]], align 8
-// CHECK9-NEXT: [[TMP143:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP143]], ptr [[N_CASTED46]], align 4
-// CHECK9-NEXT: [[TMP144:%.*]] = load i64, ptr [[N_CASTED46]], align 8
-// CHECK9-NEXT: [[TMP145:%.*]] = load ptr, ptr [[A]], align 8
-// CHECK9-NEXT: [[TMP146:%.*]] = load ptr, ptr [[B]], align 8
-// CHECK9-NEXT: [[TMP147:%.*]] = load ptr, ptr [[C]], align 8
-// CHECK9-NEXT: [[TMP148:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP142]], ptr [[TMP148]], align 8
-// CHECK9-NEXT: [[TMP149:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP142]], ptr [[TMP149]], align 8
-// CHECK9-NEXT: [[TMP150:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP150]], align 8
-// CHECK9-NEXT: [[TMP151:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP144]], ptr [[TMP151]], align 8
-// CHECK9-NEXT: [[TMP152:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP144]], ptr [[TMP152]], align 8
-// CHECK9-NEXT: [[TMP153:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP153]], align 8
-// CHECK9-NEXT: [[TMP154:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP145]], ptr [[TMP154]], align 8
-// CHECK9-NEXT: [[TMP155:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP145]], ptr [[TMP155]], align 8
-// CHECK9-NEXT: [[TMP156:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP156]], align 8
-// CHECK9-NEXT: [[TMP157:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP146]], ptr [[TMP157]], align 8
-// CHECK9-NEXT: [[TMP158:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP146]], ptr [[TMP158]], align 8
-// CHECK9-NEXT: [[TMP159:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP159]], align 8
-// CHECK9-NEXT: [[TMP160:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP147]], ptr [[TMP160]], align 8
-// CHECK9-NEXT: [[TMP161:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP147]], ptr [[TMP161]], align 8
-// CHECK9-NEXT: [[TMP162:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 4
-// CHECK9-NEXT: store ptr null, ptr [[TMP162]], align 8
-// CHECK9-NEXT: [[TMP163:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP164:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP165:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP165]], ptr [[DOTCAPTURE_EXPR_51]], align 4
-// CHECK9-NEXT: [[TMP166:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_51]], align 4
-// CHECK9-NEXT: [[SUB53:%.*]] = sub nsw i32 [[TMP166]], 0
+// CHECK9-NEXT: [[TMP157:%.*]] = load i32, ptr [[CH]], align 4
+// CHECK9-NEXT: store i32 [[TMP157]], ptr [[CH_CASTED45]], align 4
+// CHECK9-NEXT: [[TMP158:%.*]] = load i64, ptr [[CH_CASTED45]], align 8
+// CHECK9-NEXT: [[TMP159:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP159]], ptr [[N_CASTED46]], align 4
+// CHECK9-NEXT: [[TMP160:%.*]] = load i64, ptr [[N_CASTED46]], align 8
+// CHECK9-NEXT: [[TMP161:%.*]] = load ptr, ptr [[A]], align 8
+// CHECK9-NEXT: [[TMP162:%.*]] = load ptr, ptr [[B]], align 8
+// CHECK9-NEXT: [[TMP163:%.*]] = load ptr, ptr [[C]], align 8
+// CHECK9-NEXT: [[TMP164:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP158]], ptr [[TMP164]], align 8
+// CHECK9-NEXT: [[TMP165:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP158]], ptr [[TMP165]], align 8
+// CHECK9-NEXT: [[TMP166:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP166]], align 8
+// CHECK9-NEXT: [[TMP167:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP160]], ptr [[TMP167]], align 8
+// CHECK9-NEXT: [[TMP168:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP160]], ptr [[TMP168]], align 8
+// CHECK9-NEXT: [[TMP169:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP169]], align 8
+// CHECK9-NEXT: [[TMP170:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP161]], ptr [[TMP170]], align 8
+// CHECK9-NEXT: [[TMP171:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP161]], ptr [[TMP171]], align 8
+// CHECK9-NEXT: [[TMP172:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP172]], align 8
+// CHECK9-NEXT: [[TMP173:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP162]], ptr [[TMP173]], align 8
+// CHECK9-NEXT: [[TMP174:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP162]], ptr [[TMP174]], align 8
+// CHECK9-NEXT: [[TMP175:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP175]], align 8
+// CHECK9-NEXT: [[TMP176:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP163]], ptr [[TMP176]], align 8
+// CHECK9-NEXT: [[TMP177:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP163]], ptr [[TMP177]], align 8
+// CHECK9-NEXT: [[TMP178:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 4
+// CHECK9-NEXT: store ptr null, ptr [[TMP178]], align 8
+// CHECK9-NEXT: [[TMP179:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP180:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP181:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP181]], ptr [[DOTCAPTURE_EXPR_51]], align 4
+// CHECK9-NEXT: [[TMP182:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_51]], align 4
+// CHECK9-NEXT: [[SUB53:%.*]] = sub nsw i32 [[TMP182]], 0
// CHECK9-NEXT: [[DIV54:%.*]] = sdiv i32 [[SUB53]], 1
// CHECK9-NEXT: [[SUB55:%.*]] = sub nsw i32 [[DIV54]], 1
// CHECK9-NEXT: store i32 [[SUB55]], ptr [[DOTCAPTURE_EXPR_52]], align 4
-// CHECK9-NEXT: [[TMP167:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_52]], align 4
-// CHECK9-NEXT: [[ADD56:%.*]] = add nsw i32 [[TMP167]], 1
-// CHECK9-NEXT: [[TMP168:%.*]] = zext i32 [[ADD56]] to i64
+// CHECK9-NEXT: [[TMP183:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_52]], align 4
+// CHECK9-NEXT: [[ADD56:%.*]] = add nsw i32 [[TMP183]], 1
+// CHECK9-NEXT: [[TMP184:%.*]] = zext i32 [[ADD56]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS57:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP169:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP169]], align 4
-// CHECK9-NEXT: [[TMP170:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 1
-// CHECK9-NEXT: store i32 5, ptr [[TMP170]], align 4
-// CHECK9-NEXT: [[TMP171:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP163]], ptr [[TMP171]], align 8
-// CHECK9-NEXT: [[TMP172:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP164]], ptr [[TMP172]], align 8
-// CHECK9-NEXT: [[TMP173:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.16, ptr [[TMP173]], align 8
-// CHECK9-NEXT: [[TMP174:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP174]], align 8
-// CHECK9-NEXT: [[TMP175:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP175]], align 8
-// CHECK9-NEXT: [[TMP176:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP176]], align 8
-// CHECK9-NEXT: [[TMP177:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP168]], ptr [[TMP177]], align 8
-// CHECK9-NEXT: [[TMP178:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l505.region_id, ptr [[KERNEL_ARGS57]])
-// CHECK9-NEXT: [[TMP179:%.*]] = icmp ne i32 [[TMP178]], 0
-// CHECK9-NEXT: br i1 [[TMP179]], label [[OMP_OFFLOAD_FAILED58:%.*]], label [[OMP_OFFLOAD_CONT59:%.*]]
+// CHECK9-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP185]], align 4
+// CHECK9-NEXT: [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 1
+// CHECK9-NEXT: store i32 5, ptr [[TMP186]], align 4
+// CHECK9-NEXT: [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP179]], ptr [[TMP187]], align 8
+// CHECK9-NEXT: [[TMP188:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP180]], ptr [[TMP188]], align 8
+// CHECK9-NEXT: [[TMP189:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.16, ptr [[TMP189]], align 8
+// CHECK9-NEXT: [[TMP190:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP190]], align 8
+// CHECK9-NEXT: [[TMP191:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP191]], align 8
+// CHECK9-NEXT: [[TMP192:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP192]], align 8
+// CHECK9-NEXT: [[TMP193:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP184]], ptr [[TMP193]], align 8
+// CHECK9-NEXT: [[TMP194:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP194]], align 8
+// CHECK9-NEXT: [[TMP195:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP195]], align 4
+// CHECK9-NEXT: [[TMP196:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP196]], align 4
+// CHECK9-NEXT: [[TMP197:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP197]], align 4
+// CHECK9-NEXT: [[TMP198:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l505.region_id, ptr [[KERNEL_ARGS57]])
+// CHECK9-NEXT: [[TMP199:%.*]] = icmp ne i32 [[TMP198]], 0
+// CHECK9-NEXT: br i1 [[TMP199]], label [[OMP_OFFLOAD_FAILED58:%.*]], label [[OMP_OFFLOAD_CONT59:%.*]]
// CHECK9: omp_offload.failed58:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l505(i64 [[TMP142]], i64 [[TMP144]], ptr [[TMP145]], ptr [[TMP146]], ptr [[TMP147]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l505(i64 [[TMP158]], i64 [[TMP160]], ptr [[TMP161]], ptr [[TMP162]], ptr [[TMP163]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT59]]
// CHECK9: omp_offload.cont59:
-// CHECK9-NEXT: [[TMP180:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP180]], ptr [[N_CASTED60]], align 4
-// CHECK9-NEXT: [[TMP181:%.*]] = load i64, ptr [[N_CASTED60]], align 8
-// CHECK9-NEXT: [[TMP182:%.*]] = load ptr, ptr [[A]], align 8
-// CHECK9-NEXT: [[TMP183:%.*]] = load ptr, ptr [[B]], align 8
-// CHECK9-NEXT: [[TMP184:%.*]] = load ptr, ptr [[C]], align 8
-// CHECK9-NEXT: [[TMP185:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP181]], ptr [[TMP185]], align 8
-// CHECK9-NEXT: [[TMP186:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP181]], ptr [[TMP186]], align 8
-// CHECK9-NEXT: [[TMP187:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP187]], align 8
-// CHECK9-NEXT: [[TMP188:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 1
-// CHECK9-NEXT: store ptr [[TMP182]], ptr [[TMP188]], align 8
-// CHECK9-NEXT: [[TMP189:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 1
-// CHECK9-NEXT: store ptr [[TMP182]], ptr [[TMP189]], align 8
-// CHECK9-NEXT: [[TMP190:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP190]], align 8
-// CHECK9-NEXT: [[TMP191:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP183]], ptr [[TMP191]], align 8
-// CHECK9-NEXT: [[TMP192:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP183]], ptr [[TMP192]], align 8
-// CHECK9-NEXT: [[TMP193:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP193]], align 8
-// CHECK9-NEXT: [[TMP194:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP184]], ptr [[TMP194]], align 8
-// CHECK9-NEXT: [[TMP195:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP184]], ptr [[TMP195]], align 8
-// CHECK9-NEXT: [[TMP196:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP196]], align 8
-// CHECK9-NEXT: [[TMP197:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP198:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP199:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP199]], ptr [[DOTCAPTURE_EXPR_65]], align 4
-// CHECK9-NEXT: [[TMP200:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_65]], align 4
-// CHECK9-NEXT: [[SUB67:%.*]] = sub nsw i32 [[TMP200]], 0
+// CHECK9-NEXT: [[TMP200:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP200]], ptr [[N_CASTED60]], align 4
+// CHECK9-NEXT: [[TMP201:%.*]] = load i64, ptr [[N_CASTED60]], align 8
+// CHECK9-NEXT: [[TMP202:%.*]] = load ptr, ptr [[A]], align 8
+// CHECK9-NEXT: [[TMP203:%.*]] = load ptr, ptr [[B]], align 8
+// CHECK9-NEXT: [[TMP204:%.*]] = load ptr, ptr [[C]], align 8
+// CHECK9-NEXT: [[TMP205:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP201]], ptr [[TMP205]], align 8
+// CHECK9-NEXT: [[TMP206:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP201]], ptr [[TMP206]], align 8
+// CHECK9-NEXT: [[TMP207:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP207]], align 8
+// CHECK9-NEXT: [[TMP208:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 1
+// CHECK9-NEXT: store ptr [[TMP202]], ptr [[TMP208]], align 8
+// CHECK9-NEXT: [[TMP209:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 1
+// CHECK9-NEXT: store ptr [[TMP202]], ptr [[TMP209]], align 8
+// CHECK9-NEXT: [[TMP210:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP210]], align 8
+// CHECK9-NEXT: [[TMP211:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP203]], ptr [[TMP211]], align 8
+// CHECK9-NEXT: [[TMP212:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP203]], ptr [[TMP212]], align 8
+// CHECK9-NEXT: [[TMP213:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP213]], align 8
+// CHECK9-NEXT: [[TMP214:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP204]], ptr [[TMP214]], align 8
+// CHECK9-NEXT: [[TMP215:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP204]], ptr [[TMP215]], align 8
+// CHECK9-NEXT: [[TMP216:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP216]], align 8
+// CHECK9-NEXT: [[TMP217:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP218:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP219:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP219]], ptr [[DOTCAPTURE_EXPR_65]], align 4
+// CHECK9-NEXT: [[TMP220:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_65]], align 4
+// CHECK9-NEXT: [[SUB67:%.*]] = sub nsw i32 [[TMP220]], 0
// CHECK9-NEXT: [[DIV68:%.*]] = sdiv i32 [[SUB67]], 1
// CHECK9-NEXT: [[SUB69:%.*]] = sub nsw i32 [[DIV68]], 1
// CHECK9-NEXT: store i32 [[SUB69]], ptr [[DOTCAPTURE_EXPR_66]], align 4
-// CHECK9-NEXT: [[TMP201:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_66]], align 4
-// CHECK9-NEXT: [[ADD70:%.*]] = add nsw i32 [[TMP201]], 1
-// CHECK9-NEXT: [[TMP202:%.*]] = zext i32 [[ADD70]] to i64
+// CHECK9-NEXT: [[TMP221:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_66]], align 4
+// CHECK9-NEXT: [[ADD70:%.*]] = add nsw i32 [[TMP221]], 1
+// CHECK9-NEXT: [[TMP222:%.*]] = zext i32 [[ADD70]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS71:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP203:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP203]], align 4
-// CHECK9-NEXT: [[TMP204:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 1
-// CHECK9-NEXT: store i32 4, ptr [[TMP204]], align 4
-// CHECK9-NEXT: [[TMP205:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP197]], ptr [[TMP205]], align 8
-// CHECK9-NEXT: [[TMP206:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP198]], ptr [[TMP206]], align 8
-// CHECK9-NEXT: [[TMP207:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.20, ptr [[TMP207]], align 8
-// CHECK9-NEXT: [[TMP208:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.21, ptr [[TMP208]], align 8
-// CHECK9-NEXT: [[TMP209:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP209]], align 8
-// CHECK9-NEXT: [[TMP210:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP210]], align 8
-// CHECK9-NEXT: [[TMP211:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP202]], ptr [[TMP211]], align 8
-// CHECK9-NEXT: [[TMP212:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l535.region_id, ptr [[KERNEL_ARGS71]])
-// CHECK9-NEXT: [[TMP213:%.*]] = icmp ne i32 [[TMP212]], 0
-// CHECK9-NEXT: br i1 [[TMP213]], label [[OMP_OFFLOAD_FAILED72:%.*]], label [[OMP_OFFLOAD_CONT73:%.*]]
+// CHECK9-NEXT: [[TMP223:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP223]], align 4
+// CHECK9-NEXT: [[TMP224:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 1
+// CHECK9-NEXT: store i32 4, ptr [[TMP224]], align 4
+// CHECK9-NEXT: [[TMP225:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP217]], ptr [[TMP225]], align 8
+// CHECK9-NEXT: [[TMP226:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP218]], ptr [[TMP226]], align 8
+// CHECK9-NEXT: [[TMP227:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.20, ptr [[TMP227]], align 8
+// CHECK9-NEXT: [[TMP228:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.21, ptr [[TMP228]], align 8
+// CHECK9-NEXT: [[TMP229:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP229]], align 8
+// CHECK9-NEXT: [[TMP230:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP230]], align 8
+// CHECK9-NEXT: [[TMP231:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP222]], ptr [[TMP231]], align 8
+// CHECK9-NEXT: [[TMP232:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP232]], align 8
+// CHECK9-NEXT: [[TMP233:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP233]], align 4
+// CHECK9-NEXT: [[TMP234:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP234]], align 4
+// CHECK9-NEXT: [[TMP235:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP235]], align 4
+// CHECK9-NEXT: [[TMP236:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l535.region_id, ptr [[KERNEL_ARGS71]])
+// CHECK9-NEXT: [[TMP237:%.*]] = icmp ne i32 [[TMP236]], 0
+// CHECK9-NEXT: br i1 [[TMP237]], label [[OMP_OFFLOAD_FAILED72:%.*]], label [[OMP_OFFLOAD_CONT73:%.*]]
// CHECK9: omp_offload.failed72:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l535(i64 [[TMP181]], ptr [[TMP182]], ptr [[TMP183]], ptr [[TMP184]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l535(i64 [[TMP201]], ptr [[TMP202]], ptr [[TMP203]], ptr [[TMP204]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT73]]
// CHECK9: omp_offload.cont73:
-// CHECK9-NEXT: [[TMP214:%.*]] = load i32, ptr [[CH]], align 4
-// CHECK9-NEXT: store i32 [[TMP214]], ptr [[CH_CASTED74]], align 4
-// CHECK9-NEXT: [[TMP215:%.*]] = load i64, ptr [[CH_CASTED74]], align 8
-// CHECK9-NEXT: [[TMP216:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP216]], ptr [[N_CASTED75]], align 4
-// CHECK9-NEXT: [[TMP217:%.*]] = load i64, ptr [[N_CASTED75]], align 8
-// CHECK9-NEXT: [[TMP218:%.*]] = load ptr, ptr [[A]], align 8
-// CHECK9-NEXT: [[TMP219:%.*]] = load ptr, ptr [[B]], align 8
-// CHECK9-NEXT: [[TMP220:%.*]] = load ptr, ptr [[C]], align 8
-// CHECK9-NEXT: [[TMP221:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP215]], ptr [[TMP221]], align 8
-// CHECK9-NEXT: [[TMP222:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP215]], ptr [[TMP222]], align 8
-// CHECK9-NEXT: [[TMP223:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP223]], align 8
-// CHECK9-NEXT: [[TMP224:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP217]], ptr [[TMP224]], align 8
-// CHECK9-NEXT: [[TMP225:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP217]], ptr [[TMP225]], align 8
-// CHECK9-NEXT: [[TMP226:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP226]], align 8
-// CHECK9-NEXT: [[TMP227:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP218]], ptr [[TMP227]], align 8
-// CHECK9-NEXT: [[TMP228:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP218]], ptr [[TMP228]], align 8
-// CHECK9-NEXT: [[TMP229:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP229]], align 8
-// CHECK9-NEXT: [[TMP230:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP219]], ptr [[TMP230]], align 8
-// CHECK9-NEXT: [[TMP231:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP219]], ptr [[TMP231]], align 8
-// CHECK9-NEXT: [[TMP232:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP232]], align 8
-// CHECK9-NEXT: [[TMP233:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP220]], ptr [[TMP233]], align 8
-// CHECK9-NEXT: [[TMP234:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP220]], ptr [[TMP234]], align 8
-// CHECK9-NEXT: [[TMP235:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 4
-// CHECK9-NEXT: store ptr null, ptr [[TMP235]], align 8
-// CHECK9-NEXT: [[TMP236:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP237:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP238:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP238]], ptr [[DOTCAPTURE_EXPR_80]], align 4
-// CHECK9-NEXT: [[TMP239:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_80]], align 4
-// CHECK9-NEXT: [[SUB82:%.*]] = sub nsw i32 [[TMP239]], 0
+// CHECK9-NEXT: [[TMP238:%.*]] = load i32, ptr [[CH]], align 4
+// CHECK9-NEXT: store i32 [[TMP238]], ptr [[CH_CASTED74]], align 4
+// CHECK9-NEXT: [[TMP239:%.*]] = load i64, ptr [[CH_CASTED74]], align 8
+// CHECK9-NEXT: [[TMP240:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP240]], ptr [[N_CASTED75]], align 4
+// CHECK9-NEXT: [[TMP241:%.*]] = load i64, ptr [[N_CASTED75]], align 8
+// CHECK9-NEXT: [[TMP242:%.*]] = load ptr, ptr [[A]], align 8
+// CHECK9-NEXT: [[TMP243:%.*]] = load ptr, ptr [[B]], align 8
+// CHECK9-NEXT: [[TMP244:%.*]] = load ptr, ptr [[C]], align 8
+// CHECK9-NEXT: [[TMP245:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP239]], ptr [[TMP245]], align 8
+// CHECK9-NEXT: [[TMP246:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP239]], ptr [[TMP246]], align 8
+// CHECK9-NEXT: [[TMP247:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP247]], align 8
+// CHECK9-NEXT: [[TMP248:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP241]], ptr [[TMP248]], align 8
+// CHECK9-NEXT: [[TMP249:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP241]], ptr [[TMP249]], align 8
+// CHECK9-NEXT: [[TMP250:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP250]], align 8
+// CHECK9-NEXT: [[TMP251:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP242]], ptr [[TMP251]], align 8
+// CHECK9-NEXT: [[TMP252:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP242]], ptr [[TMP252]], align 8
+// CHECK9-NEXT: [[TMP253:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP253]], align 8
+// CHECK9-NEXT: [[TMP254:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP243]], ptr [[TMP254]], align 8
+// CHECK9-NEXT: [[TMP255:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP243]], ptr [[TMP255]], align 8
+// CHECK9-NEXT: [[TMP256:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP256]], align 8
+// CHECK9-NEXT: [[TMP257:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP244]], ptr [[TMP257]], align 8
+// CHECK9-NEXT: [[TMP258:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP244]], ptr [[TMP258]], align 8
+// CHECK9-NEXT: [[TMP259:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 4
+// CHECK9-NEXT: store ptr null, ptr [[TMP259]], align 8
+// CHECK9-NEXT: [[TMP260:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP261:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP262:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP262]], ptr [[DOTCAPTURE_EXPR_80]], align 4
+// CHECK9-NEXT: [[TMP263:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_80]], align 4
+// CHECK9-NEXT: [[SUB82:%.*]] = sub nsw i32 [[TMP263]], 0
// CHECK9-NEXT: [[DIV83:%.*]] = sdiv i32 [[SUB82]], 1
// CHECK9-NEXT: [[SUB84:%.*]] = sub nsw i32 [[DIV83]], 1
// CHECK9-NEXT: store i32 [[SUB84]], ptr [[DOTCAPTURE_EXPR_81]], align 4
-// CHECK9-NEXT: [[TMP240:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_81]], align 4
-// CHECK9-NEXT: [[ADD85:%.*]] = add nsw i32 [[TMP240]], 1
-// CHECK9-NEXT: [[TMP241:%.*]] = zext i32 [[ADD85]] to i64
+// CHECK9-NEXT: [[TMP264:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_81]], align 4
+// CHECK9-NEXT: [[ADD85:%.*]] = add nsw i32 [[TMP264]], 1
+// CHECK9-NEXT: [[TMP265:%.*]] = zext i32 [[ADD85]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS86:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP242:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP242]], align 4
-// CHECK9-NEXT: [[TMP243:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 1
-// CHECK9-NEXT: store i32 5, ptr [[TMP243]], align 4
-// CHECK9-NEXT: [[TMP244:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP236]], ptr [[TMP244]], align 8
-// CHECK9-NEXT: [[TMP245:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP237]], ptr [[TMP245]], align 8
-// CHECK9-NEXT: [[TMP246:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.24, ptr [[TMP246]], align 8
-// CHECK9-NEXT: [[TMP247:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP247]], align 8
-// CHECK9-NEXT: [[TMP248:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP248]], align 8
-// CHECK9-NEXT: [[TMP249:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP249]], align 8
-// CHECK9-NEXT: [[TMP250:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP241]], ptr [[TMP250]], align 8
-// CHECK9-NEXT: [[TMP251:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l561.region_id, ptr [[KERNEL_ARGS86]])
-// CHECK9-NEXT: [[TMP252:%.*]] = icmp ne i32 [[TMP251]], 0
-// CHECK9-NEXT: br i1 [[TMP252]], label [[OMP_OFFLOAD_FAILED87:%.*]], label [[OMP_OFFLOAD_CONT88:%.*]]
+// CHECK9-NEXT: [[TMP266:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP266]], align 4
+// CHECK9-NEXT: [[TMP267:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 1
+// CHECK9-NEXT: store i32 5, ptr [[TMP267]], align 4
+// CHECK9-NEXT: [[TMP268:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP260]], ptr [[TMP268]], align 8
+// CHECK9-NEXT: [[TMP269:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP261]], ptr [[TMP269]], align 8
+// CHECK9-NEXT: [[TMP270:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.24, ptr [[TMP270]], align 8
+// CHECK9-NEXT: [[TMP271:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP271]], align 8
+// CHECK9-NEXT: [[TMP272:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP272]], align 8
+// CHECK9-NEXT: [[TMP273:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP273]], align 8
+// CHECK9-NEXT: [[TMP274:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP265]], ptr [[TMP274]], align 8
+// CHECK9-NEXT: [[TMP275:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP275]], align 8
+// CHECK9-NEXT: [[TMP276:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP276]], align 4
+// CHECK9-NEXT: [[TMP277:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP277]], align 4
+// CHECK9-NEXT: [[TMP278:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP278]], align 4
+// CHECK9-NEXT: [[TMP279:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l561.region_id, ptr [[KERNEL_ARGS86]])
+// CHECK9-NEXT: [[TMP280:%.*]] = icmp ne i32 [[TMP279]], 0
+// CHECK9-NEXT: br i1 [[TMP280]], label [[OMP_OFFLOAD_FAILED87:%.*]], label [[OMP_OFFLOAD_CONT88:%.*]]
// CHECK9: omp_offload.failed87:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l561(i64 [[TMP215]], i64 [[TMP217]], ptr [[TMP218]], ptr [[TMP219]], ptr [[TMP220]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l561(i64 [[TMP239]], i64 [[TMP241]], ptr [[TMP242]], ptr [[TMP243]], ptr [[TMP244]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT88]]
// CHECK9: omp_offload.cont88:
// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiET_v()
@@ -7181,7 +7237,7 @@ int main() {
// CHECK9-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 4, ptr [[TMP24]], align 4
// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -7198,446 +7254,502 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK9-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l42.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK9-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP34]], align 4
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l42.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK9-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l42(i64 [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK9-NEXT: [[TMP35:%.*]] = load i64, ptr [[N_CASTED3]], align 8
-// CHECK9-NEXT: [[TMP36:%.*]] = load ptr, ptr [[A]], align 8
-// CHECK9-NEXT: [[TMP37:%.*]] = load ptr, ptr [[B]], align 8
-// CHECK9-NEXT: [[TMP38:%.*]] = load ptr, ptr [[C]], align 8
-// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP35]], ptr [[TMP39]], align 8
-// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP35]], ptr [[TMP40]], align 8
-// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP41]], align 8
-// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK9-NEXT: store ptr [[TMP36]], ptr [[TMP42]], align 8
-// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK9-NEXT: store ptr [[TMP36]], ptr [[TMP43]], align 8
-// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP44]], align 8
-// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP37]], ptr [[TMP45]], align 8
-// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP37]], ptr [[TMP46]], align 8
-// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP47]], align 8
-// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP38]], ptr [[TMP48]], align 8
-// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP38]], ptr [[TMP49]], align 8
-// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP50]], align 8
-// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP53:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP53]], ptr [[DOTCAPTURE_EXPR_8]], align 4
-// CHECK9-NEXT: [[TMP54:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_8]], align 4
-// CHECK9-NEXT: [[SUB10:%.*]] = sub nsw i32 [[TMP54]], 0
+// CHECK9-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK9-NEXT: [[TMP39:%.*]] = load i64, ptr [[N_CASTED3]], align 8
+// CHECK9-NEXT: [[TMP40:%.*]] = load ptr, ptr [[A]], align 8
+// CHECK9-NEXT: [[TMP41:%.*]] = load ptr, ptr [[B]], align 8
+// CHECK9-NEXT: [[TMP42:%.*]] = load ptr, ptr [[C]], align 8
+// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP39]], ptr [[TMP43]], align 8
+// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP39]], ptr [[TMP44]], align 8
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK9-NEXT: store ptr [[TMP40]], ptr [[TMP46]], align 8
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK9-NEXT: store ptr [[TMP40]], ptr [[TMP47]], align 8
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP48]], align 8
+// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP41]], ptr [[TMP49]], align 8
+// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP41]], ptr [[TMP50]], align 8
+// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP51]], align 8
+// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP42]], ptr [[TMP52]], align 8
+// CHECK9-NEXT: [[TMP53:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP42]], ptr [[TMP53]], align 8
+// CHECK9-NEXT: [[TMP54:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP54]], align 8
+// CHECK9-NEXT: [[TMP55:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP56:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP57:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP57]], ptr [[DOTCAPTURE_EXPR_8]], align 4
+// CHECK9-NEXT: [[TMP58:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_8]], align 4
+// CHECK9-NEXT: [[SUB10:%.*]] = sub nsw i32 [[TMP58]], 0
// CHECK9-NEXT: [[DIV11:%.*]] = sdiv i32 [[SUB10]], 1
// CHECK9-NEXT: [[SUB12:%.*]] = sub nsw i32 [[DIV11]], 1
// CHECK9-NEXT: store i32 [[SUB12]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK9-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK9-NEXT: [[ADD13:%.*]] = add nsw i32 [[TMP55]], 1
-// CHECK9-NEXT: [[TMP56:%.*]] = zext i32 [[ADD13]] to i64
+// CHECK9-NEXT: [[TMP59:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK9-NEXT: [[ADD13:%.*]] = add nsw i32 [[TMP59]], 1
+// CHECK9-NEXT: [[TMP60:%.*]] = zext i32 [[ADD13]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS14:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP57]], align 4
-// CHECK9-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
-// CHECK9-NEXT: store i32 4, ptr [[TMP58]], align 4
-// CHECK9-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP51]], ptr [[TMP59]], align 8
-// CHECK9-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP52]], ptr [[TMP60]], align 8
-// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.32, ptr [[TMP61]], align 8
-// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP62]], align 8
-// CHECK9-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP63]], align 8
-// CHECK9-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP64]], align 8
-// CHECK9-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP56]], ptr [[TMP65]], align 8
-// CHECK9-NEXT: [[TMP66:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l50.region_id, ptr [[KERNEL_ARGS14]])
-// CHECK9-NEXT: [[TMP67:%.*]] = icmp ne i32 [[TMP66]], 0
-// CHECK9-NEXT: br i1 [[TMP67]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
+// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP61]], align 4
+// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
+// CHECK9-NEXT: store i32 4, ptr [[TMP62]], align 4
+// CHECK9-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP55]], ptr [[TMP63]], align 8
+// CHECK9-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP56]], ptr [[TMP64]], align 8
+// CHECK9-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.32, ptr [[TMP65]], align 8
+// CHECK9-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP66]], align 8
+// CHECK9-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP67]], align 8
+// CHECK9-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP68]], align 8
+// CHECK9-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP60]], ptr [[TMP69]], align 8
+// CHECK9-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP70]], align 8
+// CHECK9-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP71]], align 4
+// CHECK9-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP72]], align 4
+// CHECK9-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP73]], align 4
+// CHECK9-NEXT: [[TMP74:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l50.region_id, ptr [[KERNEL_ARGS14]])
+// CHECK9-NEXT: [[TMP75:%.*]] = icmp ne i32 [[TMP74]], 0
+// CHECK9-NEXT: br i1 [[TMP75]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
// CHECK9: omp_offload.failed15:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l50(i64 [[TMP35]], ptr [[TMP36]], ptr [[TMP37]], ptr [[TMP38]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l50(i64 [[TMP39]], ptr [[TMP40]], ptr [[TMP41]], ptr [[TMP42]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT16]]
// CHECK9: omp_offload.cont16:
-// CHECK9-NEXT: [[TMP68:%.*]] = load i32, ptr [[CH]], align 4
-// CHECK9-NEXT: store i32 [[TMP68]], ptr [[CH_CASTED]], align 4
-// CHECK9-NEXT: [[TMP69:%.*]] = load i64, ptr [[CH_CASTED]], align 8
-// CHECK9-NEXT: [[TMP70:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP70]], ptr [[N_CASTED17]], align 4
-// CHECK9-NEXT: [[TMP71:%.*]] = load i64, ptr [[N_CASTED17]], align 8
-// CHECK9-NEXT: [[TMP72:%.*]] = load ptr, ptr [[A]], align 8
-// CHECK9-NEXT: [[TMP73:%.*]] = load ptr, ptr [[B]], align 8
-// CHECK9-NEXT: [[TMP74:%.*]] = load ptr, ptr [[C]], align 8
-// CHECK9-NEXT: [[TMP75:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP69]], ptr [[TMP75]], align 8
-// CHECK9-NEXT: [[TMP76:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP69]], ptr [[TMP76]], align 8
-// CHECK9-NEXT: [[TMP77:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP77]], align 8
-// CHECK9-NEXT: [[TMP78:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP71]], ptr [[TMP78]], align 8
-// CHECK9-NEXT: [[TMP79:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP71]], ptr [[TMP79]], align 8
-// CHECK9-NEXT: [[TMP80:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP80]], align 8
-// CHECK9-NEXT: [[TMP81:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP72]], ptr [[TMP81]], align 8
-// CHECK9-NEXT: [[TMP82:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP72]], ptr [[TMP82]], align 8
-// CHECK9-NEXT: [[TMP83:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP83]], align 8
-// CHECK9-NEXT: [[TMP84:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP73]], ptr [[TMP84]], align 8
-// CHECK9-NEXT: [[TMP85:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP73]], ptr [[TMP85]], align 8
-// CHECK9-NEXT: [[TMP86:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP86]], align 8
-// CHECK9-NEXT: [[TMP87:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP74]], ptr [[TMP87]], align 8
-// CHECK9-NEXT: [[TMP88:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP74]], ptr [[TMP88]], align 8
-// CHECK9-NEXT: [[TMP89:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 4
-// CHECK9-NEXT: store ptr null, ptr [[TMP89]], align 8
-// CHECK9-NEXT: [[TMP90:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP91:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP92:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP92]], ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK9-NEXT: [[TMP93:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK9-NEXT: [[SUB24:%.*]] = sub nsw i32 [[TMP93]], 0
+// CHECK9-NEXT: [[TMP76:%.*]] = load i32, ptr [[CH]], align 4
+// CHECK9-NEXT: store i32 [[TMP76]], ptr [[CH_CASTED]], align 4
+// CHECK9-NEXT: [[TMP77:%.*]] = load i64, ptr [[CH_CASTED]], align 8
+// CHECK9-NEXT: [[TMP78:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP78]], ptr [[N_CASTED17]], align 4
+// CHECK9-NEXT: [[TMP79:%.*]] = load i64, ptr [[N_CASTED17]], align 8
+// CHECK9-NEXT: [[TMP80:%.*]] = load ptr, ptr [[A]], align 8
+// CHECK9-NEXT: [[TMP81:%.*]] = load ptr, ptr [[B]], align 8
+// CHECK9-NEXT: [[TMP82:%.*]] = load ptr, ptr [[C]], align 8
+// CHECK9-NEXT: [[TMP83:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP77]], ptr [[TMP83]], align 8
+// CHECK9-NEXT: [[TMP84:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP77]], ptr [[TMP84]], align 8
+// CHECK9-NEXT: [[TMP85:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP85]], align 8
+// CHECK9-NEXT: [[TMP86:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP79]], ptr [[TMP86]], align 8
+// CHECK9-NEXT: [[TMP87:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP79]], ptr [[TMP87]], align 8
+// CHECK9-NEXT: [[TMP88:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP88]], align 8
+// CHECK9-NEXT: [[TMP89:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP80]], ptr [[TMP89]], align 8
+// CHECK9-NEXT: [[TMP90:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP80]], ptr [[TMP90]], align 8
+// CHECK9-NEXT: [[TMP91:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP91]], align 8
+// CHECK9-NEXT: [[TMP92:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP81]], ptr [[TMP92]], align 8
+// CHECK9-NEXT: [[TMP93:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP81]], ptr [[TMP93]], align 8
+// CHECK9-NEXT: [[TMP94:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP94]], align 8
+// CHECK9-NEXT: [[TMP95:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP82]], ptr [[TMP95]], align 8
+// CHECK9-NEXT: [[TMP96:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP82]], ptr [[TMP96]], align 8
+// CHECK9-NEXT: [[TMP97:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 4
+// CHECK9-NEXT: store ptr null, ptr [[TMP97]], align 8
+// CHECK9-NEXT: [[TMP98:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP99:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP100:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP100]], ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK9-NEXT: [[TMP101:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK9-NEXT: [[SUB24:%.*]] = sub nsw i32 [[TMP101]], 0
// CHECK9-NEXT: [[DIV25:%.*]] = sdiv i32 [[SUB24]], 1
// CHECK9-NEXT: [[SUB26:%.*]] = sub nsw i32 [[DIV25]], 1
// CHECK9-NEXT: store i32 [[SUB26]], ptr [[DOTCAPTURE_EXPR_23]], align 4
-// CHECK9-NEXT: [[TMP94:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_23]], align 4
-// CHECK9-NEXT: [[ADD27:%.*]] = add nsw i32 [[TMP94]], 1
-// CHECK9-NEXT: [[TMP95:%.*]] = zext i32 [[ADD27]] to i64
+// CHECK9-NEXT: [[TMP102:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_23]], align 4
+// CHECK9-NEXT: [[ADD27:%.*]] = add nsw i32 [[TMP102]], 1
+// CHECK9-NEXT: [[TMP103:%.*]] = zext i32 [[ADD27]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS28:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP96]], align 4
-// CHECK9-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
-// CHECK9-NEXT: store i32 5, ptr [[TMP97]], align 4
-// CHECK9-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP90]], ptr [[TMP98]], align 8
-// CHECK9-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP91]], ptr [[TMP99]], align 8
-// CHECK9-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.36, ptr [[TMP100]], align 8
-// CHECK9-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP101]], align 8
-// CHECK9-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP102]], align 8
-// CHECK9-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP103]], align 8
-// CHECK9-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP95]], ptr [[TMP104]], align 8
-// CHECK9-NEXT: [[TMP105:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l58.region_id, ptr [[KERNEL_ARGS28]])
-// CHECK9-NEXT: [[TMP106:%.*]] = icmp ne i32 [[TMP105]], 0
-// CHECK9-NEXT: br i1 [[TMP106]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
+// CHECK9-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP104]], align 4
+// CHECK9-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
+// CHECK9-NEXT: store i32 5, ptr [[TMP105]], align 4
+// CHECK9-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP98]], ptr [[TMP106]], align 8
+// CHECK9-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP99]], ptr [[TMP107]], align 8
+// CHECK9-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.36, ptr [[TMP108]], align 8
+// CHECK9-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP109]], align 8
+// CHECK9-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP110]], align 8
+// CHECK9-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP111]], align 8
+// CHECK9-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP103]], ptr [[TMP112]], align 8
+// CHECK9-NEXT: [[TMP113:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP113]], align 8
+// CHECK9-NEXT: [[TMP114:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP114]], align 4
+// CHECK9-NEXT: [[TMP115:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP115]], align 4
+// CHECK9-NEXT: [[TMP116:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP116]], align 4
+// CHECK9-NEXT: [[TMP117:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l58.region_id, ptr [[KERNEL_ARGS28]])
+// CHECK9-NEXT: [[TMP118:%.*]] = icmp ne i32 [[TMP117]], 0
+// CHECK9-NEXT: br i1 [[TMP118]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
// CHECK9: omp_offload.failed29:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l58(i64 [[TMP69]], i64 [[TMP71]], ptr [[TMP72]], ptr [[TMP73]], ptr [[TMP74]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l58(i64 [[TMP77]], i64 [[TMP79]], ptr [[TMP80]], ptr [[TMP81]], ptr [[TMP82]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT30]]
// CHECK9: omp_offload.cont30:
-// CHECK9-NEXT: [[TMP107:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP107]], ptr [[N_CASTED31]], align 4
-// CHECK9-NEXT: [[TMP108:%.*]] = load i64, ptr [[N_CASTED31]], align 8
-// CHECK9-NEXT: [[TMP109:%.*]] = load ptr, ptr [[A]], align 8
-// CHECK9-NEXT: [[TMP110:%.*]] = load ptr, ptr [[B]], align 8
-// CHECK9-NEXT: [[TMP111:%.*]] = load ptr, ptr [[C]], align 8
-// CHECK9-NEXT: [[TMP112:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP108]], ptr [[TMP112]], align 8
-// CHECK9-NEXT: [[TMP113:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP108]], ptr [[TMP113]], align 8
-// CHECK9-NEXT: [[TMP114:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP114]], align 8
-// CHECK9-NEXT: [[TMP115:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 1
-// CHECK9-NEXT: store ptr [[TMP109]], ptr [[TMP115]], align 8
-// CHECK9-NEXT: [[TMP116:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 1
-// CHECK9-NEXT: store ptr [[TMP109]], ptr [[TMP116]], align 8
-// CHECK9-NEXT: [[TMP117:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP117]], align 8
-// CHECK9-NEXT: [[TMP118:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP110]], ptr [[TMP118]], align 8
-// CHECK9-NEXT: [[TMP119:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP110]], ptr [[TMP119]], align 8
-// CHECK9-NEXT: [[TMP120:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP120]], align 8
-// CHECK9-NEXT: [[TMP121:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP111]], ptr [[TMP121]], align 8
-// CHECK9-NEXT: [[TMP122:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP111]], ptr [[TMP122]], align 8
-// CHECK9-NEXT: [[TMP123:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP123]], align 8
+// CHECK9-NEXT: [[TMP119:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP119]], ptr [[N_CASTED31]], align 4
+// CHECK9-NEXT: [[TMP120:%.*]] = load i64, ptr [[N_CASTED31]], align 8
+// CHECK9-NEXT: [[TMP121:%.*]] = load ptr, ptr [[A]], align 8
+// CHECK9-NEXT: [[TMP122:%.*]] = load ptr, ptr [[B]], align 8
+// CHECK9-NEXT: [[TMP123:%.*]] = load ptr, ptr [[C]], align 8
// CHECK9-NEXT: [[TMP124:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP120]], ptr [[TMP124]], align 8
// CHECK9-NEXT: [[TMP125:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP126:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP126]], ptr [[DOTCAPTURE_EXPR_36]], align 4
-// CHECK9-NEXT: [[TMP127:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_36]], align 4
-// CHECK9-NEXT: [[SUB38:%.*]] = sub nsw i32 [[TMP127]], 0
+// CHECK9-NEXT: store i64 [[TMP120]], ptr [[TMP125]], align 8
+// CHECK9-NEXT: [[TMP126:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP126]], align 8
+// CHECK9-NEXT: [[TMP127:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 1
+// CHECK9-NEXT: store ptr [[TMP121]], ptr [[TMP127]], align 8
+// CHECK9-NEXT: [[TMP128:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 1
+// CHECK9-NEXT: store ptr [[TMP121]], ptr [[TMP128]], align 8
+// CHECK9-NEXT: [[TMP129:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP129]], align 8
+// CHECK9-NEXT: [[TMP130:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP122]], ptr [[TMP130]], align 8
+// CHECK9-NEXT: [[TMP131:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP122]], ptr [[TMP131]], align 8
+// CHECK9-NEXT: [[TMP132:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP132]], align 8
+// CHECK9-NEXT: [[TMP133:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP123]], ptr [[TMP133]], align 8
+// CHECK9-NEXT: [[TMP134:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP123]], ptr [[TMP134]], align 8
+// CHECK9-NEXT: [[TMP135:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP135]], align 8
+// CHECK9-NEXT: [[TMP136:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP137:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP138:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP138]], ptr [[DOTCAPTURE_EXPR_36]], align 4
+// CHECK9-NEXT: [[TMP139:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_36]], align 4
+// CHECK9-NEXT: [[SUB38:%.*]] = sub nsw i32 [[TMP139]], 0
// CHECK9-NEXT: [[DIV39:%.*]] = sdiv i32 [[SUB38]], 1
// CHECK9-NEXT: [[SUB40:%.*]] = sub nsw i32 [[DIV39]], 1
// CHECK9-NEXT: store i32 [[SUB40]], ptr [[DOTCAPTURE_EXPR_37]], align 4
-// CHECK9-NEXT: [[TMP128:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_37]], align 4
-// CHECK9-NEXT: [[ADD41:%.*]] = add nsw i32 [[TMP128]], 1
-// CHECK9-NEXT: [[TMP129:%.*]] = zext i32 [[ADD41]] to i64
+// CHECK9-NEXT: [[TMP140:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_37]], align 4
+// CHECK9-NEXT: [[ADD41:%.*]] = add nsw i32 [[TMP140]], 1
+// CHECK9-NEXT: [[TMP141:%.*]] = zext i32 [[ADD41]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS42:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP130]], align 4
-// CHECK9-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 1
-// CHECK9-NEXT: store i32 4, ptr [[TMP131]], align 4
-// CHECK9-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP124]], ptr [[TMP132]], align 8
-// CHECK9-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP125]], ptr [[TMP133]], align 8
-// CHECK9-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.40, ptr [[TMP134]], align 8
-// CHECK9-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.41, ptr [[TMP135]], align 8
-// CHECK9-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP136]], align 8
-// CHECK9-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP137]], align 8
-// CHECK9-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP129]], ptr [[TMP138]], align 8
-// CHECK9-NEXT: [[TMP139:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l66.region_id, ptr [[KERNEL_ARGS42]])
-// CHECK9-NEXT: [[TMP140:%.*]] = icmp ne i32 [[TMP139]], 0
-// CHECK9-NEXT: br i1 [[TMP140]], label [[OMP_OFFLOAD_FAILED43:%.*]], label [[OMP_OFFLOAD_CONT44:%.*]]
+// CHECK9-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP142]], align 4
+// CHECK9-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 1
+// CHECK9-NEXT: store i32 4, ptr [[TMP143]], align 4
+// CHECK9-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP136]], ptr [[TMP144]], align 8
+// CHECK9-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP137]], ptr [[TMP145]], align 8
+// CHECK9-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.40, ptr [[TMP146]], align 8
+// CHECK9-NEXT: [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.41, ptr [[TMP147]], align 8
+// CHECK9-NEXT: [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP148]], align 8
+// CHECK9-NEXT: [[TMP149:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP149]], align 8
+// CHECK9-NEXT: [[TMP150:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP141]], ptr [[TMP150]], align 8
+// CHECK9-NEXT: [[TMP151:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP151]], align 8
+// CHECK9-NEXT: [[TMP152:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP152]], align 4
+// CHECK9-NEXT: [[TMP153:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP153]], align 4
+// CHECK9-NEXT: [[TMP154:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP154]], align 4
+// CHECK9-NEXT: [[TMP155:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l66.region_id, ptr [[KERNEL_ARGS42]])
+// CHECK9-NEXT: [[TMP156:%.*]] = icmp ne i32 [[TMP155]], 0
+// CHECK9-NEXT: br i1 [[TMP156]], label [[OMP_OFFLOAD_FAILED43:%.*]], label [[OMP_OFFLOAD_CONT44:%.*]]
// CHECK9: omp_offload.failed43:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l66(i64 [[TMP108]], ptr [[TMP109]], ptr [[TMP110]], ptr [[TMP111]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l66(i64 [[TMP120]], ptr [[TMP121]], ptr [[TMP122]], ptr [[TMP123]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT44]]
// CHECK9: omp_offload.cont44:
-// CHECK9-NEXT: [[TMP141:%.*]] = load i32, ptr [[CH]], align 4
-// CHECK9-NEXT: store i32 [[TMP141]], ptr [[CH_CASTED45]], align 4
-// CHECK9-NEXT: [[TMP142:%.*]] = load i64, ptr [[CH_CASTED45]], align 8
-// CHECK9-NEXT: [[TMP143:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP143]], ptr [[N_CASTED46]], align 4
-// CHECK9-NEXT: [[TMP144:%.*]] = load i64, ptr [[N_CASTED46]], align 8
-// CHECK9-NEXT: [[TMP145:%.*]] = load ptr, ptr [[A]], align 8
-// CHECK9-NEXT: [[TMP146:%.*]] = load ptr, ptr [[B]], align 8
-// CHECK9-NEXT: [[TMP147:%.*]] = load ptr, ptr [[C]], align 8
-// CHECK9-NEXT: [[TMP148:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP142]], ptr [[TMP148]], align 8
-// CHECK9-NEXT: [[TMP149:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP142]], ptr [[TMP149]], align 8
-// CHECK9-NEXT: [[TMP150:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP150]], align 8
-// CHECK9-NEXT: [[TMP151:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP144]], ptr [[TMP151]], align 8
-// CHECK9-NEXT: [[TMP152:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP144]], ptr [[TMP152]], align 8
-// CHECK9-NEXT: [[TMP153:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP153]], align 8
-// CHECK9-NEXT: [[TMP154:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP145]], ptr [[TMP154]], align 8
-// CHECK9-NEXT: [[TMP155:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP145]], ptr [[TMP155]], align 8
-// CHECK9-NEXT: [[TMP156:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP156]], align 8
-// CHECK9-NEXT: [[TMP157:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP146]], ptr [[TMP157]], align 8
-// CHECK9-NEXT: [[TMP158:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP146]], ptr [[TMP158]], align 8
-// CHECK9-NEXT: [[TMP159:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP159]], align 8
-// CHECK9-NEXT: [[TMP160:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP147]], ptr [[TMP160]], align 8
-// CHECK9-NEXT: [[TMP161:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP147]], ptr [[TMP161]], align 8
-// CHECK9-NEXT: [[TMP162:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 4
-// CHECK9-NEXT: store ptr null, ptr [[TMP162]], align 8
-// CHECK9-NEXT: [[TMP163:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP164:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP165:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP165]], ptr [[DOTCAPTURE_EXPR_51]], align 4
-// CHECK9-NEXT: [[TMP166:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_51]], align 4
-// CHECK9-NEXT: [[SUB53:%.*]] = sub nsw i32 [[TMP166]], 0
+// CHECK9-NEXT: [[TMP157:%.*]] = load i32, ptr [[CH]], align 4
+// CHECK9-NEXT: store i32 [[TMP157]], ptr [[CH_CASTED45]], align 4
+// CHECK9-NEXT: [[TMP158:%.*]] = load i64, ptr [[CH_CASTED45]], align 8
+// CHECK9-NEXT: [[TMP159:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP159]], ptr [[N_CASTED46]], align 4
+// CHECK9-NEXT: [[TMP160:%.*]] = load i64, ptr [[N_CASTED46]], align 8
+// CHECK9-NEXT: [[TMP161:%.*]] = load ptr, ptr [[A]], align 8
+// CHECK9-NEXT: [[TMP162:%.*]] = load ptr, ptr [[B]], align 8
+// CHECK9-NEXT: [[TMP163:%.*]] = load ptr, ptr [[C]], align 8
+// CHECK9-NEXT: [[TMP164:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP158]], ptr [[TMP164]], align 8
+// CHECK9-NEXT: [[TMP165:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP158]], ptr [[TMP165]], align 8
+// CHECK9-NEXT: [[TMP166:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP166]], align 8
+// CHECK9-NEXT: [[TMP167:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP160]], ptr [[TMP167]], align 8
+// CHECK9-NEXT: [[TMP168:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP160]], ptr [[TMP168]], align 8
+// CHECK9-NEXT: [[TMP169:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP169]], align 8
+// CHECK9-NEXT: [[TMP170:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP161]], ptr [[TMP170]], align 8
+// CHECK9-NEXT: [[TMP171:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP161]], ptr [[TMP171]], align 8
+// CHECK9-NEXT: [[TMP172:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP172]], align 8
+// CHECK9-NEXT: [[TMP173:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP162]], ptr [[TMP173]], align 8
+// CHECK9-NEXT: [[TMP174:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP162]], ptr [[TMP174]], align 8
+// CHECK9-NEXT: [[TMP175:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP175]], align 8
+// CHECK9-NEXT: [[TMP176:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP163]], ptr [[TMP176]], align 8
+// CHECK9-NEXT: [[TMP177:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP163]], ptr [[TMP177]], align 8
+// CHECK9-NEXT: [[TMP178:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 4
+// CHECK9-NEXT: store ptr null, ptr [[TMP178]], align 8
+// CHECK9-NEXT: [[TMP179:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP180:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP181:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP181]], ptr [[DOTCAPTURE_EXPR_51]], align 4
+// CHECK9-NEXT: [[TMP182:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_51]], align 4
+// CHECK9-NEXT: [[SUB53:%.*]] = sub nsw i32 [[TMP182]], 0
// CHECK9-NEXT: [[DIV54:%.*]] = sdiv i32 [[SUB53]], 1
// CHECK9-NEXT: [[SUB55:%.*]] = sub nsw i32 [[DIV54]], 1
// CHECK9-NEXT: store i32 [[SUB55]], ptr [[DOTCAPTURE_EXPR_52]], align 4
-// CHECK9-NEXT: [[TMP167:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_52]], align 4
-// CHECK9-NEXT: [[ADD56:%.*]] = add nsw i32 [[TMP167]], 1
-// CHECK9-NEXT: [[TMP168:%.*]] = zext i32 [[ADD56]] to i64
+// CHECK9-NEXT: [[TMP183:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_52]], align 4
+// CHECK9-NEXT: [[ADD56:%.*]] = add nsw i32 [[TMP183]], 1
+// CHECK9-NEXT: [[TMP184:%.*]] = zext i32 [[ADD56]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS57:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP169:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP169]], align 4
-// CHECK9-NEXT: [[TMP170:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 1
-// CHECK9-NEXT: store i32 5, ptr [[TMP170]], align 4
-// CHECK9-NEXT: [[TMP171:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP163]], ptr [[TMP171]], align 8
-// CHECK9-NEXT: [[TMP172:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP164]], ptr [[TMP172]], align 8
-// CHECK9-NEXT: [[TMP173:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.44, ptr [[TMP173]], align 8
-// CHECK9-NEXT: [[TMP174:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.45, ptr [[TMP174]], align 8
-// CHECK9-NEXT: [[TMP175:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP175]], align 8
-// CHECK9-NEXT: [[TMP176:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP176]], align 8
-// CHECK9-NEXT: [[TMP177:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP168]], ptr [[TMP177]], align 8
-// CHECK9-NEXT: [[TMP178:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l74.region_id, ptr [[KERNEL_ARGS57]])
-// CHECK9-NEXT: [[TMP179:%.*]] = icmp ne i32 [[TMP178]], 0
-// CHECK9-NEXT: br i1 [[TMP179]], label [[OMP_OFFLOAD_FAILED58:%.*]], label [[OMP_OFFLOAD_CONT59:%.*]]
+// CHECK9-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP185]], align 4
+// CHECK9-NEXT: [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 1
+// CHECK9-NEXT: store i32 5, ptr [[TMP186]], align 4
+// CHECK9-NEXT: [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP179]], ptr [[TMP187]], align 8
+// CHECK9-NEXT: [[TMP188:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP180]], ptr [[TMP188]], align 8
+// CHECK9-NEXT: [[TMP189:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.44, ptr [[TMP189]], align 8
+// CHECK9-NEXT: [[TMP190:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.45, ptr [[TMP190]], align 8
+// CHECK9-NEXT: [[TMP191:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP191]], align 8
+// CHECK9-NEXT: [[TMP192:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP192]], align 8
+// CHECK9-NEXT: [[TMP193:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP184]], ptr [[TMP193]], align 8
+// CHECK9-NEXT: [[TMP194:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP194]], align 8
+// CHECK9-NEXT: [[TMP195:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP195]], align 4
+// CHECK9-NEXT: [[TMP196:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP196]], align 4
+// CHECK9-NEXT: [[TMP197:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP197]], align 4
+// CHECK9-NEXT: [[TMP198:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l74.region_id, ptr [[KERNEL_ARGS57]])
+// CHECK9-NEXT: [[TMP199:%.*]] = icmp ne i32 [[TMP198]], 0
+// CHECK9-NEXT: br i1 [[TMP199]], label [[OMP_OFFLOAD_FAILED58:%.*]], label [[OMP_OFFLOAD_CONT59:%.*]]
// CHECK9: omp_offload.failed58:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l74(i64 [[TMP142]], i64 [[TMP144]], ptr [[TMP145]], ptr [[TMP146]], ptr [[TMP147]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l74(i64 [[TMP158]], i64 [[TMP160]], ptr [[TMP161]], ptr [[TMP162]], ptr [[TMP163]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT59]]
// CHECK9: omp_offload.cont59:
-// CHECK9-NEXT: [[TMP180:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP180]], ptr [[N_CASTED60]], align 4
-// CHECK9-NEXT: [[TMP181:%.*]] = load i64, ptr [[N_CASTED60]], align 8
-// CHECK9-NEXT: [[TMP182:%.*]] = load ptr, ptr [[A]], align 8
-// CHECK9-NEXT: [[TMP183:%.*]] = load ptr, ptr [[B]], align 8
-// CHECK9-NEXT: [[TMP184:%.*]] = load ptr, ptr [[C]], align 8
-// CHECK9-NEXT: [[TMP185:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP181]], ptr [[TMP185]], align 8
-// CHECK9-NEXT: [[TMP186:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP181]], ptr [[TMP186]], align 8
-// CHECK9-NEXT: [[TMP187:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP187]], align 8
-// CHECK9-NEXT: [[TMP188:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 1
-// CHECK9-NEXT: store ptr [[TMP182]], ptr [[TMP188]], align 8
-// CHECK9-NEXT: [[TMP189:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 1
-// CHECK9-NEXT: store ptr [[TMP182]], ptr [[TMP189]], align 8
-// CHECK9-NEXT: [[TMP190:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP190]], align 8
-// CHECK9-NEXT: [[TMP191:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP183]], ptr [[TMP191]], align 8
-// CHECK9-NEXT: [[TMP192:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP183]], ptr [[TMP192]], align 8
-// CHECK9-NEXT: [[TMP193:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP193]], align 8
-// CHECK9-NEXT: [[TMP194:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP184]], ptr [[TMP194]], align 8
-// CHECK9-NEXT: [[TMP195:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP184]], ptr [[TMP195]], align 8
-// CHECK9-NEXT: [[TMP196:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP196]], align 8
-// CHECK9-NEXT: [[TMP197:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP198:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP199:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP199]], ptr [[DOTCAPTURE_EXPR_65]], align 4
-// CHECK9-NEXT: [[TMP200:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_65]], align 4
-// CHECK9-NEXT: [[SUB67:%.*]] = sub nsw i32 [[TMP200]], 0
+// CHECK9-NEXT: [[TMP200:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP200]], ptr [[N_CASTED60]], align 4
+// CHECK9-NEXT: [[TMP201:%.*]] = load i64, ptr [[N_CASTED60]], align 8
+// CHECK9-NEXT: [[TMP202:%.*]] = load ptr, ptr [[A]], align 8
+// CHECK9-NEXT: [[TMP203:%.*]] = load ptr, ptr [[B]], align 8
+// CHECK9-NEXT: [[TMP204:%.*]] = load ptr, ptr [[C]], align 8
+// CHECK9-NEXT: [[TMP205:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP201]], ptr [[TMP205]], align 8
+// CHECK9-NEXT: [[TMP206:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP201]], ptr [[TMP206]], align 8
+// CHECK9-NEXT: [[TMP207:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP207]], align 8
+// CHECK9-NEXT: [[TMP208:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 1
+// CHECK9-NEXT: store ptr [[TMP202]], ptr [[TMP208]], align 8
+// CHECK9-NEXT: [[TMP209:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 1
+// CHECK9-NEXT: store ptr [[TMP202]], ptr [[TMP209]], align 8
+// CHECK9-NEXT: [[TMP210:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP210]], align 8
+// CHECK9-NEXT: [[TMP211:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP203]], ptr [[TMP211]], align 8
+// CHECK9-NEXT: [[TMP212:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP203]], ptr [[TMP212]], align 8
+// CHECK9-NEXT: [[TMP213:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP213]], align 8
+// CHECK9-NEXT: [[TMP214:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP204]], ptr [[TMP214]], align 8
+// CHECK9-NEXT: [[TMP215:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP204]], ptr [[TMP215]], align 8
+// CHECK9-NEXT: [[TMP216:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP216]], align 8
+// CHECK9-NEXT: [[TMP217:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP218:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP219:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP219]], ptr [[DOTCAPTURE_EXPR_65]], align 4
+// CHECK9-NEXT: [[TMP220:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_65]], align 4
+// CHECK9-NEXT: [[SUB67:%.*]] = sub nsw i32 [[TMP220]], 0
// CHECK9-NEXT: [[DIV68:%.*]] = sdiv i32 [[SUB67]], 1
// CHECK9-NEXT: [[SUB69:%.*]] = sub nsw i32 [[DIV68]], 1
// CHECK9-NEXT: store i32 [[SUB69]], ptr [[DOTCAPTURE_EXPR_66]], align 4
-// CHECK9-NEXT: [[TMP201:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_66]], align 4
-// CHECK9-NEXT: [[ADD70:%.*]] = add nsw i32 [[TMP201]], 1
-// CHECK9-NEXT: [[TMP202:%.*]] = zext i32 [[ADD70]] to i64
+// CHECK9-NEXT: [[TMP221:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_66]], align 4
+// CHECK9-NEXT: [[ADD70:%.*]] = add nsw i32 [[TMP221]], 1
+// CHECK9-NEXT: [[TMP222:%.*]] = zext i32 [[ADD70]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS71:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP203:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP203]], align 4
-// CHECK9-NEXT: [[TMP204:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 1
-// CHECK9-NEXT: store i32 4, ptr [[TMP204]], align 4
-// CHECK9-NEXT: [[TMP205:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP197]], ptr [[TMP205]], align 8
-// CHECK9-NEXT: [[TMP206:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP198]], ptr [[TMP206]], align 8
-// CHECK9-NEXT: [[TMP207:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.48, ptr [[TMP207]], align 8
-// CHECK9-NEXT: [[TMP208:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.49, ptr [[TMP208]], align 8
-// CHECK9-NEXT: [[TMP209:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP209]], align 8
-// CHECK9-NEXT: [[TMP210:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP210]], align 8
-// CHECK9-NEXT: [[TMP211:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP202]], ptr [[TMP211]], align 8
-// CHECK9-NEXT: [[TMP212:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l82.region_id, ptr [[KERNEL_ARGS71]])
-// CHECK9-NEXT: [[TMP213:%.*]] = icmp ne i32 [[TMP212]], 0
-// CHECK9-NEXT: br i1 [[TMP213]], label [[OMP_OFFLOAD_FAILED72:%.*]], label [[OMP_OFFLOAD_CONT73:%.*]]
+// CHECK9-NEXT: [[TMP223:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP223]], align 4
+// CHECK9-NEXT: [[TMP224:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 1
+// CHECK9-NEXT: store i32 4, ptr [[TMP224]], align 4
+// CHECK9-NEXT: [[TMP225:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP217]], ptr [[TMP225]], align 8
+// CHECK9-NEXT: [[TMP226:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP218]], ptr [[TMP226]], align 8
+// CHECK9-NEXT: [[TMP227:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.48, ptr [[TMP227]], align 8
+// CHECK9-NEXT: [[TMP228:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.49, ptr [[TMP228]], align 8
+// CHECK9-NEXT: [[TMP229:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP229]], align 8
+// CHECK9-NEXT: [[TMP230:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP230]], align 8
+// CHECK9-NEXT: [[TMP231:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP222]], ptr [[TMP231]], align 8
+// CHECK9-NEXT: [[TMP232:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP232]], align 8
+// CHECK9-NEXT: [[TMP233:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP233]], align 4
+// CHECK9-NEXT: [[TMP234:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP234]], align 4
+// CHECK9-NEXT: [[TMP235:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP235]], align 4
+// CHECK9-NEXT: [[TMP236:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l82.region_id, ptr [[KERNEL_ARGS71]])
+// CHECK9-NEXT: [[TMP237:%.*]] = icmp ne i32 [[TMP236]], 0
+// CHECK9-NEXT: br i1 [[TMP237]], label [[OMP_OFFLOAD_FAILED72:%.*]], label [[OMP_OFFLOAD_CONT73:%.*]]
// CHECK9: omp_offload.failed72:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l82(i64 [[TMP181]], ptr [[TMP182]], ptr [[TMP183]], ptr [[TMP184]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l82(i64 [[TMP201]], ptr [[TMP202]], ptr [[TMP203]], ptr [[TMP204]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT73]]
// CHECK9: omp_offload.cont73:
-// CHECK9-NEXT: [[TMP214:%.*]] = load i32, ptr [[CH]], align 4
-// CHECK9-NEXT: store i32 [[TMP214]], ptr [[CH_CASTED74]], align 4
-// CHECK9-NEXT: [[TMP215:%.*]] = load i64, ptr [[CH_CASTED74]], align 8
-// CHECK9-NEXT: [[TMP216:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP216]], ptr [[N_CASTED75]], align 4
-// CHECK9-NEXT: [[TMP217:%.*]] = load i64, ptr [[N_CASTED75]], align 8
-// CHECK9-NEXT: [[TMP218:%.*]] = load ptr, ptr [[A]], align 8
-// CHECK9-NEXT: [[TMP219:%.*]] = load ptr, ptr [[B]], align 8
-// CHECK9-NEXT: [[TMP220:%.*]] = load ptr, ptr [[C]], align 8
-// CHECK9-NEXT: [[TMP221:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP215]], ptr [[TMP221]], align 8
-// CHECK9-NEXT: [[TMP222:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP215]], ptr [[TMP222]], align 8
-// CHECK9-NEXT: [[TMP223:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP223]], align 8
-// CHECK9-NEXT: [[TMP224:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP217]], ptr [[TMP224]], align 8
-// CHECK9-NEXT: [[TMP225:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP217]], ptr [[TMP225]], align 8
-// CHECK9-NEXT: [[TMP226:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP226]], align 8
-// CHECK9-NEXT: [[TMP227:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP218]], ptr [[TMP227]], align 8
-// CHECK9-NEXT: [[TMP228:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP218]], ptr [[TMP228]], align 8
-// CHECK9-NEXT: [[TMP229:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP229]], align 8
-// CHECK9-NEXT: [[TMP230:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP219]], ptr [[TMP230]], align 8
-// CHECK9-NEXT: [[TMP231:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP219]], ptr [[TMP231]], align 8
-// CHECK9-NEXT: [[TMP232:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP232]], align 8
-// CHECK9-NEXT: [[TMP233:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP220]], ptr [[TMP233]], align 8
-// CHECK9-NEXT: [[TMP234:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP220]], ptr [[TMP234]], align 8
-// CHECK9-NEXT: [[TMP235:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 4
-// CHECK9-NEXT: store ptr null, ptr [[TMP235]], align 8
-// CHECK9-NEXT: [[TMP236:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP237:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP238:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP238]], ptr [[DOTCAPTURE_EXPR_80]], align 4
-// CHECK9-NEXT: [[TMP239:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_80]], align 4
-// CHECK9-NEXT: [[SUB82:%.*]] = sub nsw i32 [[TMP239]], 0
+// CHECK9-NEXT: [[TMP238:%.*]] = load i32, ptr [[CH]], align 4
+// CHECK9-NEXT: store i32 [[TMP238]], ptr [[CH_CASTED74]], align 4
+// CHECK9-NEXT: [[TMP239:%.*]] = load i64, ptr [[CH_CASTED74]], align 8
+// CHECK9-NEXT: [[TMP240:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP240]], ptr [[N_CASTED75]], align 4
+// CHECK9-NEXT: [[TMP241:%.*]] = load i64, ptr [[N_CASTED75]], align 8
+// CHECK9-NEXT: [[TMP242:%.*]] = load ptr, ptr [[A]], align 8
+// CHECK9-NEXT: [[TMP243:%.*]] = load ptr, ptr [[B]], align 8
+// CHECK9-NEXT: [[TMP244:%.*]] = load ptr, ptr [[C]], align 8
+// CHECK9-NEXT: [[TMP245:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP239]], ptr [[TMP245]], align 8
+// CHECK9-NEXT: [[TMP246:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP239]], ptr [[TMP246]], align 8
+// CHECK9-NEXT: [[TMP247:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP247]], align 8
+// CHECK9-NEXT: [[TMP248:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP241]], ptr [[TMP248]], align 8
+// CHECK9-NEXT: [[TMP249:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP241]], ptr [[TMP249]], align 8
+// CHECK9-NEXT: [[TMP250:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP250]], align 8
+// CHECK9-NEXT: [[TMP251:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP242]], ptr [[TMP251]], align 8
+// CHECK9-NEXT: [[TMP252:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP242]], ptr [[TMP252]], align 8
+// CHECK9-NEXT: [[TMP253:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP253]], align 8
+// CHECK9-NEXT: [[TMP254:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP243]], ptr [[TMP254]], align 8
+// CHECK9-NEXT: [[TMP255:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP243]], ptr [[TMP255]], align 8
+// CHECK9-NEXT: [[TMP256:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP256]], align 8
+// CHECK9-NEXT: [[TMP257:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP244]], ptr [[TMP257]], align 8
+// CHECK9-NEXT: [[TMP258:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP244]], ptr [[TMP258]], align 8
+// CHECK9-NEXT: [[TMP259:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 4
+// CHECK9-NEXT: store ptr null, ptr [[TMP259]], align 8
+// CHECK9-NEXT: [[TMP260:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP261:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP262:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP262]], ptr [[DOTCAPTURE_EXPR_80]], align 4
+// CHECK9-NEXT: [[TMP263:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_80]], align 4
+// CHECK9-NEXT: [[SUB82:%.*]] = sub nsw i32 [[TMP263]], 0
// CHECK9-NEXT: [[DIV83:%.*]] = sdiv i32 [[SUB82]], 1
// CHECK9-NEXT: [[SUB84:%.*]] = sub nsw i32 [[DIV83]], 1
// CHECK9-NEXT: store i32 [[SUB84]], ptr [[DOTCAPTURE_EXPR_81]], align 4
-// CHECK9-NEXT: [[TMP240:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_81]], align 4
-// CHECK9-NEXT: [[ADD85:%.*]] = add nsw i32 [[TMP240]], 1
-// CHECK9-NEXT: [[TMP241:%.*]] = zext i32 [[ADD85]] to i64
+// CHECK9-NEXT: [[TMP264:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_81]], align 4
+// CHECK9-NEXT: [[ADD85:%.*]] = add nsw i32 [[TMP264]], 1
+// CHECK9-NEXT: [[TMP265:%.*]] = zext i32 [[ADD85]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS86:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP242:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP242]], align 4
-// CHECK9-NEXT: [[TMP243:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 1
-// CHECK9-NEXT: store i32 5, ptr [[TMP243]], align 4
-// CHECK9-NEXT: [[TMP244:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP236]], ptr [[TMP244]], align 8
-// CHECK9-NEXT: [[TMP245:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP237]], ptr [[TMP245]], align 8
-// CHECK9-NEXT: [[TMP246:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.52, ptr [[TMP246]], align 8
-// CHECK9-NEXT: [[TMP247:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.53, ptr [[TMP247]], align 8
-// CHECK9-NEXT: [[TMP248:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP248]], align 8
-// CHECK9-NEXT: [[TMP249:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP249]], align 8
-// CHECK9-NEXT: [[TMP250:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP241]], ptr [[TMP250]], align 8
-// CHECK9-NEXT: [[TMP251:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l90.region_id, ptr [[KERNEL_ARGS86]])
-// CHECK9-NEXT: [[TMP252:%.*]] = icmp ne i32 [[TMP251]], 0
-// CHECK9-NEXT: br i1 [[TMP252]], label [[OMP_OFFLOAD_FAILED87:%.*]], label [[OMP_OFFLOAD_CONT88:%.*]]
+// CHECK9-NEXT: [[TMP266:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP266]], align 4
+// CHECK9-NEXT: [[TMP267:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 1
+// CHECK9-NEXT: store i32 5, ptr [[TMP267]], align 4
+// CHECK9-NEXT: [[TMP268:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP260]], ptr [[TMP268]], align 8
+// CHECK9-NEXT: [[TMP269:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP261]], ptr [[TMP269]], align 8
+// CHECK9-NEXT: [[TMP270:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.52, ptr [[TMP270]], align 8
+// CHECK9-NEXT: [[TMP271:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.53, ptr [[TMP271]], align 8
+// CHECK9-NEXT: [[TMP272:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP272]], align 8
+// CHECK9-NEXT: [[TMP273:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP273]], align 8
+// CHECK9-NEXT: [[TMP274:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP265]], ptr [[TMP274]], align 8
+// CHECK9-NEXT: [[TMP275:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP275]], align 8
+// CHECK9-NEXT: [[TMP276:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP276]], align 4
+// CHECK9-NEXT: [[TMP277:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP277]], align 4
+// CHECK9-NEXT: [[TMP278:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP278]], align 4
+// CHECK9-NEXT: [[TMP279:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l90.region_id, ptr [[KERNEL_ARGS86]])
+// CHECK9-NEXT: [[TMP280:%.*]] = icmp ne i32 [[TMP279]], 0
+// CHECK9-NEXT: br i1 [[TMP280]], label [[OMP_OFFLOAD_FAILED87:%.*]], label [[OMP_OFFLOAD_CONT88:%.*]]
// CHECK9: omp_offload.failed87:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l90(i64 [[TMP215]], i64 [[TMP217]], ptr [[TMP218]], ptr [[TMP219]], ptr [[TMP220]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l90(i64 [[TMP239]], i64 [[TMP241]], ptr [[TMP242]], ptr [[TMP243]], ptr [[TMP244]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT88]]
// CHECK9: omp_offload.cont88:
// CHECK9-NEXT: ret i32 0
@@ -9600,7 +9712,7 @@ int main() {
// CHECK11-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 4, ptr [[TMP24]], align 4
// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -9617,446 +9729,502 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK11-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l368.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK11-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP34]], align 4
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l368.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK11-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l368(i32 [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]]) #[[ATTR2:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK11-NEXT: [[TMP35:%.*]] = load i32, ptr [[N_CASTED3]], align 4
-// CHECK11-NEXT: [[TMP36:%.*]] = load ptr, ptr [[A]], align 4
-// CHECK11-NEXT: [[TMP37:%.*]] = load ptr, ptr [[B]], align 4
-// CHECK11-NEXT: [[TMP38:%.*]] = load ptr, ptr [[C]], align 4
-// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP35]], ptr [[TMP39]], align 4
-// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP35]], ptr [[TMP40]], align 4
-// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP41]], align 4
-// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK11-NEXT: store ptr [[TMP36]], ptr [[TMP42]], align 4
-// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK11-NEXT: store ptr [[TMP36]], ptr [[TMP43]], align 4
-// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP44]], align 4
-// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP37]], ptr [[TMP45]], align 4
-// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP37]], ptr [[TMP46]], align 4
-// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP47]], align 4
-// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP38]], ptr [[TMP48]], align 4
-// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP38]], ptr [[TMP49]], align 4
-// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP50]], align 4
-// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP53:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP53]], ptr [[DOTCAPTURE_EXPR_8]], align 4
-// CHECK11-NEXT: [[TMP54:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_8]], align 4
-// CHECK11-NEXT: [[SUB10:%.*]] = sub nsw i32 [[TMP54]], 0
+// CHECK11-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK11-NEXT: [[TMP39:%.*]] = load i32, ptr [[N_CASTED3]], align 4
+// CHECK11-NEXT: [[TMP40:%.*]] = load ptr, ptr [[A]], align 4
+// CHECK11-NEXT: [[TMP41:%.*]] = load ptr, ptr [[B]], align 4
+// CHECK11-NEXT: [[TMP42:%.*]] = load ptr, ptr [[C]], align 4
+// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP39]], ptr [[TMP43]], align 4
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP39]], ptr [[TMP44]], align 4
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK11-NEXT: store ptr [[TMP40]], ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK11-NEXT: store ptr [[TMP40]], ptr [[TMP47]], align 4
+// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP48]], align 4
+// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP41]], ptr [[TMP49]], align 4
+// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP41]], ptr [[TMP50]], align 4
+// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP42]], ptr [[TMP52]], align 4
+// CHECK11-NEXT: [[TMP53:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP42]], ptr [[TMP53]], align 4
+// CHECK11-NEXT: [[TMP54:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP54]], align 4
+// CHECK11-NEXT: [[TMP55:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP56:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP57:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP57]], ptr [[DOTCAPTURE_EXPR_8]], align 4
+// CHECK11-NEXT: [[TMP58:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_8]], align 4
+// CHECK11-NEXT: [[SUB10:%.*]] = sub nsw i32 [[TMP58]], 0
// CHECK11-NEXT: [[DIV11:%.*]] = sdiv i32 [[SUB10]], 1
// CHECK11-NEXT: [[SUB12:%.*]] = sub nsw i32 [[DIV11]], 1
// CHECK11-NEXT: store i32 [[SUB12]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK11-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK11-NEXT: [[ADD13:%.*]] = add nsw i32 [[TMP55]], 1
-// CHECK11-NEXT: [[TMP56:%.*]] = zext i32 [[ADD13]] to i64
+// CHECK11-NEXT: [[TMP59:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK11-NEXT: [[ADD13:%.*]] = add nsw i32 [[TMP59]], 1
+// CHECK11-NEXT: [[TMP60:%.*]] = zext i32 [[ADD13]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS14:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP57]], align 4
-// CHECK11-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
-// CHECK11-NEXT: store i32 4, ptr [[TMP58]], align 4
-// CHECK11-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP51]], ptr [[TMP59]], align 4
-// CHECK11-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP52]], ptr [[TMP60]], align 4
-// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.4, ptr [[TMP61]], align 4
-// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP62]], align 4
-// CHECK11-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP63]], align 4
-// CHECK11-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP64]], align 4
-// CHECK11-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP56]], ptr [[TMP65]], align 8
-// CHECK11-NEXT: [[TMP66:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l407.region_id, ptr [[KERNEL_ARGS14]])
-// CHECK11-NEXT: [[TMP67:%.*]] = icmp ne i32 [[TMP66]], 0
-// CHECK11-NEXT: br i1 [[TMP67]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
+// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP61]], align 4
+// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
+// CHECK11-NEXT: store i32 4, ptr [[TMP62]], align 4
+// CHECK11-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP55]], ptr [[TMP63]], align 4
+// CHECK11-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP56]], ptr [[TMP64]], align 4
+// CHECK11-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.4, ptr [[TMP65]], align 4
+// CHECK11-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP66]], align 4
+// CHECK11-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP67]], align 4
+// CHECK11-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP68]], align 4
+// CHECK11-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP60]], ptr [[TMP69]], align 8
+// CHECK11-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP70]], align 8
+// CHECK11-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP71]], align 4
+// CHECK11-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP72]], align 4
+// CHECK11-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP73]], align 4
+// CHECK11-NEXT: [[TMP74:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l407.region_id, ptr [[KERNEL_ARGS14]])
+// CHECK11-NEXT: [[TMP75:%.*]] = icmp ne i32 [[TMP74]], 0
+// CHECK11-NEXT: br i1 [[TMP75]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
// CHECK11: omp_offload.failed15:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l407(i32 [[TMP35]], ptr [[TMP36]], ptr [[TMP37]], ptr [[TMP38]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l407(i32 [[TMP39]], ptr [[TMP40]], ptr [[TMP41]], ptr [[TMP42]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT16]]
// CHECK11: omp_offload.cont16:
-// CHECK11-NEXT: [[TMP68:%.*]] = load i32, ptr [[CH]], align 4
-// CHECK11-NEXT: store i32 [[TMP68]], ptr [[CH_CASTED]], align 4
-// CHECK11-NEXT: [[TMP69:%.*]] = load i32, ptr [[CH_CASTED]], align 4
-// CHECK11-NEXT: [[TMP70:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP70]], ptr [[N_CASTED17]], align 4
-// CHECK11-NEXT: [[TMP71:%.*]] = load i32, ptr [[N_CASTED17]], align 4
-// CHECK11-NEXT: [[TMP72:%.*]] = load ptr, ptr [[A]], align 4
-// CHECK11-NEXT: [[TMP73:%.*]] = load ptr, ptr [[B]], align 4
-// CHECK11-NEXT: [[TMP74:%.*]] = load ptr, ptr [[C]], align 4
-// CHECK11-NEXT: [[TMP75:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP69]], ptr [[TMP75]], align 4
-// CHECK11-NEXT: [[TMP76:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP69]], ptr [[TMP76]], align 4
-// CHECK11-NEXT: [[TMP77:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP77]], align 4
-// CHECK11-NEXT: [[TMP78:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP71]], ptr [[TMP78]], align 4
-// CHECK11-NEXT: [[TMP79:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP71]], ptr [[TMP79]], align 4
-// CHECK11-NEXT: [[TMP80:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP80]], align 4
-// CHECK11-NEXT: [[TMP81:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP72]], ptr [[TMP81]], align 4
-// CHECK11-NEXT: [[TMP82:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP72]], ptr [[TMP82]], align 4
-// CHECK11-NEXT: [[TMP83:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP83]], align 4
-// CHECK11-NEXT: [[TMP84:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP73]], ptr [[TMP84]], align 4
-// CHECK11-NEXT: [[TMP85:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP73]], ptr [[TMP85]], align 4
-// CHECK11-NEXT: [[TMP86:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP86]], align 4
-// CHECK11-NEXT: [[TMP87:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP74]], ptr [[TMP87]], align 4
-// CHECK11-NEXT: [[TMP88:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP74]], ptr [[TMP88]], align 4
-// CHECK11-NEXT: [[TMP89:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 4
-// CHECK11-NEXT: store ptr null, ptr [[TMP89]], align 4
-// CHECK11-NEXT: [[TMP90:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP91:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP92:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP92]], ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK11-NEXT: [[TMP93:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK11-NEXT: [[SUB24:%.*]] = sub nsw i32 [[TMP93]], 0
+// CHECK11-NEXT: [[TMP76:%.*]] = load i32, ptr [[CH]], align 4
+// CHECK11-NEXT: store i32 [[TMP76]], ptr [[CH_CASTED]], align 4
+// CHECK11-NEXT: [[TMP77:%.*]] = load i32, ptr [[CH_CASTED]], align 4
+// CHECK11-NEXT: [[TMP78:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP78]], ptr [[N_CASTED17]], align 4
+// CHECK11-NEXT: [[TMP79:%.*]] = load i32, ptr [[N_CASTED17]], align 4
+// CHECK11-NEXT: [[TMP80:%.*]] = load ptr, ptr [[A]], align 4
+// CHECK11-NEXT: [[TMP81:%.*]] = load ptr, ptr [[B]], align 4
+// CHECK11-NEXT: [[TMP82:%.*]] = load ptr, ptr [[C]], align 4
+// CHECK11-NEXT: [[TMP83:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP77]], ptr [[TMP83]], align 4
+// CHECK11-NEXT: [[TMP84:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP77]], ptr [[TMP84]], align 4
+// CHECK11-NEXT: [[TMP85:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP85]], align 4
+// CHECK11-NEXT: [[TMP86:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP79]], ptr [[TMP86]], align 4
+// CHECK11-NEXT: [[TMP87:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP79]], ptr [[TMP87]], align 4
+// CHECK11-NEXT: [[TMP88:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP88]], align 4
+// CHECK11-NEXT: [[TMP89:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP80]], ptr [[TMP89]], align 4
+// CHECK11-NEXT: [[TMP90:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP80]], ptr [[TMP90]], align 4
+// CHECK11-NEXT: [[TMP91:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP91]], align 4
+// CHECK11-NEXT: [[TMP92:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP81]], ptr [[TMP92]], align 4
+// CHECK11-NEXT: [[TMP93:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP81]], ptr [[TMP93]], align 4
+// CHECK11-NEXT: [[TMP94:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP94]], align 4
+// CHECK11-NEXT: [[TMP95:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP82]], ptr [[TMP95]], align 4
+// CHECK11-NEXT: [[TMP96:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP82]], ptr [[TMP96]], align 4
+// CHECK11-NEXT: [[TMP97:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 4
+// CHECK11-NEXT: store ptr null, ptr [[TMP97]], align 4
+// CHECK11-NEXT: [[TMP98:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP99:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP100:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP100]], ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK11-NEXT: [[TMP101:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK11-NEXT: [[SUB24:%.*]] = sub nsw i32 [[TMP101]], 0
// CHECK11-NEXT: [[DIV25:%.*]] = sdiv i32 [[SUB24]], 1
// CHECK11-NEXT: [[SUB26:%.*]] = sub nsw i32 [[DIV25]], 1
// CHECK11-NEXT: store i32 [[SUB26]], ptr [[DOTCAPTURE_EXPR_23]], align 4
-// CHECK11-NEXT: [[TMP94:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_23]], align 4
-// CHECK11-NEXT: [[ADD27:%.*]] = add nsw i32 [[TMP94]], 1
-// CHECK11-NEXT: [[TMP95:%.*]] = zext i32 [[ADD27]] to i64
+// CHECK11-NEXT: [[TMP102:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_23]], align 4
+// CHECK11-NEXT: [[ADD27:%.*]] = add nsw i32 [[TMP102]], 1
+// CHECK11-NEXT: [[TMP103:%.*]] = zext i32 [[ADD27]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS28:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP96]], align 4
-// CHECK11-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
-// CHECK11-NEXT: store i32 5, ptr [[TMP97]], align 4
-// CHECK11-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP90]], ptr [[TMP98]], align 4
-// CHECK11-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP91]], ptr [[TMP99]], align 4
-// CHECK11-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.8, ptr [[TMP100]], align 4
-// CHECK11-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP101]], align 4
-// CHECK11-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP102]], align 4
-// CHECK11-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP103]], align 4
-// CHECK11-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP95]], ptr [[TMP104]], align 8
-// CHECK11-NEXT: [[TMP105:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l446.region_id, ptr [[KERNEL_ARGS28]])
-// CHECK11-NEXT: [[TMP106:%.*]] = icmp ne i32 [[TMP105]], 0
-// CHECK11-NEXT: br i1 [[TMP106]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
+// CHECK11-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP104]], align 4
+// CHECK11-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
+// CHECK11-NEXT: store i32 5, ptr [[TMP105]], align 4
+// CHECK11-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP98]], ptr [[TMP106]], align 4
+// CHECK11-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP99]], ptr [[TMP107]], align 4
+// CHECK11-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.8, ptr [[TMP108]], align 4
+// CHECK11-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP109]], align 4
+// CHECK11-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP110]], align 4
+// CHECK11-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP111]], align 4
+// CHECK11-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP103]], ptr [[TMP112]], align 8
+// CHECK11-NEXT: [[TMP113:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP113]], align 8
+// CHECK11-NEXT: [[TMP114:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP114]], align 4
+// CHECK11-NEXT: [[TMP115:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP115]], align 4
+// CHECK11-NEXT: [[TMP116:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP116]], align 4
+// CHECK11-NEXT: [[TMP117:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l446.region_id, ptr [[KERNEL_ARGS28]])
+// CHECK11-NEXT: [[TMP118:%.*]] = icmp ne i32 [[TMP117]], 0
+// CHECK11-NEXT: br i1 [[TMP118]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
// CHECK11: omp_offload.failed29:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l446(i32 [[TMP69]], i32 [[TMP71]], ptr [[TMP72]], ptr [[TMP73]], ptr [[TMP74]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l446(i32 [[TMP77]], i32 [[TMP79]], ptr [[TMP80]], ptr [[TMP81]], ptr [[TMP82]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT30]]
// CHECK11: omp_offload.cont30:
-// CHECK11-NEXT: [[TMP107:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP107]], ptr [[N_CASTED31]], align 4
-// CHECK11-NEXT: [[TMP108:%.*]] = load i32, ptr [[N_CASTED31]], align 4
-// CHECK11-NEXT: [[TMP109:%.*]] = load ptr, ptr [[A]], align 4
-// CHECK11-NEXT: [[TMP110:%.*]] = load ptr, ptr [[B]], align 4
-// CHECK11-NEXT: [[TMP111:%.*]] = load ptr, ptr [[C]], align 4
-// CHECK11-NEXT: [[TMP112:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP108]], ptr [[TMP112]], align 4
-// CHECK11-NEXT: [[TMP113:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP108]], ptr [[TMP113]], align 4
-// CHECK11-NEXT: [[TMP114:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP114]], align 4
-// CHECK11-NEXT: [[TMP115:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 1
-// CHECK11-NEXT: store ptr [[TMP109]], ptr [[TMP115]], align 4
-// CHECK11-NEXT: [[TMP116:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 1
-// CHECK11-NEXT: store ptr [[TMP109]], ptr [[TMP116]], align 4
-// CHECK11-NEXT: [[TMP117:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP117]], align 4
-// CHECK11-NEXT: [[TMP118:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP110]], ptr [[TMP118]], align 4
-// CHECK11-NEXT: [[TMP119:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP110]], ptr [[TMP119]], align 4
-// CHECK11-NEXT: [[TMP120:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP120]], align 4
-// CHECK11-NEXT: [[TMP121:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP111]], ptr [[TMP121]], align 4
-// CHECK11-NEXT: [[TMP122:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP111]], ptr [[TMP122]], align 4
-// CHECK11-NEXT: [[TMP123:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP123]], align 4
+// CHECK11-NEXT: [[TMP119:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP119]], ptr [[N_CASTED31]], align 4
+// CHECK11-NEXT: [[TMP120:%.*]] = load i32, ptr [[N_CASTED31]], align 4
+// CHECK11-NEXT: [[TMP121:%.*]] = load ptr, ptr [[A]], align 4
+// CHECK11-NEXT: [[TMP122:%.*]] = load ptr, ptr [[B]], align 4
+// CHECK11-NEXT: [[TMP123:%.*]] = load ptr, ptr [[C]], align 4
// CHECK11-NEXT: [[TMP124:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP120]], ptr [[TMP124]], align 4
// CHECK11-NEXT: [[TMP125:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP126:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP126]], ptr [[DOTCAPTURE_EXPR_36]], align 4
-// CHECK11-NEXT: [[TMP127:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_36]], align 4
-// CHECK11-NEXT: [[SUB38:%.*]] = sub nsw i32 [[TMP127]], 0
+// CHECK11-NEXT: store i32 [[TMP120]], ptr [[TMP125]], align 4
+// CHECK11-NEXT: [[TMP126:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP126]], align 4
+// CHECK11-NEXT: [[TMP127:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 1
+// CHECK11-NEXT: store ptr [[TMP121]], ptr [[TMP127]], align 4
+// CHECK11-NEXT: [[TMP128:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 1
+// CHECK11-NEXT: store ptr [[TMP121]], ptr [[TMP128]], align 4
+// CHECK11-NEXT: [[TMP129:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP129]], align 4
+// CHECK11-NEXT: [[TMP130:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP122]], ptr [[TMP130]], align 4
+// CHECK11-NEXT: [[TMP131:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP122]], ptr [[TMP131]], align 4
+// CHECK11-NEXT: [[TMP132:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP132]], align 4
+// CHECK11-NEXT: [[TMP133:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP123]], ptr [[TMP133]], align 4
+// CHECK11-NEXT: [[TMP134:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP123]], ptr [[TMP134]], align 4
+// CHECK11-NEXT: [[TMP135:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP135]], align 4
+// CHECK11-NEXT: [[TMP136:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP137:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP138:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP138]], ptr [[DOTCAPTURE_EXPR_36]], align 4
+// CHECK11-NEXT: [[TMP139:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_36]], align 4
+// CHECK11-NEXT: [[SUB38:%.*]] = sub nsw i32 [[TMP139]], 0
// CHECK11-NEXT: [[DIV39:%.*]] = sdiv i32 [[SUB38]], 1
// CHECK11-NEXT: [[SUB40:%.*]] = sub nsw i32 [[DIV39]], 1
// CHECK11-NEXT: store i32 [[SUB40]], ptr [[DOTCAPTURE_EXPR_37]], align 4
-// CHECK11-NEXT: [[TMP128:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_37]], align 4
-// CHECK11-NEXT: [[ADD41:%.*]] = add nsw i32 [[TMP128]], 1
-// CHECK11-NEXT: [[TMP129:%.*]] = zext i32 [[ADD41]] to i64
+// CHECK11-NEXT: [[TMP140:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_37]], align 4
+// CHECK11-NEXT: [[ADD41:%.*]] = add nsw i32 [[TMP140]], 1
+// CHECK11-NEXT: [[TMP141:%.*]] = zext i32 [[ADD41]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS42:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP130]], align 4
-// CHECK11-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 1
-// CHECK11-NEXT: store i32 4, ptr [[TMP131]], align 4
-// CHECK11-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP124]], ptr [[TMP132]], align 4
-// CHECK11-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP125]], ptr [[TMP133]], align 4
-// CHECK11-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.12, ptr [[TMP134]], align 4
-// CHECK11-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP135]], align 4
-// CHECK11-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP136]], align 4
-// CHECK11-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP137]], align 4
-// CHECK11-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP129]], ptr [[TMP138]], align 8
-// CHECK11-NEXT: [[TMP139:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l477.region_id, ptr [[KERNEL_ARGS42]])
-// CHECK11-NEXT: [[TMP140:%.*]] = icmp ne i32 [[TMP139]], 0
-// CHECK11-NEXT: br i1 [[TMP140]], label [[OMP_OFFLOAD_FAILED43:%.*]], label [[OMP_OFFLOAD_CONT44:%.*]]
+// CHECK11-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP142]], align 4
+// CHECK11-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 1
+// CHECK11-NEXT: store i32 4, ptr [[TMP143]], align 4
+// CHECK11-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP136]], ptr [[TMP144]], align 4
+// CHECK11-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP137]], ptr [[TMP145]], align 4
+// CHECK11-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.12, ptr [[TMP146]], align 4
+// CHECK11-NEXT: [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP147]], align 4
+// CHECK11-NEXT: [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP148]], align 4
+// CHECK11-NEXT: [[TMP149:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP149]], align 4
+// CHECK11-NEXT: [[TMP150:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP141]], ptr [[TMP150]], align 8
+// CHECK11-NEXT: [[TMP151:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP151]], align 8
+// CHECK11-NEXT: [[TMP152:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP152]], align 4
+// CHECK11-NEXT: [[TMP153:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP153]], align 4
+// CHECK11-NEXT: [[TMP154:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP154]], align 4
+// CHECK11-NEXT: [[TMP155:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l477.region_id, ptr [[KERNEL_ARGS42]])
+// CHECK11-NEXT: [[TMP156:%.*]] = icmp ne i32 [[TMP155]], 0
+// CHECK11-NEXT: br i1 [[TMP156]], label [[OMP_OFFLOAD_FAILED43:%.*]], label [[OMP_OFFLOAD_CONT44:%.*]]
// CHECK11: omp_offload.failed43:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l477(i32 [[TMP108]], ptr [[TMP109]], ptr [[TMP110]], ptr [[TMP111]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l477(i32 [[TMP120]], ptr [[TMP121]], ptr [[TMP122]], ptr [[TMP123]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT44]]
// CHECK11: omp_offload.cont44:
-// CHECK11-NEXT: [[TMP141:%.*]] = load i32, ptr [[CH]], align 4
-// CHECK11-NEXT: store i32 [[TMP141]], ptr [[CH_CASTED45]], align 4
-// CHECK11-NEXT: [[TMP142:%.*]] = load i32, ptr [[CH_CASTED45]], align 4
-// CHECK11-NEXT: [[TMP143:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP143]], ptr [[N_CASTED46]], align 4
-// CHECK11-NEXT: [[TMP144:%.*]] = load i32, ptr [[N_CASTED46]], align 4
-// CHECK11-NEXT: [[TMP145:%.*]] = load ptr, ptr [[A]], align 4
-// CHECK11-NEXT: [[TMP146:%.*]] = load ptr, ptr [[B]], align 4
-// CHECK11-NEXT: [[TMP147:%.*]] = load ptr, ptr [[C]], align 4
-// CHECK11-NEXT: [[TMP148:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP142]], ptr [[TMP148]], align 4
-// CHECK11-NEXT: [[TMP149:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP142]], ptr [[TMP149]], align 4
-// CHECK11-NEXT: [[TMP150:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP150]], align 4
-// CHECK11-NEXT: [[TMP151:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP144]], ptr [[TMP151]], align 4
-// CHECK11-NEXT: [[TMP152:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP144]], ptr [[TMP152]], align 4
-// CHECK11-NEXT: [[TMP153:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP153]], align 4
-// CHECK11-NEXT: [[TMP154:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP145]], ptr [[TMP154]], align 4
-// CHECK11-NEXT: [[TMP155:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP145]], ptr [[TMP155]], align 4
-// CHECK11-NEXT: [[TMP156:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP156]], align 4
-// CHECK11-NEXT: [[TMP157:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP146]], ptr [[TMP157]], align 4
-// CHECK11-NEXT: [[TMP158:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP146]], ptr [[TMP158]], align 4
-// CHECK11-NEXT: [[TMP159:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP159]], align 4
-// CHECK11-NEXT: [[TMP160:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP147]], ptr [[TMP160]], align 4
-// CHECK11-NEXT: [[TMP161:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP147]], ptr [[TMP161]], align 4
-// CHECK11-NEXT: [[TMP162:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 4
-// CHECK11-NEXT: store ptr null, ptr [[TMP162]], align 4
-// CHECK11-NEXT: [[TMP163:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP164:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP165:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP165]], ptr [[DOTCAPTURE_EXPR_51]], align 4
-// CHECK11-NEXT: [[TMP166:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_51]], align 4
-// CHECK11-NEXT: [[SUB53:%.*]] = sub nsw i32 [[TMP166]], 0
+// CHECK11-NEXT: [[TMP157:%.*]] = load i32, ptr [[CH]], align 4
+// CHECK11-NEXT: store i32 [[TMP157]], ptr [[CH_CASTED45]], align 4
+// CHECK11-NEXT: [[TMP158:%.*]] = load i32, ptr [[CH_CASTED45]], align 4
+// CHECK11-NEXT: [[TMP159:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP159]], ptr [[N_CASTED46]], align 4
+// CHECK11-NEXT: [[TMP160:%.*]] = load i32, ptr [[N_CASTED46]], align 4
+// CHECK11-NEXT: [[TMP161:%.*]] = load ptr, ptr [[A]], align 4
+// CHECK11-NEXT: [[TMP162:%.*]] = load ptr, ptr [[B]], align 4
+// CHECK11-NEXT: [[TMP163:%.*]] = load ptr, ptr [[C]], align 4
+// CHECK11-NEXT: [[TMP164:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP158]], ptr [[TMP164]], align 4
+// CHECK11-NEXT: [[TMP165:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP158]], ptr [[TMP165]], align 4
+// CHECK11-NEXT: [[TMP166:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP166]], align 4
+// CHECK11-NEXT: [[TMP167:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP160]], ptr [[TMP167]], align 4
+// CHECK11-NEXT: [[TMP168:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP160]], ptr [[TMP168]], align 4
+// CHECK11-NEXT: [[TMP169:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP169]], align 4
+// CHECK11-NEXT: [[TMP170:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP161]], ptr [[TMP170]], align 4
+// CHECK11-NEXT: [[TMP171:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP161]], ptr [[TMP171]], align 4
+// CHECK11-NEXT: [[TMP172:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP172]], align 4
+// CHECK11-NEXT: [[TMP173:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP162]], ptr [[TMP173]], align 4
+// CHECK11-NEXT: [[TMP174:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP162]], ptr [[TMP174]], align 4
+// CHECK11-NEXT: [[TMP175:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP175]], align 4
+// CHECK11-NEXT: [[TMP176:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP163]], ptr [[TMP176]], align 4
+// CHECK11-NEXT: [[TMP177:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP163]], ptr [[TMP177]], align 4
+// CHECK11-NEXT: [[TMP178:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 4
+// CHECK11-NEXT: store ptr null, ptr [[TMP178]], align 4
+// CHECK11-NEXT: [[TMP179:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP180:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP181:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP181]], ptr [[DOTCAPTURE_EXPR_51]], align 4
+// CHECK11-NEXT: [[TMP182:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_51]], align 4
+// CHECK11-NEXT: [[SUB53:%.*]] = sub nsw i32 [[TMP182]], 0
// CHECK11-NEXT: [[DIV54:%.*]] = sdiv i32 [[SUB53]], 1
// CHECK11-NEXT: [[SUB55:%.*]] = sub nsw i32 [[DIV54]], 1
// CHECK11-NEXT: store i32 [[SUB55]], ptr [[DOTCAPTURE_EXPR_52]], align 4
-// CHECK11-NEXT: [[TMP167:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_52]], align 4
-// CHECK11-NEXT: [[ADD56:%.*]] = add nsw i32 [[TMP167]], 1
-// CHECK11-NEXT: [[TMP168:%.*]] = zext i32 [[ADD56]] to i64
+// CHECK11-NEXT: [[TMP183:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_52]], align 4
+// CHECK11-NEXT: [[ADD56:%.*]] = add nsw i32 [[TMP183]], 1
+// CHECK11-NEXT: [[TMP184:%.*]] = zext i32 [[ADD56]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS57:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP169:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP169]], align 4
-// CHECK11-NEXT: [[TMP170:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 1
-// CHECK11-NEXT: store i32 5, ptr [[TMP170]], align 4
-// CHECK11-NEXT: [[TMP171:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP163]], ptr [[TMP171]], align 4
-// CHECK11-NEXT: [[TMP172:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP164]], ptr [[TMP172]], align 4
-// CHECK11-NEXT: [[TMP173:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.16, ptr [[TMP173]], align 4
-// CHECK11-NEXT: [[TMP174:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP174]], align 4
-// CHECK11-NEXT: [[TMP175:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP175]], align 4
-// CHECK11-NEXT: [[TMP176:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP176]], align 4
-// CHECK11-NEXT: [[TMP177:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP168]], ptr [[TMP177]], align 8
-// CHECK11-NEXT: [[TMP178:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l505.region_id, ptr [[KERNEL_ARGS57]])
-// CHECK11-NEXT: [[TMP179:%.*]] = icmp ne i32 [[TMP178]], 0
-// CHECK11-NEXT: br i1 [[TMP179]], label [[OMP_OFFLOAD_FAILED58:%.*]], label [[OMP_OFFLOAD_CONT59:%.*]]
+// CHECK11-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP185]], align 4
+// CHECK11-NEXT: [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 1
+// CHECK11-NEXT: store i32 5, ptr [[TMP186]], align 4
+// CHECK11-NEXT: [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP179]], ptr [[TMP187]], align 4
+// CHECK11-NEXT: [[TMP188:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP180]], ptr [[TMP188]], align 4
+// CHECK11-NEXT: [[TMP189:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.16, ptr [[TMP189]], align 4
+// CHECK11-NEXT: [[TMP190:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP190]], align 4
+// CHECK11-NEXT: [[TMP191:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP191]], align 4
+// CHECK11-NEXT: [[TMP192:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP192]], align 4
+// CHECK11-NEXT: [[TMP193:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP184]], ptr [[TMP193]], align 8
+// CHECK11-NEXT: [[TMP194:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP194]], align 8
+// CHECK11-NEXT: [[TMP195:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP195]], align 4
+// CHECK11-NEXT: [[TMP196:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP196]], align 4
+// CHECK11-NEXT: [[TMP197:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP197]], align 4
+// CHECK11-NEXT: [[TMP198:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l505.region_id, ptr [[KERNEL_ARGS57]])
+// CHECK11-NEXT: [[TMP199:%.*]] = icmp ne i32 [[TMP198]], 0
+// CHECK11-NEXT: br i1 [[TMP199]], label [[OMP_OFFLOAD_FAILED58:%.*]], label [[OMP_OFFLOAD_CONT59:%.*]]
// CHECK11: omp_offload.failed58:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l505(i32 [[TMP142]], i32 [[TMP144]], ptr [[TMP145]], ptr [[TMP146]], ptr [[TMP147]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l505(i32 [[TMP158]], i32 [[TMP160]], ptr [[TMP161]], ptr [[TMP162]], ptr [[TMP163]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT59]]
// CHECK11: omp_offload.cont59:
-// CHECK11-NEXT: [[TMP180:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP180]], ptr [[N_CASTED60]], align 4
-// CHECK11-NEXT: [[TMP181:%.*]] = load i32, ptr [[N_CASTED60]], align 4
-// CHECK11-NEXT: [[TMP182:%.*]] = load ptr, ptr [[A]], align 4
-// CHECK11-NEXT: [[TMP183:%.*]] = load ptr, ptr [[B]], align 4
-// CHECK11-NEXT: [[TMP184:%.*]] = load ptr, ptr [[C]], align 4
-// CHECK11-NEXT: [[TMP185:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP181]], ptr [[TMP185]], align 4
-// CHECK11-NEXT: [[TMP186:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP181]], ptr [[TMP186]], align 4
-// CHECK11-NEXT: [[TMP187:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP187]], align 4
-// CHECK11-NEXT: [[TMP188:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 1
-// CHECK11-NEXT: store ptr [[TMP182]], ptr [[TMP188]], align 4
-// CHECK11-NEXT: [[TMP189:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 1
-// CHECK11-NEXT: store ptr [[TMP182]], ptr [[TMP189]], align 4
-// CHECK11-NEXT: [[TMP190:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP190]], align 4
-// CHECK11-NEXT: [[TMP191:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP183]], ptr [[TMP191]], align 4
-// CHECK11-NEXT: [[TMP192:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP183]], ptr [[TMP192]], align 4
-// CHECK11-NEXT: [[TMP193:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP193]], align 4
-// CHECK11-NEXT: [[TMP194:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP184]], ptr [[TMP194]], align 4
-// CHECK11-NEXT: [[TMP195:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP184]], ptr [[TMP195]], align 4
-// CHECK11-NEXT: [[TMP196:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP196]], align 4
-// CHECK11-NEXT: [[TMP197:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP198:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP199:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP199]], ptr [[DOTCAPTURE_EXPR_65]], align 4
-// CHECK11-NEXT: [[TMP200:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_65]], align 4
-// CHECK11-NEXT: [[SUB67:%.*]] = sub nsw i32 [[TMP200]], 0
+// CHECK11-NEXT: [[TMP200:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP200]], ptr [[N_CASTED60]], align 4
+// CHECK11-NEXT: [[TMP201:%.*]] = load i32, ptr [[N_CASTED60]], align 4
+// CHECK11-NEXT: [[TMP202:%.*]] = load ptr, ptr [[A]], align 4
+// CHECK11-NEXT: [[TMP203:%.*]] = load ptr, ptr [[B]], align 4
+// CHECK11-NEXT: [[TMP204:%.*]] = load ptr, ptr [[C]], align 4
+// CHECK11-NEXT: [[TMP205:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP201]], ptr [[TMP205]], align 4
+// CHECK11-NEXT: [[TMP206:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP201]], ptr [[TMP206]], align 4
+// CHECK11-NEXT: [[TMP207:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP207]], align 4
+// CHECK11-NEXT: [[TMP208:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 1
+// CHECK11-NEXT: store ptr [[TMP202]], ptr [[TMP208]], align 4
+// CHECK11-NEXT: [[TMP209:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 1
+// CHECK11-NEXT: store ptr [[TMP202]], ptr [[TMP209]], align 4
+// CHECK11-NEXT: [[TMP210:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP210]], align 4
+// CHECK11-NEXT: [[TMP211:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP203]], ptr [[TMP211]], align 4
+// CHECK11-NEXT: [[TMP212:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP203]], ptr [[TMP212]], align 4
+// CHECK11-NEXT: [[TMP213:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP213]], align 4
+// CHECK11-NEXT: [[TMP214:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP204]], ptr [[TMP214]], align 4
+// CHECK11-NEXT: [[TMP215:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP204]], ptr [[TMP215]], align 4
+// CHECK11-NEXT: [[TMP216:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP216]], align 4
+// CHECK11-NEXT: [[TMP217:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP218:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP219:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP219]], ptr [[DOTCAPTURE_EXPR_65]], align 4
+// CHECK11-NEXT: [[TMP220:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_65]], align 4
+// CHECK11-NEXT: [[SUB67:%.*]] = sub nsw i32 [[TMP220]], 0
// CHECK11-NEXT: [[DIV68:%.*]] = sdiv i32 [[SUB67]], 1
// CHECK11-NEXT: [[SUB69:%.*]] = sub nsw i32 [[DIV68]], 1
// CHECK11-NEXT: store i32 [[SUB69]], ptr [[DOTCAPTURE_EXPR_66]], align 4
-// CHECK11-NEXT: [[TMP201:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_66]], align 4
-// CHECK11-NEXT: [[ADD70:%.*]] = add nsw i32 [[TMP201]], 1
-// CHECK11-NEXT: [[TMP202:%.*]] = zext i32 [[ADD70]] to i64
+// CHECK11-NEXT: [[TMP221:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_66]], align 4
+// CHECK11-NEXT: [[ADD70:%.*]] = add nsw i32 [[TMP221]], 1
+// CHECK11-NEXT: [[TMP222:%.*]] = zext i32 [[ADD70]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS71:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP203:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP203]], align 4
-// CHECK11-NEXT: [[TMP204:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 1
-// CHECK11-NEXT: store i32 4, ptr [[TMP204]], align 4
-// CHECK11-NEXT: [[TMP205:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP197]], ptr [[TMP205]], align 4
-// CHECK11-NEXT: [[TMP206:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP198]], ptr [[TMP206]], align 4
-// CHECK11-NEXT: [[TMP207:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.20, ptr [[TMP207]], align 4
-// CHECK11-NEXT: [[TMP208:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.21, ptr [[TMP208]], align 4
-// CHECK11-NEXT: [[TMP209:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP209]], align 4
-// CHECK11-NEXT: [[TMP210:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP210]], align 4
-// CHECK11-NEXT: [[TMP211:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP202]], ptr [[TMP211]], align 8
-// CHECK11-NEXT: [[TMP212:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l535.region_id, ptr [[KERNEL_ARGS71]])
-// CHECK11-NEXT: [[TMP213:%.*]] = icmp ne i32 [[TMP212]], 0
-// CHECK11-NEXT: br i1 [[TMP213]], label [[OMP_OFFLOAD_FAILED72:%.*]], label [[OMP_OFFLOAD_CONT73:%.*]]
+// CHECK11-NEXT: [[TMP223:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP223]], align 4
+// CHECK11-NEXT: [[TMP224:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 1
+// CHECK11-NEXT: store i32 4, ptr [[TMP224]], align 4
+// CHECK11-NEXT: [[TMP225:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP217]], ptr [[TMP225]], align 4
+// CHECK11-NEXT: [[TMP226:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP218]], ptr [[TMP226]], align 4
+// CHECK11-NEXT: [[TMP227:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.20, ptr [[TMP227]], align 4
+// CHECK11-NEXT: [[TMP228:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.21, ptr [[TMP228]], align 4
+// CHECK11-NEXT: [[TMP229:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP229]], align 4
+// CHECK11-NEXT: [[TMP230:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP230]], align 4
+// CHECK11-NEXT: [[TMP231:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP222]], ptr [[TMP231]], align 8
+// CHECK11-NEXT: [[TMP232:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP232]], align 8
+// CHECK11-NEXT: [[TMP233:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP233]], align 4
+// CHECK11-NEXT: [[TMP234:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP234]], align 4
+// CHECK11-NEXT: [[TMP235:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP235]], align 4
+// CHECK11-NEXT: [[TMP236:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l535.region_id, ptr [[KERNEL_ARGS71]])
+// CHECK11-NEXT: [[TMP237:%.*]] = icmp ne i32 [[TMP236]], 0
+// CHECK11-NEXT: br i1 [[TMP237]], label [[OMP_OFFLOAD_FAILED72:%.*]], label [[OMP_OFFLOAD_CONT73:%.*]]
// CHECK11: omp_offload.failed72:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l535(i32 [[TMP181]], ptr [[TMP182]], ptr [[TMP183]], ptr [[TMP184]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l535(i32 [[TMP201]], ptr [[TMP202]], ptr [[TMP203]], ptr [[TMP204]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT73]]
// CHECK11: omp_offload.cont73:
-// CHECK11-NEXT: [[TMP214:%.*]] = load i32, ptr [[CH]], align 4
-// CHECK11-NEXT: store i32 [[TMP214]], ptr [[CH_CASTED74]], align 4
-// CHECK11-NEXT: [[TMP215:%.*]] = load i32, ptr [[CH_CASTED74]], align 4
-// CHECK11-NEXT: [[TMP216:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP216]], ptr [[N_CASTED75]], align 4
-// CHECK11-NEXT: [[TMP217:%.*]] = load i32, ptr [[N_CASTED75]], align 4
-// CHECK11-NEXT: [[TMP218:%.*]] = load ptr, ptr [[A]], align 4
-// CHECK11-NEXT: [[TMP219:%.*]] = load ptr, ptr [[B]], align 4
-// CHECK11-NEXT: [[TMP220:%.*]] = load ptr, ptr [[C]], align 4
-// CHECK11-NEXT: [[TMP221:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP215]], ptr [[TMP221]], align 4
-// CHECK11-NEXT: [[TMP222:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP215]], ptr [[TMP222]], align 4
-// CHECK11-NEXT: [[TMP223:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP223]], align 4
-// CHECK11-NEXT: [[TMP224:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP217]], ptr [[TMP224]], align 4
-// CHECK11-NEXT: [[TMP225:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP217]], ptr [[TMP225]], align 4
-// CHECK11-NEXT: [[TMP226:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP226]], align 4
-// CHECK11-NEXT: [[TMP227:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP218]], ptr [[TMP227]], align 4
-// CHECK11-NEXT: [[TMP228:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP218]], ptr [[TMP228]], align 4
-// CHECK11-NEXT: [[TMP229:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP229]], align 4
-// CHECK11-NEXT: [[TMP230:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP219]], ptr [[TMP230]], align 4
-// CHECK11-NEXT: [[TMP231:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP219]], ptr [[TMP231]], align 4
-// CHECK11-NEXT: [[TMP232:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP232]], align 4
-// CHECK11-NEXT: [[TMP233:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP220]], ptr [[TMP233]], align 4
-// CHECK11-NEXT: [[TMP234:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP220]], ptr [[TMP234]], align 4
-// CHECK11-NEXT: [[TMP235:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 4
-// CHECK11-NEXT: store ptr null, ptr [[TMP235]], align 4
-// CHECK11-NEXT: [[TMP236:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP237:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP238:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP238]], ptr [[DOTCAPTURE_EXPR_80]], align 4
-// CHECK11-NEXT: [[TMP239:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_80]], align 4
-// CHECK11-NEXT: [[SUB82:%.*]] = sub nsw i32 [[TMP239]], 0
+// CHECK11-NEXT: [[TMP238:%.*]] = load i32, ptr [[CH]], align 4
+// CHECK11-NEXT: store i32 [[TMP238]], ptr [[CH_CASTED74]], align 4
+// CHECK11-NEXT: [[TMP239:%.*]] = load i32, ptr [[CH_CASTED74]], align 4
+// CHECK11-NEXT: [[TMP240:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP240]], ptr [[N_CASTED75]], align 4
+// CHECK11-NEXT: [[TMP241:%.*]] = load i32, ptr [[N_CASTED75]], align 4
+// CHECK11-NEXT: [[TMP242:%.*]] = load ptr, ptr [[A]], align 4
+// CHECK11-NEXT: [[TMP243:%.*]] = load ptr, ptr [[B]], align 4
+// CHECK11-NEXT: [[TMP244:%.*]] = load ptr, ptr [[C]], align 4
+// CHECK11-NEXT: [[TMP245:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP239]], ptr [[TMP245]], align 4
+// CHECK11-NEXT: [[TMP246:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP239]], ptr [[TMP246]], align 4
+// CHECK11-NEXT: [[TMP247:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP247]], align 4
+// CHECK11-NEXT: [[TMP248:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP241]], ptr [[TMP248]], align 4
+// CHECK11-NEXT: [[TMP249:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP241]], ptr [[TMP249]], align 4
+// CHECK11-NEXT: [[TMP250:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP250]], align 4
+// CHECK11-NEXT: [[TMP251:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP242]], ptr [[TMP251]], align 4
+// CHECK11-NEXT: [[TMP252:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP242]], ptr [[TMP252]], align 4
+// CHECK11-NEXT: [[TMP253:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP253]], align 4
+// CHECK11-NEXT: [[TMP254:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP243]], ptr [[TMP254]], align 4
+// CHECK11-NEXT: [[TMP255:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP243]], ptr [[TMP255]], align 4
+// CHECK11-NEXT: [[TMP256:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP256]], align 4
+// CHECK11-NEXT: [[TMP257:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP244]], ptr [[TMP257]], align 4
+// CHECK11-NEXT: [[TMP258:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP244]], ptr [[TMP258]], align 4
+// CHECK11-NEXT: [[TMP259:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 4
+// CHECK11-NEXT: store ptr null, ptr [[TMP259]], align 4
+// CHECK11-NEXT: [[TMP260:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP261:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP262:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP262]], ptr [[DOTCAPTURE_EXPR_80]], align 4
+// CHECK11-NEXT: [[TMP263:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_80]], align 4
+// CHECK11-NEXT: [[SUB82:%.*]] = sub nsw i32 [[TMP263]], 0
// CHECK11-NEXT: [[DIV83:%.*]] = sdiv i32 [[SUB82]], 1
// CHECK11-NEXT: [[SUB84:%.*]] = sub nsw i32 [[DIV83]], 1
// CHECK11-NEXT: store i32 [[SUB84]], ptr [[DOTCAPTURE_EXPR_81]], align 4
-// CHECK11-NEXT: [[TMP240:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_81]], align 4
-// CHECK11-NEXT: [[ADD85:%.*]] = add nsw i32 [[TMP240]], 1
-// CHECK11-NEXT: [[TMP241:%.*]] = zext i32 [[ADD85]] to i64
+// CHECK11-NEXT: [[TMP264:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_81]], align 4
+// CHECK11-NEXT: [[ADD85:%.*]] = add nsw i32 [[TMP264]], 1
+// CHECK11-NEXT: [[TMP265:%.*]] = zext i32 [[ADD85]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS86:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP242:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP242]], align 4
-// CHECK11-NEXT: [[TMP243:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 1
-// CHECK11-NEXT: store i32 5, ptr [[TMP243]], align 4
-// CHECK11-NEXT: [[TMP244:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP236]], ptr [[TMP244]], align 4
-// CHECK11-NEXT: [[TMP245:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP237]], ptr [[TMP245]], align 4
-// CHECK11-NEXT: [[TMP246:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.24, ptr [[TMP246]], align 4
-// CHECK11-NEXT: [[TMP247:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP247]], align 4
-// CHECK11-NEXT: [[TMP248:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP248]], align 4
-// CHECK11-NEXT: [[TMP249:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP249]], align 4
-// CHECK11-NEXT: [[TMP250:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP241]], ptr [[TMP250]], align 8
-// CHECK11-NEXT: [[TMP251:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l561.region_id, ptr [[KERNEL_ARGS86]])
-// CHECK11-NEXT: [[TMP252:%.*]] = icmp ne i32 [[TMP251]], 0
-// CHECK11-NEXT: br i1 [[TMP252]], label [[OMP_OFFLOAD_FAILED87:%.*]], label [[OMP_OFFLOAD_CONT88:%.*]]
+// CHECK11-NEXT: [[TMP266:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP266]], align 4
+// CHECK11-NEXT: [[TMP267:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 1
+// CHECK11-NEXT: store i32 5, ptr [[TMP267]], align 4
+// CHECK11-NEXT: [[TMP268:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP260]], ptr [[TMP268]], align 4
+// CHECK11-NEXT: [[TMP269:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP261]], ptr [[TMP269]], align 4
+// CHECK11-NEXT: [[TMP270:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.24, ptr [[TMP270]], align 4
+// CHECK11-NEXT: [[TMP271:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP271]], align 4
+// CHECK11-NEXT: [[TMP272:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP272]], align 4
+// CHECK11-NEXT: [[TMP273:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP273]], align 4
+// CHECK11-NEXT: [[TMP274:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP265]], ptr [[TMP274]], align 8
+// CHECK11-NEXT: [[TMP275:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP275]], align 8
+// CHECK11-NEXT: [[TMP276:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP276]], align 4
+// CHECK11-NEXT: [[TMP277:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP277]], align 4
+// CHECK11-NEXT: [[TMP278:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP278]], align 4
+// CHECK11-NEXT: [[TMP279:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l561.region_id, ptr [[KERNEL_ARGS86]])
+// CHECK11-NEXT: [[TMP280:%.*]] = icmp ne i32 [[TMP279]], 0
+// CHECK11-NEXT: br i1 [[TMP280]], label [[OMP_OFFLOAD_FAILED87:%.*]], label [[OMP_OFFLOAD_CONT88:%.*]]
// CHECK11: omp_offload.failed87:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l561(i32 [[TMP215]], i32 [[TMP217]], ptr [[TMP218]], ptr [[TMP219]], ptr [[TMP220]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l561(i32 [[TMP239]], i32 [[TMP241]], ptr [[TMP242]], ptr [[TMP243]], ptr [[TMP244]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT88]]
// CHECK11: omp_offload.cont88:
// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiET_v()
@@ -11960,7 +12128,7 @@ int main() {
// CHECK11-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 4, ptr [[TMP24]], align 4
// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -11977,446 +12145,502 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK11-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l42.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK11-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP34]], align 4
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l42.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK11-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l42(i32 [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK11-NEXT: [[TMP35:%.*]] = load i32, ptr [[N_CASTED3]], align 4
-// CHECK11-NEXT: [[TMP36:%.*]] = load ptr, ptr [[A]], align 4
-// CHECK11-NEXT: [[TMP37:%.*]] = load ptr, ptr [[B]], align 4
-// CHECK11-NEXT: [[TMP38:%.*]] = load ptr, ptr [[C]], align 4
-// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP35]], ptr [[TMP39]], align 4
-// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP35]], ptr [[TMP40]], align 4
-// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP41]], align 4
-// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK11-NEXT: store ptr [[TMP36]], ptr [[TMP42]], align 4
-// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK11-NEXT: store ptr [[TMP36]], ptr [[TMP43]], align 4
-// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP44]], align 4
-// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP37]], ptr [[TMP45]], align 4
-// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP37]], ptr [[TMP46]], align 4
-// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP47]], align 4
-// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP38]], ptr [[TMP48]], align 4
-// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP38]], ptr [[TMP49]], align 4
-// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP50]], align 4
-// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP53:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP53]], ptr [[DOTCAPTURE_EXPR_8]], align 4
-// CHECK11-NEXT: [[TMP54:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_8]], align 4
-// CHECK11-NEXT: [[SUB10:%.*]] = sub nsw i32 [[TMP54]], 0
+// CHECK11-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK11-NEXT: [[TMP39:%.*]] = load i32, ptr [[N_CASTED3]], align 4
+// CHECK11-NEXT: [[TMP40:%.*]] = load ptr, ptr [[A]], align 4
+// CHECK11-NEXT: [[TMP41:%.*]] = load ptr, ptr [[B]], align 4
+// CHECK11-NEXT: [[TMP42:%.*]] = load ptr, ptr [[C]], align 4
+// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP39]], ptr [[TMP43]], align 4
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP39]], ptr [[TMP44]], align 4
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK11-NEXT: store ptr [[TMP40]], ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK11-NEXT: store ptr [[TMP40]], ptr [[TMP47]], align 4
+// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP48]], align 4
+// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP41]], ptr [[TMP49]], align 4
+// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP41]], ptr [[TMP50]], align 4
+// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP42]], ptr [[TMP52]], align 4
+// CHECK11-NEXT: [[TMP53:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP42]], ptr [[TMP53]], align 4
+// CHECK11-NEXT: [[TMP54:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP54]], align 4
+// CHECK11-NEXT: [[TMP55:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP56:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP57:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP57]], ptr [[DOTCAPTURE_EXPR_8]], align 4
+// CHECK11-NEXT: [[TMP58:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_8]], align 4
+// CHECK11-NEXT: [[SUB10:%.*]] = sub nsw i32 [[TMP58]], 0
// CHECK11-NEXT: [[DIV11:%.*]] = sdiv i32 [[SUB10]], 1
// CHECK11-NEXT: [[SUB12:%.*]] = sub nsw i32 [[DIV11]], 1
// CHECK11-NEXT: store i32 [[SUB12]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK11-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK11-NEXT: [[ADD13:%.*]] = add nsw i32 [[TMP55]], 1
-// CHECK11-NEXT: [[TMP56:%.*]] = zext i32 [[ADD13]] to i64
+// CHECK11-NEXT: [[TMP59:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK11-NEXT: [[ADD13:%.*]] = add nsw i32 [[TMP59]], 1
+// CHECK11-NEXT: [[TMP60:%.*]] = zext i32 [[ADD13]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS14:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP57]], align 4
-// CHECK11-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
-// CHECK11-NEXT: store i32 4, ptr [[TMP58]], align 4
-// CHECK11-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP51]], ptr [[TMP59]], align 4
-// CHECK11-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP52]], ptr [[TMP60]], align 4
-// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.32, ptr [[TMP61]], align 4
-// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP62]], align 4
-// CHECK11-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP63]], align 4
-// CHECK11-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP64]], align 4
-// CHECK11-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP56]], ptr [[TMP65]], align 8
-// CHECK11-NEXT: [[TMP66:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l50.region_id, ptr [[KERNEL_ARGS14]])
-// CHECK11-NEXT: [[TMP67:%.*]] = icmp ne i32 [[TMP66]], 0
-// CHECK11-NEXT: br i1 [[TMP67]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
+// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP61]], align 4
+// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
+// CHECK11-NEXT: store i32 4, ptr [[TMP62]], align 4
+// CHECK11-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP55]], ptr [[TMP63]], align 4
+// CHECK11-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP56]], ptr [[TMP64]], align 4
+// CHECK11-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.32, ptr [[TMP65]], align 4
+// CHECK11-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP66]], align 4
+// CHECK11-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP67]], align 4
+// CHECK11-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP68]], align 4
+// CHECK11-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP60]], ptr [[TMP69]], align 8
+// CHECK11-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP70]], align 8
+// CHECK11-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP71]], align 4
+// CHECK11-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP72]], align 4
+// CHECK11-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP73]], align 4
+// CHECK11-NEXT: [[TMP74:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l50.region_id, ptr [[KERNEL_ARGS14]])
+// CHECK11-NEXT: [[TMP75:%.*]] = icmp ne i32 [[TMP74]], 0
+// CHECK11-NEXT: br i1 [[TMP75]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
// CHECK11: omp_offload.failed15:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l50(i32 [[TMP35]], ptr [[TMP36]], ptr [[TMP37]], ptr [[TMP38]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l50(i32 [[TMP39]], ptr [[TMP40]], ptr [[TMP41]], ptr [[TMP42]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT16]]
// CHECK11: omp_offload.cont16:
-// CHECK11-NEXT: [[TMP68:%.*]] = load i32, ptr [[CH]], align 4
-// CHECK11-NEXT: store i32 [[TMP68]], ptr [[CH_CASTED]], align 4
-// CHECK11-NEXT: [[TMP69:%.*]] = load i32, ptr [[CH_CASTED]], align 4
-// CHECK11-NEXT: [[TMP70:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP70]], ptr [[N_CASTED17]], align 4
-// CHECK11-NEXT: [[TMP71:%.*]] = load i32, ptr [[N_CASTED17]], align 4
-// CHECK11-NEXT: [[TMP72:%.*]] = load ptr, ptr [[A]], align 4
-// CHECK11-NEXT: [[TMP73:%.*]] = load ptr, ptr [[B]], align 4
-// CHECK11-NEXT: [[TMP74:%.*]] = load ptr, ptr [[C]], align 4
-// CHECK11-NEXT: [[TMP75:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP69]], ptr [[TMP75]], align 4
-// CHECK11-NEXT: [[TMP76:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP69]], ptr [[TMP76]], align 4
-// CHECK11-NEXT: [[TMP77:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP77]], align 4
-// CHECK11-NEXT: [[TMP78:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP71]], ptr [[TMP78]], align 4
-// CHECK11-NEXT: [[TMP79:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP71]], ptr [[TMP79]], align 4
-// CHECK11-NEXT: [[TMP80:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP80]], align 4
-// CHECK11-NEXT: [[TMP81:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP72]], ptr [[TMP81]], align 4
-// CHECK11-NEXT: [[TMP82:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP72]], ptr [[TMP82]], align 4
-// CHECK11-NEXT: [[TMP83:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP83]], align 4
-// CHECK11-NEXT: [[TMP84:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP73]], ptr [[TMP84]], align 4
-// CHECK11-NEXT: [[TMP85:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP73]], ptr [[TMP85]], align 4
-// CHECK11-NEXT: [[TMP86:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP86]], align 4
-// CHECK11-NEXT: [[TMP87:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP74]], ptr [[TMP87]], align 4
-// CHECK11-NEXT: [[TMP88:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP74]], ptr [[TMP88]], align 4
-// CHECK11-NEXT: [[TMP89:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 4
-// CHECK11-NEXT: store ptr null, ptr [[TMP89]], align 4
-// CHECK11-NEXT: [[TMP90:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP91:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP92:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP92]], ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK11-NEXT: [[TMP93:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK11-NEXT: [[SUB24:%.*]] = sub nsw i32 [[TMP93]], 0
+// CHECK11-NEXT: [[TMP76:%.*]] = load i32, ptr [[CH]], align 4
+// CHECK11-NEXT: store i32 [[TMP76]], ptr [[CH_CASTED]], align 4
+// CHECK11-NEXT: [[TMP77:%.*]] = load i32, ptr [[CH_CASTED]], align 4
+// CHECK11-NEXT: [[TMP78:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP78]], ptr [[N_CASTED17]], align 4
+// CHECK11-NEXT: [[TMP79:%.*]] = load i32, ptr [[N_CASTED17]], align 4
+// CHECK11-NEXT: [[TMP80:%.*]] = load ptr, ptr [[A]], align 4
+// CHECK11-NEXT: [[TMP81:%.*]] = load ptr, ptr [[B]], align 4
+// CHECK11-NEXT: [[TMP82:%.*]] = load ptr, ptr [[C]], align 4
+// CHECK11-NEXT: [[TMP83:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP77]], ptr [[TMP83]], align 4
+// CHECK11-NEXT: [[TMP84:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP77]], ptr [[TMP84]], align 4
+// CHECK11-NEXT: [[TMP85:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP85]], align 4
+// CHECK11-NEXT: [[TMP86:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP79]], ptr [[TMP86]], align 4
+// CHECK11-NEXT: [[TMP87:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP79]], ptr [[TMP87]], align 4
+// CHECK11-NEXT: [[TMP88:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP88]], align 4
+// CHECK11-NEXT: [[TMP89:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP80]], ptr [[TMP89]], align 4
+// CHECK11-NEXT: [[TMP90:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP80]], ptr [[TMP90]], align 4
+// CHECK11-NEXT: [[TMP91:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP91]], align 4
+// CHECK11-NEXT: [[TMP92:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP81]], ptr [[TMP92]], align 4
+// CHECK11-NEXT: [[TMP93:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP81]], ptr [[TMP93]], align 4
+// CHECK11-NEXT: [[TMP94:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP94]], align 4
+// CHECK11-NEXT: [[TMP95:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP82]], ptr [[TMP95]], align 4
+// CHECK11-NEXT: [[TMP96:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP82]], ptr [[TMP96]], align 4
+// CHECK11-NEXT: [[TMP97:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 4
+// CHECK11-NEXT: store ptr null, ptr [[TMP97]], align 4
+// CHECK11-NEXT: [[TMP98:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP99:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP100:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP100]], ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK11-NEXT: [[TMP101:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK11-NEXT: [[SUB24:%.*]] = sub nsw i32 [[TMP101]], 0
// CHECK11-NEXT: [[DIV25:%.*]] = sdiv i32 [[SUB24]], 1
// CHECK11-NEXT: [[SUB26:%.*]] = sub nsw i32 [[DIV25]], 1
// CHECK11-NEXT: store i32 [[SUB26]], ptr [[DOTCAPTURE_EXPR_23]], align 4
-// CHECK11-NEXT: [[TMP94:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_23]], align 4
-// CHECK11-NEXT: [[ADD27:%.*]] = add nsw i32 [[TMP94]], 1
-// CHECK11-NEXT: [[TMP95:%.*]] = zext i32 [[ADD27]] to i64
+// CHECK11-NEXT: [[TMP102:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_23]], align 4
+// CHECK11-NEXT: [[ADD27:%.*]] = add nsw i32 [[TMP102]], 1
+// CHECK11-NEXT: [[TMP103:%.*]] = zext i32 [[ADD27]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS28:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP96]], align 4
-// CHECK11-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
-// CHECK11-NEXT: store i32 5, ptr [[TMP97]], align 4
-// CHECK11-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP90]], ptr [[TMP98]], align 4
-// CHECK11-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP91]], ptr [[TMP99]], align 4
-// CHECK11-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.36, ptr [[TMP100]], align 4
-// CHECK11-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP101]], align 4
-// CHECK11-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP102]], align 4
-// CHECK11-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP103]], align 4
-// CHECK11-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP95]], ptr [[TMP104]], align 8
-// CHECK11-NEXT: [[TMP105:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l58.region_id, ptr [[KERNEL_ARGS28]])
-// CHECK11-NEXT: [[TMP106:%.*]] = icmp ne i32 [[TMP105]], 0
-// CHECK11-NEXT: br i1 [[TMP106]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
+// CHECK11-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP104]], align 4
+// CHECK11-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
+// CHECK11-NEXT: store i32 5, ptr [[TMP105]], align 4
+// CHECK11-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP98]], ptr [[TMP106]], align 4
+// CHECK11-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP99]], ptr [[TMP107]], align 4
+// CHECK11-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.36, ptr [[TMP108]], align 4
+// CHECK11-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP109]], align 4
+// CHECK11-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP110]], align 4
+// CHECK11-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP111]], align 4
+// CHECK11-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP103]], ptr [[TMP112]], align 8
+// CHECK11-NEXT: [[TMP113:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP113]], align 8
+// CHECK11-NEXT: [[TMP114:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP114]], align 4
+// CHECK11-NEXT: [[TMP115:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP115]], align 4
+// CHECK11-NEXT: [[TMP116:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP116]], align 4
+// CHECK11-NEXT: [[TMP117:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l58.region_id, ptr [[KERNEL_ARGS28]])
+// CHECK11-NEXT: [[TMP118:%.*]] = icmp ne i32 [[TMP117]], 0
+// CHECK11-NEXT: br i1 [[TMP118]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
// CHECK11: omp_offload.failed29:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l58(i32 [[TMP69]], i32 [[TMP71]], ptr [[TMP72]], ptr [[TMP73]], ptr [[TMP74]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l58(i32 [[TMP77]], i32 [[TMP79]], ptr [[TMP80]], ptr [[TMP81]], ptr [[TMP82]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT30]]
// CHECK11: omp_offload.cont30:
-// CHECK11-NEXT: [[TMP107:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP107]], ptr [[N_CASTED31]], align 4
-// CHECK11-NEXT: [[TMP108:%.*]] = load i32, ptr [[N_CASTED31]], align 4
-// CHECK11-NEXT: [[TMP109:%.*]] = load ptr, ptr [[A]], align 4
-// CHECK11-NEXT: [[TMP110:%.*]] = load ptr, ptr [[B]], align 4
-// CHECK11-NEXT: [[TMP111:%.*]] = load ptr, ptr [[C]], align 4
-// CHECK11-NEXT: [[TMP112:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP108]], ptr [[TMP112]], align 4
-// CHECK11-NEXT: [[TMP113:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP108]], ptr [[TMP113]], align 4
-// CHECK11-NEXT: [[TMP114:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP114]], align 4
-// CHECK11-NEXT: [[TMP115:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 1
-// CHECK11-NEXT: store ptr [[TMP109]], ptr [[TMP115]], align 4
-// CHECK11-NEXT: [[TMP116:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 1
-// CHECK11-NEXT: store ptr [[TMP109]], ptr [[TMP116]], align 4
-// CHECK11-NEXT: [[TMP117:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP117]], align 4
-// CHECK11-NEXT: [[TMP118:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP110]], ptr [[TMP118]], align 4
-// CHECK11-NEXT: [[TMP119:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP110]], ptr [[TMP119]], align 4
-// CHECK11-NEXT: [[TMP120:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP120]], align 4
-// CHECK11-NEXT: [[TMP121:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP111]], ptr [[TMP121]], align 4
-// CHECK11-NEXT: [[TMP122:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP111]], ptr [[TMP122]], align 4
-// CHECK11-NEXT: [[TMP123:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP123]], align 4
+// CHECK11-NEXT: [[TMP119:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP119]], ptr [[N_CASTED31]], align 4
+// CHECK11-NEXT: [[TMP120:%.*]] = load i32, ptr [[N_CASTED31]], align 4
+// CHECK11-NEXT: [[TMP121:%.*]] = load ptr, ptr [[A]], align 4
+// CHECK11-NEXT: [[TMP122:%.*]] = load ptr, ptr [[B]], align 4
+// CHECK11-NEXT: [[TMP123:%.*]] = load ptr, ptr [[C]], align 4
// CHECK11-NEXT: [[TMP124:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP120]], ptr [[TMP124]], align 4
// CHECK11-NEXT: [[TMP125:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP126:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP126]], ptr [[DOTCAPTURE_EXPR_36]], align 4
-// CHECK11-NEXT: [[TMP127:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_36]], align 4
-// CHECK11-NEXT: [[SUB38:%.*]] = sub nsw i32 [[TMP127]], 0
+// CHECK11-NEXT: store i32 [[TMP120]], ptr [[TMP125]], align 4
+// CHECK11-NEXT: [[TMP126:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP126]], align 4
+// CHECK11-NEXT: [[TMP127:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 1
+// CHECK11-NEXT: store ptr [[TMP121]], ptr [[TMP127]], align 4
+// CHECK11-NEXT: [[TMP128:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 1
+// CHECK11-NEXT: store ptr [[TMP121]], ptr [[TMP128]], align 4
+// CHECK11-NEXT: [[TMP129:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP129]], align 4
+// CHECK11-NEXT: [[TMP130:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP122]], ptr [[TMP130]], align 4
+// CHECK11-NEXT: [[TMP131:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP122]], ptr [[TMP131]], align 4
+// CHECK11-NEXT: [[TMP132:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP132]], align 4
+// CHECK11-NEXT: [[TMP133:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP123]], ptr [[TMP133]], align 4
+// CHECK11-NEXT: [[TMP134:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP123]], ptr [[TMP134]], align 4
+// CHECK11-NEXT: [[TMP135:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP135]], align 4
+// CHECK11-NEXT: [[TMP136:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP137:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP138:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP138]], ptr [[DOTCAPTURE_EXPR_36]], align 4
+// CHECK11-NEXT: [[TMP139:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_36]], align 4
+// CHECK11-NEXT: [[SUB38:%.*]] = sub nsw i32 [[TMP139]], 0
// CHECK11-NEXT: [[DIV39:%.*]] = sdiv i32 [[SUB38]], 1
// CHECK11-NEXT: [[SUB40:%.*]] = sub nsw i32 [[DIV39]], 1
// CHECK11-NEXT: store i32 [[SUB40]], ptr [[DOTCAPTURE_EXPR_37]], align 4
-// CHECK11-NEXT: [[TMP128:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_37]], align 4
-// CHECK11-NEXT: [[ADD41:%.*]] = add nsw i32 [[TMP128]], 1
-// CHECK11-NEXT: [[TMP129:%.*]] = zext i32 [[ADD41]] to i64
+// CHECK11-NEXT: [[TMP140:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_37]], align 4
+// CHECK11-NEXT: [[ADD41:%.*]] = add nsw i32 [[TMP140]], 1
+// CHECK11-NEXT: [[TMP141:%.*]] = zext i32 [[ADD41]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS42:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP130]], align 4
-// CHECK11-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 1
-// CHECK11-NEXT: store i32 4, ptr [[TMP131]], align 4
-// CHECK11-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP124]], ptr [[TMP132]], align 4
-// CHECK11-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP125]], ptr [[TMP133]], align 4
-// CHECK11-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.40, ptr [[TMP134]], align 4
-// CHECK11-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.41, ptr [[TMP135]], align 4
-// CHECK11-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP136]], align 4
-// CHECK11-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP137]], align 4
-// CHECK11-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP129]], ptr [[TMP138]], align 8
-// CHECK11-NEXT: [[TMP139:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l66.region_id, ptr [[KERNEL_ARGS42]])
-// CHECK11-NEXT: [[TMP140:%.*]] = icmp ne i32 [[TMP139]], 0
-// CHECK11-NEXT: br i1 [[TMP140]], label [[OMP_OFFLOAD_FAILED43:%.*]], label [[OMP_OFFLOAD_CONT44:%.*]]
+// CHECK11-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP142]], align 4
+// CHECK11-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 1
+// CHECK11-NEXT: store i32 4, ptr [[TMP143]], align 4
+// CHECK11-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP136]], ptr [[TMP144]], align 4
+// CHECK11-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP137]], ptr [[TMP145]], align 4
+// CHECK11-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.40, ptr [[TMP146]], align 4
+// CHECK11-NEXT: [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.41, ptr [[TMP147]], align 4
+// CHECK11-NEXT: [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP148]], align 4
+// CHECK11-NEXT: [[TMP149:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP149]], align 4
+// CHECK11-NEXT: [[TMP150:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP141]], ptr [[TMP150]], align 8
+// CHECK11-NEXT: [[TMP151:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP151]], align 8
+// CHECK11-NEXT: [[TMP152:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP152]], align 4
+// CHECK11-NEXT: [[TMP153:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP153]], align 4
+// CHECK11-NEXT: [[TMP154:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP154]], align 4
+// CHECK11-NEXT: [[TMP155:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l66.region_id, ptr [[KERNEL_ARGS42]])
+// CHECK11-NEXT: [[TMP156:%.*]] = icmp ne i32 [[TMP155]], 0
+// CHECK11-NEXT: br i1 [[TMP156]], label [[OMP_OFFLOAD_FAILED43:%.*]], label [[OMP_OFFLOAD_CONT44:%.*]]
// CHECK11: omp_offload.failed43:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l66(i32 [[TMP108]], ptr [[TMP109]], ptr [[TMP110]], ptr [[TMP111]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l66(i32 [[TMP120]], ptr [[TMP121]], ptr [[TMP122]], ptr [[TMP123]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT44]]
// CHECK11: omp_offload.cont44:
-// CHECK11-NEXT: [[TMP141:%.*]] = load i32, ptr [[CH]], align 4
-// CHECK11-NEXT: store i32 [[TMP141]], ptr [[CH_CASTED45]], align 4
-// CHECK11-NEXT: [[TMP142:%.*]] = load i32, ptr [[CH_CASTED45]], align 4
-// CHECK11-NEXT: [[TMP143:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP143]], ptr [[N_CASTED46]], align 4
-// CHECK11-NEXT: [[TMP144:%.*]] = load i32, ptr [[N_CASTED46]], align 4
-// CHECK11-NEXT: [[TMP145:%.*]] = load ptr, ptr [[A]], align 4
-// CHECK11-NEXT: [[TMP146:%.*]] = load ptr, ptr [[B]], align 4
-// CHECK11-NEXT: [[TMP147:%.*]] = load ptr, ptr [[C]], align 4
-// CHECK11-NEXT: [[TMP148:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP142]], ptr [[TMP148]], align 4
-// CHECK11-NEXT: [[TMP149:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP142]], ptr [[TMP149]], align 4
-// CHECK11-NEXT: [[TMP150:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP150]], align 4
-// CHECK11-NEXT: [[TMP151:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP144]], ptr [[TMP151]], align 4
-// CHECK11-NEXT: [[TMP152:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP144]], ptr [[TMP152]], align 4
-// CHECK11-NEXT: [[TMP153:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP153]], align 4
-// CHECK11-NEXT: [[TMP154:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP145]], ptr [[TMP154]], align 4
-// CHECK11-NEXT: [[TMP155:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP145]], ptr [[TMP155]], align 4
-// CHECK11-NEXT: [[TMP156:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP156]], align 4
-// CHECK11-NEXT: [[TMP157:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP146]], ptr [[TMP157]], align 4
-// CHECK11-NEXT: [[TMP158:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP146]], ptr [[TMP158]], align 4
-// CHECK11-NEXT: [[TMP159:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP159]], align 4
-// CHECK11-NEXT: [[TMP160:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP147]], ptr [[TMP160]], align 4
-// CHECK11-NEXT: [[TMP161:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP147]], ptr [[TMP161]], align 4
-// CHECK11-NEXT: [[TMP162:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 4
-// CHECK11-NEXT: store ptr null, ptr [[TMP162]], align 4
-// CHECK11-NEXT: [[TMP163:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP164:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP165:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP165]], ptr [[DOTCAPTURE_EXPR_51]], align 4
-// CHECK11-NEXT: [[TMP166:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_51]], align 4
-// CHECK11-NEXT: [[SUB53:%.*]] = sub nsw i32 [[TMP166]], 0
+// CHECK11-NEXT: [[TMP157:%.*]] = load i32, ptr [[CH]], align 4
+// CHECK11-NEXT: store i32 [[TMP157]], ptr [[CH_CASTED45]], align 4
+// CHECK11-NEXT: [[TMP158:%.*]] = load i32, ptr [[CH_CASTED45]], align 4
+// CHECK11-NEXT: [[TMP159:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP159]], ptr [[N_CASTED46]], align 4
+// CHECK11-NEXT: [[TMP160:%.*]] = load i32, ptr [[N_CASTED46]], align 4
+// CHECK11-NEXT: [[TMP161:%.*]] = load ptr, ptr [[A]], align 4
+// CHECK11-NEXT: [[TMP162:%.*]] = load ptr, ptr [[B]], align 4
+// CHECK11-NEXT: [[TMP163:%.*]] = load ptr, ptr [[C]], align 4
+// CHECK11-NEXT: [[TMP164:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP158]], ptr [[TMP164]], align 4
+// CHECK11-NEXT: [[TMP165:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP158]], ptr [[TMP165]], align 4
+// CHECK11-NEXT: [[TMP166:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP166]], align 4
+// CHECK11-NEXT: [[TMP167:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP160]], ptr [[TMP167]], align 4
+// CHECK11-NEXT: [[TMP168:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP160]], ptr [[TMP168]], align 4
+// CHECK11-NEXT: [[TMP169:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP169]], align 4
+// CHECK11-NEXT: [[TMP170:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP161]], ptr [[TMP170]], align 4
+// CHECK11-NEXT: [[TMP171:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP161]], ptr [[TMP171]], align 4
+// CHECK11-NEXT: [[TMP172:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP172]], align 4
+// CHECK11-NEXT: [[TMP173:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP162]], ptr [[TMP173]], align 4
+// CHECK11-NEXT: [[TMP174:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP162]], ptr [[TMP174]], align 4
+// CHECK11-NEXT: [[TMP175:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP175]], align 4
+// CHECK11-NEXT: [[TMP176:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP163]], ptr [[TMP176]], align 4
+// CHECK11-NEXT: [[TMP177:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP163]], ptr [[TMP177]], align 4
+// CHECK11-NEXT: [[TMP178:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 4
+// CHECK11-NEXT: store ptr null, ptr [[TMP178]], align 4
+// CHECK11-NEXT: [[TMP179:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP180:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP181:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP181]], ptr [[DOTCAPTURE_EXPR_51]], align 4
+// CHECK11-NEXT: [[TMP182:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_51]], align 4
+// CHECK11-NEXT: [[SUB53:%.*]] = sub nsw i32 [[TMP182]], 0
// CHECK11-NEXT: [[DIV54:%.*]] = sdiv i32 [[SUB53]], 1
// CHECK11-NEXT: [[SUB55:%.*]] = sub nsw i32 [[DIV54]], 1
// CHECK11-NEXT: store i32 [[SUB55]], ptr [[DOTCAPTURE_EXPR_52]], align 4
-// CHECK11-NEXT: [[TMP167:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_52]], align 4
-// CHECK11-NEXT: [[ADD56:%.*]] = add nsw i32 [[TMP167]], 1
-// CHECK11-NEXT: [[TMP168:%.*]] = zext i32 [[ADD56]] to i64
+// CHECK11-NEXT: [[TMP183:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_52]], align 4
+// CHECK11-NEXT: [[ADD56:%.*]] = add nsw i32 [[TMP183]], 1
+// CHECK11-NEXT: [[TMP184:%.*]] = zext i32 [[ADD56]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS57:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP169:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP169]], align 4
-// CHECK11-NEXT: [[TMP170:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 1
-// CHECK11-NEXT: store i32 5, ptr [[TMP170]], align 4
-// CHECK11-NEXT: [[TMP171:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP163]], ptr [[TMP171]], align 4
-// CHECK11-NEXT: [[TMP172:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP164]], ptr [[TMP172]], align 4
-// CHECK11-NEXT: [[TMP173:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.44, ptr [[TMP173]], align 4
-// CHECK11-NEXT: [[TMP174:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.45, ptr [[TMP174]], align 4
-// CHECK11-NEXT: [[TMP175:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP175]], align 4
-// CHECK11-NEXT: [[TMP176:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP176]], align 4
-// CHECK11-NEXT: [[TMP177:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP168]], ptr [[TMP177]], align 8
-// CHECK11-NEXT: [[TMP178:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l74.region_id, ptr [[KERNEL_ARGS57]])
-// CHECK11-NEXT: [[TMP179:%.*]] = icmp ne i32 [[TMP178]], 0
-// CHECK11-NEXT: br i1 [[TMP179]], label [[OMP_OFFLOAD_FAILED58:%.*]], label [[OMP_OFFLOAD_CONT59:%.*]]
+// CHECK11-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP185]], align 4
+// CHECK11-NEXT: [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 1
+// CHECK11-NEXT: store i32 5, ptr [[TMP186]], align 4
+// CHECK11-NEXT: [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP179]], ptr [[TMP187]], align 4
+// CHECK11-NEXT: [[TMP188:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP180]], ptr [[TMP188]], align 4
+// CHECK11-NEXT: [[TMP189:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.44, ptr [[TMP189]], align 4
+// CHECK11-NEXT: [[TMP190:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.45, ptr [[TMP190]], align 4
+// CHECK11-NEXT: [[TMP191:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP191]], align 4
+// CHECK11-NEXT: [[TMP192:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP192]], align 4
+// CHECK11-NEXT: [[TMP193:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP184]], ptr [[TMP193]], align 8
+// CHECK11-NEXT: [[TMP194:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP194]], align 8
+// CHECK11-NEXT: [[TMP195:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP195]], align 4
+// CHECK11-NEXT: [[TMP196:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP196]], align 4
+// CHECK11-NEXT: [[TMP197:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP197]], align 4
+// CHECK11-NEXT: [[TMP198:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l74.region_id, ptr [[KERNEL_ARGS57]])
+// CHECK11-NEXT: [[TMP199:%.*]] = icmp ne i32 [[TMP198]], 0
+// CHECK11-NEXT: br i1 [[TMP199]], label [[OMP_OFFLOAD_FAILED58:%.*]], label [[OMP_OFFLOAD_CONT59:%.*]]
// CHECK11: omp_offload.failed58:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l74(i32 [[TMP142]], i32 [[TMP144]], ptr [[TMP145]], ptr [[TMP146]], ptr [[TMP147]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l74(i32 [[TMP158]], i32 [[TMP160]], ptr [[TMP161]], ptr [[TMP162]], ptr [[TMP163]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT59]]
// CHECK11: omp_offload.cont59:
-// CHECK11-NEXT: [[TMP180:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP180]], ptr [[N_CASTED60]], align 4
-// CHECK11-NEXT: [[TMP181:%.*]] = load i32, ptr [[N_CASTED60]], align 4
-// CHECK11-NEXT: [[TMP182:%.*]] = load ptr, ptr [[A]], align 4
-// CHECK11-NEXT: [[TMP183:%.*]] = load ptr, ptr [[B]], align 4
-// CHECK11-NEXT: [[TMP184:%.*]] = load ptr, ptr [[C]], align 4
-// CHECK11-NEXT: [[TMP185:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP181]], ptr [[TMP185]], align 4
-// CHECK11-NEXT: [[TMP186:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP181]], ptr [[TMP186]], align 4
-// CHECK11-NEXT: [[TMP187:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP187]], align 4
-// CHECK11-NEXT: [[TMP188:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 1
-// CHECK11-NEXT: store ptr [[TMP182]], ptr [[TMP188]], align 4
-// CHECK11-NEXT: [[TMP189:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 1
-// CHECK11-NEXT: store ptr [[TMP182]], ptr [[TMP189]], align 4
-// CHECK11-NEXT: [[TMP190:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP190]], align 4
-// CHECK11-NEXT: [[TMP191:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP183]], ptr [[TMP191]], align 4
-// CHECK11-NEXT: [[TMP192:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP183]], ptr [[TMP192]], align 4
-// CHECK11-NEXT: [[TMP193:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP193]], align 4
-// CHECK11-NEXT: [[TMP194:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP184]], ptr [[TMP194]], align 4
-// CHECK11-NEXT: [[TMP195:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP184]], ptr [[TMP195]], align 4
-// CHECK11-NEXT: [[TMP196:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP196]], align 4
-// CHECK11-NEXT: [[TMP197:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP198:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP199:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP199]], ptr [[DOTCAPTURE_EXPR_65]], align 4
-// CHECK11-NEXT: [[TMP200:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_65]], align 4
-// CHECK11-NEXT: [[SUB67:%.*]] = sub nsw i32 [[TMP200]], 0
+// CHECK11-NEXT: [[TMP200:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP200]], ptr [[N_CASTED60]], align 4
+// CHECK11-NEXT: [[TMP201:%.*]] = load i32, ptr [[N_CASTED60]], align 4
+// CHECK11-NEXT: [[TMP202:%.*]] = load ptr, ptr [[A]], align 4
+// CHECK11-NEXT: [[TMP203:%.*]] = load ptr, ptr [[B]], align 4
+// CHECK11-NEXT: [[TMP204:%.*]] = load ptr, ptr [[C]], align 4
+// CHECK11-NEXT: [[TMP205:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP201]], ptr [[TMP205]], align 4
+// CHECK11-NEXT: [[TMP206:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP201]], ptr [[TMP206]], align 4
+// CHECK11-NEXT: [[TMP207:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP207]], align 4
+// CHECK11-NEXT: [[TMP208:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 1
+// CHECK11-NEXT: store ptr [[TMP202]], ptr [[TMP208]], align 4
+// CHECK11-NEXT: [[TMP209:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 1
+// CHECK11-NEXT: store ptr [[TMP202]], ptr [[TMP209]], align 4
+// CHECK11-NEXT: [[TMP210:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP210]], align 4
+// CHECK11-NEXT: [[TMP211:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP203]], ptr [[TMP211]], align 4
+// CHECK11-NEXT: [[TMP212:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP203]], ptr [[TMP212]], align 4
+// CHECK11-NEXT: [[TMP213:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP213]], align 4
+// CHECK11-NEXT: [[TMP214:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP204]], ptr [[TMP214]], align 4
+// CHECK11-NEXT: [[TMP215:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP204]], ptr [[TMP215]], align 4
+// CHECK11-NEXT: [[TMP216:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP216]], align 4
+// CHECK11-NEXT: [[TMP217:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP218:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP219:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP219]], ptr [[DOTCAPTURE_EXPR_65]], align 4
+// CHECK11-NEXT: [[TMP220:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_65]], align 4
+// CHECK11-NEXT: [[SUB67:%.*]] = sub nsw i32 [[TMP220]], 0
// CHECK11-NEXT: [[DIV68:%.*]] = sdiv i32 [[SUB67]], 1
// CHECK11-NEXT: [[SUB69:%.*]] = sub nsw i32 [[DIV68]], 1
// CHECK11-NEXT: store i32 [[SUB69]], ptr [[DOTCAPTURE_EXPR_66]], align 4
-// CHECK11-NEXT: [[TMP201:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_66]], align 4
-// CHECK11-NEXT: [[ADD70:%.*]] = add nsw i32 [[TMP201]], 1
-// CHECK11-NEXT: [[TMP202:%.*]] = zext i32 [[ADD70]] to i64
+// CHECK11-NEXT: [[TMP221:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_66]], align 4
+// CHECK11-NEXT: [[ADD70:%.*]] = add nsw i32 [[TMP221]], 1
+// CHECK11-NEXT: [[TMP222:%.*]] = zext i32 [[ADD70]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS71:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP203:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP203]], align 4
-// CHECK11-NEXT: [[TMP204:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 1
-// CHECK11-NEXT: store i32 4, ptr [[TMP204]], align 4
-// CHECK11-NEXT: [[TMP205:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP197]], ptr [[TMP205]], align 4
-// CHECK11-NEXT: [[TMP206:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP198]], ptr [[TMP206]], align 4
-// CHECK11-NEXT: [[TMP207:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.48, ptr [[TMP207]], align 4
-// CHECK11-NEXT: [[TMP208:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.49, ptr [[TMP208]], align 4
-// CHECK11-NEXT: [[TMP209:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP209]], align 4
-// CHECK11-NEXT: [[TMP210:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP210]], align 4
-// CHECK11-NEXT: [[TMP211:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP202]], ptr [[TMP211]], align 8
-// CHECK11-NEXT: [[TMP212:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l82.region_id, ptr [[KERNEL_ARGS71]])
-// CHECK11-NEXT: [[TMP213:%.*]] = icmp ne i32 [[TMP212]], 0
-// CHECK11-NEXT: br i1 [[TMP213]], label [[OMP_OFFLOAD_FAILED72:%.*]], label [[OMP_OFFLOAD_CONT73:%.*]]
+// CHECK11-NEXT: [[TMP223:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP223]], align 4
+// CHECK11-NEXT: [[TMP224:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 1
+// CHECK11-NEXT: store i32 4, ptr [[TMP224]], align 4
+// CHECK11-NEXT: [[TMP225:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP217]], ptr [[TMP225]], align 4
+// CHECK11-NEXT: [[TMP226:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP218]], ptr [[TMP226]], align 4
+// CHECK11-NEXT: [[TMP227:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.48, ptr [[TMP227]], align 4
+// CHECK11-NEXT: [[TMP228:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.49, ptr [[TMP228]], align 4
+// CHECK11-NEXT: [[TMP229:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP229]], align 4
+// CHECK11-NEXT: [[TMP230:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP230]], align 4
+// CHECK11-NEXT: [[TMP231:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP222]], ptr [[TMP231]], align 8
+// CHECK11-NEXT: [[TMP232:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP232]], align 8
+// CHECK11-NEXT: [[TMP233:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP233]], align 4
+// CHECK11-NEXT: [[TMP234:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP234]], align 4
+// CHECK11-NEXT: [[TMP235:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP235]], align 4
+// CHECK11-NEXT: [[TMP236:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l82.region_id, ptr [[KERNEL_ARGS71]])
+// CHECK11-NEXT: [[TMP237:%.*]] = icmp ne i32 [[TMP236]], 0
+// CHECK11-NEXT: br i1 [[TMP237]], label [[OMP_OFFLOAD_FAILED72:%.*]], label [[OMP_OFFLOAD_CONT73:%.*]]
// CHECK11: omp_offload.failed72:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l82(i32 [[TMP181]], ptr [[TMP182]], ptr [[TMP183]], ptr [[TMP184]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l82(i32 [[TMP201]], ptr [[TMP202]], ptr [[TMP203]], ptr [[TMP204]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT73]]
// CHECK11: omp_offload.cont73:
-// CHECK11-NEXT: [[TMP214:%.*]] = load i32, ptr [[CH]], align 4
-// CHECK11-NEXT: store i32 [[TMP214]], ptr [[CH_CASTED74]], align 4
-// CHECK11-NEXT: [[TMP215:%.*]] = load i32, ptr [[CH_CASTED74]], align 4
-// CHECK11-NEXT: [[TMP216:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP216]], ptr [[N_CASTED75]], align 4
-// CHECK11-NEXT: [[TMP217:%.*]] = load i32, ptr [[N_CASTED75]], align 4
-// CHECK11-NEXT: [[TMP218:%.*]] = load ptr, ptr [[A]], align 4
-// CHECK11-NEXT: [[TMP219:%.*]] = load ptr, ptr [[B]], align 4
-// CHECK11-NEXT: [[TMP220:%.*]] = load ptr, ptr [[C]], align 4
-// CHECK11-NEXT: [[TMP221:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP215]], ptr [[TMP221]], align 4
-// CHECK11-NEXT: [[TMP222:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP215]], ptr [[TMP222]], align 4
-// CHECK11-NEXT: [[TMP223:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP223]], align 4
-// CHECK11-NEXT: [[TMP224:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP217]], ptr [[TMP224]], align 4
-// CHECK11-NEXT: [[TMP225:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP217]], ptr [[TMP225]], align 4
-// CHECK11-NEXT: [[TMP226:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP226]], align 4
-// CHECK11-NEXT: [[TMP227:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP218]], ptr [[TMP227]], align 4
-// CHECK11-NEXT: [[TMP228:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP218]], ptr [[TMP228]], align 4
-// CHECK11-NEXT: [[TMP229:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP229]], align 4
-// CHECK11-NEXT: [[TMP230:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP219]], ptr [[TMP230]], align 4
-// CHECK11-NEXT: [[TMP231:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP219]], ptr [[TMP231]], align 4
-// CHECK11-NEXT: [[TMP232:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP232]], align 4
-// CHECK11-NEXT: [[TMP233:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP220]], ptr [[TMP233]], align 4
-// CHECK11-NEXT: [[TMP234:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP220]], ptr [[TMP234]], align 4
-// CHECK11-NEXT: [[TMP235:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 4
-// CHECK11-NEXT: store ptr null, ptr [[TMP235]], align 4
-// CHECK11-NEXT: [[TMP236:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP237:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP238:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP238]], ptr [[DOTCAPTURE_EXPR_80]], align 4
-// CHECK11-NEXT: [[TMP239:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_80]], align 4
-// CHECK11-NEXT: [[SUB82:%.*]] = sub nsw i32 [[TMP239]], 0
+// CHECK11-NEXT: [[TMP238:%.*]] = load i32, ptr [[CH]], align 4
+// CHECK11-NEXT: store i32 [[TMP238]], ptr [[CH_CASTED74]], align 4
+// CHECK11-NEXT: [[TMP239:%.*]] = load i32, ptr [[CH_CASTED74]], align 4
+// CHECK11-NEXT: [[TMP240:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP240]], ptr [[N_CASTED75]], align 4
+// CHECK11-NEXT: [[TMP241:%.*]] = load i32, ptr [[N_CASTED75]], align 4
+// CHECK11-NEXT: [[TMP242:%.*]] = load ptr, ptr [[A]], align 4
+// CHECK11-NEXT: [[TMP243:%.*]] = load ptr, ptr [[B]], align 4
+// CHECK11-NEXT: [[TMP244:%.*]] = load ptr, ptr [[C]], align 4
+// CHECK11-NEXT: [[TMP245:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP239]], ptr [[TMP245]], align 4
+// CHECK11-NEXT: [[TMP246:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP239]], ptr [[TMP246]], align 4
+// CHECK11-NEXT: [[TMP247:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP247]], align 4
+// CHECK11-NEXT: [[TMP248:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP241]], ptr [[TMP248]], align 4
+// CHECK11-NEXT: [[TMP249:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP241]], ptr [[TMP249]], align 4
+// CHECK11-NEXT: [[TMP250:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP250]], align 4
+// CHECK11-NEXT: [[TMP251:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP242]], ptr [[TMP251]], align 4
+// CHECK11-NEXT: [[TMP252:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP242]], ptr [[TMP252]], align 4
+// CHECK11-NEXT: [[TMP253:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP253]], align 4
+// CHECK11-NEXT: [[TMP254:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP243]], ptr [[TMP254]], align 4
+// CHECK11-NEXT: [[TMP255:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP243]], ptr [[TMP255]], align 4
+// CHECK11-NEXT: [[TMP256:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP256]], align 4
+// CHECK11-NEXT: [[TMP257:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP244]], ptr [[TMP257]], align 4
+// CHECK11-NEXT: [[TMP258:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP244]], ptr [[TMP258]], align 4
+// CHECK11-NEXT: [[TMP259:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 4
+// CHECK11-NEXT: store ptr null, ptr [[TMP259]], align 4
+// CHECK11-NEXT: [[TMP260:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP261:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP262:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP262]], ptr [[DOTCAPTURE_EXPR_80]], align 4
+// CHECK11-NEXT: [[TMP263:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_80]], align 4
+// CHECK11-NEXT: [[SUB82:%.*]] = sub nsw i32 [[TMP263]], 0
// CHECK11-NEXT: [[DIV83:%.*]] = sdiv i32 [[SUB82]], 1
// CHECK11-NEXT: [[SUB84:%.*]] = sub nsw i32 [[DIV83]], 1
// CHECK11-NEXT: store i32 [[SUB84]], ptr [[DOTCAPTURE_EXPR_81]], align 4
-// CHECK11-NEXT: [[TMP240:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_81]], align 4
-// CHECK11-NEXT: [[ADD85:%.*]] = add nsw i32 [[TMP240]], 1
-// CHECK11-NEXT: [[TMP241:%.*]] = zext i32 [[ADD85]] to i64
+// CHECK11-NEXT: [[TMP264:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_81]], align 4
+// CHECK11-NEXT: [[ADD85:%.*]] = add nsw i32 [[TMP264]], 1
+// CHECK11-NEXT: [[TMP265:%.*]] = zext i32 [[ADD85]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS86:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP242:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP242]], align 4
-// CHECK11-NEXT: [[TMP243:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 1
-// CHECK11-NEXT: store i32 5, ptr [[TMP243]], align 4
-// CHECK11-NEXT: [[TMP244:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP236]], ptr [[TMP244]], align 4
-// CHECK11-NEXT: [[TMP245:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP237]], ptr [[TMP245]], align 4
-// CHECK11-NEXT: [[TMP246:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.52, ptr [[TMP246]], align 4
-// CHECK11-NEXT: [[TMP247:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.53, ptr [[TMP247]], align 4
-// CHECK11-NEXT: [[TMP248:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP248]], align 4
-// CHECK11-NEXT: [[TMP249:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP249]], align 4
-// CHECK11-NEXT: [[TMP250:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP241]], ptr [[TMP250]], align 8
-// CHECK11-NEXT: [[TMP251:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l90.region_id, ptr [[KERNEL_ARGS86]])
-// CHECK11-NEXT: [[TMP252:%.*]] = icmp ne i32 [[TMP251]], 0
-// CHECK11-NEXT: br i1 [[TMP252]], label [[OMP_OFFLOAD_FAILED87:%.*]], label [[OMP_OFFLOAD_CONT88:%.*]]
+// CHECK11-NEXT: [[TMP266:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP266]], align 4
+// CHECK11-NEXT: [[TMP267:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 1
+// CHECK11-NEXT: store i32 5, ptr [[TMP267]], align 4
+// CHECK11-NEXT: [[TMP268:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP260]], ptr [[TMP268]], align 4
+// CHECK11-NEXT: [[TMP269:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP261]], ptr [[TMP269]], align 4
+// CHECK11-NEXT: [[TMP270:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.52, ptr [[TMP270]], align 4
+// CHECK11-NEXT: [[TMP271:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.53, ptr [[TMP271]], align 4
+// CHECK11-NEXT: [[TMP272:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP272]], align 4
+// CHECK11-NEXT: [[TMP273:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP273]], align 4
+// CHECK11-NEXT: [[TMP274:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP265]], ptr [[TMP274]], align 8
+// CHECK11-NEXT: [[TMP275:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP275]], align 8
+// CHECK11-NEXT: [[TMP276:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP276]], align 4
+// CHECK11-NEXT: [[TMP277:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP277]], align 4
+// CHECK11-NEXT: [[TMP278:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP278]], align 4
+// CHECK11-NEXT: [[TMP279:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l90.region_id, ptr [[KERNEL_ARGS86]])
+// CHECK11-NEXT: [[TMP280:%.*]] = icmp ne i32 [[TMP279]], 0
+// CHECK11-NEXT: br i1 [[TMP280]], label [[OMP_OFFLOAD_FAILED87:%.*]], label [[OMP_OFFLOAD_CONT88:%.*]]
// CHECK11: omp_offload.failed87:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l90(i32 [[TMP215]], i32 [[TMP217]], ptr [[TMP218]], ptr [[TMP219]], ptr [[TMP220]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l90(i32 [[TMP239]], i32 [[TMP241]], ptr [[TMP242]], ptr [[TMP243]], ptr [[TMP244]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT88]]
// CHECK11: omp_offload.cont88:
// CHECK11-NEXT: ret i32 0
diff --git a/clang/test/OpenMP/distribute_parallel_for_simd_firstprivate_codegen.cpp b/clang/test/OpenMP/distribute_parallel_for_simd_firstprivate_codegen.cpp
index e0cd9f79fa484..af7cddc3114c1 100644
--- a/clang/test/OpenMP/distribute_parallel_for_simd_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/distribute_parallel_for_simd_firstprivate_codegen.cpp
@@ -902,7 +902,7 @@ int main() {
// CHECK8-NEXT: [[TMP24:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK8-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK8-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK8-NEXT: store i32 1, ptr [[TMP25]], align 4
+// CHECK8-NEXT: store i32 2, ptr [[TMP25]], align 4
// CHECK8-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK8-NEXT: store i32 5, ptr [[TMP26]], align 4
// CHECK8-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -919,9 +919,17 @@ int main() {
// CHECK8-NEXT: store ptr null, ptr [[TMP32]], align 8
// CHECK8-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK8-NEXT: store i64 2, ptr [[TMP33]], align 8
-// CHECK8-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l138.region_id, ptr [[KERNEL_ARGS]])
-// CHECK8-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
-// CHECK8-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK8-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK8-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK8-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK8-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK8-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK8-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP36]], align 4
+// CHECK8-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK8-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK8-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l138.region_id, ptr [[KERNEL_ARGS]])
+// CHECK8-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK8-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK8: omp_offload.failed:
// CHECK8-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l138(i64 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]], i64 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK8-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -929,18 +937,18 @@ int main() {
// CHECK8-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiET_v()
// CHECK8-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK8-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK8-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK8-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK8-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK8: arraydestroy.body:
-// CHECK8-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP36]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK8-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP40]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK8-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK8-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK8-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK8-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK8: arraydestroy.done2:
// CHECK8-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK8-NEXT: [[TMP37:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK8-NEXT: ret i32 [[TMP37]]
+// CHECK8-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK8-NEXT: ret i32 [[TMP41]]
//
//
// CHECK8-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -1334,7 +1342,7 @@ int main() {
// CHECK8-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK8-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK8-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK8-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK8-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK8-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK8-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK8-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1351,27 +1359,35 @@ int main() {
// CHECK8-NEXT: store ptr null, ptr [[TMP27]], align 8
// CHECK8-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK8-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK8-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l48.region_id, ptr [[KERNEL_ARGS]])
-// CHECK8-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK8-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK8-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK8-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK8-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK8-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK8-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK8-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP31]], align 4
+// CHECK8-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK8-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK8-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l48.region_id, ptr [[KERNEL_ARGS]])
+// CHECK8-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK8-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK8: omp_offload.failed:
// CHECK8-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l48(i64 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK8-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK8: omp_offload.cont:
// CHECK8-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK8-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK8-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK8-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK8-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK8: arraydestroy.body:
-// CHECK8-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK8-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK8-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK8-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK8-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK8-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK8: arraydestroy.done2:
// CHECK8-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK8-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK8-NEXT: ret i32 [[TMP32]]
+// CHECK8-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK8-NEXT: ret i32 [[TMP36]]
//
//
// CHECK8-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
@@ -1840,7 +1856,7 @@ int main() {
// CHECK10-NEXT: [[TMP24:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK10-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK10-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK10-NEXT: store i32 1, ptr [[TMP25]], align 4
+// CHECK10-NEXT: store i32 2, ptr [[TMP25]], align 4
// CHECK10-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK10-NEXT: store i32 5, ptr [[TMP26]], align 4
// CHECK10-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1857,9 +1873,17 @@ int main() {
// CHECK10-NEXT: store ptr null, ptr [[TMP32]], align 4
// CHECK10-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK10-NEXT: store i64 2, ptr [[TMP33]], align 8
-// CHECK10-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l138.region_id, ptr [[KERNEL_ARGS]])
-// CHECK10-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
-// CHECK10-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK10-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK10-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK10-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK10-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK10-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK10-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP36]], align 4
+// CHECK10-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK10-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK10-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l138.region_id, ptr [[KERNEL_ARGS]])
+// CHECK10-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK10-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK10: omp_offload.failed:
// CHECK10-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l138(i32 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]], i32 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK10-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1867,18 +1891,18 @@ int main() {
// CHECK10-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiET_v()
// CHECK10-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK10-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK10-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK10-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK10-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK10: arraydestroy.body:
-// CHECK10-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP36]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK10-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP40]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK10-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK10-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK10-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK10-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK10: arraydestroy.done2:
// CHECK10-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK10-NEXT: [[TMP37:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK10-NEXT: ret i32 [[TMP37]]
+// CHECK10-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK10-NEXT: ret i32 [[TMP41]]
//
//
// CHECK10-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -2266,7 +2290,7 @@ int main() {
// CHECK10-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK10-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK10-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK10-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK10-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK10-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK10-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK10-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2283,27 +2307,35 @@ int main() {
// CHECK10-NEXT: store ptr null, ptr [[TMP27]], align 4
// CHECK10-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK10-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK10-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l48.region_id, ptr [[KERNEL_ARGS]])
-// CHECK10-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK10-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK10-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK10-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK10-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK10-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK10-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK10-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP31]], align 4
+// CHECK10-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK10-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK10-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l48.region_id, ptr [[KERNEL_ARGS]])
+// CHECK10-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK10-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK10: omp_offload.failed:
// CHECK10-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l48(i32 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK10-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK10: omp_offload.cont:
// CHECK10-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK10-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK10-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK10-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK10-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK10: arraydestroy.body:
-// CHECK10-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK10-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK10-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK10-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK10-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK10-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK10: arraydestroy.done2:
// CHECK10-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK10-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK10-NEXT: ret i32 [[TMP32]]
+// CHECK10-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK10-NEXT: ret i32 [[TMP36]]
//
//
// CHECK10-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
diff --git a/clang/test/OpenMP/distribute_parallel_for_simd_if_codegen.cpp b/clang/test/OpenMP/distribute_parallel_for_simd_if_codegen.cpp
index 3d8bc48ccc660..ad23bf5bdc640 100644
--- a/clang/test/OpenMP/distribute_parallel_for_simd_if_codegen.cpp
+++ b/clang/test/OpenMP/distribute_parallel_for_simd_if_codegen.cpp
@@ -124,7 +124,7 @@ int main() {
// CHECK1-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -141,35 +141,51 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l43.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l43.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l43() #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l48.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l48.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK1-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l48() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -499,7 +515,7 @@ int main() {
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -516,78 +532,102 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l90.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l90.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK1-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l90() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK1: omp_offload.cont4:
-// CHECK1-NEXT: [[TMP22:%.*]] = load i32, ptr @Arg, align 4
-// CHECK1-NEXT: store i32 [[TMP22]], ptr [[ARG_CASTED]], align 4
-// CHECK1-NEXT: [[TMP23:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP23]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP23]], ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP30:%.*]] = load i32, ptr @Arg, align 4
+// CHECK1-NEXT: store i32 [[TMP30]], ptr [[ARG_CASTED]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP31]], ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP31]], ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP29]], align 4
-// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP30]], align 4
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP27]], ptr [[TMP31]], align 8
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP28]], ptr [[TMP32]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP35]], align 8
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP36]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l99.region_id, ptr [[KERNEL_ARGS6]])
-// CHECK1-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK1-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP38]], align 4
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP44]], align 8
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP45]], align 8
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP46]], align 8
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP47]], align 4
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP48]], align 4
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP49]], align 4
+// CHECK1-NEXT: [[TMP50:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l99.region_id, ptr [[KERNEL_ARGS6]])
+// CHECK1-NEXT: [[TMP51:%.*]] = icmp ne i32 [[TMP50]], 0
+// CHECK1-NEXT: br i1 [[TMP51]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK1: omp_offload.failed7:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l99(i64 [[TMP23]]) #[[ATTR2]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l99(i64 [[TMP31]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK1: omp_offload.cont8:
-// CHECK1-NEXT: [[TMP40:%.*]] = load i32, ptr @Arg, align 4
-// CHECK1-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP40]])
+// CHECK1-NEXT: [[TMP52:%.*]] = load i32, ptr @Arg, align 4
+// CHECK1-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP52]])
// CHECK1-NEXT: ret i32 [[CALL]]
//
//
@@ -1085,7 +1125,7 @@ int main() {
// CHECK1-NEXT: store i32 [[ARG]], ptr [[ARG_ADDR]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1102,74 +1142,98 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l59.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l59.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l59() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l65.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l65.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK1-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l65() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK1: omp_offload.cont4:
-// CHECK1-NEXT: [[TMP22:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
-// CHECK1-NEXT: store i32 [[TMP22]], ptr [[ARG_CASTED]], align 4
-// CHECK1-NEXT: [[TMP23:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP23]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP23]], ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP30:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
+// CHECK1-NEXT: store i32 [[TMP30]], ptr [[ARG_CASTED]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP31]], ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP31]], ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP29]], align 4
-// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP30]], align 4
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP27]], ptr [[TMP31]], align 8
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP28]], ptr [[TMP32]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.16, ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP35]], align 8
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP36]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l71.region_id, ptr [[KERNEL_ARGS6]])
-// CHECK1-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK1-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP38]], align 4
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.16, ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP44]], align 8
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP45]], align 8
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP46]], align 8
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP47]], align 4
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP48]], align 4
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP49]], align 4
+// CHECK1-NEXT: [[TMP50:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l71.region_id, ptr [[KERNEL_ARGS6]])
+// CHECK1-NEXT: [[TMP51:%.*]] = icmp ne i32 [[TMP50]], 0
+// CHECK1-NEXT: br i1 [[TMP51]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK1: omp_offload.failed7:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l71(i64 [[TMP23]]) #[[ATTR2]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l71(i64 [[TMP31]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK1: omp_offload.cont8:
// CHECK1-NEXT: ret i32 0
@@ -1669,7 +1733,7 @@ int main() {
// CHECK3-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1686,35 +1750,51 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l43.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK3-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l43.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK3-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l43() #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK3-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK3-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK3-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK3-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK3-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK3-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK3-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK3-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK3-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l48.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK3-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK3-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK3-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK3-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK3-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l48.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK3-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK3-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK3: omp_offload.failed3:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l48() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -2044,7 +2124,7 @@ int main() {
// CHECK3-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2061,78 +2141,102 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK3-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK3-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK3-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK3-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK3-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK3-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK3-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK3-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK3-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK3-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK3-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l90.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK3-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK3-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK3-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK3-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK3-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l90.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK3-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK3-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK3: omp_offload.failed3:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l90() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK3: omp_offload.cont4:
-// CHECK3-NEXT: [[TMP22:%.*]] = load i32, ptr @Arg, align 4
-// CHECK3-NEXT: store i32 [[TMP22]], ptr [[ARG_CASTED]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK3-NEXT: store i64 [[TMP23]], ptr [[TMP24]], align 8
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK3-NEXT: store i64 [[TMP23]], ptr [[TMP25]], align 8
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP26]], align 8
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP30:%.*]] = load i32, ptr @Arg, align 4
+// CHECK3-NEXT: store i32 [[TMP30]], ptr [[ARG_CASTED]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK3-NEXT: store i64 [[TMP31]], ptr [[TMP32]], align 8
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK3-NEXT: store i64 [[TMP31]], ptr [[TMP33]], align 8
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP29]], align 4
-// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP30]], align 4
-// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP27]], ptr [[TMP31]], align 8
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP28]], ptr [[TMP32]], align 8
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes, ptr [[TMP33]], align 8
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP34]], align 8
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP35]], align 8
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP36]], align 8
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK3-NEXT: store i64 100, ptr [[TMP37]], align 8
-// CHECK3-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l99.region_id, ptr [[KERNEL_ARGS6]])
-// CHECK3-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK3-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP38]], align 4
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes, ptr [[TMP41]], align 8
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP42]], align 8
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP44]], align 8
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK3-NEXT: store i64 100, ptr [[TMP45]], align 8
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP46]], align 8
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP47]], align 4
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP48]], align 4
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP49]], align 4
+// CHECK3-NEXT: [[TMP50:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l99.region_id, ptr [[KERNEL_ARGS6]])
+// CHECK3-NEXT: [[TMP51:%.*]] = icmp ne i32 [[TMP50]], 0
+// CHECK3-NEXT: br i1 [[TMP51]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK3: omp_offload.failed7:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l99(i64 [[TMP23]]) #[[ATTR2]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l99(i64 [[TMP31]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK3: omp_offload.cont8:
-// CHECK3-NEXT: [[TMP40:%.*]] = load i32, ptr @Arg, align 4
-// CHECK3-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP40]])
+// CHECK3-NEXT: [[TMP52:%.*]] = load i32, ptr @Arg, align 4
+// CHECK3-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP52]])
// CHECK3-NEXT: ret i32 [[CALL]]
//
//
@@ -2865,7 +2969,7 @@ int main() {
// CHECK3-NEXT: store i32 [[ARG]], ptr [[ARG_ADDR]], align 4
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2882,74 +2986,98 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l59.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK3-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l59.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK3-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l59() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK3-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK3-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK3-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK3-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK3-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK3-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK3-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK3-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK3-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l65.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK3-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK3-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK3-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK3-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK3-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l65.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK3-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK3-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK3: omp_offload.failed3:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l65() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK3: omp_offload.cont4:
-// CHECK3-NEXT: [[TMP22:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
-// CHECK3-NEXT: store i32 [[TMP22]], ptr [[ARG_CASTED]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK3-NEXT: store i64 [[TMP23]], ptr [[TMP24]], align 8
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK3-NEXT: store i64 [[TMP23]], ptr [[TMP25]], align 8
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP26]], align 8
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP30:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
+// CHECK3-NEXT: store i32 [[TMP30]], ptr [[ARG_CASTED]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK3-NEXT: store i64 [[TMP31]], ptr [[TMP32]], align 8
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK3-NEXT: store i64 [[TMP31]], ptr [[TMP33]], align 8
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP29]], align 4
-// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP30]], align 4
-// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP27]], ptr [[TMP31]], align 8
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP28]], ptr [[TMP32]], align 8
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.17, ptr [[TMP33]], align 8
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.18, ptr [[TMP34]], align 8
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP35]], align 8
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP36]], align 8
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK3-NEXT: store i64 100, ptr [[TMP37]], align 8
-// CHECK3-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l71.region_id, ptr [[KERNEL_ARGS6]])
-// CHECK3-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK3-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP38]], align 4
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.17, ptr [[TMP41]], align 8
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.18, ptr [[TMP42]], align 8
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP44]], align 8
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK3-NEXT: store i64 100, ptr [[TMP45]], align 8
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP46]], align 8
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP47]], align 4
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP48]], align 4
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP49]], align 4
+// CHECK3-NEXT: [[TMP50:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l71.region_id, ptr [[KERNEL_ARGS6]])
+// CHECK3-NEXT: [[TMP51:%.*]] = icmp ne i32 [[TMP50]], 0
+// CHECK3-NEXT: br i1 [[TMP51]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK3: omp_offload.failed7:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l71(i64 [[TMP23]]) #[[ATTR2]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l71(i64 [[TMP31]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK3: omp_offload.cont8:
// CHECK3-NEXT: ret i32 0
@@ -4025,7 +4153,7 @@ int main() {
// CHECK9-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK9-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK9-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -4042,35 +4170,51 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK9-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK9-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l43.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK9-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK9-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK9-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l43.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK9-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l43() #[[ATTR2:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK9-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK9-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK9-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK9-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK9-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK9-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK9-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l48.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK9-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK9-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK9-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK9-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK9-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK9-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l48.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK9-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK9-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK9: omp_offload.failed3:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l48() #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -4400,7 +4544,7 @@ int main() {
// CHECK9-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK9-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK9-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -4417,78 +4561,102 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK9-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK9-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK9-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK9-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK9-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK9-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81() #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK9-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK9-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK9-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK9-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK9-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK9-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK9-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l90.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK9-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK9-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK9-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK9-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK9-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK9-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l90.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK9-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK9-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK9: omp_offload.failed3:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l90() #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK9: omp_offload.cont4:
-// CHECK9-NEXT: [[TMP22:%.*]] = load i32, ptr @Arg, align 4
-// CHECK9-NEXT: store i32 [[TMP22]], ptr [[ARG_CASTED]], align 4
-// CHECK9-NEXT: [[TMP23:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
-// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP23]], ptr [[TMP24]], align 8
-// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP23]], ptr [[TMP25]], align 8
-// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP26]], align 8
-// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP30:%.*]] = load i32, ptr @Arg, align 4
+// CHECK9-NEXT: store i32 [[TMP30]], ptr [[ARG_CASTED]], align 4
+// CHECK9-NEXT: [[TMP31:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP31]], ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP31]], ptr [[TMP33]], align 8
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP29]], align 4
-// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK9-NEXT: store i32 1, ptr [[TMP30]], align 4
-// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP27]], ptr [[TMP31]], align 8
-// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP28]], ptr [[TMP32]], align 8
-// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes, ptr [[TMP33]], align 8
-// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes, ptr [[TMP34]], align 8
-// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP35]], align 8
-// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP36]], align 8
-// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK9-NEXT: store i64 100, ptr [[TMP37]], align 8
-// CHECK9-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l99.region_id, ptr [[KERNEL_ARGS6]])
-// CHECK9-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK9-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP37]], align 4
+// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK9-NEXT: store i32 1, ptr [[TMP38]], align 4
+// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
+// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
+// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes, ptr [[TMP41]], align 8
+// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes, ptr [[TMP42]], align 8
+// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP44]], align 8
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK9-NEXT: store i64 100, ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP46]], align 8
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP47]], align 4
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP48]], align 4
+// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP49]], align 4
+// CHECK9-NEXT: [[TMP50:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l99.region_id, ptr [[KERNEL_ARGS6]])
+// CHECK9-NEXT: [[TMP51:%.*]] = icmp ne i32 [[TMP50]], 0
+// CHECK9-NEXT: br i1 [[TMP51]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK9: omp_offload.failed7:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l99(i64 [[TMP23]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l99(i64 [[TMP31]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK9: omp_offload.cont8:
-// CHECK9-NEXT: [[TMP40:%.*]] = load i32, ptr @Arg, align 4
-// CHECK9-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP40]])
+// CHECK9-NEXT: [[TMP52:%.*]] = load i32, ptr @Arg, align 4
+// CHECK9-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP52]])
// CHECK9-NEXT: ret i32 [[CALL]]
//
//
@@ -4986,7 +5154,7 @@ int main() {
// CHECK9-NEXT: store i32 [[ARG]], ptr [[ARG_ADDR]], align 4
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK9-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK9-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -5003,74 +5171,98 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK9-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK9-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l59.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK9-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK9-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK9-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l59.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK9-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l59() #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK9-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK9-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK9-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK9-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK9-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK9-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK9-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l65.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK9-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK9-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK9-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK9-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK9-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK9-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l65.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK9-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK9-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK9: omp_offload.failed3:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l65() #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK9: omp_offload.cont4:
-// CHECK9-NEXT: [[TMP22:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
-// CHECK9-NEXT: store i32 [[TMP22]], ptr [[ARG_CASTED]], align 4
-// CHECK9-NEXT: [[TMP23:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
-// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP23]], ptr [[TMP24]], align 8
-// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP23]], ptr [[TMP25]], align 8
-// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP26]], align 8
-// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP30:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
+// CHECK9-NEXT: store i32 [[TMP30]], ptr [[ARG_CASTED]], align 4
+// CHECK9-NEXT: [[TMP31:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP31]], ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP31]], ptr [[TMP33]], align 8
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP29]], align 4
-// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK9-NEXT: store i32 1, ptr [[TMP30]], align 4
-// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP27]], ptr [[TMP31]], align 8
-// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP28]], ptr [[TMP32]], align 8
-// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.16, ptr [[TMP33]], align 8
-// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP34]], align 8
-// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP35]], align 8
-// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP36]], align 8
-// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK9-NEXT: store i64 100, ptr [[TMP37]], align 8
-// CHECK9-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l71.region_id, ptr [[KERNEL_ARGS6]])
-// CHECK9-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK9-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP37]], align 4
+// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK9-NEXT: store i32 1, ptr [[TMP38]], align 4
+// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
+// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
+// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.16, ptr [[TMP41]], align 8
+// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP42]], align 8
+// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP44]], align 8
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK9-NEXT: store i64 100, ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP46]], align 8
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP47]], align 4
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP48]], align 4
+// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP49]], align 4
+// CHECK9-NEXT: [[TMP50:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l71.region_id, ptr [[KERNEL_ARGS6]])
+// CHECK9-NEXT: [[TMP51:%.*]] = icmp ne i32 [[TMP50]], 0
+// CHECK9-NEXT: br i1 [[TMP51]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK9: omp_offload.failed7:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l71(i64 [[TMP23]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l71(i64 [[TMP31]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK9: omp_offload.cont8:
// CHECK9-NEXT: ret i32 0
@@ -5570,7 +5762,7 @@ int main() {
// CHECK11-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK11-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK11-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -5587,35 +5779,51 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK11-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK11-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l43.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK11-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK11-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK11-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK11-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l43.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK11-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l43() #[[ATTR2:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK11-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK11-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK11-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK11-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK11-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK11-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK11-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK11-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK11-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK11-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK11-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l48.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK11-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK11-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK11-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK11-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK11-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK11-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l48.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK11-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK11-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK11: omp_offload.failed3:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l48() #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -5945,7 +6153,7 @@ int main() {
// CHECK11-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK11-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK11-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -5962,78 +6170,102 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK11-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK11-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK11-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK11-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK11-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK11-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK11-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81() #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK11-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK11-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK11-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK11-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK11-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK11-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK11-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK11-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK11-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK11-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK11-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l90.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK11-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK11-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK11-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK11-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK11-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK11-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l90.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK11-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK11-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK11: omp_offload.failed3:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l90() #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK11: omp_offload.cont4:
-// CHECK11-NEXT: [[TMP22:%.*]] = load i32, ptr @Arg, align 4
-// CHECK11-NEXT: store i32 [[TMP22]], ptr [[ARG_CASTED]], align 4
-// CHECK11-NEXT: [[TMP23:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
-// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK11-NEXT: store i64 [[TMP23]], ptr [[TMP24]], align 8
-// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK11-NEXT: store i64 [[TMP23]], ptr [[TMP25]], align 8
-// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP26]], align 8
-// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP30:%.*]] = load i32, ptr @Arg, align 4
+// CHECK11-NEXT: store i32 [[TMP30]], ptr [[ARG_CASTED]], align 4
+// CHECK11-NEXT: [[TMP31:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK11-NEXT: store i64 [[TMP31]], ptr [[TMP32]], align 8
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK11-NEXT: store i64 [[TMP31]], ptr [[TMP33]], align 8
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP29]], align 4
-// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK11-NEXT: store i32 1, ptr [[TMP30]], align 4
-// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP27]], ptr [[TMP31]], align 8
-// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP28]], ptr [[TMP32]], align 8
-// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes, ptr [[TMP33]], align 8
-// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes, ptr [[TMP34]], align 8
-// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP35]], align 8
-// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP36]], align 8
-// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK11-NEXT: store i64 100, ptr [[TMP37]], align 8
-// CHECK11-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l99.region_id, ptr [[KERNEL_ARGS6]])
-// CHECK11-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK11-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP37]], align 4
+// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK11-NEXT: store i32 1, ptr [[TMP38]], align 4
+// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
+// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
+// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes, ptr [[TMP41]], align 8
+// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes, ptr [[TMP42]], align 8
+// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP44]], align 8
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK11-NEXT: store i64 100, ptr [[TMP45]], align 8
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP46]], align 8
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP47]], align 4
+// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP48]], align 4
+// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP49]], align 4
+// CHECK11-NEXT: [[TMP50:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l99.region_id, ptr [[KERNEL_ARGS6]])
+// CHECK11-NEXT: [[TMP51:%.*]] = icmp ne i32 [[TMP50]], 0
+// CHECK11-NEXT: br i1 [[TMP51]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK11: omp_offload.failed7:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l99(i64 [[TMP23]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l99(i64 [[TMP31]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK11: omp_offload.cont8:
-// CHECK11-NEXT: [[TMP40:%.*]] = load i32, ptr @Arg, align 4
-// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP40]])
+// CHECK11-NEXT: [[TMP52:%.*]] = load i32, ptr @Arg, align 4
+// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP52]])
// CHECK11-NEXT: ret i32 [[CALL]]
//
//
@@ -6766,7 +6998,7 @@ int main() {
// CHECK11-NEXT: store i32 [[ARG]], ptr [[ARG_ADDR]], align 4
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK11-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK11-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -6783,74 +7015,98 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK11-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK11-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l59.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK11-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK11-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK11-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK11-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l59.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK11-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l59() #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK11-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK11-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK11-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK11-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK11-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK11-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK11-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK11-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK11-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK11-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK11-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l65.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK11-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK11-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK11-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK11-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK11-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK11-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l65.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK11-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK11-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK11: omp_offload.failed3:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l65() #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK11: omp_offload.cont4:
-// CHECK11-NEXT: [[TMP22:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
-// CHECK11-NEXT: store i32 [[TMP22]], ptr [[ARG_CASTED]], align 4
-// CHECK11-NEXT: [[TMP23:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
-// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK11-NEXT: store i64 [[TMP23]], ptr [[TMP24]], align 8
-// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK11-NEXT: store i64 [[TMP23]], ptr [[TMP25]], align 8
-// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP26]], align 8
-// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP30:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
+// CHECK11-NEXT: store i32 [[TMP30]], ptr [[ARG_CASTED]], align 4
+// CHECK11-NEXT: [[TMP31:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK11-NEXT: store i64 [[TMP31]], ptr [[TMP32]], align 8
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK11-NEXT: store i64 [[TMP31]], ptr [[TMP33]], align 8
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP29]], align 4
-// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK11-NEXT: store i32 1, ptr [[TMP30]], align 4
-// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP27]], ptr [[TMP31]], align 8
-// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP28]], ptr [[TMP32]], align 8
-// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.17, ptr [[TMP33]], align 8
-// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.18, ptr [[TMP34]], align 8
-// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP35]], align 8
-// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP36]], align 8
-// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK11-NEXT: store i64 100, ptr [[TMP37]], align 8
-// CHECK11-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l71.region_id, ptr [[KERNEL_ARGS6]])
-// CHECK11-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK11-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP37]], align 4
+// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK11-NEXT: store i32 1, ptr [[TMP38]], align 4
+// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
+// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
+// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.17, ptr [[TMP41]], align 8
+// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.18, ptr [[TMP42]], align 8
+// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP44]], align 8
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK11-NEXT: store i64 100, ptr [[TMP45]], align 8
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP46]], align 8
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP47]], align 4
+// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP48]], align 4
+// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP49]], align 4
+// CHECK11-NEXT: [[TMP50:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l71.region_id, ptr [[KERNEL_ARGS6]])
+// CHECK11-NEXT: [[TMP51:%.*]] = icmp ne i32 [[TMP50]], 0
+// CHECK11-NEXT: br i1 [[TMP51]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK11: omp_offload.failed7:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l71(i64 [[TMP23]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l71(i64 [[TMP31]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK11: omp_offload.cont8:
// CHECK11-NEXT: ret i32 0
diff --git a/clang/test/OpenMP/distribute_parallel_for_simd_lastprivate_codegen.cpp b/clang/test/OpenMP/distribute_parallel_for_simd_lastprivate_codegen.cpp
index 857e5ebca0ba0..78e461c7ce40e 100644
--- a/clang/test/OpenMP/distribute_parallel_for_simd_lastprivate_codegen.cpp
+++ b/clang/test/OpenMP/distribute_parallel_for_simd_lastprivate_codegen.cpp
@@ -882,7 +882,7 @@ int main() {
// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP25]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP25]], align 4
// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 5, ptr [[TMP26]], align 4
// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -899,9 +899,17 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP32]], align 8
// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP33]], align 8
-// CHECK9-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l125.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
-// CHECK9-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP36]], align 4
+// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK9-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l125.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK9-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l125(i64 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]], i64 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -909,18 +917,18 @@ int main() {
// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiET_v()
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP36]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP40]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP37:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP37]]
+// CHECK9-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP41]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -1350,7 +1358,7 @@ int main() {
// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1367,27 +1375,35 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP27]], align 8
// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK9-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK9-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP31]], align 4
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK9-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK9-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49(i64 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP32]]
+// CHECK9-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP36]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
@@ -1893,7 +1909,7 @@ int main() {
// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP25]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP25]], align 4
// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 5, ptr [[TMP26]], align 4
// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1910,9 +1926,17 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP32]], align 4
// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP33]], align 8
-// CHECK11-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l125.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
-// CHECK11-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP36]], align 4
+// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK11-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l125.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK11-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l125(i32 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]], i32 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1920,18 +1944,18 @@ int main() {
// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiET_v()
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP36]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP40]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP37:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP37]]
+// CHECK11-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP41]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -2355,7 +2379,7 @@ int main() {
// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2372,27 +2396,35 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP27]], align 4
// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK11-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK11-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP31]], align 4
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK11-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK11-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49(i32 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP32]]
+// CHECK11-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP36]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
diff --git a/clang/test/OpenMP/distribute_parallel_for_simd_num_threads_codegen.cpp b/clang/test/OpenMP/distribute_parallel_for_simd_num_threads_codegen.cpp
index a00ad0bf04f7c..c2d6814d62ee2 100644
--- a/clang/test/OpenMP/distribute_parallel_for_simd_num_threads_codegen.cpp
+++ b/clang/test/OpenMP/distribute_parallel_for_simd_num_threads_codegen.cpp
@@ -115,7 +115,7 @@ int main() {
// CHECK1-NEXT: store i8 [[CALL]], ptr [[A]], align 1
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -132,61 +132,77 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68() #[[ATTR6:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: lpad:
-// CHECK1-NEXT: [[TMP11:%.*]] = landingpad { ptr, i32 }
+// CHECK1-NEXT: [[TMP15:%.*]] = landingpad { ptr, i32 }
// CHECK1-NEXT: cleanup
-// CHECK1-NEXT: [[TMP12:%.*]] = extractvalue { ptr, i32 } [[TMP11]], 0
-// CHECK1-NEXT: store ptr [[TMP12]], ptr [[EXN_SLOT]], align 8
-// CHECK1-NEXT: [[TMP13:%.*]] = extractvalue { ptr, i32 } [[TMP11]], 1
-// CHECK1-NEXT: store i32 [[TMP13]], ptr [[EHSELECTOR_SLOT]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = extractvalue { ptr, i32 } [[TMP15]], 0
+// CHECK1-NEXT: store ptr [[TMP16]], ptr [[EXN_SLOT]], align 8
+// CHECK1-NEXT: [[TMP17:%.*]] = extractvalue { ptr, i32 } [[TMP15]], 1
+// CHECK1-NEXT: store i32 [[TMP17]], ptr [[EHSELECTOR_SLOT]], align 4
// CHECK1-NEXT: call void @_ZN1SD1Ev(ptr nonnull align 8 dereferenceable(24) [[S]]) #[[ATTR6]]
// CHECK1-NEXT: br label [[EH_RESUME:%.*]]
// CHECK1: omp_offload.cont:
-// CHECK1-NEXT: [[TMP14:%.*]] = load i8, ptr [[A]], align 1
-// CHECK1-NEXT: store i8 [[TMP14]], ptr [[A_CASTED]], align 1
-// CHECK1-NEXT: [[TMP15:%.*]] = load i64, ptr [[A_CASTED]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP15]], ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP15]], ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP18:%.*]] = load i8, ptr [[A]], align 1
+// CHECK1-NEXT: store i8 [[TMP18]], ptr [[A_CASTED]], align 1
+// CHECK1-NEXT: [[TMP19:%.*]] = load i64, ptr [[A_CASTED]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP19]], ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP19]], ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP36]], align 4
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK1-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74(i64 [[TMP15]]) #[[ATTR6]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74(i64 [[TMP19]]) #[[ATTR6]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK1: omp_offload.cont4:
-// CHECK1-NEXT: [[TMP32:%.*]] = load i8, ptr [[A]], align 1
-// CHECK1-NEXT: [[CONV:%.*]] = sext i8 [[TMP32]] to i32
+// CHECK1-NEXT: [[TMP40:%.*]] = load i8, ptr [[A]], align 1
+// CHECK1-NEXT: [[CONV:%.*]] = sext i8 [[TMP40]] to i32
// CHECK1-NEXT: [[CALL6:%.*]] = invoke i32 @_Z5tmainIcLi5EEiv()
// CHECK1-NEXT: to label [[INVOKE_CONT5:%.*]] unwind label [[LPAD]]
// CHECK1: invoke.cont5:
@@ -197,8 +213,8 @@ int main() {
// CHECK1-NEXT: [[ADD9:%.*]] = add nsw i32 [[ADD]], [[CALL8]]
// CHECK1-NEXT: store i32 [[ADD9]], ptr [[RETVAL]], align 4
// CHECK1-NEXT: call void @_ZN1SD1Ev(ptr nonnull align 8 dereferenceable(24) [[S]]) #[[ATTR6]]
-// CHECK1-NEXT: [[TMP33:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK1-NEXT: ret i32 [[TMP33]]
+// CHECK1-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK1-NEXT: ret i32 [[TMP41]]
// CHECK1: eh.resume:
// CHECK1-NEXT: [[EXN:%.*]] = load ptr, ptr [[EXN_SLOT]], align 8
// CHECK1-NEXT: [[SEL:%.*]] = load i32, ptr [[EHSELECTOR_SLOT]], align 4
@@ -576,7 +592,7 @@ int main() {
// CHECK1-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -593,35 +609,51 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l52() #[[ATTR6]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK1-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l57() #[[ATTR6]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -636,7 +668,7 @@ int main() {
// CHECK1-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -653,35 +685,51 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l52() #[[ATTR6]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK1-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l57() #[[ATTR6]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -1749,7 +1797,7 @@ int main() {
// CHECK5-NEXT: store i8 [[CALL]], ptr [[A]], align 1
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK5-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK5-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1766,61 +1814,77 @@ int main() {
// CHECK5-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK5-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK5-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK5-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK5-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK5-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK5-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68() #[[ATTR6:[0-9]+]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK5: lpad:
-// CHECK5-NEXT: [[TMP11:%.*]] = landingpad { ptr, i32 }
+// CHECK5-NEXT: [[TMP15:%.*]] = landingpad { ptr, i32 }
// CHECK5-NEXT: cleanup
-// CHECK5-NEXT: [[TMP12:%.*]] = extractvalue { ptr, i32 } [[TMP11]], 0
-// CHECK5-NEXT: store ptr [[TMP12]], ptr [[EXN_SLOT]], align 8
-// CHECK5-NEXT: [[TMP13:%.*]] = extractvalue { ptr, i32 } [[TMP11]], 1
-// CHECK5-NEXT: store i32 [[TMP13]], ptr [[EHSELECTOR_SLOT]], align 4
+// CHECK5-NEXT: [[TMP16:%.*]] = extractvalue { ptr, i32 } [[TMP15]], 0
+// CHECK5-NEXT: store ptr [[TMP16]], ptr [[EXN_SLOT]], align 8
+// CHECK5-NEXT: [[TMP17:%.*]] = extractvalue { ptr, i32 } [[TMP15]], 1
+// CHECK5-NEXT: store i32 [[TMP17]], ptr [[EHSELECTOR_SLOT]], align 4
// CHECK5-NEXT: call void @_ZN1SD1Ev(ptr nonnull align 8 dereferenceable(24) [[S]]) #[[ATTR6]]
// CHECK5-NEXT: br label [[EH_RESUME:%.*]]
// CHECK5: omp_offload.cont:
-// CHECK5-NEXT: [[TMP14:%.*]] = load i8, ptr [[A]], align 1
-// CHECK5-NEXT: store i8 [[TMP14]], ptr [[A_CASTED]], align 1
-// CHECK5-NEXT: [[TMP15:%.*]] = load i64, ptr [[A_CASTED]], align 8
-// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK5-NEXT: store i64 [[TMP15]], ptr [[TMP16]], align 8
-// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK5-NEXT: store i64 [[TMP15]], ptr [[TMP17]], align 8
-// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK5-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP18:%.*]] = load i8, ptr [[A]], align 1
+// CHECK5-NEXT: store i8 [[TMP18]], ptr [[A_CASTED]], align 1
+// CHECK5-NEXT: [[TMP19:%.*]] = load i64, ptr [[A_CASTED]], align 8
+// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK5-NEXT: store i64 [[TMP19]], ptr [[TMP20]], align 8
+// CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK5-NEXT: store i64 [[TMP19]], ptr [[TMP21]], align 8
+// CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK5-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK5-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK5-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK5-NEXT: store ptr @.offload_sizes, ptr [[TMP25]], align 8
-// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK5-NEXT: store ptr @.offload_maptypes, ptr [[TMP26]], align 8
-// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK5-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
-// CHECK5-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK5-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK5-NEXT: store i64 100, ptr [[TMP29]], align 8
-// CHECK5-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK5-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK5-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK5-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
+// CHECK5-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK5-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK5-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK5-NEXT: store ptr @.offload_sizes, ptr [[TMP29]], align 8
+// CHECK5-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK5-NEXT: store ptr @.offload_maptypes, ptr [[TMP30]], align 8
+// CHECK5-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK5-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK5-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK5-NEXT: store i64 100, ptr [[TMP33]], align 8
+// CHECK5-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK5-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK5-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP36]], align 4
+// CHECK5-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK5-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK5-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK5-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK5: omp_offload.failed3:
-// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74(i64 [[TMP15]]) #[[ATTR6]]
+// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74(i64 [[TMP19]]) #[[ATTR6]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK5: omp_offload.cont4:
-// CHECK5-NEXT: [[TMP32:%.*]] = load i8, ptr [[A]], align 1
-// CHECK5-NEXT: [[CONV:%.*]] = sext i8 [[TMP32]] to i32
+// CHECK5-NEXT: [[TMP40:%.*]] = load i8, ptr [[A]], align 1
+// CHECK5-NEXT: [[CONV:%.*]] = sext i8 [[TMP40]] to i32
// CHECK5-NEXT: [[CALL6:%.*]] = invoke i32 @_Z5tmainIcLi5EEiv()
// CHECK5-NEXT: to label [[INVOKE_CONT5:%.*]] unwind label [[LPAD]]
// CHECK5: invoke.cont5:
@@ -1831,8 +1895,8 @@ int main() {
// CHECK5-NEXT: [[ADD9:%.*]] = add nsw i32 [[ADD]], [[CALL8]]
// CHECK5-NEXT: store i32 [[ADD9]], ptr [[RETVAL]], align 4
// CHECK5-NEXT: call void @_ZN1SD1Ev(ptr nonnull align 8 dereferenceable(24) [[S]]) #[[ATTR6]]
-// CHECK5-NEXT: [[TMP33:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK5-NEXT: ret i32 [[TMP33]]
+// CHECK5-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK5-NEXT: ret i32 [[TMP41]]
// CHECK5: eh.resume:
// CHECK5-NEXT: [[EXN:%.*]] = load ptr, ptr [[EXN_SLOT]], align 8
// CHECK5-NEXT: [[SEL:%.*]] = load i32, ptr [[EHSELECTOR_SLOT]], align 4
@@ -2210,7 +2274,7 @@ int main() {
// CHECK5-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK5-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK5-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2227,35 +2291,51 @@ int main() {
// CHECK5-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK5-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK5-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK5-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK5-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK5-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK5-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l52() #[[ATTR6]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK5: omp_offload.cont:
// CHECK5-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK5-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK5-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK5-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK5-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK5-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK5-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK5-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK5-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK5-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK5-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK5-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK5-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK5-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK5-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK5-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK5-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK5-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK5-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK5-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK5-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK5-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK5-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK5-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK5: omp_offload.failed3:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l57() #[[ATTR6]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -2270,7 +2350,7 @@ int main() {
// CHECK5-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK5-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK5-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2287,35 +2367,51 @@ int main() {
// CHECK5-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK5-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK5-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK5-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK5-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK5-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK5-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l52() #[[ATTR6]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK5: omp_offload.cont:
// CHECK5-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK5-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK5-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK5-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK5-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK5-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK5-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK5-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK5-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK5-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK5-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK5-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK5-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK5-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK5-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK5-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK5-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK5-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK5-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK5-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK5-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK5-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK5-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK5-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK5: omp_offload.failed3:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l57() #[[ATTR6]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -3044,7 +3140,7 @@ int main() {
// CHECK9-NEXT: store i8 [[CALL]], ptr [[A]], align 1
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK9-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK9-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3061,61 +3157,77 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK9-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK9-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK9-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK9-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK9-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK9-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68() #[[ATTR6:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: lpad:
-// CHECK9-NEXT: [[TMP11:%.*]] = landingpad { ptr, i32 }
+// CHECK9-NEXT: [[TMP15:%.*]] = landingpad { ptr, i32 }
// CHECK9-NEXT: cleanup
-// CHECK9-NEXT: [[TMP12:%.*]] = extractvalue { ptr, i32 } [[TMP11]], 0
-// CHECK9-NEXT: store ptr [[TMP12]], ptr [[EXN_SLOT]], align 8
-// CHECK9-NEXT: [[TMP13:%.*]] = extractvalue { ptr, i32 } [[TMP11]], 1
-// CHECK9-NEXT: store i32 [[TMP13]], ptr [[EHSELECTOR_SLOT]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = extractvalue { ptr, i32 } [[TMP15]], 0
+// CHECK9-NEXT: store ptr [[TMP16]], ptr [[EXN_SLOT]], align 8
+// CHECK9-NEXT: [[TMP17:%.*]] = extractvalue { ptr, i32 } [[TMP15]], 1
+// CHECK9-NEXT: store i32 [[TMP17]], ptr [[EHSELECTOR_SLOT]], align 4
// CHECK9-NEXT: call void @_ZN1SD1Ev(ptr nonnull align 8 dereferenceable(24) [[S]]) #[[ATTR6]]
// CHECK9-NEXT: br label [[EH_RESUME:%.*]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP14:%.*]] = load i8, ptr [[A]], align 1
-// CHECK9-NEXT: store i8 [[TMP14]], ptr [[A_CASTED]], align 1
-// CHECK9-NEXT: [[TMP15:%.*]] = load i64, ptr [[A_CASTED]], align 8
-// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP15]], ptr [[TMP16]], align 8
-// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP15]], ptr [[TMP17]], align 8
-// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP18:%.*]] = load i8, ptr [[A]], align 1
+// CHECK9-NEXT: store i8 [[TMP18]], ptr [[A_CASTED]], align 1
+// CHECK9-NEXT: [[TMP19:%.*]] = load i64, ptr [[A_CASTED]], align 8
+// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP19]], ptr [[TMP20]], align 8
+// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP19]], ptr [[TMP21]], align 8
+// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK9-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes, ptr [[TMP25]], align 8
-// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes, ptr [[TMP26]], align 8
-// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK9-NEXT: store i64 100, ptr [[TMP29]], align 8
-// CHECK9-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK9-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK9-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK9-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes, ptr [[TMP29]], align 8
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes, ptr [[TMP30]], align 8
+// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK9-NEXT: store i64 100, ptr [[TMP33]], align 8
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP36]], align 4
+// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK9-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK9-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK9-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK9: omp_offload.failed3:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74(i64 [[TMP15]]) #[[ATTR6]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74(i64 [[TMP19]]) #[[ATTR6]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK9: omp_offload.cont4:
-// CHECK9-NEXT: [[TMP32:%.*]] = load i8, ptr [[A]], align 1
-// CHECK9-NEXT: [[CONV:%.*]] = sext i8 [[TMP32]] to i32
+// CHECK9-NEXT: [[TMP40:%.*]] = load i8, ptr [[A]], align 1
+// CHECK9-NEXT: [[CONV:%.*]] = sext i8 [[TMP40]] to i32
// CHECK9-NEXT: [[CALL6:%.*]] = invoke i32 @_Z5tmainIcLi5EEiv()
// CHECK9-NEXT: to label [[INVOKE_CONT5:%.*]] unwind label [[LPAD]]
// CHECK9: invoke.cont5:
@@ -3126,8 +3238,8 @@ int main() {
// CHECK9-NEXT: [[ADD9:%.*]] = add nsw i32 [[ADD]], [[CALL8]]
// CHECK9-NEXT: store i32 [[ADD9]], ptr [[RETVAL]], align 4
// CHECK9-NEXT: call void @_ZN1SD1Ev(ptr nonnull align 8 dereferenceable(24) [[S]]) #[[ATTR6]]
-// CHECK9-NEXT: [[TMP33:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP33]]
+// CHECK9-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP41]]
// CHECK9: eh.resume:
// CHECK9-NEXT: [[EXN:%.*]] = load ptr, ptr [[EXN_SLOT]], align 8
// CHECK9-NEXT: [[SEL:%.*]] = load i32, ptr [[EHSELECTOR_SLOT]], align 4
@@ -3505,7 +3617,7 @@ int main() {
// CHECK9-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK9-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK9-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3522,35 +3634,51 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK9-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK9-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK9-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK9-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK9-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK9-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l52() #[[ATTR6]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK9-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK9-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK9-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK9-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK9-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK9-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK9-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK9-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK9-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK9-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK9-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK9-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK9-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK9-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK9-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK9: omp_offload.failed3:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l57() #[[ATTR6]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -3565,7 +3693,7 @@ int main() {
// CHECK9-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK9-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK9-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3582,35 +3710,51 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK9-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK9-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK9-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK9-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK9-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK9-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l52() #[[ATTR6]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK9-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK9-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK9-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK9-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK9-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK9-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK9-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK9-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK9-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK9-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK9-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK9-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK9-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK9-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK9-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK9: omp_offload.failed3:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l57() #[[ATTR6]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -4678,7 +4822,7 @@ int main() {
// CHECK13-NEXT: store i8 [[CALL]], ptr [[A]], align 1
// CHECK13-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK13-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK13-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK13-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK13-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK13-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -4695,61 +4839,77 @@ int main() {
// CHECK13-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK13-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK13-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK13-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68.region_id, ptr [[KERNEL_ARGS]])
-// CHECK13-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK13-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK13-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK13-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK13-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK13-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK13-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68.region_id, ptr [[KERNEL_ARGS]])
+// CHECK13-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK13-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK13: omp_offload.failed:
// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68() #[[ATTR6:[0-9]+]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK13: lpad:
-// CHECK13-NEXT: [[TMP11:%.*]] = landingpad { ptr, i32 }
+// CHECK13-NEXT: [[TMP15:%.*]] = landingpad { ptr, i32 }
// CHECK13-NEXT: cleanup
-// CHECK13-NEXT: [[TMP12:%.*]] = extractvalue { ptr, i32 } [[TMP11]], 0
-// CHECK13-NEXT: store ptr [[TMP12]], ptr [[EXN_SLOT]], align 8
-// CHECK13-NEXT: [[TMP13:%.*]] = extractvalue { ptr, i32 } [[TMP11]], 1
-// CHECK13-NEXT: store i32 [[TMP13]], ptr [[EHSELECTOR_SLOT]], align 4
+// CHECK13-NEXT: [[TMP16:%.*]] = extractvalue { ptr, i32 } [[TMP15]], 0
+// CHECK13-NEXT: store ptr [[TMP16]], ptr [[EXN_SLOT]], align 8
+// CHECK13-NEXT: [[TMP17:%.*]] = extractvalue { ptr, i32 } [[TMP15]], 1
+// CHECK13-NEXT: store i32 [[TMP17]], ptr [[EHSELECTOR_SLOT]], align 4
// CHECK13-NEXT: call void @_ZN1SD1Ev(ptr nonnull align 8 dereferenceable(24) [[S]]) #[[ATTR6]]
// CHECK13-NEXT: br label [[EH_RESUME:%.*]]
// CHECK13: omp_offload.cont:
-// CHECK13-NEXT: [[TMP14:%.*]] = load i8, ptr [[A]], align 1
-// CHECK13-NEXT: store i8 [[TMP14]], ptr [[A_CASTED]], align 1
-// CHECK13-NEXT: [[TMP15:%.*]] = load i64, ptr [[A_CASTED]], align 8
-// CHECK13-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP15]], ptr [[TMP16]], align 8
-// CHECK13-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP15]], ptr [[TMP17]], align 8
-// CHECK13-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK13-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP18:%.*]] = load i8, ptr [[A]], align 1
+// CHECK13-NEXT: store i8 [[TMP18]], ptr [[A_CASTED]], align 1
+// CHECK13-NEXT: [[TMP19:%.*]] = load i64, ptr [[A_CASTED]], align 8
+// CHECK13-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP19]], ptr [[TMP20]], align 8
+// CHECK13-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP19]], ptr [[TMP21]], align 8
+// CHECK13-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK13-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK13-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK13-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK13-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK13-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK13-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK13-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK13-NEXT: store ptr @.offload_sizes, ptr [[TMP25]], align 8
-// CHECK13-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes, ptr [[TMP26]], align 8
-// CHECK13-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK13-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK13-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK13-NEXT: store i64 100, ptr [[TMP29]], align 8
-// CHECK13-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK13-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK13-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK13-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK13-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK13-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK13-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK13-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK13-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK13-NEXT: store ptr @.offload_sizes, ptr [[TMP29]], align 8
+// CHECK13-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes, ptr [[TMP30]], align 8
+// CHECK13-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK13-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK13-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK13-NEXT: store i64 100, ptr [[TMP33]], align 8
+// CHECK13-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK13-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK13-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP36]], align 4
+// CHECK13-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK13-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK13-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK13-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK13: omp_offload.failed3:
-// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74(i64 [[TMP15]]) #[[ATTR6]]
+// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l74(i64 [[TMP19]]) #[[ATTR6]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK13: omp_offload.cont4:
-// CHECK13-NEXT: [[TMP32:%.*]] = load i8, ptr [[A]], align 1
-// CHECK13-NEXT: [[CONV:%.*]] = sext i8 [[TMP32]] to i32
+// CHECK13-NEXT: [[TMP40:%.*]] = load i8, ptr [[A]], align 1
+// CHECK13-NEXT: [[CONV:%.*]] = sext i8 [[TMP40]] to i32
// CHECK13-NEXT: [[CALL6:%.*]] = invoke i32 @_Z5tmainIcLi5EEiv()
// CHECK13-NEXT: to label [[INVOKE_CONT5:%.*]] unwind label [[LPAD]]
// CHECK13: invoke.cont5:
@@ -4760,8 +4920,8 @@ int main() {
// CHECK13-NEXT: [[ADD9:%.*]] = add nsw i32 [[ADD]], [[CALL8]]
// CHECK13-NEXT: store i32 [[ADD9]], ptr [[RETVAL]], align 4
// CHECK13-NEXT: call void @_ZN1SD1Ev(ptr nonnull align 8 dereferenceable(24) [[S]]) #[[ATTR6]]
-// CHECK13-NEXT: [[TMP33:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK13-NEXT: ret i32 [[TMP33]]
+// CHECK13-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK13-NEXT: ret i32 [[TMP41]]
// CHECK13: eh.resume:
// CHECK13-NEXT: [[EXN:%.*]] = load ptr, ptr [[EXN_SLOT]], align 8
// CHECK13-NEXT: [[SEL:%.*]] = load i32, ptr [[EHSELECTOR_SLOT]], align 4
@@ -5139,7 +5299,7 @@ int main() {
// CHECK13-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK13-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK13-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK13-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK13-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK13-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK13-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -5156,35 +5316,51 @@ int main() {
// CHECK13-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK13-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK13-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK13-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
-// CHECK13-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK13-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK13-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK13-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK13-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK13-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK13-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
+// CHECK13-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK13-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK13: omp_offload.failed:
// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l52() #[[ATTR6]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK13: omp_offload.cont:
// CHECK13-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK13-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK13-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK13-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK13-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK13-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK13-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK13-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK13-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK13-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK13-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK13-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK13-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK13-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK13-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK13-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK13-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK13-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK13-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK13-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK13-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK13-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK13-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK13-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK13-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK13-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK13-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK13-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK13-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK13-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK13-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK13-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK13-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK13-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK13-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK13-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK13-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK13-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK13-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK13-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK13: omp_offload.failed3:
// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l57() #[[ATTR6]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -5199,7 +5375,7 @@ int main() {
// CHECK13-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK13-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK13-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK13-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK13-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK13-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK13-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -5216,35 +5392,51 @@ int main() {
// CHECK13-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK13-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK13-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK13-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
-// CHECK13-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK13-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK13-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK13-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK13-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK13-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK13-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l52.region_id, ptr [[KERNEL_ARGS]])
+// CHECK13-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK13-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK13: omp_offload.failed:
// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l52() #[[ATTR6]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK13: omp_offload.cont:
// CHECK13-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK13-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK13-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK13-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK13-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK13-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK13-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK13-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK13-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK13-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK13-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK13-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK13-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK13-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK13-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK13-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK13-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK13-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK13-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK13-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK13-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK13-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK13-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK13-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK13-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK13-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK13-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK13-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK13-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK13-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK13-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK13-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK13-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK13-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK13-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK13-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK13-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK13-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l57.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK13-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK13-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK13: omp_offload.failed3:
// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l57() #[[ATTR6]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT4]]
diff --git a/clang/test/OpenMP/distribute_parallel_for_simd_private_codegen.cpp b/clang/test/OpenMP/distribute_parallel_for_simd_private_codegen.cpp
index 5a999951823e7..0816e513772fe 100644
--- a/clang/test/OpenMP/distribute_parallel_for_simd_private_codegen.cpp
+++ b/clang/test/OpenMP/distribute_parallel_for_simd_private_codegen.cpp
@@ -589,7 +589,7 @@ int main() {
// CHECK9-NEXT: store ptr undef, ptr [[_TMP1]], align 8
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK9-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK9-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -606,9 +606,17 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK9-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK9-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l95.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK9-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK9-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK9-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l95.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK9-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l95() #[[ATTR4:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -616,18 +624,18 @@ int main() {
// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiET_v()
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP12]]
+// CHECK9-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP16]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -916,7 +924,7 @@ int main() {
// CHECK9-NEXT: store ptr undef, ptr [[_TMP1]], align 8
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK9-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK9-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -933,27 +941,35 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK9-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK9-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK9-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK9-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK9-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK9-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49() #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP12]]
+// CHECK9-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP16]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
@@ -1319,7 +1335,7 @@ int main() {
// CHECK11-NEXT: store ptr undef, ptr [[_TMP1]], align 4
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK11-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK11-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1336,9 +1352,17 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK11-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK11-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l95.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK11-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK11-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK11-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK11-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l95.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK11-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l95() #[[ATTR4:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1346,18 +1370,18 @@ int main() {
// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiET_v()
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP12]]
+// CHECK11-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP16]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -1640,7 +1664,7 @@ int main() {
// CHECK11-NEXT: store ptr undef, ptr [[_TMP1]], align 4
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK11-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK11-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1657,27 +1681,35 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK11-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK11-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK11-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK11-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK11-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK11-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK11-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49() #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP12]]
+// CHECK11-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP16]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
diff --git a/clang/test/OpenMP/distribute_parallel_for_simd_proc_bind_codegen.cpp b/clang/test/OpenMP/distribute_parallel_for_simd_proc_bind_codegen.cpp
index 674afcc83bd95..d992f389100e3 100644
--- a/clang/test/OpenMP/distribute_parallel_for_simd_proc_bind_codegen.cpp
+++ b/clang/test/OpenMP/distribute_parallel_for_simd_proc_bind_codegen.cpp
@@ -62,7 +62,7 @@ int main() {
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -79,35 +79,51 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 1000, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l37.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l37.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l37() #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 1000, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l41.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 1000, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l41.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK1-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l41() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -426,7 +442,7 @@ int main() {
// CHECK1-NEXT: [[TMP:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -443,9 +459,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 1000, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l29.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l29.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l29() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/distribute_private_codegen.cpp b/clang/test/OpenMP/distribute_private_codegen.cpp
index 115b190633cdc..ab654e7ab3fa1 100644
--- a/clang/test/OpenMP/distribute_private_codegen.cpp
+++ b/clang/test/OpenMP/distribute_private_codegen.cpp
@@ -364,7 +364,7 @@ int main() {
// CHECK9-NEXT: store ptr undef, ptr [[_TMP1]], align 8
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK9-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK9-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -381,35 +381,51 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK9-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK9-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l93.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK9-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK9-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK9-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l93.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK9-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l93() #[[ATTR4:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: [[KERNEL_ARGS3:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 1
-// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 4
-// CHECK9-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 5
-// CHECK9-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 6
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 1
+// CHECK9-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 2
// CHECK9-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 7
+// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 3
// CHECK9-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 8
-// CHECK9-NEXT: store i64 2, ptr [[TMP19]], align 8
-// CHECK9-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102.region_id, ptr [[KERNEL_ARGS3]])
-// CHECK9-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK9-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED4:%.*]], label [[OMP_OFFLOAD_CONT5:%.*]]
+// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 4
+// CHECK9-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 5
+// CHECK9-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 8
+// CHECK9-NEXT: store i64 2, ptr [[TMP23]], align 8
+// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK9-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102.region_id, ptr [[KERNEL_ARGS3]])
+// CHECK9-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK9-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED4:%.*]], label [[OMP_OFFLOAD_CONT5:%.*]]
// CHECK9: omp_offload.failed4:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102() #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT5]]
@@ -417,18 +433,18 @@ int main() {
// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiET_v()
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP22]], [[OMP_OFFLOAD_CONT5]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP30]], [[OMP_OFFLOAD_CONT5]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE6:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done6:
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP23:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP23]]
+// CHECK9-NEXT: [[TMP31:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP31]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -666,7 +682,7 @@ int main() {
// CHECK9-NEXT: store ptr undef, ptr [[_TMP1]], align 8
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK9-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK9-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -683,27 +699,35 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK9-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK9-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK9-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK9-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK9-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK9-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49() #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP12]]
+// CHECK9-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP16]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
@@ -953,7 +977,7 @@ int main() {
// CHECK11-NEXT: store ptr undef, ptr [[_TMP1]], align 4
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK11-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK11-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -970,35 +994,51 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK11-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK11-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l93.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK11-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK11-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK11-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK11-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l93.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK11-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l93() #[[ATTR4:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: [[KERNEL_ARGS3:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK11-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 1
-// CHECK11-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK11-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP13]], align 4
-// CHECK11-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP14]], align 4
-// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 4
-// CHECK11-NEXT: store ptr null, ptr [[TMP15]], align 4
-// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 5
-// CHECK11-NEXT: store ptr null, ptr [[TMP16]], align 4
-// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 6
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 1
+// CHECK11-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 2
// CHECK11-NEXT: store ptr null, ptr [[TMP17]], align 4
-// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 7
+// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 3
// CHECK11-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 8
-// CHECK11-NEXT: store i64 2, ptr [[TMP19]], align 8
-// CHECK11-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102.region_id, ptr [[KERNEL_ARGS3]])
-// CHECK11-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK11-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED4:%.*]], label [[OMP_OFFLOAD_CONT5:%.*]]
+// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 4
+// CHECK11-NEXT: store ptr null, ptr [[TMP19]], align 4
+// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 5
+// CHECK11-NEXT: store ptr null, ptr [[TMP20]], align 4
+// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP21]], align 4
+// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 8
+// CHECK11-NEXT: store i64 2, ptr [[TMP23]], align 8
+// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK11-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102.region_id, ptr [[KERNEL_ARGS3]])
+// CHECK11-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK11-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED4:%.*]], label [[OMP_OFFLOAD_CONT5:%.*]]
// CHECK11: omp_offload.failed4:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102() #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT5]]
@@ -1006,18 +1046,18 @@ int main() {
// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiET_v()
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP22]], [[OMP_OFFLOAD_CONT5]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP30]], [[OMP_OFFLOAD_CONT5]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE6:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done6:
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP23:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP23]]
+// CHECK11-NEXT: [[TMP31:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP31]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -1253,7 +1293,7 @@ int main() {
// CHECK11-NEXT: store ptr undef, ptr [[_TMP1]], align 4
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK11-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK11-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1270,27 +1310,35 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK11-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK11-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK11-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK11-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK11-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK11-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK11-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49() #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP12]]
+// CHECK11-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP16]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
diff --git a/clang/test/OpenMP/distribute_simd_codegen.cpp b/clang/test/OpenMP/distribute_simd_codegen.cpp
index c3f50de0fc1d6..bcf49a213f2ab 100644
--- a/clang/test/OpenMP/distribute_simd_codegen.cpp
+++ b/clang/test/OpenMP/distribute_simd_codegen.cpp
@@ -191,7 +191,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -208,9 +208,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK1-NEXT: store ptr null, ptr [[TMP25]], align 8
// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 4571424, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z23without_schedule_clausePfS_S_S__l70.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK1-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP29]], align 4
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z23without_schedule_clausePfS_S_S__l70.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK1-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z23without_schedule_clausePfS_S_S__l70(ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]]) #[[ATTR3:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -383,7 +391,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -400,9 +408,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK1-NEXT: store ptr null, ptr [[TMP25]], align 8
// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 4571424, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18static_not_chunkedPfS_S_S__l86.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK1-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP29]], align 4
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18static_not_chunkedPfS_S_S__l86.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK1-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18static_not_chunkedPfS_S_S__l86(ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -573,7 +589,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -590,9 +606,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK1-NEXT: store ptr null, ptr [[TMP25]], align 8
// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 16908289, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z14static_chunkedPfS_S_S__l103.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK1-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP29]], align 4
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z14static_chunkedPfS_S_S__l103.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK1-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z14static_chunkedPfS_S_S__l103(ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -782,7 +806,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK1-NEXT: [[TMP15:%.*]] = zext i32 [[ADD4]] to i64
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP16]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP16]], align 4
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 2, ptr [[TMP17]], align 4
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -799,9 +823,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK1-NEXT: store ptr null, ptr [[TMP23]], align 8
// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 [[TMP15]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z12test_precondv_l115.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK1-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP25]], align 8
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP28]], align 4
+// CHECK1-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z12test_precondv_l115.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK1-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z12test_precondv_l115(i64 [[TMP1]], i64 [[TMP3]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -964,7 +996,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -981,9 +1013,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_v_l135.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK1-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP18]], align 4
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_v_l135.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_v_l135(i64 [[TMP1]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1143,7 +1183,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1160,9 +1200,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK3-NEXT: store ptr null, ptr [[TMP25]], align 4
// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 4571424, ptr [[TMP26]], align 8
-// CHECK3-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z23without_schedule_clausePfS_S_S__l70.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK3-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z23without_schedule_clausePfS_S_S__l70.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK3-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z23without_schedule_clausePfS_S_S__l70(ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]]) #[[ATTR3:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1331,7 +1379,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1348,9 +1396,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK3-NEXT: store ptr null, ptr [[TMP25]], align 4
// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 4571424, ptr [[TMP26]], align 8
-// CHECK3-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18static_not_chunkedPfS_S_S__l86.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK3-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18static_not_chunkedPfS_S_S__l86.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK3-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18static_not_chunkedPfS_S_S__l86(ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1517,7 +1573,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1534,9 +1590,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK3-NEXT: store ptr null, ptr [[TMP25]], align 4
// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 16908289, ptr [[TMP26]], align 8
-// CHECK3-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z14static_chunkedPfS_S_S__l103.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK3-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z14static_chunkedPfS_S_S__l103.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK3-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z14static_chunkedPfS_S_S__l103(ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1722,7 +1786,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK3-NEXT: [[TMP15:%.*]] = zext i32 [[ADD4]] to i64
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP16]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP16]], align 4
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 2, ptr [[TMP17]], align 4
// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1739,9 +1803,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK3-NEXT: store ptr null, ptr [[TMP23]], align 4
// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 [[TMP15]], ptr [[TMP24]], align 8
-// CHECK3-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z12test_precondv_l115.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK3-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP25]], align 8
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z12test_precondv_l115.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK3-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z12test_precondv_l115(i32 [[TMP1]], i32 [[TMP3]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1904,7 +1976,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1921,9 +1993,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK3-NEXT: store ptr null, ptr [[TMP14]], align 4
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 100, ptr [[TMP15]], align 8
-// CHECK3-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_v_l135.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK3-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP18]], align 4
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK3-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_v_l135.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK3-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_v_l135(i32 [[TMP1]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -2083,7 +2163,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2100,9 +2180,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK5-NEXT: store ptr null, ptr [[TMP25]], align 8
// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 4571424, ptr [[TMP26]], align 8
-// CHECK5-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z23without_schedule_clausePfS_S_S__l70.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK5-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK5-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK5-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP29]], align 4
+// CHECK5-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK5-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z23without_schedule_clausePfS_S_S__l70.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK5-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z23without_schedule_clausePfS_S_S__l70(ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]]) #[[ATTR3:[0-9]+]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -2275,7 +2363,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2292,9 +2380,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK5-NEXT: store ptr null, ptr [[TMP25]], align 8
// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 4571424, ptr [[TMP26]], align 8
-// CHECK5-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18static_not_chunkedPfS_S_S__l86.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK5-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK5-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK5-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP29]], align 4
+// CHECK5-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK5-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18static_not_chunkedPfS_S_S__l86.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK5-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18static_not_chunkedPfS_S_S__l86(ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]]) #[[ATTR3]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -2465,7 +2561,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2482,9 +2578,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK5-NEXT: store ptr null, ptr [[TMP25]], align 8
// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 16908289, ptr [[TMP26]], align 8
-// CHECK5-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z14static_chunkedPfS_S_S__l103.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK5-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK5-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK5-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP29]], align 4
+// CHECK5-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK5-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z14static_chunkedPfS_S_S__l103.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK5-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z14static_chunkedPfS_S_S__l103(ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]]) #[[ATTR3]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -2674,7 +2778,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK5-NEXT: [[TMP15:%.*]] = zext i32 [[ADD4]] to i64
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP16]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP16]], align 4
// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 2, ptr [[TMP17]], align 4
// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2691,9 +2795,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK5-NEXT: store ptr null, ptr [[TMP23]], align 8
// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 [[TMP15]], ptr [[TMP24]], align 8
-// CHECK5-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z12test_precondv_l115.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK5-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP25]], align 8
+// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP27]], align 4
+// CHECK5-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP28]], align 4
+// CHECK5-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z12test_precondv_l115.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK5-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z12test_precondv_l115(i64 [[TMP1]], i64 [[TMP3]]) #[[ATTR3]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -2887,7 +2999,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK5-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2904,9 +3016,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK5-NEXT: store ptr null, ptr [[TMP14]], align 8
// CHECK5-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 100, ptr [[TMP15]], align 8
-// CHECK5-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_v_l135.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK5-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP18]], align 4
+// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK5-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_v_l135.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK5-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_v_l135(i64 [[TMP1]]) #[[ATTR3]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -3066,7 +3186,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK7-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK7-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK7-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK7-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK7-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK7-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3083,9 +3203,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK7-NEXT: store ptr null, ptr [[TMP25]], align 4
// CHECK7-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK7-NEXT: store i64 4571424, ptr [[TMP26]], align 8
-// CHECK7-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z23without_schedule_clausePfS_S_S__l70.region_id, ptr [[KERNEL_ARGS]])
-// CHECK7-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK7-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK7-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK7-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK7-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP29]], align 4
+// CHECK7-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK7-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z23without_schedule_clausePfS_S_S__l70.region_id, ptr [[KERNEL_ARGS]])
+// CHECK7-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK7-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK7: omp_offload.failed:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z23without_schedule_clausePfS_S_S__l70(ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]]) #[[ATTR3:[0-9]+]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -3254,7 +3382,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK7-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK7-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK7-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK7-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK7-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK7-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3271,9 +3399,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK7-NEXT: store ptr null, ptr [[TMP25]], align 4
// CHECK7-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK7-NEXT: store i64 4571424, ptr [[TMP26]], align 8
-// CHECK7-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18static_not_chunkedPfS_S_S__l86.region_id, ptr [[KERNEL_ARGS]])
-// CHECK7-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK7-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK7-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK7-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK7-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP29]], align 4
+// CHECK7-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK7-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18static_not_chunkedPfS_S_S__l86.region_id, ptr [[KERNEL_ARGS]])
+// CHECK7-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK7-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK7: omp_offload.failed:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18static_not_chunkedPfS_S_S__l86(ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]]) #[[ATTR3]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -3440,7 +3576,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK7-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK7-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK7-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK7-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK7-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK7-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3457,9 +3593,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK7-NEXT: store ptr null, ptr [[TMP25]], align 4
// CHECK7-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK7-NEXT: store i64 16908289, ptr [[TMP26]], align 8
-// CHECK7-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z14static_chunkedPfS_S_S__l103.region_id, ptr [[KERNEL_ARGS]])
-// CHECK7-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK7-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK7-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK7-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK7-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP29]], align 4
+// CHECK7-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK7-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z14static_chunkedPfS_S_S__l103.region_id, ptr [[KERNEL_ARGS]])
+// CHECK7-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK7-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK7: omp_offload.failed:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z14static_chunkedPfS_S_S__l103(ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]]) #[[ATTR3]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -3645,7 +3789,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK7-NEXT: [[TMP15:%.*]] = zext i32 [[ADD4]] to i64
// CHECK7-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK7-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP16]], align 4
+// CHECK7-NEXT: store i32 2, ptr [[TMP16]], align 4
// CHECK7-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK7-NEXT: store i32 2, ptr [[TMP17]], align 4
// CHECK7-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3662,9 +3806,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK7-NEXT: store ptr null, ptr [[TMP23]], align 4
// CHECK7-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK7-NEXT: store i64 [[TMP15]], ptr [[TMP24]], align 8
-// CHECK7-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z12test_precondv_l115.region_id, ptr [[KERNEL_ARGS]])
-// CHECK7-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK7-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK7-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP25]], align 8
+// CHECK7-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK7-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP27]], align 4
+// CHECK7-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP28]], align 4
+// CHECK7-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z12test_precondv_l115.region_id, ptr [[KERNEL_ARGS]])
+// CHECK7-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK7-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK7: omp_offload.failed:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z12test_precondv_l115(i32 [[TMP1]], i32 [[TMP3]]) #[[ATTR3]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -3858,7 +4010,7 @@ int fint(void) { return ftemplate<int>(); }
// CHECK7-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK7-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK7-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK7-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK7-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK7-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3875,9 +4027,17 @@ int fint(void) { return ftemplate<int>(); }
// CHECK7-NEXT: store ptr null, ptr [[TMP14]], align 4
// CHECK7-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK7-NEXT: store i64 100, ptr [[TMP15]], align 8
-// CHECK7-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_v_l135.region_id, ptr [[KERNEL_ARGS]])
-// CHECK7-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK7-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK7-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK7-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK7-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP18]], align 4
+// CHECK7-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK7-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_v_l135.region_id, ptr [[KERNEL_ARGS]])
+// CHECK7-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK7-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK7: omp_offload.failed:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_v_l135(i32 [[TMP1]]) #[[ATTR3]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/distribute_simd_firstprivate_codegen.cpp b/clang/test/OpenMP/distribute_simd_firstprivate_codegen.cpp
index fa9180e8f7932..60ed8f0e9eb68 100644
--- a/clang/test/OpenMP/distribute_simd_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/distribute_simd_firstprivate_codegen.cpp
@@ -609,7 +609,7 @@ int main() {
// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP25]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP25]], align 4
// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 5, ptr [[TMP26]], align 4
// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -626,9 +626,17 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP32]], align 8
// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP33]], align 8
-// CHECK9-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l105.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
-// CHECK9-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP36]], align 4
+// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK9-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l105.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK9-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l105(i64 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]], i64 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -636,18 +644,18 @@ int main() {
// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiET_v()
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP36]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP40]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP37:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP37]]
+// CHECK9-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP41]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -904,7 +912,7 @@ int main() {
// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -921,27 +929,35 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP27]], align 8
// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK9-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK9-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP31]], align 4
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK9-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK9-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49(i64 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP32]]
+// CHECK9-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP36]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
@@ -1279,7 +1295,7 @@ int main() {
// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP25]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP25]], align 4
// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 5, ptr [[TMP26]], align 4
// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1296,9 +1312,17 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP32]], align 4
// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP33]], align 8
-// CHECK11-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l105.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
-// CHECK11-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP36]], align 4
+// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK11-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l105.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK11-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l105(i32 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]], i32 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1306,18 +1330,18 @@ int main() {
// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiET_v()
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP36]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP40]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP37:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP37]]
+// CHECK11-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP41]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -1572,7 +1596,7 @@ int main() {
// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1589,27 +1613,35 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP27]], align 4
// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK11-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK11-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP31]], align 4
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK11-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK11-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49(i32 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP32]]
+// CHECK11-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP36]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
diff --git a/clang/test/OpenMP/distribute_simd_lastprivate_codegen.cpp b/clang/test/OpenMP/distribute_simd_lastprivate_codegen.cpp
index 5166d35e52d73..d540eda5dbfc5 100644
--- a/clang/test/OpenMP/distribute_simd_lastprivate_codegen.cpp
+++ b/clang/test/OpenMP/distribute_simd_lastprivate_codegen.cpp
@@ -596,7 +596,7 @@ int main() {
// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP25]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP25]], align 4
// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 5, ptr [[TMP26]], align 4
// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -613,9 +613,17 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP32]], align 8
// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP33]], align 8
-// CHECK9-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
-// CHECK9-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP36]], align 4
+// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK9-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK9-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102(i64 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]], i64 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -623,18 +631,18 @@ int main() {
// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiET_v()
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP36]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP40]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP37:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP37]]
+// CHECK9-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP41]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -909,7 +917,7 @@ int main() {
// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -926,27 +934,35 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP27]], align 8
// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK9-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK9-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP31]], align 4
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK9-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK9-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49(i64 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP32]]
+// CHECK9-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP36]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
@@ -1303,7 +1319,7 @@ int main() {
// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP25]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP25]], align 4
// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 5, ptr [[TMP26]], align 4
// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1320,9 +1336,17 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP32]], align 4
// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP33]], align 8
-// CHECK11-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
-// CHECK11-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP36]], align 4
+// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK11-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK11-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102(i32 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]], i32 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1330,18 +1354,18 @@ int main() {
// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiET_v()
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP36]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP40]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP37:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP37]]
+// CHECK11-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP41]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -1614,7 +1638,7 @@ int main() {
// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1631,27 +1655,35 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP27]], align 4
// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK11-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK11-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP31]], align 4
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK11-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK11-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49(i32 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP32]]
+// CHECK11-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP36]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
diff --git a/clang/test/OpenMP/distribute_simd_private_codegen.cpp b/clang/test/OpenMP/distribute_simd_private_codegen.cpp
index c0e80a80fe005..824a96642f046 100644
--- a/clang/test/OpenMP/distribute_simd_private_codegen.cpp
+++ b/clang/test/OpenMP/distribute_simd_private_codegen.cpp
@@ -409,7 +409,7 @@ int main() {
// CHECK9-NEXT: store ptr undef, ptr [[_TMP1]], align 8
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK9-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK9-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -426,65 +426,81 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK9-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK9-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l93.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK9-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK9-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK9-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l93.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK9-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l93() #[[ATTR4:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP11:%.*]] = load i32, ptr [[I]], align 4
-// CHECK9-NEXT: store i32 [[TMP11]], ptr [[I_CASTED]], align 4
-// CHECK9-NEXT: [[TMP12:%.*]] = load i64, ptr [[I_CASTED]], align 8
-// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP12]], ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP12]], ptr [[TMP14]], align 8
-// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP15:%.*]] = load i32, ptr [[I]], align 4
+// CHECK9-NEXT: store i32 [[TMP15]], ptr [[I_CASTED]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = load i64, ptr [[I_CASTED]], align 8
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP16]], ptr [[TMP17]], align 8
+// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP16]], ptr [[TMP18]], align 8
+// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS3:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP18]], align 4
-// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 1
-// CHECK9-NEXT: store i32 1, ptr [[TMP19]], align 4
-// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP16]], ptr [[TMP20]], align 8
-// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP17]], ptr [[TMP21]], align 8
-// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes, ptr [[TMP22]], align 8
-// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes, ptr [[TMP23]], align 8
-// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP24]], align 8
-// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP25]], align 8
-// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 8
-// CHECK9-NEXT: store i64 2, ptr [[TMP26]], align 8
-// CHECK9-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102.region_id, ptr [[KERNEL_ARGS3]])
-// CHECK9-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK9-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED4:%.*]], label [[OMP_OFFLOAD_CONT5:%.*]]
+// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP22]], align 4
+// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 1
+// CHECK9-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
+// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP21]], ptr [[TMP25]], align 8
+// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes, ptr [[TMP26]], align 8
+// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes, ptr [[TMP27]], align 8
+// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP28]], align 8
+// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP29]], align 8
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 8
+// CHECK9-NEXT: store i64 2, ptr [[TMP30]], align 8
+// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP31]], align 8
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP33]], align 4
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP34]], align 4
+// CHECK9-NEXT: [[TMP35:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102.region_id, ptr [[KERNEL_ARGS3]])
+// CHECK9-NEXT: [[TMP36:%.*]] = icmp ne i32 [[TMP35]], 0
+// CHECK9-NEXT: br i1 [[TMP36]], label [[OMP_OFFLOAD_FAILED4:%.*]], label [[OMP_OFFLOAD_CONT5:%.*]]
// CHECK9: omp_offload.failed4:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102(i64 [[TMP12]]) #[[ATTR4]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102(i64 [[TMP16]]) #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT5]]
// CHECK9: omp_offload.cont5:
// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiET_v()
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP29]], [[OMP_OFFLOAD_CONT5]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP37]], [[OMP_OFFLOAD_CONT5]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE6:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done6:
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP30:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP30]]
+// CHECK9-NEXT: [[TMP38:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP38]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -740,7 +756,7 @@ int main() {
// CHECK9-NEXT: store ptr undef, ptr [[_TMP1]], align 8
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK9-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK9-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -757,27 +773,35 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK9-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK9-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK9-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK9-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK9-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK9-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49() #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP12]]
+// CHECK9-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP16]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
@@ -1038,7 +1062,7 @@ int main() {
// CHECK11-NEXT: store ptr undef, ptr [[_TMP1]], align 4
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK11-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK11-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1055,65 +1079,81 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK11-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK11-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l93.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK11-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK11-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK11-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK11-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l93.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK11-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l93() #[[ATTR4:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP11:%.*]] = load i32, ptr [[I]], align 4
-// CHECK11-NEXT: store i32 [[TMP11]], ptr [[I_CASTED]], align 4
-// CHECK11-NEXT: [[TMP12:%.*]] = load i32, ptr [[I_CASTED]], align 4
-// CHECK11-NEXT: [[TMP13:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP12]], ptr [[TMP13]], align 4
-// CHECK11-NEXT: [[TMP14:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP12]], ptr [[TMP14]], align 4
-// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP15]], align 4
-// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP15:%.*]] = load i32, ptr [[I]], align 4
+// CHECK11-NEXT: store i32 [[TMP15]], ptr [[I_CASTED]], align 4
+// CHECK11-NEXT: [[TMP16:%.*]] = load i32, ptr [[I_CASTED]], align 4
+// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP16]], ptr [[TMP17]], align 4
+// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP16]], ptr [[TMP18]], align 4
+// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP19]], align 4
+// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS3:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP18]], align 4
-// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 1
-// CHECK11-NEXT: store i32 1, ptr [[TMP19]], align 4
-// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP16]], ptr [[TMP20]], align 4
-// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP17]], ptr [[TMP21]], align 4
-// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes, ptr [[TMP22]], align 4
-// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes, ptr [[TMP23]], align 4
-// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP24]], align 4
-// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP25]], align 4
-// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 8
-// CHECK11-NEXT: store i64 2, ptr [[TMP26]], align 8
-// CHECK11-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102.region_id, ptr [[KERNEL_ARGS3]])
-// CHECK11-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK11-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED4:%.*]], label [[OMP_OFFLOAD_CONT5:%.*]]
+// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP22]], align 4
+// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 1
+// CHECK11-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
+// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP21]], ptr [[TMP25]], align 4
+// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes, ptr [[TMP26]], align 4
+// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes, ptr [[TMP27]], align 4
+// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP28]], align 4
+// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP29]], align 4
+// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 8
+// CHECK11-NEXT: store i64 2, ptr [[TMP30]], align 8
+// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP31]], align 8
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP33]], align 4
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS3]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP34]], align 4
+// CHECK11-NEXT: [[TMP35:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102.region_id, ptr [[KERNEL_ARGS3]])
+// CHECK11-NEXT: [[TMP36:%.*]] = icmp ne i32 [[TMP35]], 0
+// CHECK11-NEXT: br i1 [[TMP36]], label [[OMP_OFFLOAD_FAILED4:%.*]], label [[OMP_OFFLOAD_CONT5:%.*]]
// CHECK11: omp_offload.failed4:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102(i32 [[TMP12]]) #[[ATTR4]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102(i32 [[TMP16]]) #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT5]]
// CHECK11: omp_offload.cont5:
// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiET_v()
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP29]], [[OMP_OFFLOAD_CONT5]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP37]], [[OMP_OFFLOAD_CONT5]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE6:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done6:
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP30:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP30]]
+// CHECK11-NEXT: [[TMP38:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP38]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -1367,7 +1407,7 @@ int main() {
// CHECK11-NEXT: store ptr undef, ptr [[_TMP1]], align 4
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK11-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK11-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1384,27 +1424,35 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK11-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK11-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK11-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK11-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK11-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK11-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK11-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49() #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP12]]
+// CHECK11-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP16]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
diff --git a/clang/test/OpenMP/distribute_simd_reduction_codegen.cpp b/clang/test/OpenMP/distribute_simd_reduction_codegen.cpp
index d0eae4fd8de72..72249ee7a719b 100644
--- a/clang/test/OpenMP/distribute_simd_reduction_codegen.cpp
+++ b/clang/test/OpenMP/distribute_simd_reduction_codegen.cpp
@@ -105,7 +105,7 @@ int main() {
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -122,9 +122,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l64.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK1-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP18]], align 4
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l64.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l64(i64 [[TMP1]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -247,7 +255,7 @@ int main() {
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -264,9 +272,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK1-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP18]], align 4
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32(i64 [[TMP1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -393,7 +409,7 @@ int main() {
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -410,9 +426,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP14]], align 4
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP15]], align 8
-// CHECK3-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l64.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK3-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP18]], align 4
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK3-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l64.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK3-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l64(i32 [[TMP1]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -535,7 +559,7 @@ int main() {
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -552,9 +576,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP14]], align 4
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP15]], align 8
-// CHECK3-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK3-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP18]], align 4
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK3-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK3-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32(i32 [[TMP1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/interop_irbuilder.cpp b/clang/test/OpenMP/interop_irbuilder.cpp
index e1a96dad65de7..337af16c5c90a 100644
--- a/clang/test/OpenMP/interop_irbuilder.cpp
+++ b/clang/test/OpenMP/interop_irbuilder.cpp
@@ -98,26 +98,26 @@ void S::member_test() {
// CHECK-NEXT: store i64 2, ptr [[DEP_COUNTER_ADDR]], align 8
// CHECK-NEXT: [[OMP_GLOBAL_THREAD_NUM4:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-NEXT: call void @__tgt_interop_use(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM4]], ptr [[INTEROP]], i32 -1, i32 2, ptr [[TMP2]], i32 1)
-// CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds [2 x %struct.kmp_depend_info], ptr [[DOTDEP_ARR_ADDR5]], i64 0, i64 0
-// CHECK-NEXT: [[TMP15:%.*]] = ptrtoint ptr [[D0]] to i64
-// CHECK-NEXT: [[TMP16:%.*]] = getelementptr [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP14]], i64 0
-// CHECK-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP16]], i32 0, i32 0
-// CHECK-NEXT: store i64 [[TMP15]], ptr [[TMP17]], align 8
-// CHECK-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP16]], i32 0, i32 1
-// CHECK-NEXT: store i64 4, ptr [[TMP18]], align 8
-// CHECK-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP16]], i32 0, i32 2
-// CHECK-NEXT: store i8 1, ptr [[TMP19]], align 8
-// CHECK-NEXT: [[TMP20:%.*]] = ptrtoint ptr [[D1]] to i64
-// CHECK-NEXT: [[TMP21:%.*]] = getelementptr [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP14]], i64 1
-// CHECK-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP21]], i32 0, i32 0
-// CHECK-NEXT: store i64 [[TMP20]], ptr [[TMP22]], align 8
-// CHECK-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP21]], i32 0, i32 1
-// CHECK-NEXT: store i64 4, ptr [[TMP23]], align 8
-// CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP21]], i32 0, i32 2
-// CHECK-NEXT: store i8 1, ptr [[TMP24]], align 8
+// CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds [2 x %struct.kmp_depend_info], ptr [[DOTDEP_ARR_ADDR5]], i64 0, i64 0
+// CHECK-NEXT: [[TMP14:%.*]] = ptrtoint ptr [[D0]] to i64
+// CHECK-NEXT: [[TMP15:%.*]] = getelementptr [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP13]], i64 0
+// CHECK-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP15]], i32 0, i32 0
+// CHECK-NEXT: store i64 [[TMP14]], ptr [[TMP16]], align 8
+// CHECK-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP15]], i32 0, i32 1
+// CHECK-NEXT: store i64 4, ptr [[TMP17]], align 8
+// CHECK-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP15]], i32 0, i32 2
+// CHECK-NEXT: store i8 1, ptr [[TMP18]], align 8
+// CHECK-NEXT: [[TMP19:%.*]] = ptrtoint ptr [[D1]] to i64
+// CHECK-NEXT: [[TMP20:%.*]] = getelementptr [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP13]], i64 1
+// CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP20]], i32 0, i32 0
+// CHECK-NEXT: store i64 [[TMP19]], ptr [[TMP21]], align 8
+// CHECK-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP20]], i32 0, i32 1
+// CHECK-NEXT: store i64 4, ptr [[TMP22]], align 8
+// CHECK-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP20]], i32 0, i32 2
+// CHECK-NEXT: store i8 1, ptr [[TMP23]], align 8
// CHECK-NEXT: store i64 2, ptr [[DEP_COUNTER_ADDR6]], align 8
// CHECK-NEXT: [[OMP_GLOBAL_THREAD_NUM7:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
-// CHECK-NEXT: call void @__tgt_interop_destroy(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM7]], ptr [[INTEROP]], i32 -1, i32 2, ptr [[TMP14]], i32 0)
+// CHECK-NEXT: call void @__tgt_interop_destroy(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM7]], ptr [[INTEROP]], i32 -1, i32 2, ptr [[TMP13]], i32 0)
// CHECK-NEXT: ret void
//
//
@@ -169,26 +169,26 @@ void S::member_test() {
// CHECK-NEXT: [[INTEROP8:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[THIS1]], i32 0, i32 0
// CHECK-NEXT: [[OMP_GLOBAL_THREAD_NUM9:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-NEXT: call void @__tgt_interop_use(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM9]], ptr [[INTEROP8]], i32 -1, i32 2, ptr [[TMP2]], i32 1)
-// CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds [2 x %struct.kmp_depend_info], ptr [[DOTDEP_ARR_ADDR10]], i64 0, i64 0
-// CHECK-NEXT: [[TMP15:%.*]] = ptrtoint ptr [[D0]] to i64
-// CHECK-NEXT: [[TMP16:%.*]] = getelementptr [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP14]], i64 0
-// CHECK-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP16]], i32 0, i32 0
-// CHECK-NEXT: store i64 [[TMP15]], ptr [[TMP17]], align 8
-// CHECK-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP16]], i32 0, i32 1
-// CHECK-NEXT: store i64 4, ptr [[TMP18]], align 8
-// CHECK-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP16]], i32 0, i32 2
-// CHECK-NEXT: store i8 1, ptr [[TMP19]], align 8
-// CHECK-NEXT: [[TMP20:%.*]] = ptrtoint ptr [[D1]] to i64
-// CHECK-NEXT: [[TMP21:%.*]] = getelementptr [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP14]], i64 1
-// CHECK-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP21]], i32 0, i32 0
-// CHECK-NEXT: store i64 [[TMP20]], ptr [[TMP22]], align 8
-// CHECK-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP21]], i32 0, i32 1
-// CHECK-NEXT: store i64 4, ptr [[TMP23]], align 8
-// CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP21]], i32 0, i32 2
-// CHECK-NEXT: store i8 1, ptr [[TMP24]], align 8
+// CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds [2 x %struct.kmp_depend_info], ptr [[DOTDEP_ARR_ADDR10]], i64 0, i64 0
+// CHECK-NEXT: [[TMP14:%.*]] = ptrtoint ptr [[D0]] to i64
+// CHECK-NEXT: [[TMP15:%.*]] = getelementptr [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP13]], i64 0
+// CHECK-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP15]], i32 0, i32 0
+// CHECK-NEXT: store i64 [[TMP14]], ptr [[TMP16]], align 8
+// CHECK-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP15]], i32 0, i32 1
+// CHECK-NEXT: store i64 4, ptr [[TMP17]], align 8
+// CHECK-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP15]], i32 0, i32 2
+// CHECK-NEXT: store i8 1, ptr [[TMP18]], align 8
+// CHECK-NEXT: [[TMP19:%.*]] = ptrtoint ptr [[D1]] to i64
+// CHECK-NEXT: [[TMP20:%.*]] = getelementptr [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP13]], i64 1
+// CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP20]], i32 0, i32 0
+// CHECK-NEXT: store i64 [[TMP19]], ptr [[TMP21]], align 8
+// CHECK-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP20]], i32 0, i32 1
+// CHECK-NEXT: store i64 4, ptr [[TMP22]], align 8
+// CHECK-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT_KMP_DEPEND_INFO]], ptr [[TMP20]], i32 0, i32 2
+// CHECK-NEXT: store i8 1, ptr [[TMP23]], align 8
// CHECK-NEXT: store i64 2, ptr [[DEP_COUNTER_ADDR11]], align 8
// CHECK-NEXT: [[INTEROP12:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[THIS1]], i32 0, i32 0
// CHECK-NEXT: [[OMP_GLOBAL_THREAD_NUM13:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
-// CHECK-NEXT: call void @__tgt_interop_destroy(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM13]], ptr [[INTEROP12]], i32 -1, i32 2, ptr [[TMP14]], i32 0)
+// CHECK-NEXT: call void @__tgt_interop_destroy(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM13]], ptr [[INTEROP12]], i32 -1, i32 2, ptr [[TMP13]], i32 0)
// CHECK-NEXT: ret void
//
diff --git a/clang/test/OpenMP/nvptx_lambda_capturing.cpp b/clang/test/OpenMP/nvptx_lambda_capturing.cpp
index 32e5c3e9f6701..4a6de49e78cf0 100644
--- a/clang/test/OpenMP/nvptx_lambda_capturing.cpp
+++ b/clang/test/OpenMP/nvptx_lambda_capturing.cpp
@@ -190,7 +190,7 @@ int main(int argc, char **argv) {
// CHECK1-NEXT: [[TMP65:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP66]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP66]], align 4
// CHECK1-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 11, ptr [[TMP67]], align 4
// CHECK1-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -207,135 +207,151 @@ int main(int argc, char **argv) {
// CHECK1-NEXT: store ptr null, ptr [[TMP73]], align 8
// CHECK1-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP74]], align 8
-// CHECK1-NEXT: [[TMP75:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l41.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP76:%.*]] = icmp ne i32 [[TMP75]], 0
-// CHECK1-NEXT: br i1 [[TMP76]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP75]], align 8
+// CHECK1-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP76]], align 4
+// CHECK1-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP77]], align 4
+// CHECK1-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP78]], align 4
+// CHECK1-NEXT: [[TMP79:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l41.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP80:%.*]] = icmp ne i32 [[TMP79]], 0
+// CHECK1-NEXT: br i1 [[TMP80]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l41(i64 [[TMP11]], ptr [[TMP12]], ptr [[TMP13]], ptr [[TMP14]], ptr [[A]], ptr [[TMP15]]) #[[ATTR4:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
-// CHECK1-NEXT: [[TMP77:%.*]] = load ptr, ptr [[B]], align 8
-// CHECK1-NEXT: store ptr [[TMP77]], ptr [[_TMP4]], align 8
-// CHECK1-NEXT: [[TMP78:%.*]] = load ptr, ptr [[C]], align 8
-// CHECK1-NEXT: store ptr [[TMP78]], ptr [[_TMP5]], align 8
-// CHECK1-NEXT: [[TMP79:%.*]] = load ptr, ptr [[L]], align 8
-// CHECK1-NEXT: store ptr [[TMP79]], ptr [[_TMP6]], align 8
-// CHECK1-NEXT: [[TMP80:%.*]] = load ptr, ptr [[_TMP4]], align 8
-// CHECK1-NEXT: [[TMP81:%.*]] = load ptr, ptr [[_TMP5]], align 8
-// CHECK1-NEXT: [[TMP82:%.*]] = load ptr, ptr [[D]], align 8
-// CHECK1-NEXT: [[TMP83:%.*]] = load ptr, ptr [[_TMP6]], align 8
-// CHECK1-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[TMP83]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[TMP83]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP86:%.*]] = load ptr, ptr [[TMP85]], align 8
-// CHECK1-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[TMP83]], i32 0, i32 1
-// CHECK1-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[TMP83]], i32 0, i32 1
-// CHECK1-NEXT: [[TMP89:%.*]] = load ptr, ptr [[TMP88]], align 8
-// CHECK1-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[TMP83]], i32 0, i32 2
-// CHECK1-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[TMP83]], i32 0, i32 2
-// CHECK1-NEXT: [[TMP92:%.*]] = load ptr, ptr [[TMP91]], align 8
-// CHECK1-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[TMP83]], i32 0, i32 3
-// CHECK1-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[TMP83]], i32 0, i32 3
-// CHECK1-NEXT: [[TMP95:%.*]] = load ptr, ptr [[TMP94]], align 8
-// CHECK1-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[TMP83]], i32 0, i32 4
-// CHECK1-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[TMP83]], i32 0, i32 4
-// CHECK1-NEXT: [[TMP98:%.*]] = load ptr, ptr [[TMP97]], align 8
-// CHECK1-NEXT: [[TMP99:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[ARGC_ADDR]], ptr [[TMP99]], align 8
-// CHECK1-NEXT: [[TMP100:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[ARGC_ADDR]], ptr [[TMP100]], align 8
-// CHECK1-NEXT: [[TMP101:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP101]], align 8
-// CHECK1-NEXT: [[TMP102:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 1
-// CHECK1-NEXT: store ptr [[TMP80]], ptr [[TMP102]], align 8
-// CHECK1-NEXT: [[TMP103:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 1
-// CHECK1-NEXT: store ptr [[TMP80]], ptr [[TMP103]], align 8
-// CHECK1-NEXT: [[TMP104:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP104]], align 8
-// CHECK1-NEXT: [[TMP105:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP81]], ptr [[TMP105]], align 8
-// CHECK1-NEXT: [[TMP106:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP81]], ptr [[TMP106]], align 8
-// CHECK1-NEXT: [[TMP107:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i64 0, i64 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP107]], align 8
-// CHECK1-NEXT: [[TMP108:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP82]], ptr [[TMP108]], align 8
-// CHECK1-NEXT: [[TMP109:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP82]], ptr [[TMP109]], align 8
-// CHECK1-NEXT: [[TMP110:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i64 0, i64 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP110]], align 8
-// CHECK1-NEXT: [[TMP111:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 4
-// CHECK1-NEXT: store ptr [[A]], ptr [[TMP111]], align 8
-// CHECK1-NEXT: [[TMP112:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 4
-// CHECK1-NEXT: store ptr [[A]], ptr [[TMP112]], align 8
-// CHECK1-NEXT: [[TMP113:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i64 0, i64 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP113]], align 8
-// CHECK1-NEXT: [[TMP114:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 5
-// CHECK1-NEXT: store ptr [[TMP83]], ptr [[TMP114]], align 8
-// CHECK1-NEXT: [[TMP115:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 5
-// CHECK1-NEXT: store ptr [[TMP83]], ptr [[TMP115]], align 8
-// CHECK1-NEXT: [[TMP116:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i64 0, i64 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP116]], align 8
-// CHECK1-NEXT: [[TMP117:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 6
-// CHECK1-NEXT: store ptr [[TMP84]], ptr [[TMP117]], align 8
-// CHECK1-NEXT: [[TMP118:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 6
-// CHECK1-NEXT: store ptr [[TMP86]], ptr [[TMP118]], align 8
-// CHECK1-NEXT: [[TMP119:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i64 0, i64 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP119]], align 8
-// CHECK1-NEXT: [[TMP120:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 7
-// CHECK1-NEXT: store ptr [[TMP87]], ptr [[TMP120]], align 8
-// CHECK1-NEXT: [[TMP121:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 7
-// CHECK1-NEXT: store ptr [[TMP89]], ptr [[TMP121]], align 8
-// CHECK1-NEXT: [[TMP122:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i64 0, i64 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP122]], align 8
-// CHECK1-NEXT: [[TMP123:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 8
-// CHECK1-NEXT: store ptr [[TMP90]], ptr [[TMP123]], align 8
-// CHECK1-NEXT: [[TMP124:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 8
-// CHECK1-NEXT: store ptr [[TMP92]], ptr [[TMP124]], align 8
-// CHECK1-NEXT: [[TMP125:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i64 0, i64 8
-// CHECK1-NEXT: store ptr null, ptr [[TMP125]], align 8
-// CHECK1-NEXT: [[TMP126:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 9
-// CHECK1-NEXT: store ptr [[TMP93]], ptr [[TMP126]], align 8
-// CHECK1-NEXT: [[TMP127:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 9
-// CHECK1-NEXT: store ptr [[TMP95]], ptr [[TMP127]], align 8
-// CHECK1-NEXT: [[TMP128:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i64 0, i64 9
-// CHECK1-NEXT: store ptr null, ptr [[TMP128]], align 8
-// CHECK1-NEXT: [[TMP129:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 10
-// CHECK1-NEXT: store ptr [[TMP96]], ptr [[TMP129]], align 8
-// CHECK1-NEXT: [[TMP130:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 10
-// CHECK1-NEXT: store ptr [[TMP98]], ptr [[TMP130]], align 8
-// CHECK1-NEXT: [[TMP131:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i64 0, i64 10
-// CHECK1-NEXT: store ptr null, ptr [[TMP131]], align 8
-// CHECK1-NEXT: [[TMP132:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP133:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP81:%.*]] = load ptr, ptr [[B]], align 8
+// CHECK1-NEXT: store ptr [[TMP81]], ptr [[_TMP4]], align 8
+// CHECK1-NEXT: [[TMP82:%.*]] = load ptr, ptr [[C]], align 8
+// CHECK1-NEXT: store ptr [[TMP82]], ptr [[_TMP5]], align 8
+// CHECK1-NEXT: [[TMP83:%.*]] = load ptr, ptr [[L]], align 8
+// CHECK1-NEXT: store ptr [[TMP83]], ptr [[_TMP6]], align 8
+// CHECK1-NEXT: [[TMP84:%.*]] = load ptr, ptr [[_TMP4]], align 8
+// CHECK1-NEXT: [[TMP85:%.*]] = load ptr, ptr [[_TMP5]], align 8
+// CHECK1-NEXT: [[TMP86:%.*]] = load ptr, ptr [[D]], align 8
+// CHECK1-NEXT: [[TMP87:%.*]] = load ptr, ptr [[_TMP6]], align 8
+// CHECK1-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[TMP87]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[TMP87]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP90:%.*]] = load ptr, ptr [[TMP89]], align 8
+// CHECK1-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[TMP87]], i32 0, i32 1
+// CHECK1-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[TMP87]], i32 0, i32 1
+// CHECK1-NEXT: [[TMP93:%.*]] = load ptr, ptr [[TMP92]], align 8
+// CHECK1-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[TMP87]], i32 0, i32 2
+// CHECK1-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[TMP87]], i32 0, i32 2
+// CHECK1-NEXT: [[TMP96:%.*]] = load ptr, ptr [[TMP95]], align 8
+// CHECK1-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[TMP87]], i32 0, i32 3
+// CHECK1-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[TMP87]], i32 0, i32 3
+// CHECK1-NEXT: [[TMP99:%.*]] = load ptr, ptr [[TMP98]], align 8
+// CHECK1-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[TMP87]], i32 0, i32 4
+// CHECK1-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[TMP87]], i32 0, i32 4
+// CHECK1-NEXT: [[TMP102:%.*]] = load ptr, ptr [[TMP101]], align 8
+// CHECK1-NEXT: [[TMP103:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[ARGC_ADDR]], ptr [[TMP103]], align 8
+// CHECK1-NEXT: [[TMP104:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[ARGC_ADDR]], ptr [[TMP104]], align 8
+// CHECK1-NEXT: [[TMP105:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP105]], align 8
+// CHECK1-NEXT: [[TMP106:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 1
+// CHECK1-NEXT: store ptr [[TMP84]], ptr [[TMP106]], align 8
+// CHECK1-NEXT: [[TMP107:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 1
+// CHECK1-NEXT: store ptr [[TMP84]], ptr [[TMP107]], align 8
+// CHECK1-NEXT: [[TMP108:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP108]], align 8
+// CHECK1-NEXT: [[TMP109:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP85]], ptr [[TMP109]], align 8
+// CHECK1-NEXT: [[TMP110:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP85]], ptr [[TMP110]], align 8
+// CHECK1-NEXT: [[TMP111:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i64 0, i64 2
+// CHECK1-NEXT: store ptr null, ptr [[TMP111]], align 8
+// CHECK1-NEXT: [[TMP112:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP86]], ptr [[TMP112]], align 8
+// CHECK1-NEXT: [[TMP113:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP86]], ptr [[TMP113]], align 8
+// CHECK1-NEXT: [[TMP114:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i64 0, i64 3
+// CHECK1-NEXT: store ptr null, ptr [[TMP114]], align 8
+// CHECK1-NEXT: [[TMP115:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 4
+// CHECK1-NEXT: store ptr [[A]], ptr [[TMP115]], align 8
+// CHECK1-NEXT: [[TMP116:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 4
+// CHECK1-NEXT: store ptr [[A]], ptr [[TMP116]], align 8
+// CHECK1-NEXT: [[TMP117:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i64 0, i64 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP117]], align 8
+// CHECK1-NEXT: [[TMP118:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 5
+// CHECK1-NEXT: store ptr [[TMP87]], ptr [[TMP118]], align 8
+// CHECK1-NEXT: [[TMP119:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 5
+// CHECK1-NEXT: store ptr [[TMP87]], ptr [[TMP119]], align 8
+// CHECK1-NEXT: [[TMP120:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i64 0, i64 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP120]], align 8
+// CHECK1-NEXT: [[TMP121:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 6
+// CHECK1-NEXT: store ptr [[TMP88]], ptr [[TMP121]], align 8
+// CHECK1-NEXT: [[TMP122:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 6
+// CHECK1-NEXT: store ptr [[TMP90]], ptr [[TMP122]], align 8
+// CHECK1-NEXT: [[TMP123:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i64 0, i64 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP123]], align 8
+// CHECK1-NEXT: [[TMP124:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 7
+// CHECK1-NEXT: store ptr [[TMP91]], ptr [[TMP124]], align 8
+// CHECK1-NEXT: [[TMP125:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 7
+// CHECK1-NEXT: store ptr [[TMP93]], ptr [[TMP125]], align 8
+// CHECK1-NEXT: [[TMP126:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i64 0, i64 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP126]], align 8
+// CHECK1-NEXT: [[TMP127:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 8
+// CHECK1-NEXT: store ptr [[TMP94]], ptr [[TMP127]], align 8
+// CHECK1-NEXT: [[TMP128:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 8
+// CHECK1-NEXT: store ptr [[TMP96]], ptr [[TMP128]], align 8
+// CHECK1-NEXT: [[TMP129:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i64 0, i64 8
+// CHECK1-NEXT: store ptr null, ptr [[TMP129]], align 8
+// CHECK1-NEXT: [[TMP130:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 9
+// CHECK1-NEXT: store ptr [[TMP97]], ptr [[TMP130]], align 8
+// CHECK1-NEXT: [[TMP131:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 9
+// CHECK1-NEXT: store ptr [[TMP99]], ptr [[TMP131]], align 8
+// CHECK1-NEXT: [[TMP132:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i64 0, i64 9
+// CHECK1-NEXT: store ptr null, ptr [[TMP132]], align 8
+// CHECK1-NEXT: [[TMP133:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 10
+// CHECK1-NEXT: store ptr [[TMP100]], ptr [[TMP133]], align 8
+// CHECK1-NEXT: [[TMP134:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 10
+// CHECK1-NEXT: store ptr [[TMP102]], ptr [[TMP134]], align 8
+// CHECK1-NEXT: [[TMP135:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i64 0, i64 10
+// CHECK1-NEXT: store ptr null, ptr [[TMP135]], align 8
+// CHECK1-NEXT: [[TMP136:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP137:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS10:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP134]], align 4
-// CHECK1-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 1
-// CHECK1-NEXT: store i32 11, ptr [[TMP135]], align 4
-// CHECK1-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP132]], ptr [[TMP136]], align 8
-// CHECK1-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP133]], ptr [[TMP137]], align 8
-// CHECK1-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.1, ptr [[TMP138]], align 8
-// CHECK1-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.2, ptr [[TMP139]], align 8
-// CHECK1-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP140]], align 8
-// CHECK1-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP141]], align 8
-// CHECK1-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP142]], align 8
-// CHECK1-NEXT: [[TMP143:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l43.region_id, ptr [[KERNEL_ARGS10]])
-// CHECK1-NEXT: [[TMP144:%.*]] = icmp ne i32 [[TMP143]], 0
-// CHECK1-NEXT: br i1 [[TMP144]], label [[OMP_OFFLOAD_FAILED11:%.*]], label [[OMP_OFFLOAD_CONT12:%.*]]
+// CHECK1-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP138]], align 4
+// CHECK1-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 1
+// CHECK1-NEXT: store i32 11, ptr [[TMP139]], align 4
+// CHECK1-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP136]], ptr [[TMP140]], align 8
+// CHECK1-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP137]], ptr [[TMP141]], align 8
+// CHECK1-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.1, ptr [[TMP142]], align 8
+// CHECK1-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.2, ptr [[TMP143]], align 8
+// CHECK1-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP144]], align 8
+// CHECK1-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP145]], align 8
+// CHECK1-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP146]], align 8
+// CHECK1-NEXT: [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP147]], align 8
+// CHECK1-NEXT: [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP148]], align 4
+// CHECK1-NEXT: [[TMP149:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP149]], align 4
+// CHECK1-NEXT: [[TMP150:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP150]], align 4
+// CHECK1-NEXT: [[TMP151:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l43.region_id, ptr [[KERNEL_ARGS10]])
+// CHECK1-NEXT: [[TMP152:%.*]] = icmp ne i32 [[TMP151]], 0
+// CHECK1-NEXT: br i1 [[TMP152]], label [[OMP_OFFLOAD_FAILED11:%.*]], label [[OMP_OFFLOAD_CONT12:%.*]]
// CHECK1: omp_offload.failed11:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l43(ptr [[ARGC_ADDR]], ptr [[TMP80]], ptr [[TMP81]], ptr [[TMP82]], ptr [[A]], ptr [[TMP83]]) #[[ATTR4]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l43(ptr [[ARGC_ADDR]], ptr [[TMP84]], ptr [[TMP85]], ptr [[TMP86]], ptr [[A]], ptr [[TMP87]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT12]]
// CHECK1: omp_offload.cont12:
-// CHECK1-NEXT: [[TMP145:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK1-NEXT: [[TMP153:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
// CHECK1-NEXT: [[CALL:%.*]] = call noundef signext i32 @_ZN1S3fooEv(ptr noundef nonnull align 4 dereferenceable(4) @s)
-// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP145]], [[CALL]]
+// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP153]], [[CALL]]
// CHECK1-NEXT: ret i32 [[ADD]]
//
//
@@ -524,7 +540,7 @@ int main(int argc, char **argv) {
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP16]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP16]], align 4
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 3, ptr [[TMP17]], align 4
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -541,69 +557,85 @@ int main(int argc, char **argv) {
// CHECK1-NEXT: store ptr null, ptr [[TMP23]], align 8
// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN1S3fooEv_l27.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK1-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP25]], align 8
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP28]], align 4
+// CHECK1-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN1S3fooEv_l27.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK1-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN1S3fooEv_l27(ptr [[THIS1]], ptr [[TMP2]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
-// CHECK1-NEXT: [[TMP27:%.*]] = load ptr, ptr [[L]], align 8
-// CHECK1-NEXT: store ptr [[TMP27]], ptr [[_TMP2]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = load ptr, ptr [[_TMP2]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], ptr [[TMP28]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], ptr [[TMP28]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP31]], align 8
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 1
-// CHECK1-NEXT: store ptr [[TMP28]], ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 1
-// CHECK1-NEXT: store ptr [[TMP28]], ptr [[TMP35]], align 8
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP36]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP29]], ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP30]], ptr [[TMP38]], align 8
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP31:%.*]] = load ptr, ptr [[L]], align 8
+// CHECK1-NEXT: store ptr [[TMP31]], ptr [[_TMP2]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = load ptr, ptr [[_TMP2]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], ptr [[TMP32]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], ptr [[TMP32]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP35]], align 8
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP36]], align 8
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP37]], align 8
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 1
+// CHECK1-NEXT: store ptr [[TMP32]], ptr [[TMP38]], align 8
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 1
+// CHECK1-NEXT: store ptr [[TMP32]], ptr [[TMP39]], align 8
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP33]], ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP34]], ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 2
+// CHECK1-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP42]], align 4
-// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK1-NEXT: store i32 3, ptr [[TMP43]], align 4
-// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP40]], ptr [[TMP44]], align 8
-// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP41]], ptr [[TMP45]], align 8
-// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.6, ptr [[TMP46]], align 8
-// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.7, ptr [[TMP47]], align 8
-// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP48]], align 8
-// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP49]], align 8
-// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP50]], align 8
-// CHECK1-NEXT: [[TMP51:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN1S3fooEv_l29.region_id, ptr [[KERNEL_ARGS6]])
-// CHECK1-NEXT: [[TMP52:%.*]] = icmp ne i32 [[TMP51]], 0
-// CHECK1-NEXT: br i1 [[TMP52]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP46]], align 4
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK1-NEXT: store i32 3, ptr [[TMP47]], align 4
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 8
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP45]], ptr [[TMP49]], align 8
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.6, ptr [[TMP50]], align 8
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.7, ptr [[TMP51]], align 8
+// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP52]], align 8
+// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP53]], align 8
+// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP55]], align 8
+// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP56]], align 4
+// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP57]], align 4
+// CHECK1-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP58]], align 4
+// CHECK1-NEXT: [[TMP59:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN1S3fooEv_l29.region_id, ptr [[KERNEL_ARGS6]])
+// CHECK1-NEXT: [[TMP60:%.*]] = icmp ne i32 [[TMP59]], 0
+// CHECK1-NEXT: br i1 [[TMP60]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK1: omp_offload.failed7:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN1S3fooEv_l29(ptr [[THIS1]], ptr [[TMP28]]) #[[ATTR4]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN1S3fooEv_l29(ptr [[THIS1]], ptr [[TMP32]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK1: omp_offload.cont8:
// CHECK1-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP53:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: [[TMP54:%.*]] = load ptr, ptr [[L]], align 8
-// CHECK1-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z3fooIZN1S3fooEvEUlvE_EiRKT_(ptr noundef nonnull align 8 dereferenceable(8) [[TMP54]])
-// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP53]], [[CALL]]
+// CHECK1-NEXT: [[TMP61:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: [[TMP62:%.*]] = load ptr, ptr [[L]], align 8
+// CHECK1-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z3fooIZN1S3fooEvEUlvE_EiRKT_(ptr noundef nonnull align 8 dereferenceable(8) [[TMP62]])
+// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP61]], [[CALL]]
// CHECK1-NEXT: ret i32 [[ADD]]
//
//
@@ -712,7 +744,7 @@ int main(int argc, char **argv) {
// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP12]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP12]], align 4
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 2, ptr [[TMP13]], align 4
// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -729,9 +761,17 @@ int main(int argc, char **argv) {
// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP20]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooIZN1S3fooEvEUlvE_EiRKT__l18.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP22:%.*]] = icmp ne i32 [[TMP21]], 0
-// CHECK1-NEXT: br i1 [[TMP22]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP22]], align 4
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP23]], align 4
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP24]], align 4
+// CHECK1-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooIZN1S3fooEvEUlvE_EiRKT__l18.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
+// CHECK1-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooIZN1S3fooEvEUlvE_EiRKT__l18(ptr [[TMP1]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/reduction_implicit_map.cpp b/clang/test/OpenMP/reduction_implicit_map.cpp
index 73c384a47bea6..610b9e737ca93 100644
--- a/clang/test/OpenMP/reduction_implicit_map.cpp
+++ b/clang/test/OpenMP/reduction_implicit_map.cpp
@@ -318,7 +318,7 @@ int main()
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -335,43 +335,59 @@ int main()
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3barv_l50.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3barv_l50.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3barv_l50(ptr [[O]]) #[[ATTR8:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[B]], ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[B]], ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[B]], ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[B]], ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS4:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.3, ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.4, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3barv_l55.region_id, ptr [[KERNEL_ARGS4]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED5:%.*]], label [[OMP_OFFLOAD_CONT6:%.*]]
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.3, ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.4, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP35]], align 4
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3barv_l55.region_id, ptr [[KERNEL_ARGS4]])
+// CHECK1-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK1-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED5:%.*]], label [[OMP_OFFLOAD_CONT6:%.*]]
// CHECK1: omp_offload.failed5:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3barv_l55(ptr [[B]]) #[[ATTR8]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT6]]
@@ -879,7 +895,7 @@ int main()
// CHECK2-NEXT: [[TMP27:%.*]] = zext i32 [[ADD]] to i64
// CHECK2-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK2-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK2-NEXT: store i32 1, ptr [[TMP28]], align 4
+// CHECK2-NEXT: store i32 2, ptr [[TMP28]], align 4
// CHECK2-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK2-NEXT: store i32 3, ptr [[TMP29]], align 4
// CHECK2-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -896,173 +912,205 @@ int main()
// CHECK2-NEXT: store ptr null, ptr [[TMP35]], align 4
// CHECK2-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK2-NEXT: store i64 [[TMP27]], ptr [[TMP36]], align 8
-// CHECK2-NEXT: [[TMP37:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3sumPiiS__l69.region_id, ptr [[KERNEL_ARGS]])
-// CHECK2-NEXT: [[TMP38:%.*]] = icmp ne i32 [[TMP37]], 0
-// CHECK2-NEXT: br i1 [[TMP38]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK2-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK2-NEXT: store i64 0, ptr [[TMP37]], align 8
+// CHECK2-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK2-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP38]], align 4
+// CHECK2-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK2-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP39]], align 4
+// CHECK2-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK2-NEXT: store i32 0, ptr [[TMP40]], align 4
+// CHECK2-NEXT: [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3sumPiiS__l69.region_id, ptr [[KERNEL_ARGS]])
+// CHECK2-NEXT: [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
+// CHECK2-NEXT: br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK2: omp_offload.failed:
// CHECK2-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3sumPiiS__l69(i32 [[TMP1]], ptr [[TMP2]], ptr [[TMP3]]) #[[ATTR2:[0-9]+]]
// CHECK2-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK2: omp_offload.cont:
-// CHECK2-NEXT: [[TMP39:%.*]] = load i32, ptr [[SIZE_ADDR]], align 4
-// CHECK2-NEXT: store i32 [[TMP39]], ptr [[SIZE_CASTED4]], align 4
-// CHECK2-NEXT: [[TMP40:%.*]] = load i32, ptr [[SIZE_CASTED4]], align 4
-// CHECK2-NEXT: [[TMP41:%.*]] = load ptr, ptr [[OUTPUT_ADDR]], align 4
-// CHECK2-NEXT: [[TMP42:%.*]] = load ptr, ptr [[INPUT_ADDR]], align 4
-// CHECK2-NEXT: [[TMP43:%.*]] = load ptr, ptr [[OUTPUT_ADDR]], align 4
-// CHECK2-NEXT: [[TMP44:%.*]] = load ptr, ptr [[OUTPUT_ADDR]], align 4
-// CHECK2-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds i32, ptr [[TMP44]], i32 0
-// CHECK2-NEXT: [[TMP45:%.*]] = load ptr, ptr [[INPUT_ADDR]], align 4
+// CHECK2-NEXT: [[TMP43:%.*]] = load i32, ptr [[SIZE_ADDR]], align 4
+// CHECK2-NEXT: store i32 [[TMP43]], ptr [[SIZE_CASTED4]], align 4
+// CHECK2-NEXT: [[TMP44:%.*]] = load i32, ptr [[SIZE_CASTED4]], align 4
+// CHECK2-NEXT: [[TMP45:%.*]] = load ptr, ptr [[OUTPUT_ADDR]], align 4
// CHECK2-NEXT: [[TMP46:%.*]] = load ptr, ptr [[INPUT_ADDR]], align 4
-// CHECK2-NEXT: [[ARRAYIDX6:%.*]] = getelementptr inbounds i32, ptr [[TMP46]], i32 0
-// CHECK2-NEXT: [[TMP47:%.*]] = load i32, ptr [[SIZE_ADDR]], align 4
-// CHECK2-NEXT: [[TMP48:%.*]] = mul nuw i32 [[TMP47]], 4
-// CHECK2-NEXT: [[TMP49:%.*]] = sext i32 [[TMP48]] to i64
+// CHECK2-NEXT: [[TMP47:%.*]] = load ptr, ptr [[OUTPUT_ADDR]], align 4
+// CHECK2-NEXT: [[TMP48:%.*]] = load ptr, ptr [[OUTPUT_ADDR]], align 4
+// CHECK2-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds i32, ptr [[TMP48]], i32 0
+// CHECK2-NEXT: [[TMP49:%.*]] = load ptr, ptr [[INPUT_ADDR]], align 4
+// CHECK2-NEXT: [[TMP50:%.*]] = load ptr, ptr [[INPUT_ADDR]], align 4
+// CHECK2-NEXT: [[ARRAYIDX6:%.*]] = getelementptr inbounds i32, ptr [[TMP50]], i32 0
+// CHECK2-NEXT: [[TMP51:%.*]] = load i32, ptr [[SIZE_ADDR]], align 4
+// CHECK2-NEXT: [[TMP52:%.*]] = mul nuw i32 [[TMP51]], 4
+// CHECK2-NEXT: [[TMP53:%.*]] = sext i32 [[TMP52]] to i64
// CHECK2-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES10]], ptr align 4 @.offload_sizes.7, i32 24, i1 false)
-// CHECK2-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 0
-// CHECK2-NEXT: store i32 [[TMP40]], ptr [[TMP50]], align 4
-// CHECK2-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 0
-// CHECK2-NEXT: store i32 [[TMP40]], ptr [[TMP51]], align 4
-// CHECK2-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i32 0, i32 0
-// CHECK2-NEXT: store ptr null, ptr [[TMP52]], align 4
-// CHECK2-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 1
-// CHECK2-NEXT: store ptr [[TMP43]], ptr [[TMP53]], align 4
-// CHECK2-NEXT: [[TMP54:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 1
-// CHECK2-NEXT: store ptr [[ARRAYIDX5]], ptr [[TMP54]], align 4
-// CHECK2-NEXT: [[TMP55:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i32 0, i32 1
-// CHECK2-NEXT: store ptr null, ptr [[TMP55]], align 4
-// CHECK2-NEXT: [[TMP56:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 2
-// CHECK2-NEXT: store ptr [[TMP45]], ptr [[TMP56]], align 4
-// CHECK2-NEXT: [[TMP57:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 2
-// CHECK2-NEXT: store ptr [[ARRAYIDX6]], ptr [[TMP57]], align 4
-// CHECK2-NEXT: [[TMP58:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES10]], i32 0, i32 2
-// CHECK2-NEXT: store i64 [[TMP49]], ptr [[TMP58]], align 4
-// CHECK2-NEXT: [[TMP59:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i32 0, i32 2
+// CHECK2-NEXT: [[TMP54:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 0
+// CHECK2-NEXT: store i32 [[TMP44]], ptr [[TMP54]], align 4
+// CHECK2-NEXT: [[TMP55:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 0
+// CHECK2-NEXT: store i32 [[TMP44]], ptr [[TMP55]], align 4
+// CHECK2-NEXT: [[TMP56:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i32 0, i32 0
+// CHECK2-NEXT: store ptr null, ptr [[TMP56]], align 4
+// CHECK2-NEXT: [[TMP57:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 1
+// CHECK2-NEXT: store ptr [[TMP47]], ptr [[TMP57]], align 4
+// CHECK2-NEXT: [[TMP58:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 1
+// CHECK2-NEXT: store ptr [[ARRAYIDX5]], ptr [[TMP58]], align 4
+// CHECK2-NEXT: [[TMP59:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i32 0, i32 1
// CHECK2-NEXT: store ptr null, ptr [[TMP59]], align 4
-// CHECK2-NEXT: [[TMP60:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 0
-// CHECK2-NEXT: [[TMP61:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 0
-// CHECK2-NEXT: [[TMP62:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES10]], i32 0, i32 0
-// CHECK2-NEXT: [[TMP63:%.*]] = load i32, ptr [[SIZE_ADDR]], align 4
-// CHECK2-NEXT: store i32 [[TMP63]], ptr [[DOTCAPTURE_EXPR_12]], align 4
-// CHECK2-NEXT: [[TMP64:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_12]], align 4
-// CHECK2-NEXT: [[SUB14:%.*]] = sub nsw i32 [[TMP64]], 0
+// CHECK2-NEXT: [[TMP60:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 2
+// CHECK2-NEXT: store ptr [[TMP49]], ptr [[TMP60]], align 4
+// CHECK2-NEXT: [[TMP61:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 2
+// CHECK2-NEXT: store ptr [[ARRAYIDX6]], ptr [[TMP61]], align 4
+// CHECK2-NEXT: [[TMP62:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES10]], i32 0, i32 2
+// CHECK2-NEXT: store i64 [[TMP53]], ptr [[TMP62]], align 4
+// CHECK2-NEXT: [[TMP63:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i32 0, i32 2
+// CHECK2-NEXT: store ptr null, ptr [[TMP63]], align 4
+// CHECK2-NEXT: [[TMP64:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 0
+// CHECK2-NEXT: [[TMP65:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 0
+// CHECK2-NEXT: [[TMP66:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES10]], i32 0, i32 0
+// CHECK2-NEXT: [[TMP67:%.*]] = load i32, ptr [[SIZE_ADDR]], align 4
+// CHECK2-NEXT: store i32 [[TMP67]], ptr [[DOTCAPTURE_EXPR_12]], align 4
+// CHECK2-NEXT: [[TMP68:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_12]], align 4
+// CHECK2-NEXT: [[SUB14:%.*]] = sub nsw i32 [[TMP68]], 0
// CHECK2-NEXT: [[DIV15:%.*]] = sdiv i32 [[SUB14]], 1
// CHECK2-NEXT: [[SUB16:%.*]] = sub nsw i32 [[DIV15]], 1
// CHECK2-NEXT: store i32 [[SUB16]], ptr [[DOTCAPTURE_EXPR_13]], align 4
-// CHECK2-NEXT: [[TMP65:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_13]], align 4
-// CHECK2-NEXT: [[ADD17:%.*]] = add nsw i32 [[TMP65]], 1
-// CHECK2-NEXT: [[TMP66:%.*]] = zext i32 [[ADD17]] to i64
+// CHECK2-NEXT: [[TMP69:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_13]], align 4
+// CHECK2-NEXT: [[ADD17:%.*]] = add nsw i32 [[TMP69]], 1
+// CHECK2-NEXT: [[TMP70:%.*]] = zext i32 [[ADD17]] to i64
// CHECK2-NEXT: [[KERNEL_ARGS18:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK2-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 0
-// CHECK2-NEXT: store i32 1, ptr [[TMP67]], align 4
-// CHECK2-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 1
-// CHECK2-NEXT: store i32 3, ptr [[TMP68]], align 4
-// CHECK2-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 2
-// CHECK2-NEXT: store ptr [[TMP60]], ptr [[TMP69]], align 4
-// CHECK2-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 3
-// CHECK2-NEXT: store ptr [[TMP61]], ptr [[TMP70]], align 4
-// CHECK2-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 4
-// CHECK2-NEXT: store ptr [[TMP62]], ptr [[TMP71]], align 4
-// CHECK2-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 5
-// CHECK2-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP72]], align 4
-// CHECK2-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 6
-// CHECK2-NEXT: store ptr null, ptr [[TMP73]], align 4
-// CHECK2-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 7
-// CHECK2-NEXT: store ptr null, ptr [[TMP74]], align 4
-// CHECK2-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 8
-// CHECK2-NEXT: store i64 [[TMP66]], ptr [[TMP75]], align 8
-// CHECK2-NEXT: [[TMP76:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3sumPiiS__l73.region_id, ptr [[KERNEL_ARGS18]])
-// CHECK2-NEXT: [[TMP77:%.*]] = icmp ne i32 [[TMP76]], 0
-// CHECK2-NEXT: br i1 [[TMP77]], label [[OMP_OFFLOAD_FAILED19:%.*]], label [[OMP_OFFLOAD_CONT20:%.*]]
+// CHECK2-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 0
+// CHECK2-NEXT: store i32 2, ptr [[TMP71]], align 4
+// CHECK2-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 1
+// CHECK2-NEXT: store i32 3, ptr [[TMP72]], align 4
+// CHECK2-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 2
+// CHECK2-NEXT: store ptr [[TMP64]], ptr [[TMP73]], align 4
+// CHECK2-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 3
+// CHECK2-NEXT: store ptr [[TMP65]], ptr [[TMP74]], align 4
+// CHECK2-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 4
+// CHECK2-NEXT: store ptr [[TMP66]], ptr [[TMP75]], align 4
+// CHECK2-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 5
+// CHECK2-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP76]], align 4
+// CHECK2-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 6
+// CHECK2-NEXT: store ptr null, ptr [[TMP77]], align 4
+// CHECK2-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 7
+// CHECK2-NEXT: store ptr null, ptr [[TMP78]], align 4
+// CHECK2-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 8
+// CHECK2-NEXT: store i64 [[TMP70]], ptr [[TMP79]], align 8
+// CHECK2-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 9
+// CHECK2-NEXT: store i64 0, ptr [[TMP80]], align 8
+// CHECK2-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 10
+// CHECK2-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP81]], align 4
+// CHECK2-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 11
+// CHECK2-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP82]], align 4
+// CHECK2-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 12
+// CHECK2-NEXT: store i32 0, ptr [[TMP83]], align 4
+// CHECK2-NEXT: [[TMP84:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3sumPiiS__l73.region_id, ptr [[KERNEL_ARGS18]])
+// CHECK2-NEXT: [[TMP85:%.*]] = icmp ne i32 [[TMP84]], 0
+// CHECK2-NEXT: br i1 [[TMP85]], label [[OMP_OFFLOAD_FAILED19:%.*]], label [[OMP_OFFLOAD_CONT20:%.*]]
// CHECK2: omp_offload.failed19:
-// CHECK2-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3sumPiiS__l73(i32 [[TMP40]], ptr [[TMP41]], ptr [[TMP42]]) #[[ATTR2]]
+// CHECK2-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3sumPiiS__l73(i32 [[TMP44]], ptr [[TMP45]], ptr [[TMP46]]) #[[ATTR2]]
// CHECK2-NEXT: br label [[OMP_OFFLOAD_CONT20]]
// CHECK2: omp_offload.cont20:
-// CHECK2-NEXT: [[TMP78:%.*]] = load i32, ptr [[SIZE_ADDR]], align 4
-// CHECK2-NEXT: store i32 [[TMP78]], ptr [[SIZE_CASTED21]], align 4
-// CHECK2-NEXT: [[TMP79:%.*]] = load i32, ptr [[SIZE_CASTED21]], align 4
+// CHECK2-NEXT: [[TMP86:%.*]] = load i32, ptr [[SIZE_ADDR]], align 4
+// CHECK2-NEXT: store i32 [[TMP86]], ptr [[SIZE_CASTED21]], align 4
+// CHECK2-NEXT: [[TMP87:%.*]] = load i32, ptr [[SIZE_CASTED21]], align 4
// CHECK2-NEXT: [[ARRAYIDX22:%.*]] = getelementptr inbounds [10 x i32], ptr [[A]], i32 0, i32 0
-// CHECK2-NEXT: [[TMP80:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
-// CHECK2-NEXT: store i32 [[TMP79]], ptr [[TMP80]], align 4
-// CHECK2-NEXT: [[TMP81:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
-// CHECK2-NEXT: store i32 [[TMP79]], ptr [[TMP81]], align 4
-// CHECK2-NEXT: [[TMP82:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i32 0, i32 0
-// CHECK2-NEXT: store ptr null, ptr [[TMP82]], align 4
-// CHECK2-NEXT: [[TMP83:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 1
-// CHECK2-NEXT: store ptr [[A]], ptr [[TMP83]], align 4
-// CHECK2-NEXT: [[TMP84:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 1
-// CHECK2-NEXT: store ptr [[ARRAYIDX22]], ptr [[TMP84]], align 4
-// CHECK2-NEXT: [[TMP85:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i32 0, i32 1
-// CHECK2-NEXT: store ptr null, ptr [[TMP85]], align 4
-// CHECK2-NEXT: [[TMP86:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
-// CHECK2-NEXT: [[TMP87:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
+// CHECK2-NEXT: [[TMP88:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
+// CHECK2-NEXT: store i32 [[TMP87]], ptr [[TMP88]], align 4
+// CHECK2-NEXT: [[TMP89:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
+// CHECK2-NEXT: store i32 [[TMP87]], ptr [[TMP89]], align 4
+// CHECK2-NEXT: [[TMP90:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i32 0, i32 0
+// CHECK2-NEXT: store ptr null, ptr [[TMP90]], align 4
+// CHECK2-NEXT: [[TMP91:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 1
+// CHECK2-NEXT: store ptr [[A]], ptr [[TMP91]], align 4
+// CHECK2-NEXT: [[TMP92:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 1
+// CHECK2-NEXT: store ptr [[ARRAYIDX22]], ptr [[TMP92]], align 4
+// CHECK2-NEXT: [[TMP93:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i32 0, i32 1
+// CHECK2-NEXT: store ptr null, ptr [[TMP93]], align 4
+// CHECK2-NEXT: [[TMP94:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
+// CHECK2-NEXT: [[TMP95:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
// CHECK2-NEXT: [[KERNEL_ARGS26:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK2-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 0
-// CHECK2-NEXT: store i32 1, ptr [[TMP88]], align 4
-// CHECK2-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 1
-// CHECK2-NEXT: store i32 2, ptr [[TMP89]], align 4
-// CHECK2-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 2
-// CHECK2-NEXT: store ptr [[TMP86]], ptr [[TMP90]], align 4
-// CHECK2-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 3
-// CHECK2-NEXT: store ptr [[TMP87]], ptr [[TMP91]], align 4
-// CHECK2-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 4
-// CHECK2-NEXT: store ptr @.offload_sizes.11, ptr [[TMP92]], align 4
-// CHECK2-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 5
-// CHECK2-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP93]], align 4
-// CHECK2-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 6
-// CHECK2-NEXT: store ptr null, ptr [[TMP94]], align 4
-// CHECK2-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 7
-// CHECK2-NEXT: store ptr null, ptr [[TMP95]], align 4
-// CHECK2-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 8
-// CHECK2-NEXT: store i64 0, ptr [[TMP96]], align 8
-// CHECK2-NEXT: [[TMP97:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3sumPiiS__l78.region_id, ptr [[KERNEL_ARGS26]])
-// CHECK2-NEXT: [[TMP98:%.*]] = icmp ne i32 [[TMP97]], 0
-// CHECK2-NEXT: br i1 [[TMP98]], label [[OMP_OFFLOAD_FAILED27:%.*]], label [[OMP_OFFLOAD_CONT28:%.*]]
+// CHECK2-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 0
+// CHECK2-NEXT: store i32 2, ptr [[TMP96]], align 4
+// CHECK2-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 1
+// CHECK2-NEXT: store i32 2, ptr [[TMP97]], align 4
+// CHECK2-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 2
+// CHECK2-NEXT: store ptr [[TMP94]], ptr [[TMP98]], align 4
+// CHECK2-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 3
+// CHECK2-NEXT: store ptr [[TMP95]], ptr [[TMP99]], align 4
+// CHECK2-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 4
+// CHECK2-NEXT: store ptr @.offload_sizes.11, ptr [[TMP100]], align 4
+// CHECK2-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 5
+// CHECK2-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP101]], align 4
+// CHECK2-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 6
+// CHECK2-NEXT: store ptr null, ptr [[TMP102]], align 4
+// CHECK2-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 7
+// CHECK2-NEXT: store ptr null, ptr [[TMP103]], align 4
+// CHECK2-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 8
+// CHECK2-NEXT: store i64 0, ptr [[TMP104]], align 8
+// CHECK2-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 9
+// CHECK2-NEXT: store i64 0, ptr [[TMP105]], align 8
+// CHECK2-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 10
+// CHECK2-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP106]], align 4
+// CHECK2-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 11
+// CHECK2-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP107]], align 4
+// CHECK2-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 12
+// CHECK2-NEXT: store i32 0, ptr [[TMP108]], align 4
+// CHECK2-NEXT: [[TMP109:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3sumPiiS__l78.region_id, ptr [[KERNEL_ARGS26]])
+// CHECK2-NEXT: [[TMP110:%.*]] = icmp ne i32 [[TMP109]], 0
+// CHECK2-NEXT: br i1 [[TMP110]], label [[OMP_OFFLOAD_FAILED27:%.*]], label [[OMP_OFFLOAD_CONT28:%.*]]
// CHECK2: omp_offload.failed27:
-// CHECK2-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3sumPiiS__l78(i32 [[TMP79]], ptr [[A]]) #[[ATTR2]]
+// CHECK2-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3sumPiiS__l78(i32 [[TMP87]], ptr [[A]]) #[[ATTR2]]
// CHECK2-NEXT: br label [[OMP_OFFLOAD_CONT28]]
// CHECK2: omp_offload.cont28:
-// CHECK2-NEXT: [[TMP99:%.*]] = load i32, ptr [[SIZE_ADDR]], align 4
-// CHECK2-NEXT: store i32 [[TMP99]], ptr [[SIZE_CASTED29]], align 4
-// CHECK2-NEXT: [[TMP100:%.*]] = load i32, ptr [[SIZE_CASTED29]], align 4
+// CHECK2-NEXT: [[TMP111:%.*]] = load i32, ptr [[SIZE_ADDR]], align 4
+// CHECK2-NEXT: store i32 [[TMP111]], ptr [[SIZE_CASTED29]], align 4
+// CHECK2-NEXT: [[TMP112:%.*]] = load i32, ptr [[SIZE_CASTED29]], align 4
// CHECK2-NEXT: [[ARRAYIDX30:%.*]] = getelementptr inbounds [10 x i32], ptr [[A]], i32 0, i32 3
-// CHECK2-NEXT: [[TMP101:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS31]], i32 0, i32 0
-// CHECK2-NEXT: store i32 [[TMP100]], ptr [[TMP101]], align 4
-// CHECK2-NEXT: [[TMP102:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS32]], i32 0, i32 0
-// CHECK2-NEXT: store i32 [[TMP100]], ptr [[TMP102]], align 4
-// CHECK2-NEXT: [[TMP103:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS33]], i32 0, i32 0
-// CHECK2-NEXT: store ptr null, ptr [[TMP103]], align 4
-// CHECK2-NEXT: [[TMP104:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS31]], i32 0, i32 1
-// CHECK2-NEXT: store ptr [[A]], ptr [[TMP104]], align 4
-// CHECK2-NEXT: [[TMP105:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS32]], i32 0, i32 1
-// CHECK2-NEXT: store ptr [[ARRAYIDX30]], ptr [[TMP105]], align 4
-// CHECK2-NEXT: [[TMP106:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS33]], i32 0, i32 1
-// CHECK2-NEXT: store ptr null, ptr [[TMP106]], align 4
-// CHECK2-NEXT: [[TMP107:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS31]], i32 0, i32 0
-// CHECK2-NEXT: [[TMP108:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS32]], i32 0, i32 0
-// CHECK2-NEXT: [[KERNEL_ARGS34:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK2-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS34]], i32 0, i32 0
-// CHECK2-NEXT: store i32 1, ptr [[TMP109]], align 4
-// CHECK2-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS34]], i32 0, i32 1
-// CHECK2-NEXT: store i32 2, ptr [[TMP110]], align 4
-// CHECK2-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS34]], i32 0, i32 2
-// CHECK2-NEXT: store ptr [[TMP107]], ptr [[TMP111]], align 4
-// CHECK2-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS34]], i32 0, i32 3
-// CHECK2-NEXT: store ptr [[TMP108]], ptr [[TMP112]], align 4
-// CHECK2-NEXT: [[TMP113:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS34]], i32 0, i32 4
-// CHECK2-NEXT: store ptr @.offload_sizes.15, ptr [[TMP113]], align 4
-// CHECK2-NEXT: [[TMP114:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS34]], i32 0, i32 5
-// CHECK2-NEXT: store ptr @.offload_maptypes.16, ptr [[TMP114]], align 4
-// CHECK2-NEXT: [[TMP115:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS34]], i32 0, i32 6
+// CHECK2-NEXT: [[TMP113:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS31]], i32 0, i32 0
+// CHECK2-NEXT: store i32 [[TMP112]], ptr [[TMP113]], align 4
+// CHECK2-NEXT: [[TMP114:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS32]], i32 0, i32 0
+// CHECK2-NEXT: store i32 [[TMP112]], ptr [[TMP114]], align 4
+// CHECK2-NEXT: [[TMP115:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS33]], i32 0, i32 0
// CHECK2-NEXT: store ptr null, ptr [[TMP115]], align 4
-// CHECK2-NEXT: [[TMP116:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS34]], i32 0, i32 7
-// CHECK2-NEXT: store ptr null, ptr [[TMP116]], align 4
-// CHECK2-NEXT: [[TMP117:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS34]], i32 0, i32 8
-// CHECK2-NEXT: store i64 0, ptr [[TMP117]], align 8
-// CHECK2-NEXT: [[TMP118:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3sumPiiS__l81.region_id, ptr [[KERNEL_ARGS34]])
-// CHECK2-NEXT: [[TMP119:%.*]] = icmp ne i32 [[TMP118]], 0
-// CHECK2-NEXT: br i1 [[TMP119]], label [[OMP_OFFLOAD_FAILED35:%.*]], label [[OMP_OFFLOAD_CONT36:%.*]]
+// CHECK2-NEXT: [[TMP116:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS31]], i32 0, i32 1
+// CHECK2-NEXT: store ptr [[A]], ptr [[TMP116]], align 4
+// CHECK2-NEXT: [[TMP117:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS32]], i32 0, i32 1
+// CHECK2-NEXT: store ptr [[ARRAYIDX30]], ptr [[TMP117]], align 4
+// CHECK2-NEXT: [[TMP118:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS33]], i32 0, i32 1
+// CHECK2-NEXT: store ptr null, ptr [[TMP118]], align 4
+// CHECK2-NEXT: [[TMP119:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS31]], i32 0, i32 0
+// CHECK2-NEXT: [[TMP120:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS32]], i32 0, i32 0
+// CHECK2-NEXT: [[KERNEL_ARGS34:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
+// CHECK2-NEXT: [[TMP121:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS34]], i32 0, i32 0
+// CHECK2-NEXT: store i32 2, ptr [[TMP121]], align 4
+// CHECK2-NEXT: [[TMP122:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS34]], i32 0, i32 1
+// CHECK2-NEXT: store i32 2, ptr [[TMP122]], align 4
+// CHECK2-NEXT: [[TMP123:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS34]], i32 0, i32 2
+// CHECK2-NEXT: store ptr [[TMP119]], ptr [[TMP123]], align 4
+// CHECK2-NEXT: [[TMP124:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS34]], i32 0, i32 3
+// CHECK2-NEXT: store ptr [[TMP120]], ptr [[TMP124]], align 4
+// CHECK2-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS34]], i32 0, i32 4
+// CHECK2-NEXT: store ptr @.offload_sizes.15, ptr [[TMP125]], align 4
+// CHECK2-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS34]], i32 0, i32 5
+// CHECK2-NEXT: store ptr @.offload_maptypes.16, ptr [[TMP126]], align 4
+// CHECK2-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS34]], i32 0, i32 6
+// CHECK2-NEXT: store ptr null, ptr [[TMP127]], align 4
+// CHECK2-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS34]], i32 0, i32 7
+// CHECK2-NEXT: store ptr null, ptr [[TMP128]], align 4
+// CHECK2-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS34]], i32 0, i32 8
+// CHECK2-NEXT: store i64 0, ptr [[TMP129]], align 8
+// CHECK2-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS34]], i32 0, i32 9
+// CHECK2-NEXT: store i64 0, ptr [[TMP130]], align 8
+// CHECK2-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS34]], i32 0, i32 10
+// CHECK2-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP131]], align 4
+// CHECK2-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS34]], i32 0, i32 11
+// CHECK2-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP132]], align 4
+// CHECK2-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS34]], i32 0, i32 12
+// CHECK2-NEXT: store i32 0, ptr [[TMP133]], align 4
+// CHECK2-NEXT: [[TMP134:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3sumPiiS__l81.region_id, ptr [[KERNEL_ARGS34]])
+// CHECK2-NEXT: [[TMP135:%.*]] = icmp ne i32 [[TMP134]], 0
+// CHECK2-NEXT: br i1 [[TMP135]], label [[OMP_OFFLOAD_FAILED35:%.*]], label [[OMP_OFFLOAD_CONT36:%.*]]
// CHECK2: omp_offload.failed35:
-// CHECK2-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3sumPiiS__l81(i32 [[TMP100]], ptr [[A]]) #[[ATTR2]]
+// CHECK2-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3sumPiiS__l81(i32 [[TMP112]], ptr [[A]]) #[[ATTR2]]
// CHECK2-NEXT: br label [[OMP_OFFLOAD_CONT36]]
// CHECK2: omp_offload.cont36:
// CHECK2-NEXT: ret void
diff --git a/clang/test/OpenMP/target_codegen.cpp b/clang/test/OpenMP/target_codegen.cpp
index 5a283338eaf05..af5ad7303de75 100644
--- a/clang/test/OpenMP/target_codegen.cpp
+++ b/clang/test/OpenMP/target_codegen.cpp
@@ -403,7 +403,7 @@ int foo(int n) {
// CHECK: define internal void [[HVT0_:@.+]](i[[SZ]]* noundef {{%[^,]+}}, i[[SZ]] noundef {{%[^,]+}})
// CHECK: define internal {{.*}}i32 [[OMP_TASK_ENTRY]](i32 {{.*}}%0, [[KMP_TASK_T_WITH_PRIVATES]]* noalias noundef %1)
-// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_kernel_nowait(%struct.ident_t* @{{.+}}, i64 [[DEVICE:%.+]], i32 {{.+}}, i32 {{.+}}, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]], i32 0, i8* null, i32 0, i8* null)
+// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 [[DEVICE:%.+]], i32 {{.+}}, i32 {{.+}}, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]])
// CHECK-DAG: [[BPARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 2
// CHECK-DAG: store i8** [[BPR:%.+]], i8*** [[BPARG]]
// CHECK-DAG: [[PARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 3
diff --git a/clang/test/OpenMP/target_codegen_global_capture.cpp b/clang/test/OpenMP/target_codegen_global_capture.cpp
index 330ee8e786465..4c9556c5aa908 100644
--- a/clang/test/OpenMP/target_codegen_global_capture.cpp
+++ b/clang/test/OpenMP/target_codegen_global_capture.cpp
@@ -288,7 +288,7 @@ int tbar2(short a, short b, short c, short d){
// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP50]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP50]], align 4
// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 9, ptr [[TMP51]], align 4
// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -305,9 +305,17 @@ int tbar2(short a, short b, short c, short d){
// CHECK1-NEXT: store ptr null, ptr [[TMP57]], align 8
// CHECK1-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP58]], align 8
-// CHECK1-NEXT: [[TMP59:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foossss_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP60:%.*]] = icmp ne i32 [[TMP59]], 0
-// CHECK1-NEXT: br i1 [[TMP60]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP59]], align 8
+// CHECK1-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP60]], align 4
+// CHECK1-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4
+// CHECK1-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP62]], align 4
+// CHECK1-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foossss_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
+// CHECK1-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foossss_l49(i64 [[TMP1]], i64 [[TMP3]], i64 [[TMP5]], i64 [[TMP7]], i64 [[TMP9]], i64 [[TMP11]], i64 [[TMP13]], i64 [[TMP15]], i64 [[TMP17]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -317,28 +325,28 @@ int tbar2(short a, short b, short c, short d){
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foossss_l49(i64 [[TMP1]], i64 [[TMP3]], i64 [[TMP5]], i64 [[TMP7]], i64 [[TMP9]], i64 [[TMP11]], i64 [[TMP13]], i64 [[TMP15]], i64 [[TMP17]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
-// CHECK1-NEXT: [[TMP61:%.*]] = load i16, ptr [[A_ADDR]], align 2
-// CHECK1-NEXT: [[CONV5:%.*]] = sext i16 [[TMP61]] to i32
-// CHECK1-NEXT: [[TMP62:%.*]] = load i16, ptr [[B_ADDR]], align 2
-// CHECK1-NEXT: [[CONV6:%.*]] = sext i16 [[TMP62]] to i32
+// CHECK1-NEXT: [[TMP65:%.*]] = load i16, ptr [[A_ADDR]], align 2
+// CHECK1-NEXT: [[CONV5:%.*]] = sext i16 [[TMP65]] to i32
+// CHECK1-NEXT: [[TMP66:%.*]] = load i16, ptr [[B_ADDR]], align 2
+// CHECK1-NEXT: [[CONV6:%.*]] = sext i16 [[TMP66]] to i32
// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[CONV5]], [[CONV6]]
-// CHECK1-NEXT: [[TMP63:%.*]] = load i16, ptr [[C_ADDR]], align 2
-// CHECK1-NEXT: [[CONV7:%.*]] = sext i16 [[TMP63]] to i32
+// CHECK1-NEXT: [[TMP67:%.*]] = load i16, ptr [[C_ADDR]], align 2
+// CHECK1-NEXT: [[CONV7:%.*]] = sext i16 [[TMP67]] to i32
// CHECK1-NEXT: [[ADD8:%.*]] = add nsw i32 [[ADD]], [[CONV7]]
-// CHECK1-NEXT: [[TMP64:%.*]] = load i16, ptr [[D_ADDR]], align 2
-// CHECK1-NEXT: [[CONV9:%.*]] = sext i16 [[TMP64]] to i32
+// CHECK1-NEXT: [[TMP68:%.*]] = load i16, ptr [[D_ADDR]], align 2
+// CHECK1-NEXT: [[CONV9:%.*]] = sext i16 [[TMP68]] to i32
// CHECK1-NEXT: [[ADD10:%.*]] = add nsw i32 [[ADD8]], [[CONV9]]
-// CHECK1-NEXT: [[TMP65:%.*]] = load float, ptr @_ZZ3foossssE2Sa, align 4
-// CHECK1-NEXT: [[CONV11:%.*]] = fptosi float [[TMP65]] to i32
+// CHECK1-NEXT: [[TMP69:%.*]] = load float, ptr @_ZZ3foossssE2Sa, align 4
+// CHECK1-NEXT: [[CONV11:%.*]] = fptosi float [[TMP69]] to i32
// CHECK1-NEXT: [[ADD12:%.*]] = add nsw i32 [[ADD10]], [[CONV11]]
-// CHECK1-NEXT: [[TMP66:%.*]] = load float, ptr @_ZZ3foossssE2Sb, align 4
-// CHECK1-NEXT: [[CONV13:%.*]] = fptosi float [[TMP66]] to i32
+// CHECK1-NEXT: [[TMP70:%.*]] = load float, ptr @_ZZ3foossssE2Sb, align 4
+// CHECK1-NEXT: [[CONV13:%.*]] = fptosi float [[TMP70]] to i32
// CHECK1-NEXT: [[ADD14:%.*]] = add nsw i32 [[ADD12]], [[CONV13]]
-// CHECK1-NEXT: [[TMP67:%.*]] = load float, ptr @_ZZ3foossssE2Sc, align 4
-// CHECK1-NEXT: [[CONV15:%.*]] = fptosi float [[TMP67]] to i32
+// CHECK1-NEXT: [[TMP71:%.*]] = load float, ptr @_ZZ3foossssE2Sc, align 4
+// CHECK1-NEXT: [[CONV15:%.*]] = fptosi float [[TMP71]] to i32
// CHECK1-NEXT: [[ADD16:%.*]] = add nsw i32 [[ADD14]], [[CONV15]]
-// CHECK1-NEXT: [[TMP68:%.*]] = load float, ptr @_ZZ3foossssE2Sd, align 4
-// CHECK1-NEXT: [[CONV17:%.*]] = fptosi float [[TMP68]] to i32
+// CHECK1-NEXT: [[TMP72:%.*]] = load float, ptr @_ZZ3foossssE2Sd, align 4
+// CHECK1-NEXT: [[CONV17:%.*]] = fptosi float [[TMP72]] to i32
// CHECK1-NEXT: [[ADD18:%.*]] = add nsw i32 [[ADD16]], [[CONV17]]
// CHECK1-NEXT: ret i32 [[ADD18]]
//
@@ -607,7 +615,7 @@ int tbar2(short a, short b, short c, short d){
// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP54]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP54]], align 4
// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 9, ptr [[TMP55]], align 4
// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -624,9 +632,17 @@ int tbar2(short a, short b, short c, short d){
// CHECK1-NEXT: store ptr null, ptr [[TMP61]], align 8
// CHECK1-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP62]], align 8
-// CHECK1-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3barssss_l94.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
-// CHECK1-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP63]], align 8
+// CHECK1-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP64]], align 4
+// CHECK1-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP65]], align 4
+// CHECK1-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP66]], align 4
+// CHECK1-NEXT: [[TMP67:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3barssss_l94.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP68:%.*]] = icmp ne i32 [[TMP67]], 0
+// CHECK1-NEXT: br i1 [[TMP68]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3barssss_l94(i64 [[TMP5]], i64 [[TMP7]], i64 [[TMP9]], i64 [[TMP11]], i64 [[TMP13]], i64 [[TMP15]], i64 [[TMP17]], i64 [[TMP19]], i64 [[TMP21]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -922,7 +938,7 @@ int tbar2(short a, short b, short c, short d){
// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP54]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP54]], align 4
// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 9, ptr [[TMP55]], align 4
// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -939,9 +955,17 @@ int tbar2(short a, short b, short c, short d){
// CHECK1-NEXT: store ptr null, ptr [[TMP61]], align 8
// CHECK1-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP62]], align 8
-// CHECK1-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z4tbarIsEiT_S0_S0_S0__l145.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
-// CHECK1-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP63]], align 8
+// CHECK1-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP64]], align 4
+// CHECK1-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP65]], align 4
+// CHECK1-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP66]], align 4
+// CHECK1-NEXT: [[TMP67:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z4tbarIsEiT_S0_S0_S0__l145.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP68:%.*]] = icmp ne i32 [[TMP67]], 0
+// CHECK1-NEXT: br i1 [[TMP68]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z4tbarIsEiT_S0_S0_S0__l145(i64 [[TMP5]], i64 [[TMP7]], i64 [[TMP9]], i64 [[TMP11]], i64 [[TMP13]], i64 [[TMP15]], i64 [[TMP17]], i64 [[TMP19]], i64 [[TMP21]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1167,7 +1191,7 @@ int tbar2(short a, short b, short c, short d){
// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP44]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP44]], align 4
// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 9, ptr [[TMP45]], align 4
// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1184,9 +1208,17 @@ int tbar2(short a, short b, short c, short d){
// CHECK3-NEXT: store ptr null, ptr [[TMP51]], align 4
// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP52]], align 8
-// CHECK3-NEXT: [[TMP53:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foossss_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP54:%.*]] = icmp ne i32 [[TMP53]], 0
-// CHECK3-NEXT: br i1 [[TMP54]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP53]], align 8
+// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP54]], align 4
+// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP56]], align 4
+// CHECK3-NEXT: [[TMP57:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foossss_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP58:%.*]] = icmp ne i32 [[TMP57]], 0
+// CHECK3-NEXT: br i1 [[TMP58]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foossss_l49(i32 [[TMP1]], ptr @Gb, i32 [[TMP3]], ptr @Gc, i32 [[TMP5]], i32 [[TMP7]], i32 [[TMP9]], ptr @Gd, i32 [[TMP11]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1196,28 +1228,28 @@ int tbar2(short a, short b, short c, short d){
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foossss_l49(i32 [[TMP1]], ptr @Gb, i32 [[TMP3]], ptr @Gc, i32 [[TMP5]], i32 [[TMP7]], i32 [[TMP9]], ptr @Gd, i32 [[TMP11]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
-// CHECK3-NEXT: [[TMP55:%.*]] = load i16, ptr [[A_ADDR]], align 2
-// CHECK3-NEXT: [[CONV5:%.*]] = sext i16 [[TMP55]] to i32
-// CHECK3-NEXT: [[TMP56:%.*]] = load i16, ptr [[B_ADDR]], align 2
-// CHECK3-NEXT: [[CONV6:%.*]] = sext i16 [[TMP56]] to i32
+// CHECK3-NEXT: [[TMP59:%.*]] = load i16, ptr [[A_ADDR]], align 2
+// CHECK3-NEXT: [[CONV5:%.*]] = sext i16 [[TMP59]] to i32
+// CHECK3-NEXT: [[TMP60:%.*]] = load i16, ptr [[B_ADDR]], align 2
+// CHECK3-NEXT: [[CONV6:%.*]] = sext i16 [[TMP60]] to i32
// CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 [[CONV5]], [[CONV6]]
-// CHECK3-NEXT: [[TMP57:%.*]] = load i16, ptr [[C_ADDR]], align 2
-// CHECK3-NEXT: [[CONV7:%.*]] = sext i16 [[TMP57]] to i32
+// CHECK3-NEXT: [[TMP61:%.*]] = load i16, ptr [[C_ADDR]], align 2
+// CHECK3-NEXT: [[CONV7:%.*]] = sext i16 [[TMP61]] to i32
// CHECK3-NEXT: [[ADD8:%.*]] = add nsw i32 [[ADD]], [[CONV7]]
-// CHECK3-NEXT: [[TMP58:%.*]] = load i16, ptr [[D_ADDR]], align 2
-// CHECK3-NEXT: [[CONV9:%.*]] = sext i16 [[TMP58]] to i32
+// CHECK3-NEXT: [[TMP62:%.*]] = load i16, ptr [[D_ADDR]], align 2
+// CHECK3-NEXT: [[CONV9:%.*]] = sext i16 [[TMP62]] to i32
// CHECK3-NEXT: [[ADD10:%.*]] = add nsw i32 [[ADD8]], [[CONV9]]
-// CHECK3-NEXT: [[TMP59:%.*]] = load float, ptr @_ZZ3foossssE2Sa, align 4
-// CHECK3-NEXT: [[CONV11:%.*]] = fptosi float [[TMP59]] to i32
+// CHECK3-NEXT: [[TMP63:%.*]] = load float, ptr @_ZZ3foossssE2Sa, align 4
+// CHECK3-NEXT: [[CONV11:%.*]] = fptosi float [[TMP63]] to i32
// CHECK3-NEXT: [[ADD12:%.*]] = add nsw i32 [[ADD10]], [[CONV11]]
-// CHECK3-NEXT: [[TMP60:%.*]] = load float, ptr @_ZZ3foossssE2Sb, align 4
-// CHECK3-NEXT: [[CONV13:%.*]] = fptosi float [[TMP60]] to i32
+// CHECK3-NEXT: [[TMP64:%.*]] = load float, ptr @_ZZ3foossssE2Sb, align 4
+// CHECK3-NEXT: [[CONV13:%.*]] = fptosi float [[TMP64]] to i32
// CHECK3-NEXT: [[ADD14:%.*]] = add nsw i32 [[ADD12]], [[CONV13]]
-// CHECK3-NEXT: [[TMP61:%.*]] = load float, ptr @_ZZ3foossssE2Sc, align 4
-// CHECK3-NEXT: [[CONV15:%.*]] = fptosi float [[TMP61]] to i32
+// CHECK3-NEXT: [[TMP65:%.*]] = load float, ptr @_ZZ3foossssE2Sc, align 4
+// CHECK3-NEXT: [[CONV15:%.*]] = fptosi float [[TMP65]] to i32
// CHECK3-NEXT: [[ADD16:%.*]] = add nsw i32 [[ADD14]], [[CONV15]]
-// CHECK3-NEXT: [[TMP62:%.*]] = load float, ptr @_ZZ3foossssE2Sd, align 4
-// CHECK3-NEXT: [[CONV17:%.*]] = fptosi float [[TMP62]] to i32
+// CHECK3-NEXT: [[TMP66:%.*]] = load float, ptr @_ZZ3foossssE2Sd, align 4
+// CHECK3-NEXT: [[CONV17:%.*]] = fptosi float [[TMP66]] to i32
// CHECK3-NEXT: [[ADD18:%.*]] = add nsw i32 [[ADD16]], [[CONV17]]
// CHECK3-NEXT: ret i32 [[ADD18]]
//
@@ -1486,7 +1518,7 @@ int tbar2(short a, short b, short c, short d){
// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP48]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP48]], align 4
// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 9, ptr [[TMP49]], align 4
// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1503,9 +1535,17 @@ int tbar2(short a, short b, short c, short d){
// CHECK3-NEXT: store ptr null, ptr [[TMP55]], align 4
// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP56]], align 8
-// CHECK3-NEXT: [[TMP57:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3barssss_l94.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP58:%.*]] = icmp ne i32 [[TMP57]], 0
-// CHECK3-NEXT: br i1 [[TMP58]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP57]], align 8
+// CHECK3-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP58]], align 4
+// CHECK3-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP59]], align 4
+// CHECK3-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP60]], align 4
+// CHECK3-NEXT: [[TMP61:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3barssss_l94.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP62:%.*]] = icmp ne i32 [[TMP61]], 0
+// CHECK3-NEXT: br i1 [[TMP62]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3barssss_l94(i32 [[TMP5]], ptr @Gb, i32 [[TMP7]], ptr @Gc, i32 [[TMP9]], i32 [[TMP11]], i32 [[TMP13]], ptr @Gd, i32 [[TMP15]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1801,7 +1841,7 @@ int tbar2(short a, short b, short c, short d){
// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP48]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP48]], align 4
// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 9, ptr [[TMP49]], align 4
// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1818,9 +1858,17 @@ int tbar2(short a, short b, short c, short d){
// CHECK3-NEXT: store ptr null, ptr [[TMP55]], align 4
// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP56]], align 8
-// CHECK3-NEXT: [[TMP57:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z4tbarIsEiT_S0_S0_S0__l145.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP58:%.*]] = icmp ne i32 [[TMP57]], 0
-// CHECK3-NEXT: br i1 [[TMP58]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP57]], align 8
+// CHECK3-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP58]], align 4
+// CHECK3-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP59]], align 4
+// CHECK3-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP60]], align 4
+// CHECK3-NEXT: [[TMP61:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z4tbarIsEiT_S0_S0_S0__l145.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP62:%.*]] = icmp ne i32 [[TMP61]], 0
+// CHECK3-NEXT: br i1 [[TMP62]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z4tbarIsEiT_S0_S0_S0__l145(i32 [[TMP5]], ptr @Gb, i32 [[TMP7]], ptr @Gc, i32 [[TMP9]], i32 [[TMP11]], i32 [[TMP13]], ptr @Gd, i32 [[TMP15]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/target_data_map_codegen_hold.cpp b/clang/test/OpenMP/target_data_map_codegen_hold.cpp
index 78605c4e33fe2..f267833232fed 100644
--- a/clang/test/OpenMP/target_data_map_codegen_hold.cpp
+++ b/clang/test/OpenMP/target_data_map_codegen_hold.cpp
@@ -54,22 +54,26 @@ struct S2 {
//.
// CHECK-PPC64LE: @.offload_sizes = private unnamed_addr constant [1 x i64] [i64 20]
-// CHECK-PPC64LE: @.offload_maptypes = private unnamed_addr constant [1 x i64] [i64 [[#0x2001]]]
+// CHECK-PPC64LE: @.offload_maptypes = private unnamed_addr constant [1 x i64] [i64 8193]
+// CHECK-PPC64LE: @0 = private unnamed_addr constant [23 x i8] c"
+// CHECK-PPC64LE: @1 = private unnamed_addr constant %struct.ident_t { i32 0, i32 2, i32 0, i32 22, ptr @0 }, align 8
// CHECK-PPC64LE: @.offload_sizes.1 = private unnamed_addr constant [1 x i64] [i64 20]
-// CHECK-PPC64LE: @.offload_maptypes.2 = private unnamed_addr constant [1 x i64] [i64 [[#0x2405]]]
+// CHECK-PPC64LE: @.offload_maptypes.2 = private unnamed_addr constant [1 x i64] [i64 9221]
// CHECK-PPC64LE: @.offload_sizes.3 = private unnamed_addr constant [1 x i64] [i64 4]
-// CHECK-PPC64LE: @.offload_maptypes.4 = private unnamed_addr constant [1 x i64] [i64 [[#0x2003]]]
+// CHECK-PPC64LE: @.offload_maptypes.4 = private unnamed_addr constant [1 x i64] [i64 8195]
// CHECK-PPC64LE: @.offload_sizes.5 = private unnamed_addr constant [11 x i64] [i64 0, i64 4, i64 8, i64 8, i64 4, i64 4, i64 0, i64 4, i64 8, i64 8, i64 4]
-// CHECK-PPC64LE: @.offload_maptypes.6 = private unnamed_addr constant [11 x i64] [i64 [[#0x2000]], i64 [[#0x1000000002003]], i64 [[#0x1000000002010]], i64 [[#0x2010]], i64 [[#0x2013]], i64 [[#0x3]], i64 [[#0x2000]], i64 [[#0x7000000002003]], i64 [[#0x7000000002010]], i64 [[#0x2010]], i64 [[#0x2013]]]
+// CHECK-PPC64LE: @.offload_maptypes.6 = private unnamed_addr constant [11 x i64] [i64 8192, i64 281474976718851, i64 281474976718864, i64 8208, i64 8211, i64 3, i64 8192, i64 1970324836982787, i64 1970324836982800, i64 8208, i64 8211]
//.
// CHECK-I386: @.offload_sizes = private unnamed_addr constant [1 x i64] [i64 20]
-// CHECK-I386: @.offload_maptypes = private unnamed_addr constant [1 x i64] [i64 [[#0x2001]]]
+// CHECK-I386: @.offload_maptypes = private unnamed_addr constant [1 x i64] [i64 8193]
+// CHECK-I386: @0 = private unnamed_addr constant [23 x i8] c"
+// CHECK-I386: @1 = private unnamed_addr constant %struct.ident_t { i32 0, i32 2, i32 0, i32 22, ptr @0 }, align 8
// CHECK-I386: @.offload_sizes.1 = private unnamed_addr constant [1 x i64] [i64 20]
-// CHECK-I386: @.offload_maptypes.2 = private unnamed_addr constant [1 x i64] [i64 [[#0x2405]]]
+// CHECK-I386: @.offload_maptypes.2 = private unnamed_addr constant [1 x i64] [i64 9221]
// CHECK-I386: @.offload_sizes.3 = private unnamed_addr constant [1 x i64] [i64 4]
-// CHECK-I386: @.offload_maptypes.4 = private unnamed_addr constant [1 x i64] [i64 [[#0x2003]]]
+// CHECK-I386: @.offload_maptypes.4 = private unnamed_addr constant [1 x i64] [i64 8195]
// CHECK-I386: @.offload_sizes.5 = private unnamed_addr constant [11 x i64] [i64 0, i64 4, i64 4, i64 4, i64 4, i64 4, i64 0, i64 4, i64 4, i64 4, i64 4]
-// CHECK-I386: @.offload_maptypes.6 = private unnamed_addr constant [11 x i64] [i64 [[#0x2000]], i64 [[#0x1000000002003]], i64 [[#0x1000000002010]], i64 [[#0x2010]], i64 [[#0x2013]], i64 [[#0x3]], i64 [[#0x2000]], i64 [[#0x7000000002003]], i64 [[#0x7000000002010]], i64 [[#0x2010]], i64 [[#0x2013]]]
+// CHECK-I386: @.offload_maptypes.6 = private unnamed_addr constant [11 x i64] [i64 8192, i64 281474976718851, i64 281474976718864, i64 8208, i64 8211, i64 3, i64 8192, i64 1970324836982787, i64 1970324836982800, i64 8208, i64 8211]
//.
// CHECK-PPC64LE-LABEL: @_Z3fooi(
// CHECK-PPC64LE-NEXT: entry:
@@ -93,189 +97,189 @@ struct S2 {
// CHECK-PPC64LE-NEXT: store i32 [[ARG:%.*]], ptr [[ARG_ADDR]], align 4
// CHECK-PPC64LE-NEXT: [[TMP0:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK-PPC64LE-NEXT: store ptr [[LB]], ptr [[TMP0]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP2:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: store ptr [[LB]], ptr [[TMP2]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK-PPC64LE-NEXT: store ptr null, ptr [[TMP4]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP5:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: call void @__tgt_target_data_begin_mapper(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 1, ptr [[TMP5]], ptr [[TMP6]], ptr @.offload_sizes, ptr @.offload_maptypes, ptr null, ptr null)
-// CHECK-PPC64LE-NEXT: [[TMP7:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
-// CHECK-PPC64LE-NEXT: [[INC:%.*]] = add nsw i32 [[TMP7]], 1
+// CHECK-PPC64LE-NEXT: [[TMP1:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK-PPC64LE-NEXT: store ptr [[LB]], ptr [[TMP1]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP2:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK-PPC64LE-NEXT: store ptr null, ptr [[TMP2]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK-PPC64LE-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK-PPC64LE-NEXT: call void @__tgt_target_data_begin_mapper(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 1, ptr [[TMP3]], ptr [[TMP4]], ptr @.offload_sizes, ptr @.offload_maptypes, ptr null, ptr null)
+// CHECK-PPC64LE-NEXT: [[TMP5:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
+// CHECK-PPC64LE-NEXT: [[INC:%.*]] = add nsw i32 [[TMP5]], 1
// CHECK-PPC64LE-NEXT: store i32 [[INC]], ptr [[ARG_ADDR]], align 4
-// CHECK-PPC64LE-NEXT: [[TMP8:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: call void @__tgt_target_data_end_mapper(ptr @[[GLOB1]], i64 -1, i32 1, ptr [[TMP8]], ptr [[TMP9]], ptr @.offload_sizes, ptr @.offload_maptypes, ptr null, ptr null)
-// CHECK-PPC64LE-NEXT: [[TMP10:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: store ptr [[LB]], ptr [[TMP10]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK-PPC64LE-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK-PPC64LE-NEXT: call void @__tgt_target_data_end_mapper(ptr @[[GLOB1]], i64 -1, i32 1, ptr [[TMP6]], ptr [[TMP7]], ptr @.offload_sizes, ptr @.offload_maptypes, ptr null, ptr null)
+// CHECK-PPC64LE-NEXT: [[TMP8:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK-PPC64LE-NEXT: store ptr [[LB]], ptr [[TMP8]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK-PPC64LE-NEXT: store ptr [[LB]], ptr [[TMP9]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP10:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
+// CHECK-PPC64LE-NEXT: store ptr null, ptr [[TMP10]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP11:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
// CHECK-PPC64LE-NEXT: [[TMP12:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: store ptr [[LB]], ptr [[TMP12]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP14:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
-// CHECK-PPC64LE-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: call void @__tgt_target_data_begin_mapper(ptr @[[GLOB1]], i64 -1, i32 1, ptr [[TMP15]], ptr [[TMP16]], ptr @.offload_sizes.1, ptr @.offload_maptypes.2, ptr null, ptr null)
-// CHECK-PPC64LE-NEXT: [[TMP17:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
-// CHECK-PPC64LE-NEXT: [[INC4:%.*]] = add nsw i32 [[TMP17]], 1
+// CHECK-PPC64LE-NEXT: call void @__tgt_target_data_begin_mapper(ptr @[[GLOB1]], i64 -1, i32 1, ptr [[TMP11]], ptr [[TMP12]], ptr @.offload_sizes.1, ptr @.offload_maptypes.2, ptr null, ptr null)
+// CHECK-PPC64LE-NEXT: [[TMP13:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
+// CHECK-PPC64LE-NEXT: [[INC4:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK-PPC64LE-NEXT: store i32 [[INC4]], ptr [[ARG_ADDR]], align 4
-// CHECK-PPC64LE-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: call void @__tgt_target_data_end_mapper(ptr @[[GLOB1]], i64 -1, i32 1, ptr [[TMP18]], ptr [[TMP19]], ptr @.offload_sizes.1, ptr @.offload_maptypes.2, ptr null, ptr null)
-// CHECK-PPC64LE-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: store ptr [[ARG_ADDR]], ptr [[TMP20]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: store ptr [[ARG_ADDR]], ptr [[TMP22]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 0
-// CHECK-PPC64LE-NEXT: store ptr null, ptr [[TMP24]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: call void @__tgt_target_data_begin_mapper(ptr @[[GLOB1]], i64 -1, i32 1, ptr [[TMP25]], ptr [[TMP26]], ptr @.offload_sizes.3, ptr @.offload_maptypes.4, ptr null, ptr null)
-// CHECK-PPC64LE-NEXT: [[TMP27:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
-// CHECK-PPC64LE-NEXT: [[INC8:%.*]] = add nsw i32 [[TMP27]], 1
+// CHECK-PPC64LE-NEXT: [[TMP14:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK-PPC64LE-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK-PPC64LE-NEXT: call void @__tgt_target_data_end_mapper(ptr @[[GLOB1]], i64 -1, i32 1, ptr [[TMP14]], ptr [[TMP15]], ptr @.offload_sizes.1, ptr @.offload_maptypes.2, ptr null, ptr null)
+// CHECK-PPC64LE-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK-PPC64LE-NEXT: store ptr [[ARG_ADDR]], ptr [[TMP16]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK-PPC64LE-NEXT: store ptr [[ARG_ADDR]], ptr [[TMP17]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 0
+// CHECK-PPC64LE-NEXT: store ptr null, ptr [[TMP18]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK-PPC64LE-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK-PPC64LE-NEXT: call void @__tgt_target_data_begin_mapper(ptr @[[GLOB1]], i64 -1, i32 1, ptr [[TMP19]], ptr [[TMP20]], ptr @.offload_sizes.3, ptr @.offload_maptypes.4, ptr null, ptr null)
+// CHECK-PPC64LE-NEXT: [[TMP21:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
+// CHECK-PPC64LE-NEXT: [[INC8:%.*]] = add nsw i32 [[TMP21]], 1
// CHECK-PPC64LE-NEXT: store i32 [[INC8]], ptr [[ARG_ADDR]], align 4
-// CHECK-PPC64LE-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: [[TMP29:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: call void @__tgt_target_data_end_mapper(ptr @[[GLOB1]], i64 -1, i32 1, ptr [[TMP28]], ptr [[TMP29]], ptr @.offload_sizes.3, ptr @.offload_maptypes.4, ptr null, ptr null)
+// CHECK-PPC64LE-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK-PPC64LE-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK-PPC64LE-NEXT: call void @__tgt_target_data_end_mapper(ptr @[[GLOB1]], i64 -1, i32 1, ptr [[TMP22]], ptr [[TMP23]], ptr @.offload_sizes.3, ptr @.offload_maptypes.4, ptr null, ptr null)
+// CHECK-PPC64LE-NEXT: [[TMP24:%.*]] = load ptr, ptr [[PS1]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP25:%.*]] = load ptr, ptr [[PS1]], align 8
+// CHECK-PPC64LE-NEXT: [[S:%.*]] = getelementptr inbounds [[STRUCT_S2:%.*]], ptr [[TMP25]], i32 0, i32 0
+// CHECK-PPC64LE-NEXT: [[TMP26:%.*]] = load ptr, ptr [[PS1]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP27:%.*]] = load ptr, ptr [[PS1]], align 8
+// CHECK-PPC64LE-NEXT: [[PS:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP27]], i32 0, i32 1
+// CHECK-PPC64LE-NEXT: [[TMP28:%.*]] = load ptr, ptr [[PS1]], align 8
+// CHECK-PPC64LE-NEXT: [[PS9:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP28]], i32 0, i32 1
+// CHECK-PPC64LE-NEXT: [[TMP29:%.*]] = load ptr, ptr [[PS9]], align 8
+// CHECK-PPC64LE-NEXT: [[PS10:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP29]], i32 0, i32 1
// CHECK-PPC64LE-NEXT: [[TMP30:%.*]] = load ptr, ptr [[PS1]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP31:%.*]] = load ptr, ptr [[PS1]], align 8
-// CHECK-PPC64LE-NEXT: [[S:%.*]] = getelementptr inbounds [[STRUCT_S2:%.*]], ptr [[TMP31]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: [[TMP32:%.*]] = load ptr, ptr [[PS1]], align 8
+// CHECK-PPC64LE-NEXT: [[PS11:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP30]], i32 0, i32 1
+// CHECK-PPC64LE-NEXT: [[TMP31:%.*]] = load ptr, ptr [[PS11]], align 8
+// CHECK-PPC64LE-NEXT: [[PS12:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP31]], i32 0, i32 1
+// CHECK-PPC64LE-NEXT: [[TMP32:%.*]] = load ptr, ptr [[PS12]], align 8
+// CHECK-PPC64LE-NEXT: [[PS13:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP32]], i32 0, i32 1
// CHECK-PPC64LE-NEXT: [[TMP33:%.*]] = load ptr, ptr [[PS1]], align 8
-// CHECK-PPC64LE-NEXT: [[PS:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP33]], i32 0, i32 1
-// CHECK-PPC64LE-NEXT: [[TMP34:%.*]] = load ptr, ptr [[PS1]], align 8
-// CHECK-PPC64LE-NEXT: [[PS9:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP34]], i32 0, i32 1
-// CHECK-PPC64LE-NEXT: [[TMP35:%.*]] = load ptr, ptr [[PS9]], align 8
-// CHECK-PPC64LE-NEXT: [[PS10:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP35]], i32 0, i32 1
-// CHECK-PPC64LE-NEXT: [[TMP36:%.*]] = load ptr, ptr [[PS1]], align 8
-// CHECK-PPC64LE-NEXT: [[PS11:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP36]], i32 0, i32 1
-// CHECK-PPC64LE-NEXT: [[TMP37:%.*]] = load ptr, ptr [[PS11]], align 8
-// CHECK-PPC64LE-NEXT: [[PS12:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP37]], i32 0, i32 1
-// CHECK-PPC64LE-NEXT: [[TMP38:%.*]] = load ptr, ptr [[PS12]], align 8
-// CHECK-PPC64LE-NEXT: [[PS13:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP38]], i32 0, i32 1
-// CHECK-PPC64LE-NEXT: [[TMP39:%.*]] = load ptr, ptr [[PS1]], align 8
-// CHECK-PPC64LE-NEXT: [[PS14:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP39]], i32 0, i32 1
-// CHECK-PPC64LE-NEXT: [[TMP40:%.*]] = load ptr, ptr [[PS14]], align 8
-// CHECK-PPC64LE-NEXT: [[PS15:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP40]], i32 0, i32 1
-// CHECK-PPC64LE-NEXT: [[TMP41:%.*]] = load ptr, ptr [[PS15]], align 8
-// CHECK-PPC64LE-NEXT: [[PS16:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP41]], i32 0, i32 1
-// CHECK-PPC64LE-NEXT: [[TMP42:%.*]] = load ptr, ptr [[PS16]], align 8
-// CHECK-PPC64LE-NEXT: [[S17:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP42]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: [[TMP43:%.*]] = getelementptr ptr, ptr [[PS]], i32 1
-// CHECK-PPC64LE-NEXT: [[TMP46:%.*]] = ptrtoint ptr [[TMP43]] to i64
-// CHECK-PPC64LE-NEXT: [[TMP47:%.*]] = ptrtoint ptr [[S]] to i64
-// CHECK-PPC64LE-NEXT: [[TMP48:%.*]] = sub i64 [[TMP46]], [[TMP47]]
-// CHECK-PPC64LE-NEXT: [[TMP49:%.*]] = sdiv exact i64 [[TMP48]], ptrtoint (ptr getelementptr (i8, ptr null, i32 1) to i64)
-// CHECK-PPC64LE-NEXT: [[TMP50:%.*]] = load ptr, ptr [[PS2]], align 8
+// CHECK-PPC64LE-NEXT: [[PS14:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP33]], i32 0, i32 1
+// CHECK-PPC64LE-NEXT: [[TMP34:%.*]] = load ptr, ptr [[PS14]], align 8
+// CHECK-PPC64LE-NEXT: [[PS15:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP34]], i32 0, i32 1
+// CHECK-PPC64LE-NEXT: [[TMP35:%.*]] = load ptr, ptr [[PS15]], align 8
+// CHECK-PPC64LE-NEXT: [[PS16:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP35]], i32 0, i32 1
+// CHECK-PPC64LE-NEXT: [[TMP36:%.*]] = load ptr, ptr [[PS16]], align 8
+// CHECK-PPC64LE-NEXT: [[S17:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP36]], i32 0, i32 0
+// CHECK-PPC64LE-NEXT: [[TMP37:%.*]] = getelementptr ptr, ptr [[PS]], i32 1
+// CHECK-PPC64LE-NEXT: [[TMP38:%.*]] = ptrtoint ptr [[TMP37]] to i64
+// CHECK-PPC64LE-NEXT: [[TMP39:%.*]] = ptrtoint ptr [[S]] to i64
+// CHECK-PPC64LE-NEXT: [[TMP40:%.*]] = sub i64 [[TMP38]], [[TMP39]]
+// CHECK-PPC64LE-NEXT: [[TMP41:%.*]] = sdiv exact i64 [[TMP40]], ptrtoint (ptr getelementptr (i8, ptr null, i32 1) to i64)
+// CHECK-PPC64LE-NEXT: [[TMP42:%.*]] = load ptr, ptr [[PS2]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP43:%.*]] = load ptr, ptr [[PS2]], align 8
+// CHECK-PPC64LE-NEXT: [[S18:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP43]], i32 0, i32 0
+// CHECK-PPC64LE-NEXT: [[TMP44:%.*]] = load ptr, ptr [[PS2]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP45:%.*]] = load ptr, ptr [[PS2]], align 8
+// CHECK-PPC64LE-NEXT: [[PS19:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP45]], i32 0, i32 1
+// CHECK-PPC64LE-NEXT: [[TMP46:%.*]] = load ptr, ptr [[PS2]], align 8
+// CHECK-PPC64LE-NEXT: [[PS20:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP46]], i32 0, i32 1
+// CHECK-PPC64LE-NEXT: [[TMP47:%.*]] = load ptr, ptr [[PS20]], align 8
+// CHECK-PPC64LE-NEXT: [[PS21:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP47]], i32 0, i32 1
+// CHECK-PPC64LE-NEXT: [[TMP48:%.*]] = load ptr, ptr [[PS2]], align 8
+// CHECK-PPC64LE-NEXT: [[PS22:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP48]], i32 0, i32 1
+// CHECK-PPC64LE-NEXT: [[TMP49:%.*]] = load ptr, ptr [[PS22]], align 8
+// CHECK-PPC64LE-NEXT: [[PS23:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP49]], i32 0, i32 1
+// CHECK-PPC64LE-NEXT: [[TMP50:%.*]] = load ptr, ptr [[PS23]], align 8
+// CHECK-PPC64LE-NEXT: [[PS24:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP50]], i32 0, i32 1
// CHECK-PPC64LE-NEXT: [[TMP51:%.*]] = load ptr, ptr [[PS2]], align 8
-// CHECK-PPC64LE-NEXT: [[S18:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP51]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: [[TMP52:%.*]] = load ptr, ptr [[PS2]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP53:%.*]] = load ptr, ptr [[PS2]], align 8
-// CHECK-PPC64LE-NEXT: [[PS19:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP53]], i32 0, i32 1
-// CHECK-PPC64LE-NEXT: [[TMP54:%.*]] = load ptr, ptr [[PS2]], align 8
-// CHECK-PPC64LE-NEXT: [[PS20:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP54]], i32 0, i32 1
-// CHECK-PPC64LE-NEXT: [[TMP55:%.*]] = load ptr, ptr [[PS20]], align 8
-// CHECK-PPC64LE-NEXT: [[PS21:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP55]], i32 0, i32 1
-// CHECK-PPC64LE-NEXT: [[TMP56:%.*]] = load ptr, ptr [[PS2]], align 8
-// CHECK-PPC64LE-NEXT: [[PS22:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP56]], i32 0, i32 1
-// CHECK-PPC64LE-NEXT: [[TMP57:%.*]] = load ptr, ptr [[PS22]], align 8
-// CHECK-PPC64LE-NEXT: [[PS23:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP57]], i32 0, i32 1
-// CHECK-PPC64LE-NEXT: [[TMP58:%.*]] = load ptr, ptr [[PS23]], align 8
-// CHECK-PPC64LE-NEXT: [[PS24:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP58]], i32 0, i32 1
-// CHECK-PPC64LE-NEXT: [[TMP59:%.*]] = load ptr, ptr [[PS2]], align 8
-// CHECK-PPC64LE-NEXT: [[PS25:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP59]], i32 0, i32 1
-// CHECK-PPC64LE-NEXT: [[TMP60:%.*]] = load ptr, ptr [[PS25]], align 8
-// CHECK-PPC64LE-NEXT: [[PS26:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP60]], i32 0, i32 1
-// CHECK-PPC64LE-NEXT: [[TMP61:%.*]] = load ptr, ptr [[PS26]], align 8
-// CHECK-PPC64LE-NEXT: [[PS27:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP61]], i32 0, i32 1
-// CHECK-PPC64LE-NEXT: [[TMP62:%.*]] = load ptr, ptr [[PS27]], align 8
-// CHECK-PPC64LE-NEXT: [[S28:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP62]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: [[TMP63:%.*]] = getelementptr ptr, ptr [[PS19]], i32 1
-// CHECK-PPC64LE-NEXT: [[TMP66:%.*]] = ptrtoint ptr [[TMP63]] to i64
-// CHECK-PPC64LE-NEXT: [[TMP67:%.*]] = ptrtoint ptr [[S18]] to i64
-// CHECK-PPC64LE-NEXT: [[TMP68:%.*]] = sub i64 [[TMP66]], [[TMP67]]
-// CHECK-PPC64LE-NEXT: [[TMP69:%.*]] = sdiv exact i64 [[TMP68]], ptrtoint (ptr getelementptr (i8, ptr null, i32 1) to i64)
+// CHECK-PPC64LE-NEXT: [[PS25:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP51]], i32 0, i32 1
+// CHECK-PPC64LE-NEXT: [[TMP52:%.*]] = load ptr, ptr [[PS25]], align 8
+// CHECK-PPC64LE-NEXT: [[PS26:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP52]], i32 0, i32 1
+// CHECK-PPC64LE-NEXT: [[TMP53:%.*]] = load ptr, ptr [[PS26]], align 8
+// CHECK-PPC64LE-NEXT: [[PS27:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP53]], i32 0, i32 1
+// CHECK-PPC64LE-NEXT: [[TMP54:%.*]] = load ptr, ptr [[PS27]], align 8
+// CHECK-PPC64LE-NEXT: [[S28:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP54]], i32 0, i32 0
+// CHECK-PPC64LE-NEXT: [[TMP55:%.*]] = getelementptr ptr, ptr [[PS19]], i32 1
+// CHECK-PPC64LE-NEXT: [[TMP56:%.*]] = ptrtoint ptr [[TMP55]] to i64
+// CHECK-PPC64LE-NEXT: [[TMP57:%.*]] = ptrtoint ptr [[S18]] to i64
+// CHECK-PPC64LE-NEXT: [[TMP58:%.*]] = sub i64 [[TMP56]], [[TMP57]]
+// CHECK-PPC64LE-NEXT: [[TMP59:%.*]] = sdiv exact i64 [[TMP58]], ptrtoint (ptr getelementptr (i8, ptr null, i32 1) to i64)
// CHECK-PPC64LE-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES]], ptr align 8 @.offload_sizes.5, i64 88, i1 false)
-// CHECK-PPC64LE-NEXT: [[TMP71:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: store ptr [[TMP30]], ptr [[TMP71]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP73:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: store ptr [[S]], ptr [[TMP73]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP75:%.*]] = getelementptr inbounds [11 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: store i64 [[TMP49]], ptr [[TMP75]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP76:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i64 0, i64 0
-// CHECK-PPC64LE-NEXT: store ptr null, ptr [[TMP76]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP77:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 1
-// CHECK-PPC64LE-NEXT: store ptr [[TMP30]], ptr [[TMP77]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP79:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 1
-// CHECK-PPC64LE-NEXT: store ptr [[S]], ptr [[TMP79]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP81:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i64 0, i64 1
-// CHECK-PPC64LE-NEXT: store ptr null, ptr [[TMP81]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP82:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 2
-// CHECK-PPC64LE-NEXT: store ptr [[PS]], ptr [[TMP82]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP84:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 2
-// CHECK-PPC64LE-NEXT: store ptr [[PS10]], ptr [[TMP84]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP86:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i64 0, i64 2
-// CHECK-PPC64LE-NEXT: store ptr null, ptr [[TMP86]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP87:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 3
-// CHECK-PPC64LE-NEXT: store ptr [[PS10]], ptr [[TMP87]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP89:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 3
-// CHECK-PPC64LE-NEXT: store ptr [[PS13]], ptr [[TMP89]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP91:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i64 0, i64 3
+// CHECK-PPC64LE-NEXT: [[TMP60:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 0
+// CHECK-PPC64LE-NEXT: store ptr [[TMP24]], ptr [[TMP60]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP61:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 0
+// CHECK-PPC64LE-NEXT: store ptr [[S]], ptr [[TMP61]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP62:%.*]] = getelementptr inbounds [11 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK-PPC64LE-NEXT: store i64 [[TMP41]], ptr [[TMP62]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP63:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i64 0, i64 0
+// CHECK-PPC64LE-NEXT: store ptr null, ptr [[TMP63]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP64:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 1
+// CHECK-PPC64LE-NEXT: store ptr [[TMP24]], ptr [[TMP64]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP65:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 1
+// CHECK-PPC64LE-NEXT: store ptr [[S]], ptr [[TMP65]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP66:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i64 0, i64 1
+// CHECK-PPC64LE-NEXT: store ptr null, ptr [[TMP66]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP67:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 2
+// CHECK-PPC64LE-NEXT: store ptr [[PS]], ptr [[TMP67]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP68:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 2
+// CHECK-PPC64LE-NEXT: store ptr [[PS10]], ptr [[TMP68]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP69:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i64 0, i64 2
+// CHECK-PPC64LE-NEXT: store ptr null, ptr [[TMP69]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP70:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 3
+// CHECK-PPC64LE-NEXT: store ptr [[PS10]], ptr [[TMP70]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP71:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 3
+// CHECK-PPC64LE-NEXT: store ptr [[PS13]], ptr [[TMP71]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP72:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i64 0, i64 3
+// CHECK-PPC64LE-NEXT: store ptr null, ptr [[TMP72]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP73:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 4
+// CHECK-PPC64LE-NEXT: store ptr [[PS13]], ptr [[TMP73]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP74:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 4
+// CHECK-PPC64LE-NEXT: store ptr [[S17]], ptr [[TMP74]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP75:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i64 0, i64 4
+// CHECK-PPC64LE-NEXT: store ptr null, ptr [[TMP75]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP76:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 5
+// CHECK-PPC64LE-NEXT: store ptr [[ARG_ADDR]], ptr [[TMP76]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP77:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 5
+// CHECK-PPC64LE-NEXT: store ptr [[ARG_ADDR]], ptr [[TMP77]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP78:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i64 0, i64 5
+// CHECK-PPC64LE-NEXT: store ptr null, ptr [[TMP78]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP79:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 6
+// CHECK-PPC64LE-NEXT: store ptr [[TMP42]], ptr [[TMP79]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP80:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 6
+// CHECK-PPC64LE-NEXT: store ptr [[S18]], ptr [[TMP80]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP81:%.*]] = getelementptr inbounds [11 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 6
+// CHECK-PPC64LE-NEXT: store i64 [[TMP59]], ptr [[TMP81]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP82:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i64 0, i64 6
+// CHECK-PPC64LE-NEXT: store ptr null, ptr [[TMP82]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP83:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 7
+// CHECK-PPC64LE-NEXT: store ptr [[TMP42]], ptr [[TMP83]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP84:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 7
+// CHECK-PPC64LE-NEXT: store ptr [[S18]], ptr [[TMP84]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP85:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i64 0, i64 7
+// CHECK-PPC64LE-NEXT: store ptr null, ptr [[TMP85]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP86:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 8
+// CHECK-PPC64LE-NEXT: store ptr [[PS19]], ptr [[TMP86]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP87:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 8
+// CHECK-PPC64LE-NEXT: store ptr [[PS21]], ptr [[TMP87]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP88:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i64 0, i64 8
+// CHECK-PPC64LE-NEXT: store ptr null, ptr [[TMP88]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP89:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 9
+// CHECK-PPC64LE-NEXT: store ptr [[PS21]], ptr [[TMP89]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP90:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 9
+// CHECK-PPC64LE-NEXT: store ptr [[PS24]], ptr [[TMP90]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP91:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i64 0, i64 9
// CHECK-PPC64LE-NEXT: store ptr null, ptr [[TMP91]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP92:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 4
-// CHECK-PPC64LE-NEXT: store ptr [[PS13]], ptr [[TMP92]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP94:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 4
-// CHECK-PPC64LE-NEXT: store ptr [[S17]], ptr [[TMP94]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP96:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i64 0, i64 4
-// CHECK-PPC64LE-NEXT: store ptr null, ptr [[TMP96]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP97:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 5
-// CHECK-PPC64LE-NEXT: store ptr [[ARG_ADDR]], ptr [[TMP97]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP99:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 5
-// CHECK-PPC64LE-NEXT: store ptr [[ARG_ADDR]], ptr [[TMP99]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP101:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i64 0, i64 5
-// CHECK-PPC64LE-NEXT: store ptr null, ptr [[TMP101]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP102:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 6
-// CHECK-PPC64LE-NEXT: store ptr [[TMP50]], ptr [[TMP102]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP104:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 6
-// CHECK-PPC64LE-NEXT: store ptr [[S18]], ptr [[TMP104]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP106:%.*]] = getelementptr inbounds [11 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 6
-// CHECK-PPC64LE-NEXT: store i64 [[TMP69]], ptr [[TMP106]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP107:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i64 0, i64 6
-// CHECK-PPC64LE-NEXT: store ptr null, ptr [[TMP107]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP108:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 7
-// CHECK-PPC64LE-NEXT: store ptr [[TMP50]], ptr [[TMP108]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP110:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 7
-// CHECK-PPC64LE-NEXT: store ptr [[S18]], ptr [[TMP110]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP112:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i64 0, i64 7
-// CHECK-PPC64LE-NEXT: store ptr null, ptr [[TMP112]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP113:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 8
-// CHECK-PPC64LE-NEXT: store ptr [[PS19]], ptr [[TMP113]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP115:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 8
-// CHECK-PPC64LE-NEXT: store ptr [[PS21]], ptr [[TMP115]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP117:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i64 0, i64 8
-// CHECK-PPC64LE-NEXT: store ptr null, ptr [[TMP117]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP118:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 9
-// CHECK-PPC64LE-NEXT: store ptr [[PS21]], ptr [[TMP118]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP120:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 9
-// CHECK-PPC64LE-NEXT: store ptr [[PS24]], ptr [[TMP120]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP122:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i64 0, i64 9
-// CHECK-PPC64LE-NEXT: store ptr null, ptr [[TMP122]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP123:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 10
-// CHECK-PPC64LE-NEXT: store ptr [[PS24]], ptr [[TMP123]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP125:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 10
-// CHECK-PPC64LE-NEXT: store ptr [[S28]], ptr [[TMP125]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP127:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i64 0, i64 10
-// CHECK-PPC64LE-NEXT: store ptr null, ptr [[TMP127]], align 8
-// CHECK-PPC64LE-NEXT: [[TMP128:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: [[TMP129:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: [[TMP130:%.*]] = getelementptr inbounds [11 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: call void @__tgt_target_data_begin_mapper(ptr @[[GLOB1]], i64 -1, i32 11, ptr [[TMP128]], ptr [[TMP129]], ptr [[TMP130]], ptr @.offload_maptypes.6, ptr null, ptr null)
-// CHECK-PPC64LE-NEXT: [[TMP131:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
-// CHECK-PPC64LE-NEXT: [[INC32:%.*]] = add nsw i32 [[TMP131]], 1
+// CHECK-PPC64LE-NEXT: [[TMP92:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 10
+// CHECK-PPC64LE-NEXT: store ptr [[PS24]], ptr [[TMP92]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP93:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 10
+// CHECK-PPC64LE-NEXT: store ptr [[S28]], ptr [[TMP93]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP94:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i64 0, i64 10
+// CHECK-PPC64LE-NEXT: store ptr null, ptr [[TMP94]], align 8
+// CHECK-PPC64LE-NEXT: [[TMP95:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 0
+// CHECK-PPC64LE-NEXT: [[TMP96:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 0
+// CHECK-PPC64LE-NEXT: [[TMP97:%.*]] = getelementptr inbounds [11 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK-PPC64LE-NEXT: call void @__tgt_target_data_begin_mapper(ptr @[[GLOB1]], i64 -1, i32 11, ptr [[TMP95]], ptr [[TMP96]], ptr [[TMP97]], ptr @.offload_maptypes.6, ptr null, ptr null)
+// CHECK-PPC64LE-NEXT: [[TMP98:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
+// CHECK-PPC64LE-NEXT: [[INC32:%.*]] = add nsw i32 [[TMP98]], 1
// CHECK-PPC64LE-NEXT: store i32 [[INC32]], ptr [[ARG_ADDR]], align 4
-// CHECK-PPC64LE-NEXT: [[TMP132:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: [[TMP133:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: [[TMP134:%.*]] = getelementptr inbounds [11 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
-// CHECK-PPC64LE-NEXT: call void @__tgt_target_data_end_mapper(ptr @[[GLOB1]], i64 -1, i32 11, ptr [[TMP132]], ptr [[TMP133]], ptr [[TMP134]], ptr @.offload_maptypes.6, ptr null, ptr null)
+// CHECK-PPC64LE-NEXT: [[TMP99:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 0
+// CHECK-PPC64LE-NEXT: [[TMP100:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 0
+// CHECK-PPC64LE-NEXT: [[TMP101:%.*]] = getelementptr inbounds [11 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK-PPC64LE-NEXT: call void @__tgt_target_data_end_mapper(ptr @[[GLOB1]], i64 -1, i32 11, ptr [[TMP99]], ptr [[TMP100]], ptr [[TMP101]], ptr @.offload_maptypes.6, ptr null, ptr null)
// CHECK-PPC64LE-NEXT: ret void
//
// CHECK-I386-LABEL: @_Z3fooi(
@@ -300,189 +304,189 @@ struct S2 {
// CHECK-I386-NEXT: store i32 [[ARG:%.*]], ptr [[ARG_ADDR]], align 4
// CHECK-I386-NEXT: [[TMP0:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK-I386-NEXT: store ptr [[LB]], ptr [[TMP0]], align 4
-// CHECK-I386-NEXT: [[TMP2:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK-I386-NEXT: store ptr [[LB]], ptr [[TMP2]], align 4
-// CHECK-I386-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
-// CHECK-I386-NEXT: store ptr null, ptr [[TMP4]], align 4
-// CHECK-I386-NEXT: [[TMP5:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK-I386-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK-I386-NEXT: call void @__tgt_target_data_begin_mapper(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 1, ptr [[TMP5]], ptr [[TMP6]], ptr @.offload_sizes, ptr @.offload_maptypes, ptr null, ptr null)
-// CHECK-I386-NEXT: [[TMP7:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
-// CHECK-I386-NEXT: [[INC:%.*]] = add nsw i32 [[TMP7]], 1
+// CHECK-I386-NEXT: [[TMP1:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK-I386-NEXT: store ptr [[LB]], ptr [[TMP1]], align 4
+// CHECK-I386-NEXT: [[TMP2:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
+// CHECK-I386-NEXT: store ptr null, ptr [[TMP2]], align 4
+// CHECK-I386-NEXT: [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK-I386-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK-I386-NEXT: call void @__tgt_target_data_begin_mapper(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 1, ptr [[TMP3]], ptr [[TMP4]], ptr @.offload_sizes, ptr @.offload_maptypes, ptr null, ptr null)
+// CHECK-I386-NEXT: [[TMP5:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
+// CHECK-I386-NEXT: [[INC:%.*]] = add nsw i32 [[TMP5]], 1
// CHECK-I386-NEXT: store i32 [[INC]], ptr [[ARG_ADDR]], align 4
-// CHECK-I386-NEXT: [[TMP8:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK-I386-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK-I386-NEXT: call void @__tgt_target_data_end_mapper(ptr @[[GLOB1]], i64 -1, i32 1, ptr [[TMP8]], ptr [[TMP9]], ptr @.offload_sizes, ptr @.offload_maptypes, ptr null, ptr null)
-// CHECK-I386-NEXT: [[TMP10:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK-I386-NEXT: store ptr [[LB]], ptr [[TMP10]], align 4
+// CHECK-I386-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK-I386-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK-I386-NEXT: call void @__tgt_target_data_end_mapper(ptr @[[GLOB1]], i64 -1, i32 1, ptr [[TMP6]], ptr [[TMP7]], ptr @.offload_sizes, ptr @.offload_maptypes, ptr null, ptr null)
+// CHECK-I386-NEXT: [[TMP8:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK-I386-NEXT: store ptr [[LB]], ptr [[TMP8]], align 4
+// CHECK-I386-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK-I386-NEXT: store ptr [[LB]], ptr [[TMP9]], align 4
+// CHECK-I386-NEXT: [[TMP10:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
+// CHECK-I386-NEXT: store ptr null, ptr [[TMP10]], align 4
+// CHECK-I386-NEXT: [[TMP11:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
// CHECK-I386-NEXT: [[TMP12:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK-I386-NEXT: store ptr [[LB]], ptr [[TMP12]], align 4
-// CHECK-I386-NEXT: [[TMP14:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
-// CHECK-I386-NEXT: store ptr null, ptr [[TMP14]], align 4
-// CHECK-I386-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK-I386-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK-I386-NEXT: call void @__tgt_target_data_begin_mapper(ptr @[[GLOB1]], i64 -1, i32 1, ptr [[TMP15]], ptr [[TMP16]], ptr @.offload_sizes.1, ptr @.offload_maptypes.2, ptr null, ptr null)
-// CHECK-I386-NEXT: [[TMP17:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
-// CHECK-I386-NEXT: [[INC4:%.*]] = add nsw i32 [[TMP17]], 1
+// CHECK-I386-NEXT: call void @__tgt_target_data_begin_mapper(ptr @[[GLOB1]], i64 -1, i32 1, ptr [[TMP11]], ptr [[TMP12]], ptr @.offload_sizes.1, ptr @.offload_maptypes.2, ptr null, ptr null)
+// CHECK-I386-NEXT: [[TMP13:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
+// CHECK-I386-NEXT: [[INC4:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK-I386-NEXT: store i32 [[INC4]], ptr [[ARG_ADDR]], align 4
-// CHECK-I386-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK-I386-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK-I386-NEXT: call void @__tgt_target_data_end_mapper(ptr @[[GLOB1]], i64 -1, i32 1, ptr [[TMP18]], ptr [[TMP19]], ptr @.offload_sizes.1, ptr @.offload_maptypes.2, ptr null, ptr null)
-// CHECK-I386-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK-I386-NEXT: store ptr [[ARG_ADDR]], ptr [[TMP20]], align 4
-// CHECK-I386-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK-I386-NEXT: store ptr [[ARG_ADDR]], ptr [[TMP22]], align 4
-// CHECK-I386-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 0
-// CHECK-I386-NEXT: store ptr null, ptr [[TMP24]], align 4
-// CHECK-I386-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK-I386-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK-I386-NEXT: call void @__tgt_target_data_begin_mapper(ptr @[[GLOB1]], i64 -1, i32 1, ptr [[TMP25]], ptr [[TMP26]], ptr @.offload_sizes.3, ptr @.offload_maptypes.4, ptr null, ptr null)
-// CHECK-I386-NEXT: [[TMP27:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
-// CHECK-I386-NEXT: [[INC8:%.*]] = add nsw i32 [[TMP27]], 1
+// CHECK-I386-NEXT: [[TMP14:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK-I386-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK-I386-NEXT: call void @__tgt_target_data_end_mapper(ptr @[[GLOB1]], i64 -1, i32 1, ptr [[TMP14]], ptr [[TMP15]], ptr @.offload_sizes.1, ptr @.offload_maptypes.2, ptr null, ptr null)
+// CHECK-I386-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK-I386-NEXT: store ptr [[ARG_ADDR]], ptr [[TMP16]], align 4
+// CHECK-I386-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK-I386-NEXT: store ptr [[ARG_ADDR]], ptr [[TMP17]], align 4
+// CHECK-I386-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 0
+// CHECK-I386-NEXT: store ptr null, ptr [[TMP18]], align 4
+// CHECK-I386-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK-I386-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK-I386-NEXT: call void @__tgt_target_data_begin_mapper(ptr @[[GLOB1]], i64 -1, i32 1, ptr [[TMP19]], ptr [[TMP20]], ptr @.offload_sizes.3, ptr @.offload_maptypes.4, ptr null, ptr null)
+// CHECK-I386-NEXT: [[TMP21:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
+// CHECK-I386-NEXT: [[INC8:%.*]] = add nsw i32 [[TMP21]], 1
// CHECK-I386-NEXT: store i32 [[INC8]], ptr [[ARG_ADDR]], align 4
-// CHECK-I386-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK-I386-NEXT: [[TMP29:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK-I386-NEXT: call void @__tgt_target_data_end_mapper(ptr @[[GLOB1]], i64 -1, i32 1, ptr [[TMP28]], ptr [[TMP29]], ptr @.offload_sizes.3, ptr @.offload_maptypes.4, ptr null, ptr null)
+// CHECK-I386-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK-I386-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK-I386-NEXT: call void @__tgt_target_data_end_mapper(ptr @[[GLOB1]], i64 -1, i32 1, ptr [[TMP22]], ptr [[TMP23]], ptr @.offload_sizes.3, ptr @.offload_maptypes.4, ptr null, ptr null)
+// CHECK-I386-NEXT: [[TMP24:%.*]] = load ptr, ptr [[PS1]], align 4
+// CHECK-I386-NEXT: [[TMP25:%.*]] = load ptr, ptr [[PS1]], align 4
+// CHECK-I386-NEXT: [[S:%.*]] = getelementptr inbounds [[STRUCT_S2:%.*]], ptr [[TMP25]], i32 0, i32 0
+// CHECK-I386-NEXT: [[TMP26:%.*]] = load ptr, ptr [[PS1]], align 4
+// CHECK-I386-NEXT: [[TMP27:%.*]] = load ptr, ptr [[PS1]], align 4
+// CHECK-I386-NEXT: [[PS:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP27]], i32 0, i32 1
+// CHECK-I386-NEXT: [[TMP28:%.*]] = load ptr, ptr [[PS1]], align 4
+// CHECK-I386-NEXT: [[PS9:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP28]], i32 0, i32 1
+// CHECK-I386-NEXT: [[TMP29:%.*]] = load ptr, ptr [[PS9]], align 4
+// CHECK-I386-NEXT: [[PS10:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP29]], i32 0, i32 1
// CHECK-I386-NEXT: [[TMP30:%.*]] = load ptr, ptr [[PS1]], align 4
-// CHECK-I386-NEXT: [[TMP31:%.*]] = load ptr, ptr [[PS1]], align 4
-// CHECK-I386-NEXT: [[S:%.*]] = getelementptr inbounds [[STRUCT_S2:%.*]], ptr [[TMP31]], i32 0, i32 0
-// CHECK-I386-NEXT: [[TMP32:%.*]] = load ptr, ptr [[PS1]], align 4
+// CHECK-I386-NEXT: [[PS11:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP30]], i32 0, i32 1
+// CHECK-I386-NEXT: [[TMP31:%.*]] = load ptr, ptr [[PS11]], align 4
+// CHECK-I386-NEXT: [[PS12:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP31]], i32 0, i32 1
+// CHECK-I386-NEXT: [[TMP32:%.*]] = load ptr, ptr [[PS12]], align 4
+// CHECK-I386-NEXT: [[PS13:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP32]], i32 0, i32 1
// CHECK-I386-NEXT: [[TMP33:%.*]] = load ptr, ptr [[PS1]], align 4
-// CHECK-I386-NEXT: [[PS:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP33]], i32 0, i32 1
-// CHECK-I386-NEXT: [[TMP34:%.*]] = load ptr, ptr [[PS1]], align 4
-// CHECK-I386-NEXT: [[PS9:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP34]], i32 0, i32 1
-// CHECK-I386-NEXT: [[TMP35:%.*]] = load ptr, ptr [[PS9]], align 4
-// CHECK-I386-NEXT: [[PS10:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP35]], i32 0, i32 1
-// CHECK-I386-NEXT: [[TMP36:%.*]] = load ptr, ptr [[PS1]], align 4
-// CHECK-I386-NEXT: [[PS11:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP36]], i32 0, i32 1
-// CHECK-I386-NEXT: [[TMP37:%.*]] = load ptr, ptr [[PS11]], align 4
-// CHECK-I386-NEXT: [[PS12:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP37]], i32 0, i32 1
-// CHECK-I386-NEXT: [[TMP38:%.*]] = load ptr, ptr [[PS12]], align 4
-// CHECK-I386-NEXT: [[PS13:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP38]], i32 0, i32 1
-// CHECK-I386-NEXT: [[TMP39:%.*]] = load ptr, ptr [[PS1]], align 4
-// CHECK-I386-NEXT: [[PS14:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP39]], i32 0, i32 1
-// CHECK-I386-NEXT: [[TMP40:%.*]] = load ptr, ptr [[PS14]], align 4
-// CHECK-I386-NEXT: [[PS15:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP40]], i32 0, i32 1
-// CHECK-I386-NEXT: [[TMP41:%.*]] = load ptr, ptr [[PS15]], align 4
-// CHECK-I386-NEXT: [[PS16:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP41]], i32 0, i32 1
-// CHECK-I386-NEXT: [[TMP42:%.*]] = load ptr, ptr [[PS16]], align 4
-// CHECK-I386-NEXT: [[S17:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP42]], i32 0, i32 0
-// CHECK-I386-NEXT: [[TMP43:%.*]] = getelementptr ptr, ptr [[PS]], i32 1
-// CHECK-I386-NEXT: [[TMP46:%.*]] = ptrtoint ptr [[TMP43]] to i64
-// CHECK-I386-NEXT: [[TMP47:%.*]] = ptrtoint ptr [[S]] to i64
-// CHECK-I386-NEXT: [[TMP48:%.*]] = sub i64 [[TMP46]], [[TMP47]]
-// CHECK-I386-NEXT: [[TMP49:%.*]] = sdiv exact i64 [[TMP48]], ptrtoint (ptr getelementptr (i8, ptr null, i32 1) to i64)
-// CHECK-I386-NEXT: [[TMP50:%.*]] = load ptr, ptr [[PS2]], align 4
+// CHECK-I386-NEXT: [[PS14:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP33]], i32 0, i32 1
+// CHECK-I386-NEXT: [[TMP34:%.*]] = load ptr, ptr [[PS14]], align 4
+// CHECK-I386-NEXT: [[PS15:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP34]], i32 0, i32 1
+// CHECK-I386-NEXT: [[TMP35:%.*]] = load ptr, ptr [[PS15]], align 4
+// CHECK-I386-NEXT: [[PS16:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP35]], i32 0, i32 1
+// CHECK-I386-NEXT: [[TMP36:%.*]] = load ptr, ptr [[PS16]], align 4
+// CHECK-I386-NEXT: [[S17:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP36]], i32 0, i32 0
+// CHECK-I386-NEXT: [[TMP37:%.*]] = getelementptr ptr, ptr [[PS]], i32 1
+// CHECK-I386-NEXT: [[TMP38:%.*]] = ptrtoint ptr [[TMP37]] to i64
+// CHECK-I386-NEXT: [[TMP39:%.*]] = ptrtoint ptr [[S]] to i64
+// CHECK-I386-NEXT: [[TMP40:%.*]] = sub i64 [[TMP38]], [[TMP39]]
+// CHECK-I386-NEXT: [[TMP41:%.*]] = sdiv exact i64 [[TMP40]], ptrtoint (ptr getelementptr (i8, ptr null, i32 1) to i64)
+// CHECK-I386-NEXT: [[TMP42:%.*]] = load ptr, ptr [[PS2]], align 4
+// CHECK-I386-NEXT: [[TMP43:%.*]] = load ptr, ptr [[PS2]], align 4
+// CHECK-I386-NEXT: [[S18:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP43]], i32 0, i32 0
+// CHECK-I386-NEXT: [[TMP44:%.*]] = load ptr, ptr [[PS2]], align 4
+// CHECK-I386-NEXT: [[TMP45:%.*]] = load ptr, ptr [[PS2]], align 4
+// CHECK-I386-NEXT: [[PS19:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP45]], i32 0, i32 1
+// CHECK-I386-NEXT: [[TMP46:%.*]] = load ptr, ptr [[PS2]], align 4
+// CHECK-I386-NEXT: [[PS20:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP46]], i32 0, i32 1
+// CHECK-I386-NEXT: [[TMP47:%.*]] = load ptr, ptr [[PS20]], align 4
+// CHECK-I386-NEXT: [[PS21:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP47]], i32 0, i32 1
+// CHECK-I386-NEXT: [[TMP48:%.*]] = load ptr, ptr [[PS2]], align 4
+// CHECK-I386-NEXT: [[PS22:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP48]], i32 0, i32 1
+// CHECK-I386-NEXT: [[TMP49:%.*]] = load ptr, ptr [[PS22]], align 4
+// CHECK-I386-NEXT: [[PS23:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP49]], i32 0, i32 1
+// CHECK-I386-NEXT: [[TMP50:%.*]] = load ptr, ptr [[PS23]], align 4
+// CHECK-I386-NEXT: [[PS24:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP50]], i32 0, i32 1
// CHECK-I386-NEXT: [[TMP51:%.*]] = load ptr, ptr [[PS2]], align 4
-// CHECK-I386-NEXT: [[S18:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP51]], i32 0, i32 0
-// CHECK-I386-NEXT: [[TMP52:%.*]] = load ptr, ptr [[PS2]], align 4
-// CHECK-I386-NEXT: [[TMP53:%.*]] = load ptr, ptr [[PS2]], align 4
-// CHECK-I386-NEXT: [[PS19:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP53]], i32 0, i32 1
-// CHECK-I386-NEXT: [[TMP54:%.*]] = load ptr, ptr [[PS2]], align 4
-// CHECK-I386-NEXT: [[PS20:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP54]], i32 0, i32 1
-// CHECK-I386-NEXT: [[TMP55:%.*]] = load ptr, ptr [[PS20]], align 4
-// CHECK-I386-NEXT: [[PS21:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP55]], i32 0, i32 1
-// CHECK-I386-NEXT: [[TMP56:%.*]] = load ptr, ptr [[PS2]], align 4
-// CHECK-I386-NEXT: [[PS22:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP56]], i32 0, i32 1
-// CHECK-I386-NEXT: [[TMP57:%.*]] = load ptr, ptr [[PS22]], align 4
-// CHECK-I386-NEXT: [[PS23:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP57]], i32 0, i32 1
-// CHECK-I386-NEXT: [[TMP58:%.*]] = load ptr, ptr [[PS23]], align 4
-// CHECK-I386-NEXT: [[PS24:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP58]], i32 0, i32 1
-// CHECK-I386-NEXT: [[TMP59:%.*]] = load ptr, ptr [[PS2]], align 4
-// CHECK-I386-NEXT: [[PS25:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP59]], i32 0, i32 1
-// CHECK-I386-NEXT: [[TMP60:%.*]] = load ptr, ptr [[PS25]], align 4
-// CHECK-I386-NEXT: [[PS26:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP60]], i32 0, i32 1
-// CHECK-I386-NEXT: [[TMP61:%.*]] = load ptr, ptr [[PS26]], align 4
-// CHECK-I386-NEXT: [[PS27:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP61]], i32 0, i32 1
-// CHECK-I386-NEXT: [[TMP62:%.*]] = load ptr, ptr [[PS27]], align 4
-// CHECK-I386-NEXT: [[S28:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP62]], i32 0, i32 0
-// CHECK-I386-NEXT: [[TMP63:%.*]] = getelementptr ptr, ptr [[PS19]], i32 1
-// CHECK-I386-NEXT: [[TMP66:%.*]] = ptrtoint ptr [[TMP63]] to i64
-// CHECK-I386-NEXT: [[TMP67:%.*]] = ptrtoint ptr [[S18]] to i64
-// CHECK-I386-NEXT: [[TMP68:%.*]] = sub i64 [[TMP66]], [[TMP67]]
-// CHECK-I386-NEXT: [[TMP69:%.*]] = sdiv exact i64 [[TMP68]], ptrtoint (ptr getelementptr (i8, ptr null, i32 1) to i64)
+// CHECK-I386-NEXT: [[PS25:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP51]], i32 0, i32 1
+// CHECK-I386-NEXT: [[TMP52:%.*]] = load ptr, ptr [[PS25]], align 4
+// CHECK-I386-NEXT: [[PS26:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP52]], i32 0, i32 1
+// CHECK-I386-NEXT: [[TMP53:%.*]] = load ptr, ptr [[PS26]], align 4
+// CHECK-I386-NEXT: [[PS27:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP53]], i32 0, i32 1
+// CHECK-I386-NEXT: [[TMP54:%.*]] = load ptr, ptr [[PS27]], align 4
+// CHECK-I386-NEXT: [[S28:%.*]] = getelementptr inbounds [[STRUCT_S2]], ptr [[TMP54]], i32 0, i32 0
+// CHECK-I386-NEXT: [[TMP55:%.*]] = getelementptr ptr, ptr [[PS19]], i32 1
+// CHECK-I386-NEXT: [[TMP56:%.*]] = ptrtoint ptr [[TMP55]] to i64
+// CHECK-I386-NEXT: [[TMP57:%.*]] = ptrtoint ptr [[S18]] to i64
+// CHECK-I386-NEXT: [[TMP58:%.*]] = sub i64 [[TMP56]], [[TMP57]]
+// CHECK-I386-NEXT: [[TMP59:%.*]] = sdiv exact i64 [[TMP58]], ptrtoint (ptr getelementptr (i8, ptr null, i32 1) to i64)
// CHECK-I386-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES]], ptr align 4 @.offload_sizes.5, i32 88, i1 false)
-// CHECK-I386-NEXT: [[TMP71:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 0
-// CHECK-I386-NEXT: store ptr [[TMP30]], ptr [[TMP71]], align 4
-// CHECK-I386-NEXT: [[TMP73:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 0
-// CHECK-I386-NEXT: store ptr [[S]], ptr [[TMP73]], align 4
-// CHECK-I386-NEXT: [[TMP75:%.*]] = getelementptr inbounds [11 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
-// CHECK-I386-NEXT: store i64 [[TMP49]], ptr [[TMP75]], align 4
-// CHECK-I386-NEXT: [[TMP76:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i32 0, i32 0
-// CHECK-I386-NEXT: store ptr null, ptr [[TMP76]], align 4
-// CHECK-I386-NEXT: [[TMP77:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 1
-// CHECK-I386-NEXT: store ptr [[TMP30]], ptr [[TMP77]], align 4
-// CHECK-I386-NEXT: [[TMP79:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 1
-// CHECK-I386-NEXT: store ptr [[S]], ptr [[TMP79]], align 4
-// CHECK-I386-NEXT: [[TMP81:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i32 0, i32 1
-// CHECK-I386-NEXT: store ptr null, ptr [[TMP81]], align 4
-// CHECK-I386-NEXT: [[TMP82:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 2
-// CHECK-I386-NEXT: store ptr [[PS]], ptr [[TMP82]], align 4
-// CHECK-I386-NEXT: [[TMP84:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 2
-// CHECK-I386-NEXT: store ptr [[PS10]], ptr [[TMP84]], align 4
-// CHECK-I386-NEXT: [[TMP86:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i32 0, i32 2
-// CHECK-I386-NEXT: store ptr null, ptr [[TMP86]], align 4
-// CHECK-I386-NEXT: [[TMP87:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 3
-// CHECK-I386-NEXT: store ptr [[PS10]], ptr [[TMP87]], align 4
-// CHECK-I386-NEXT: [[TMP89:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 3
-// CHECK-I386-NEXT: store ptr [[PS13]], ptr [[TMP89]], align 4
-// CHECK-I386-NEXT: [[TMP91:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i32 0, i32 3
+// CHECK-I386-NEXT: [[TMP60:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 0
+// CHECK-I386-NEXT: store ptr [[TMP24]], ptr [[TMP60]], align 4
+// CHECK-I386-NEXT: [[TMP61:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 0
+// CHECK-I386-NEXT: store ptr [[S]], ptr [[TMP61]], align 4
+// CHECK-I386-NEXT: [[TMP62:%.*]] = getelementptr inbounds [11 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK-I386-NEXT: store i64 [[TMP41]], ptr [[TMP62]], align 4
+// CHECK-I386-NEXT: [[TMP63:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i32 0, i32 0
+// CHECK-I386-NEXT: store ptr null, ptr [[TMP63]], align 4
+// CHECK-I386-NEXT: [[TMP64:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 1
+// CHECK-I386-NEXT: store ptr [[TMP24]], ptr [[TMP64]], align 4
+// CHECK-I386-NEXT: [[TMP65:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 1
+// CHECK-I386-NEXT: store ptr [[S]], ptr [[TMP65]], align 4
+// CHECK-I386-NEXT: [[TMP66:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i32 0, i32 1
+// CHECK-I386-NEXT: store ptr null, ptr [[TMP66]], align 4
+// CHECK-I386-NEXT: [[TMP67:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 2
+// CHECK-I386-NEXT: store ptr [[PS]], ptr [[TMP67]], align 4
+// CHECK-I386-NEXT: [[TMP68:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 2
+// CHECK-I386-NEXT: store ptr [[PS10]], ptr [[TMP68]], align 4
+// CHECK-I386-NEXT: [[TMP69:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i32 0, i32 2
+// CHECK-I386-NEXT: store ptr null, ptr [[TMP69]], align 4
+// CHECK-I386-NEXT: [[TMP70:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 3
+// CHECK-I386-NEXT: store ptr [[PS10]], ptr [[TMP70]], align 4
+// CHECK-I386-NEXT: [[TMP71:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 3
+// CHECK-I386-NEXT: store ptr [[PS13]], ptr [[TMP71]], align 4
+// CHECK-I386-NEXT: [[TMP72:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i32 0, i32 3
+// CHECK-I386-NEXT: store ptr null, ptr [[TMP72]], align 4
+// CHECK-I386-NEXT: [[TMP73:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 4
+// CHECK-I386-NEXT: store ptr [[PS13]], ptr [[TMP73]], align 4
+// CHECK-I386-NEXT: [[TMP74:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 4
+// CHECK-I386-NEXT: store ptr [[S17]], ptr [[TMP74]], align 4
+// CHECK-I386-NEXT: [[TMP75:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i32 0, i32 4
+// CHECK-I386-NEXT: store ptr null, ptr [[TMP75]], align 4
+// CHECK-I386-NEXT: [[TMP76:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 5
+// CHECK-I386-NEXT: store ptr [[ARG_ADDR]], ptr [[TMP76]], align 4
+// CHECK-I386-NEXT: [[TMP77:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 5
+// CHECK-I386-NEXT: store ptr [[ARG_ADDR]], ptr [[TMP77]], align 4
+// CHECK-I386-NEXT: [[TMP78:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i32 0, i32 5
+// CHECK-I386-NEXT: store ptr null, ptr [[TMP78]], align 4
+// CHECK-I386-NEXT: [[TMP79:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 6
+// CHECK-I386-NEXT: store ptr [[TMP42]], ptr [[TMP79]], align 4
+// CHECK-I386-NEXT: [[TMP80:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 6
+// CHECK-I386-NEXT: store ptr [[S18]], ptr [[TMP80]], align 4
+// CHECK-I386-NEXT: [[TMP81:%.*]] = getelementptr inbounds [11 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 6
+// CHECK-I386-NEXT: store i64 [[TMP59]], ptr [[TMP81]], align 4
+// CHECK-I386-NEXT: [[TMP82:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i32 0, i32 6
+// CHECK-I386-NEXT: store ptr null, ptr [[TMP82]], align 4
+// CHECK-I386-NEXT: [[TMP83:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 7
+// CHECK-I386-NEXT: store ptr [[TMP42]], ptr [[TMP83]], align 4
+// CHECK-I386-NEXT: [[TMP84:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 7
+// CHECK-I386-NEXT: store ptr [[S18]], ptr [[TMP84]], align 4
+// CHECK-I386-NEXT: [[TMP85:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i32 0, i32 7
+// CHECK-I386-NEXT: store ptr null, ptr [[TMP85]], align 4
+// CHECK-I386-NEXT: [[TMP86:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 8
+// CHECK-I386-NEXT: store ptr [[PS19]], ptr [[TMP86]], align 4
+// CHECK-I386-NEXT: [[TMP87:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 8
+// CHECK-I386-NEXT: store ptr [[PS21]], ptr [[TMP87]], align 4
+// CHECK-I386-NEXT: [[TMP88:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i32 0, i32 8
+// CHECK-I386-NEXT: store ptr null, ptr [[TMP88]], align 4
+// CHECK-I386-NEXT: [[TMP89:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 9
+// CHECK-I386-NEXT: store ptr [[PS21]], ptr [[TMP89]], align 4
+// CHECK-I386-NEXT: [[TMP90:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 9
+// CHECK-I386-NEXT: store ptr [[PS24]], ptr [[TMP90]], align 4
+// CHECK-I386-NEXT: [[TMP91:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i32 0, i32 9
// CHECK-I386-NEXT: store ptr null, ptr [[TMP91]], align 4
-// CHECK-I386-NEXT: [[TMP92:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 4
-// CHECK-I386-NEXT: store ptr [[PS13]], ptr [[TMP92]], align 4
-// CHECK-I386-NEXT: [[TMP94:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 4
-// CHECK-I386-NEXT: store ptr [[S17]], ptr [[TMP94]], align 4
-// CHECK-I386-NEXT: [[TMP96:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i32 0, i32 4
-// CHECK-I386-NEXT: store ptr null, ptr [[TMP96]], align 4
-// CHECK-I386-NEXT: [[TMP97:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 5
-// CHECK-I386-NEXT: store ptr [[ARG_ADDR]], ptr [[TMP97]], align 4
-// CHECK-I386-NEXT: [[TMP99:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 5
-// CHECK-I386-NEXT: store ptr [[ARG_ADDR]], ptr [[TMP99]], align 4
-// CHECK-I386-NEXT: [[TMP101:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i32 0, i32 5
-// CHECK-I386-NEXT: store ptr null, ptr [[TMP101]], align 4
-// CHECK-I386-NEXT: [[TMP102:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 6
-// CHECK-I386-NEXT: store ptr [[TMP50]], ptr [[TMP102]], align 4
-// CHECK-I386-NEXT: [[TMP104:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 6
-// CHECK-I386-NEXT: store ptr [[S18]], ptr [[TMP104]], align 4
-// CHECK-I386-NEXT: [[TMP106:%.*]] = getelementptr inbounds [11 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 6
-// CHECK-I386-NEXT: store i64 [[TMP69]], ptr [[TMP106]], align 4
-// CHECK-I386-NEXT: [[TMP107:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i32 0, i32 6
-// CHECK-I386-NEXT: store ptr null, ptr [[TMP107]], align 4
-// CHECK-I386-NEXT: [[TMP108:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 7
-// CHECK-I386-NEXT: store ptr [[TMP50]], ptr [[TMP108]], align 4
-// CHECK-I386-NEXT: [[TMP110:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 7
-// CHECK-I386-NEXT: store ptr [[S18]], ptr [[TMP110]], align 4
-// CHECK-I386-NEXT: [[TMP112:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i32 0, i32 7
-// CHECK-I386-NEXT: store ptr null, ptr [[TMP112]], align 4
-// CHECK-I386-NEXT: [[TMP113:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 8
-// CHECK-I386-NEXT: store ptr [[PS19]], ptr [[TMP113]], align 4
-// CHECK-I386-NEXT: [[TMP115:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 8
-// CHECK-I386-NEXT: store ptr [[PS21]], ptr [[TMP115]], align 4
-// CHECK-I386-NEXT: [[TMP117:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i32 0, i32 8
-// CHECK-I386-NEXT: store ptr null, ptr [[TMP117]], align 4
-// CHECK-I386-NEXT: [[TMP118:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 9
-// CHECK-I386-NEXT: store ptr [[PS21]], ptr [[TMP118]], align 4
-// CHECK-I386-NEXT: [[TMP120:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 9
-// CHECK-I386-NEXT: store ptr [[PS24]], ptr [[TMP120]], align 4
-// CHECK-I386-NEXT: [[TMP122:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i32 0, i32 9
-// CHECK-I386-NEXT: store ptr null, ptr [[TMP122]], align 4
-// CHECK-I386-NEXT: [[TMP123:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 10
-// CHECK-I386-NEXT: store ptr [[PS24]], ptr [[TMP123]], align 4
-// CHECK-I386-NEXT: [[TMP125:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 10
-// CHECK-I386-NEXT: store ptr [[S28]], ptr [[TMP125]], align 4
-// CHECK-I386-NEXT: [[TMP127:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i32 0, i32 10
-// CHECK-I386-NEXT: store ptr null, ptr [[TMP127]], align 4
-// CHECK-I386-NEXT: [[TMP128:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 0
-// CHECK-I386-NEXT: [[TMP129:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 0
-// CHECK-I386-NEXT: [[TMP130:%.*]] = getelementptr inbounds [11 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
-// CHECK-I386-NEXT: call void @__tgt_target_data_begin_mapper(ptr @[[GLOB1]], i64 -1, i32 11, ptr [[TMP128]], ptr [[TMP129]], ptr [[TMP130]], ptr @.offload_maptypes.6, ptr null, ptr null)
-// CHECK-I386-NEXT: [[TMP131:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
-// CHECK-I386-NEXT: [[INC32:%.*]] = add nsw i32 [[TMP131]], 1
+// CHECK-I386-NEXT: [[TMP92:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 10
+// CHECK-I386-NEXT: store ptr [[PS24]], ptr [[TMP92]], align 4
+// CHECK-I386-NEXT: [[TMP93:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 10
+// CHECK-I386-NEXT: store ptr [[S28]], ptr [[TMP93]], align 4
+// CHECK-I386-NEXT: [[TMP94:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_MAPPERS31]], i32 0, i32 10
+// CHECK-I386-NEXT: store ptr null, ptr [[TMP94]], align 4
+// CHECK-I386-NEXT: [[TMP95:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 0
+// CHECK-I386-NEXT: [[TMP96:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 0
+// CHECK-I386-NEXT: [[TMP97:%.*]] = getelementptr inbounds [11 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK-I386-NEXT: call void @__tgt_target_data_begin_mapper(ptr @[[GLOB1]], i64 -1, i32 11, ptr [[TMP95]], ptr [[TMP96]], ptr [[TMP97]], ptr @.offload_maptypes.6, ptr null, ptr null)
+// CHECK-I386-NEXT: [[TMP98:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
+// CHECK-I386-NEXT: [[INC32:%.*]] = add nsw i32 [[TMP98]], 1
// CHECK-I386-NEXT: store i32 [[INC32]], ptr [[ARG_ADDR]], align 4
-// CHECK-I386-NEXT: [[TMP132:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 0
-// CHECK-I386-NEXT: [[TMP133:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 0
-// CHECK-I386-NEXT: [[TMP134:%.*]] = getelementptr inbounds [11 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
-// CHECK-I386-NEXT: call void @__tgt_target_data_end_mapper(ptr @[[GLOB1]], i64 -1, i32 11, ptr [[TMP132]], ptr [[TMP133]], ptr [[TMP134]], ptr @.offload_maptypes.6, ptr null, ptr null)
+// CHECK-I386-NEXT: [[TMP99:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_BASEPTRS29]], i32 0, i32 0
+// CHECK-I386-NEXT: [[TMP100:%.*]] = getelementptr inbounds [11 x ptr], ptr [[DOTOFFLOAD_PTRS30]], i32 0, i32 0
+// CHECK-I386-NEXT: [[TMP101:%.*]] = getelementptr inbounds [11 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK-I386-NEXT: call void @__tgt_target_data_end_mapper(ptr @[[GLOB1]], i64 -1, i32 11, ptr [[TMP99]], ptr [[TMP100]], ptr [[TMP101]], ptr @.offload_maptypes.6, ptr null, ptr null)
// CHECK-I386-NEXT: ret void
//
void foo(int arg) {
@@ -510,3 +514,21 @@ void foo(int arg) {
}
#endif
+//.
+// CHECK-PPC64LE: attributes #0 = { mustprogress noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="-altivec,-bpermd,-crbits,-crypto,-direct-move,-extdiv,-htm,-isa-v206-instructions,-isa-v207-instructions,-isa-v30-instructions,-power8-vector,-power9-vector,-privileged,-quadword-atomics,-rop-protect,-spe,-vsx" }
+// CHECK-PPC64LE: attributes #1 = { nounwind }
+// CHECK-PPC64LE: attributes #2 = { nocallback nofree nounwind willreturn memory(argmem: readwrite) }
+//.
+// CHECK-I386: attributes #0 = { mustprogress noinline nounwind optnone "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+cx8,+x87" }
+// CHECK-I386: attributes #1 = { nounwind }
+// CHECK-I386: attributes #2 = { nocallback nofree nounwind willreturn memory(argmem: readwrite) }
+//.
+// CHECK-PPC64LE: !0 = !{i32 1, !"wchar_size", i32 4}
+// CHECK-PPC64LE: !1 = !{i32 7, !"openmp", i32 50}
+// CHECK-PPC64LE: !2 = !{!"clang version 16.0.0"}
+//.
+// CHECK-I386: !0 = !{i32 1, !"NumRegisterParameters", i32 0}
+// CHECK-I386: !1 = !{i32 1, !"wchar_size", i32 4}
+// CHECK-I386: !2 = !{i32 7, !"openmp", i32 50}
+// CHECK-I386: !3 = !{!"clang version 16.0.0"}
+//.
diff --git a/clang/test/OpenMP/target_depend_codegen.cpp b/clang/test/OpenMP/target_depend_codegen.cpp
index 3b09a2ed13444..488f88a62ad97 100644
--- a/clang/test/OpenMP/target_depend_codegen.cpp
+++ b/clang/test/OpenMP/target_depend_codegen.cpp
@@ -259,7 +259,7 @@ int foo(int n) {
// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2
// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]],
// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64
-// CHECK: [[RET:%.+]] = call i32 @__tgt_target_kernel_nowait(%struct.ident_t* @{{.+}}, i64 [[DEVICE]], i32 -1, i32 0, i8* @.[[TGT_REGION:.+]].region_id, %struct.__tgt_kernel_arguments* %[[KERNEL_ARGS:.+]])
+// CHECK: [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 [[DEVICE]], i32 -1, i32 0, i8* @.[[TGT_REGION:.+]].region_id, %struct.__tgt_kernel_arguments* %[[KERNEL_ARGS:.+]])
// CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
// CHECK: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]]
diff --git a/clang/test/OpenMP/target_has_device_addr_codegen.cpp b/clang/test/OpenMP/target_has_device_addr_codegen.cpp
index eb95b84afe47f..c1111b31602e5 100644
--- a/clang/test/OpenMP/target_has_device_addr_codegen.cpp
+++ b/clang/test/OpenMP/target_has_device_addr_codegen.cpp
@@ -323,7 +323,7 @@ void use_template() {
// CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK-NEXT: store i32 1, ptr [[TMP6]], align 4
+// CHECK-NEXT: store i32 2, ptr [[TMP6]], align 4
// CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK-NEXT: store i32 1, ptr [[TMP7]], align 4
// CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -340,194 +340,242 @@ void use_template() {
// CHECK-NEXT: store ptr null, ptr [[TMP13]], align 8
// CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK-NEXT: store i64 0, ptr [[TMP14]], align 8
-// CHECK-NEXT: [[TMP15:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l145.region_id, ptr [[KERNEL_ARGS]])
-// CHECK-NEXT: [[TMP16:%.*]] = icmp ne i32 [[TMP15]], 0
-// CHECK-NEXT: br i1 [[TMP16]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK-NEXT: store i64 0, ptr [[TMP15]], align 8
+// CHECK-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP16]], align 4
+// CHECK-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK-NEXT: store i32 0, ptr [[TMP18]], align 4
+// CHECK-NEXT: [[TMP19:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l145.region_id, ptr [[KERNEL_ARGS]])
+// CHECK-NEXT: [[TMP20:%.*]] = icmp ne i32 [[TMP19]], 0
+// CHECK-NEXT: br i1 [[TMP20]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK: omp_offload.failed:
// CHECK-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l145(ptr [[K]]) #[[ATTR5:[0-9]+]]
// CHECK-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK: omp_offload.cont:
-// CHECK-NEXT: [[TMP17:%.*]] = load ptr, ptr [[Z]], align 8
-// CHECK-NEXT: store ptr [[TMP17]], ptr [[TMP]], align 8
-// CHECK-NEXT: [[TMP18:%.*]] = load ptr, ptr [[TMP]], align 8
-// CHECK-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK-NEXT: store ptr [[TMP18]], ptr [[TMP19]], align 8
-// CHECK-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK-NEXT: store ptr [[TMP18]], ptr [[TMP20]], align 8
-// CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
-// CHECK-NEXT: store ptr null, ptr [[TMP21]], align 8
-// CHECK-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK-NEXT: [[TMP21:%.*]] = load ptr, ptr [[Z]], align 8
+// CHECK-NEXT: store ptr [[TMP21]], ptr [[TMP]], align 8
+// CHECK-NEXT: [[TMP22:%.*]] = load ptr, ptr [[TMP]], align 8
+// CHECK-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK-NEXT: store ptr [[TMP22]], ptr [[TMP23]], align 8
+// CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK-NEXT: store ptr [[TMP22]], ptr [[TMP24]], align 8
+// CHECK-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
+// CHECK-NEXT: store ptr null, ptr [[TMP25]], align 8
+// CHECK-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK-NEXT: [[KERNEL_ARGS4:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 0
-// CHECK-NEXT: store i32 1, ptr [[TMP24]], align 4
-// CHECK-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 1
-// CHECK-NEXT: store i32 1, ptr [[TMP25]], align 4
-// CHECK-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 2
-// CHECK-NEXT: store ptr [[TMP22]], ptr [[TMP26]], align 8
-// CHECK-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 3
-// CHECK-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
-// CHECK-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 4
-// CHECK-NEXT: store ptr @.offload_sizes.5, ptr [[TMP28]], align 8
-// CHECK-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 5
-// CHECK-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP29]], align 8
-// CHECK-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 6
-// CHECK-NEXT: store ptr null, ptr [[TMP30]], align 8
-// CHECK-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 7
-// CHECK-NEXT: store ptr null, ptr [[TMP31]], align 8
-// CHECK-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 8
-// CHECK-NEXT: store i64 0, ptr [[TMP32]], align 8
-// CHECK-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.region_id, ptr [[KERNEL_ARGS4]])
-// CHECK-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
-// CHECK-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED5:%.*]], label [[OMP_OFFLOAD_CONT6:%.*]]
+// CHECK-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 0
+// CHECK-NEXT: store i32 2, ptr [[TMP28]], align 4
+// CHECK-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 1
+// CHECK-NEXT: store i32 1, ptr [[TMP29]], align 4
+// CHECK-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 2
+// CHECK-NEXT: store ptr [[TMP26]], ptr [[TMP30]], align 8
+// CHECK-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 3
+// CHECK-NEXT: store ptr [[TMP27]], ptr [[TMP31]], align 8
+// CHECK-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 4
+// CHECK-NEXT: store ptr @.offload_sizes.5, ptr [[TMP32]], align 8
+// CHECK-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 5
+// CHECK-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP33]], align 8
+// CHECK-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 6
+// CHECK-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 7
+// CHECK-NEXT: store ptr null, ptr [[TMP35]], align 8
+// CHECK-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 8
+// CHECK-NEXT: store i64 0, ptr [[TMP36]], align 8
+// CHECK-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 9
+// CHECK-NEXT: store i64 0, ptr [[TMP37]], align 8
+// CHECK-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 10
+// CHECK-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP38]], align 4
+// CHECK-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 11
+// CHECK-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP39]], align 4
+// CHECK-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 12
+// CHECK-NEXT: store i32 0, ptr [[TMP40]], align 4
+// CHECK-NEXT: [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.region_id, ptr [[KERNEL_ARGS4]])
+// CHECK-NEXT: [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
+// CHECK-NEXT: br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED5:%.*]], label [[OMP_OFFLOAD_CONT6:%.*]]
// CHECK: omp_offload.failed5:
-// CHECK-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147(ptr [[TMP18]]) #[[ATTR5]]
+// CHECK-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147(ptr [[TMP22]]) #[[ATTR5]]
// CHECK-NEXT: br label [[OMP_OFFLOAD_CONT6]]
// CHECK: omp_offload.cont6:
-// CHECK-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 0
-// CHECK-NEXT: store ptr [[AA]], ptr [[TMP35]], align 8
-// CHECK-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 0
-// CHECK-NEXT: store ptr [[AA]], ptr [[TMP36]], align 8
-// CHECK-NEXT: [[TMP37:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i64 0, i64 0
-// CHECK-NEXT: store ptr null, ptr [[TMP37]], align 8
-// CHECK-NEXT: [[TMP38:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 0
-// CHECK-NEXT: [[TMP39:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 0
+// CHECK-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 0
+// CHECK-NEXT: store ptr [[AA]], ptr [[TMP43]], align 8
+// CHECK-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 0
+// CHECK-NEXT: store ptr [[AA]], ptr [[TMP44]], align 8
+// CHECK-NEXT: [[TMP45:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i64 0, i64 0
+// CHECK-NEXT: store ptr null, ptr [[TMP45]], align 8
+// CHECK-NEXT: [[TMP46:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 0
+// CHECK-NEXT: [[TMP47:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 0
// CHECK-NEXT: [[KERNEL_ARGS10:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 0
-// CHECK-NEXT: store i32 1, ptr [[TMP40]], align 4
-// CHECK-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 1
-// CHECK-NEXT: store i32 1, ptr [[TMP41]], align 4
-// CHECK-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 2
-// CHECK-NEXT: store ptr [[TMP38]], ptr [[TMP42]], align 8
-// CHECK-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 3
-// CHECK-NEXT: store ptr [[TMP39]], ptr [[TMP43]], align 8
-// CHECK-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 4
-// CHECK-NEXT: store ptr @.offload_sizes.7, ptr [[TMP44]], align 8
-// CHECK-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 5
-// CHECK-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP45]], align 8
-// CHECK-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 6
-// CHECK-NEXT: store ptr null, ptr [[TMP46]], align 8
-// CHECK-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 7
-// CHECK-NEXT: store ptr null, ptr [[TMP47]], align 8
-// CHECK-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 8
-// CHECK-NEXT: store i64 0, ptr [[TMP48]], align 8
-// CHECK-NEXT: [[TMP49:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l149.region_id, ptr [[KERNEL_ARGS10]])
-// CHECK-NEXT: [[TMP50:%.*]] = icmp ne i32 [[TMP49]], 0
-// CHECK-NEXT: br i1 [[TMP50]], label [[OMP_OFFLOAD_FAILED11:%.*]], label [[OMP_OFFLOAD_CONT12:%.*]]
+// CHECK-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 0
+// CHECK-NEXT: store i32 2, ptr [[TMP48]], align 4
+// CHECK-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 1
+// CHECK-NEXT: store i32 1, ptr [[TMP49]], align 4
+// CHECK-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 2
+// CHECK-NEXT: store ptr [[TMP46]], ptr [[TMP50]], align 8
+// CHECK-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 3
+// CHECK-NEXT: store ptr [[TMP47]], ptr [[TMP51]], align 8
+// CHECK-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 4
+// CHECK-NEXT: store ptr @.offload_sizes.7, ptr [[TMP52]], align 8
+// CHECK-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 5
+// CHECK-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP53]], align 8
+// CHECK-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 6
+// CHECK-NEXT: store ptr null, ptr [[TMP54]], align 8
+// CHECK-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 7
+// CHECK-NEXT: store ptr null, ptr [[TMP55]], align 8
+// CHECK-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 8
+// CHECK-NEXT: store i64 0, ptr [[TMP56]], align 8
+// CHECK-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 9
+// CHECK-NEXT: store i64 0, ptr [[TMP57]], align 8
+// CHECK-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 10
+// CHECK-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP58]], align 4
+// CHECK-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 11
+// CHECK-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP59]], align 4
+// CHECK-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 12
+// CHECK-NEXT: store i32 0, ptr [[TMP60]], align 4
+// CHECK-NEXT: [[TMP61:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l149.region_id, ptr [[KERNEL_ARGS10]])
+// CHECK-NEXT: [[TMP62:%.*]] = icmp ne i32 [[TMP61]], 0
+// CHECK-NEXT: br i1 [[TMP62]], label [[OMP_OFFLOAD_FAILED11:%.*]], label [[OMP_OFFLOAD_CONT12:%.*]]
// CHECK: omp_offload.failed11:
// CHECK-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l149(ptr [[AA]]) #[[ATTR5]]
// CHECK-NEXT: br label [[OMP_OFFLOAD_CONT12]]
// CHECK: omp_offload.cont12:
-// CHECK-NEXT: [[TMP51:%.*]] = load ptr, ptr [[RAA]], align 8
-// CHECK-NEXT: store ptr [[TMP51]], ptr [[_TMP13]], align 8
-// CHECK-NEXT: [[TMP52:%.*]] = load ptr, ptr [[_TMP13]], align 8
-// CHECK-NEXT: [[TMP53:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
-// CHECK-NEXT: store ptr [[TMP52]], ptr [[TMP53]], align 8
-// CHECK-NEXT: [[TMP54:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
-// CHECK-NEXT: store ptr [[TMP52]], ptr [[TMP54]], align 8
-// CHECK-NEXT: [[TMP55:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 0
-// CHECK-NEXT: store ptr null, ptr [[TMP55]], align 8
-// CHECK-NEXT: [[TMP56:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
-// CHECK-NEXT: [[TMP57:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
+// CHECK-NEXT: [[TMP63:%.*]] = load ptr, ptr [[RAA]], align 8
+// CHECK-NEXT: store ptr [[TMP63]], ptr [[_TMP13]], align 8
+// CHECK-NEXT: [[TMP64:%.*]] = load ptr, ptr [[_TMP13]], align 8
+// CHECK-NEXT: [[TMP65:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
+// CHECK-NEXT: store ptr [[TMP64]], ptr [[TMP65]], align 8
+// CHECK-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
+// CHECK-NEXT: store ptr [[TMP64]], ptr [[TMP66]], align 8
+// CHECK-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 0
+// CHECK-NEXT: store ptr null, ptr [[TMP67]], align 8
+// CHECK-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
+// CHECK-NEXT: [[TMP69:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
// CHECK-NEXT: [[KERNEL_ARGS17:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 0
-// CHECK-NEXT: store i32 1, ptr [[TMP58]], align 4
-// CHECK-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 1
-// CHECK-NEXT: store i32 1, ptr [[TMP59]], align 4
-// CHECK-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 2
-// CHECK-NEXT: store ptr [[TMP56]], ptr [[TMP60]], align 8
-// CHECK-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 3
-// CHECK-NEXT: store ptr [[TMP57]], ptr [[TMP61]], align 8
-// CHECK-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 4
-// CHECK-NEXT: store ptr @.offload_sizes.9, ptr [[TMP62]], align 8
-// CHECK-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 5
-// CHECK-NEXT: store ptr @.offload_maptypes.10, ptr [[TMP63]], align 8
-// CHECK-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 6
-// CHECK-NEXT: store ptr null, ptr [[TMP64]], align 8
-// CHECK-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 7
-// CHECK-NEXT: store ptr null, ptr [[TMP65]], align 8
-// CHECK-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 8
-// CHECK-NEXT: store i64 0, ptr [[TMP66]], align 8
-// CHECK-NEXT: [[TMP67:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.region_id, ptr [[KERNEL_ARGS17]])
-// CHECK-NEXT: [[TMP68:%.*]] = icmp ne i32 [[TMP67]], 0
-// CHECK-NEXT: br i1 [[TMP68]], label [[OMP_OFFLOAD_FAILED18:%.*]], label [[OMP_OFFLOAD_CONT19:%.*]]
+// CHECK-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 0
+// CHECK-NEXT: store i32 2, ptr [[TMP70]], align 4
+// CHECK-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 1
+// CHECK-NEXT: store i32 1, ptr [[TMP71]], align 4
+// CHECK-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 2
+// CHECK-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 8
+// CHECK-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 3
+// CHECK-NEXT: store ptr [[TMP69]], ptr [[TMP73]], align 8
+// CHECK-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 4
+// CHECK-NEXT: store ptr @.offload_sizes.9, ptr [[TMP74]], align 8
+// CHECK-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 5
+// CHECK-NEXT: store ptr @.offload_maptypes.10, ptr [[TMP75]], align 8
+// CHECK-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 6
+// CHECK-NEXT: store ptr null, ptr [[TMP76]], align 8
+// CHECK-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 7
+// CHECK-NEXT: store ptr null, ptr [[TMP77]], align 8
+// CHECK-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 8
+// CHECK-NEXT: store i64 0, ptr [[TMP78]], align 8
+// CHECK-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 9
+// CHECK-NEXT: store i64 0, ptr [[TMP79]], align 8
+// CHECK-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 10
+// CHECK-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP80]], align 4
+// CHECK-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 11
+// CHECK-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP81]], align 4
+// CHECK-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 12
+// CHECK-NEXT: store i32 0, ptr [[TMP82]], align 4
+// CHECK-NEXT: [[TMP83:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.region_id, ptr [[KERNEL_ARGS17]])
+// CHECK-NEXT: [[TMP84:%.*]] = icmp ne i32 [[TMP83]], 0
+// CHECK-NEXT: br i1 [[TMP84]], label [[OMP_OFFLOAD_FAILED18:%.*]], label [[OMP_OFFLOAD_CONT19:%.*]]
// CHECK: omp_offload.failed18:
-// CHECK-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151(ptr [[TMP52]]) #[[ATTR5]]
+// CHECK-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151(ptr [[TMP64]]) #[[ATTR5]]
// CHECK-NEXT: br label [[OMP_OFFLOAD_CONT19]]
// CHECK: omp_offload.cont19:
-// CHECK-NEXT: [[TMP69:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK-NEXT: store ptr [[H]], ptr [[TMP69]], align 8
-// CHECK-NEXT: [[TMP70:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK-NEXT: store ptr [[H]], ptr [[TMP70]], align 8
-// CHECK-NEXT: [[TMP71:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 0
-// CHECK-NEXT: store ptr null, ptr [[TMP71]], align 8
-// CHECK-NEXT: [[TMP72:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK-NEXT: [[TMP73:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK-NEXT: [[KERNEL_ARGS23:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
-// CHECK-NEXT: store i32 1, ptr [[TMP74]], align 4
-// CHECK-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
-// CHECK-NEXT: store i32 1, ptr [[TMP75]], align 4
-// CHECK-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
-// CHECK-NEXT: store ptr [[TMP72]], ptr [[TMP76]], align 8
-// CHECK-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
-// CHECK-NEXT: store ptr [[TMP73]], ptr [[TMP77]], align 8
-// CHECK-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
-// CHECK-NEXT: store ptr @.offload_sizes.11, ptr [[TMP78]], align 8
-// CHECK-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
-// CHECK-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP79]], align 8
-// CHECK-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
-// CHECK-NEXT: store ptr null, ptr [[TMP80]], align 8
-// CHECK-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
-// CHECK-NEXT: store ptr null, ptr [[TMP81]], align 8
-// CHECK-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
-// CHECK-NEXT: store i64 0, ptr [[TMP82]], align 8
-// CHECK-NEXT: [[TMP83:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l153.region_id, ptr [[KERNEL_ARGS23]])
-// CHECK-NEXT: [[TMP84:%.*]] = icmp ne i32 [[TMP83]], 0
-// CHECK-NEXT: br i1 [[TMP84]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
-// CHECK: omp_offload.failed24:
-// CHECK-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l153(ptr [[H]]) #[[ATTR5]]
-// CHECK-NEXT: br label [[OMP_OFFLOAD_CONT25]]
-// CHECK: omp_offload.cont25:
-// CHECK-NEXT: [[TMP85:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS26]], i32 0, i32 0
-// CHECK-NEXT: store ptr [[DA]], ptr [[TMP85]], align 8
-// CHECK-NEXT: [[TMP86:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS27]], i32 0, i32 0
-// CHECK-NEXT: store ptr [[DA]], ptr [[TMP86]], align 8
-// CHECK-NEXT: [[TMP87:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS28]], i64 0, i64 0
+// CHECK-NEXT: [[TMP85:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK-NEXT: store ptr [[H]], ptr [[TMP85]], align 8
+// CHECK-NEXT: [[TMP86:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK-NEXT: store ptr [[H]], ptr [[TMP86]], align 8
+// CHECK-NEXT: [[TMP87:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 0
// CHECK-NEXT: store ptr null, ptr [[TMP87]], align 8
-// CHECK-NEXT: [[TMP88:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS26]], i32 0, i32 0
-// CHECK-NEXT: [[TMP89:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS27]], i32 0, i32 0
-// CHECK-NEXT: [[KERNEL_ARGS29:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS29]], i32 0, i32 0
-// CHECK-NEXT: store i32 1, ptr [[TMP90]], align 4
-// CHECK-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS29]], i32 0, i32 1
+// CHECK-NEXT: [[TMP88:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK-NEXT: [[TMP89:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK-NEXT: [[KERNEL_ARGS23:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
+// CHECK-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
+// CHECK-NEXT: store i32 2, ptr [[TMP90]], align 4
+// CHECK-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
// CHECK-NEXT: store i32 1, ptr [[TMP91]], align 4
-// CHECK-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS29]], i32 0, i32 2
+// CHECK-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
// CHECK-NEXT: store ptr [[TMP88]], ptr [[TMP92]], align 8
-// CHECK-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS29]], i32 0, i32 3
+// CHECK-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
// CHECK-NEXT: store ptr [[TMP89]], ptr [[TMP93]], align 8
-// CHECK-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS29]], i32 0, i32 4
-// CHECK-NEXT: store ptr @.offload_sizes.13, ptr [[TMP94]], align 8
-// CHECK-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS29]], i32 0, i32 5
-// CHECK-NEXT: store ptr @.offload_maptypes.14, ptr [[TMP95]], align 8
-// CHECK-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS29]], i32 0, i32 6
+// CHECK-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
+// CHECK-NEXT: store ptr @.offload_sizes.11, ptr [[TMP94]], align 8
+// CHECK-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
+// CHECK-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP95]], align 8
+// CHECK-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
// CHECK-NEXT: store ptr null, ptr [[TMP96]], align 8
-// CHECK-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS29]], i32 0, i32 7
+// CHECK-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
// CHECK-NEXT: store ptr null, ptr [[TMP97]], align 8
-// CHECK-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS29]], i32 0, i32 8
+// CHECK-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
// CHECK-NEXT: store i64 0, ptr [[TMP98]], align 8
-// CHECK-NEXT: [[TMP99:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.region_id, ptr [[KERNEL_ARGS29]])
-// CHECK-NEXT: [[TMP100:%.*]] = icmp ne i32 [[TMP99]], 0
-// CHECK-NEXT: br i1 [[TMP100]], label [[OMP_OFFLOAD_FAILED30:%.*]], label [[OMP_OFFLOAD_CONT31:%.*]]
+// CHECK-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 9
+// CHECK-NEXT: store i64 0, ptr [[TMP99]], align 8
+// CHECK-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 10
+// CHECK-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP100]], align 4
+// CHECK-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 11
+// CHECK-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP101]], align 4
+// CHECK-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 12
+// CHECK-NEXT: store i32 0, ptr [[TMP102]], align 4
+// CHECK-NEXT: [[TMP103:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l153.region_id, ptr [[KERNEL_ARGS23]])
+// CHECK-NEXT: [[TMP104:%.*]] = icmp ne i32 [[TMP103]], 0
+// CHECK-NEXT: br i1 [[TMP104]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
+// CHECK: omp_offload.failed24:
+// CHECK-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l153(ptr [[H]]) #[[ATTR5]]
+// CHECK-NEXT: br label [[OMP_OFFLOAD_CONT25]]
+// CHECK: omp_offload.cont25:
+// CHECK-NEXT: [[TMP105:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS26]], i32 0, i32 0
+// CHECK-NEXT: store ptr [[DA]], ptr [[TMP105]], align 8
+// CHECK-NEXT: [[TMP106:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS27]], i32 0, i32 0
+// CHECK-NEXT: store ptr [[DA]], ptr [[TMP106]], align 8
+// CHECK-NEXT: [[TMP107:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS28]], i64 0, i64 0
+// CHECK-NEXT: store ptr null, ptr [[TMP107]], align 8
+// CHECK-NEXT: [[TMP108:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS26]], i32 0, i32 0
+// CHECK-NEXT: [[TMP109:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS27]], i32 0, i32 0
+// CHECK-NEXT: [[KERNEL_ARGS29:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
+// CHECK-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS29]], i32 0, i32 0
+// CHECK-NEXT: store i32 2, ptr [[TMP110]], align 4
+// CHECK-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS29]], i32 0, i32 1
+// CHECK-NEXT: store i32 1, ptr [[TMP111]], align 4
+// CHECK-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS29]], i32 0, i32 2
+// CHECK-NEXT: store ptr [[TMP108]], ptr [[TMP112]], align 8
+// CHECK-NEXT: [[TMP113:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS29]], i32 0, i32 3
+// CHECK-NEXT: store ptr [[TMP109]], ptr [[TMP113]], align 8
+// CHECK-NEXT: [[TMP114:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS29]], i32 0, i32 4
+// CHECK-NEXT: store ptr @.offload_sizes.13, ptr [[TMP114]], align 8
+// CHECK-NEXT: [[TMP115:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS29]], i32 0, i32 5
+// CHECK-NEXT: store ptr @.offload_maptypes.14, ptr [[TMP115]], align 8
+// CHECK-NEXT: [[TMP116:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS29]], i32 0, i32 6
+// CHECK-NEXT: store ptr null, ptr [[TMP116]], align 8
+// CHECK-NEXT: [[TMP117:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS29]], i32 0, i32 7
+// CHECK-NEXT: store ptr null, ptr [[TMP117]], align 8
+// CHECK-NEXT: [[TMP118:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS29]], i32 0, i32 8
+// CHECK-NEXT: store i64 0, ptr [[TMP118]], align 8
+// CHECK-NEXT: [[TMP119:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS29]], i32 0, i32 9
+// CHECK-NEXT: store i64 0, ptr [[TMP119]], align 8
+// CHECK-NEXT: [[TMP120:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS29]], i32 0, i32 10
+// CHECK-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP120]], align 4
+// CHECK-NEXT: [[TMP121:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS29]], i32 0, i32 11
+// CHECK-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP121]], align 4
+// CHECK-NEXT: [[TMP122:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS29]], i32 0, i32 12
+// CHECK-NEXT: store i32 0, ptr [[TMP122]], align 4
+// CHECK-NEXT: [[TMP123:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.region_id, ptr [[KERNEL_ARGS29]])
+// CHECK-NEXT: [[TMP124:%.*]] = icmp ne i32 [[TMP123]], 0
+// CHECK-NEXT: br i1 [[TMP124]], label [[OMP_OFFLOAD_FAILED30:%.*]], label [[OMP_OFFLOAD_CONT31:%.*]]
// CHECK: omp_offload.failed30:
// CHECK-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155(ptr [[DA]]) #[[ATTR5]]
// CHECK-NEXT: br label [[OMP_OFFLOAD_CONT31]]
// CHECK: omp_offload.cont31:
-// CHECK-NEXT: [[TMP101:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiET_S0_(i32 noundef signext [[TMP101]])
+// CHECK-NEXT: [[TMP125:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiET_S0_(i32 noundef signext [[TMP125]])
// CHECK-NEXT: [[CALL32:%.*]] = call noundef ptr @_Z5tmainIPiET_S1_(ptr noundef [[ARGC_ADDR]])
-// CHECK-NEXT: [[TMP102:%.*]] = load i32, ptr [[CALL32]], align 4
-// CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[CALL]], [[TMP102]]
+// CHECK-NEXT: [[TMP126:%.*]] = load i32, ptr [[CALL32]], align 4
+// CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[CALL]], [[TMP126]]
// CHECK-NEXT: ret i32 [[ADD]]
//
//
@@ -654,7 +702,7 @@ void use_template() {
// CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK-NEXT: store i32 1, ptr [[TMP6]], align 4
+// CHECK-NEXT: store i32 2, ptr [[TMP6]], align 4
// CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK-NEXT: store i32 1, ptr [[TMP7]], align 4
// CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -671,114 +719,146 @@ void use_template() {
// CHECK-NEXT: store ptr null, ptr [[TMP13]], align 8
// CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK-NEXT: store i64 0, ptr [[TMP14]], align 8
-// CHECK-NEXT: [[TMP15:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_S0__l123.region_id, ptr [[KERNEL_ARGS]])
-// CHECK-NEXT: [[TMP16:%.*]] = icmp ne i32 [[TMP15]], 0
-// CHECK-NEXT: br i1 [[TMP16]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK-NEXT: store i64 0, ptr [[TMP15]], align 8
+// CHECK-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP16]], align 4
+// CHECK-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK-NEXT: store i32 0, ptr [[TMP18]], align 4
+// CHECK-NEXT: [[TMP19:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_S0__l123.region_id, ptr [[KERNEL_ARGS]])
+// CHECK-NEXT: [[TMP20:%.*]] = icmp ne i32 [[TMP19]], 0
+// CHECK-NEXT: br i1 [[TMP20]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK: omp_offload.failed:
// CHECK-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_S0__l123(ptr [[K]]) #[[ATTR5]]
// CHECK-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK: omp_offload.cont:
-// CHECK-NEXT: [[TMP17:%.*]] = load ptr, ptr [[Z]], align 8
-// CHECK-NEXT: store ptr [[TMP17]], ptr [[TMP]], align 8
-// CHECK-NEXT: [[TMP18:%.*]] = load ptr, ptr [[TMP]], align 8
-// CHECK-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK-NEXT: store ptr [[TMP18]], ptr [[TMP19]], align 8
-// CHECK-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK-NEXT: store ptr [[TMP18]], ptr [[TMP20]], align 8
-// CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
-// CHECK-NEXT: store ptr null, ptr [[TMP21]], align 8
-// CHECK-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK-NEXT: [[TMP21:%.*]] = load ptr, ptr [[Z]], align 8
+// CHECK-NEXT: store ptr [[TMP21]], ptr [[TMP]], align 8
+// CHECK-NEXT: [[TMP22:%.*]] = load ptr, ptr [[TMP]], align 8
+// CHECK-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK-NEXT: store ptr [[TMP22]], ptr [[TMP23]], align 8
+// CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK-NEXT: store ptr [[TMP22]], ptr [[TMP24]], align 8
+// CHECK-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
+// CHECK-NEXT: store ptr null, ptr [[TMP25]], align 8
+// CHECK-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK-NEXT: [[KERNEL_ARGS4:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 0
-// CHECK-NEXT: store i32 1, ptr [[TMP24]], align 4
-// CHECK-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 1
-// CHECK-NEXT: store i32 1, ptr [[TMP25]], align 4
-// CHECK-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 2
-// CHECK-NEXT: store ptr [[TMP22]], ptr [[TMP26]], align 8
-// CHECK-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 3
-// CHECK-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
-// CHECK-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 4
-// CHECK-NEXT: store ptr @.offload_sizes.17, ptr [[TMP28]], align 8
-// CHECK-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 5
-// CHECK-NEXT: store ptr @.offload_maptypes.18, ptr [[TMP29]], align 8
-// CHECK-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 6
-// CHECK-NEXT: store ptr null, ptr [[TMP30]], align 8
-// CHECK-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 7
-// CHECK-NEXT: store ptr null, ptr [[TMP31]], align 8
-// CHECK-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 8
-// CHECK-NEXT: store i64 0, ptr [[TMP32]], align 8
-// CHECK-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_S0__l125.region_id, ptr [[KERNEL_ARGS4]])
-// CHECK-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
-// CHECK-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED5:%.*]], label [[OMP_OFFLOAD_CONT6:%.*]]
+// CHECK-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 0
+// CHECK-NEXT: store i32 2, ptr [[TMP28]], align 4
+// CHECK-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 1
+// CHECK-NEXT: store i32 1, ptr [[TMP29]], align 4
+// CHECK-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 2
+// CHECK-NEXT: store ptr [[TMP26]], ptr [[TMP30]], align 8
+// CHECK-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 3
+// CHECK-NEXT: store ptr [[TMP27]], ptr [[TMP31]], align 8
+// CHECK-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 4
+// CHECK-NEXT: store ptr @.offload_sizes.17, ptr [[TMP32]], align 8
+// CHECK-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 5
+// CHECK-NEXT: store ptr @.offload_maptypes.18, ptr [[TMP33]], align 8
+// CHECK-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 6
+// CHECK-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 7
+// CHECK-NEXT: store ptr null, ptr [[TMP35]], align 8
+// CHECK-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 8
+// CHECK-NEXT: store i64 0, ptr [[TMP36]], align 8
+// CHECK-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 9
+// CHECK-NEXT: store i64 0, ptr [[TMP37]], align 8
+// CHECK-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 10
+// CHECK-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP38]], align 4
+// CHECK-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 11
+// CHECK-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP39]], align 4
+// CHECK-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 12
+// CHECK-NEXT: store i32 0, ptr [[TMP40]], align 4
+// CHECK-NEXT: [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_S0__l125.region_id, ptr [[KERNEL_ARGS4]])
+// CHECK-NEXT: [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
+// CHECK-NEXT: br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED5:%.*]], label [[OMP_OFFLOAD_CONT6:%.*]]
// CHECK: omp_offload.failed5:
-// CHECK-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_S0__l125(ptr [[TMP18]]) #[[ATTR5]]
+// CHECK-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_S0__l125(ptr [[TMP22]]) #[[ATTR5]]
// CHECK-NEXT: br label [[OMP_OFFLOAD_CONT6]]
// CHECK: omp_offload.cont6:
-// CHECK-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 0
-// CHECK-NEXT: store ptr [[AA]], ptr [[TMP35]], align 8
-// CHECK-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 0
-// CHECK-NEXT: store ptr [[AA]], ptr [[TMP36]], align 8
-// CHECK-NEXT: [[TMP37:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i64 0, i64 0
-// CHECK-NEXT: store ptr null, ptr [[TMP37]], align 8
-// CHECK-NEXT: [[TMP38:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 0
-// CHECK-NEXT: [[TMP39:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 0
+// CHECK-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 0
+// CHECK-NEXT: store ptr [[AA]], ptr [[TMP43]], align 8
+// CHECK-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 0
+// CHECK-NEXT: store ptr [[AA]], ptr [[TMP44]], align 8
+// CHECK-NEXT: [[TMP45:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i64 0, i64 0
+// CHECK-NEXT: store ptr null, ptr [[TMP45]], align 8
+// CHECK-NEXT: [[TMP46:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 0
+// CHECK-NEXT: [[TMP47:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 0
// CHECK-NEXT: [[KERNEL_ARGS10:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 0
-// CHECK-NEXT: store i32 1, ptr [[TMP40]], align 4
-// CHECK-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 1
-// CHECK-NEXT: store i32 1, ptr [[TMP41]], align 4
-// CHECK-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 2
-// CHECK-NEXT: store ptr [[TMP38]], ptr [[TMP42]], align 8
-// CHECK-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 3
-// CHECK-NEXT: store ptr [[TMP39]], ptr [[TMP43]], align 8
-// CHECK-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 4
-// CHECK-NEXT: store ptr @.offload_sizes.19, ptr [[TMP44]], align 8
-// CHECK-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 5
-// CHECK-NEXT: store ptr @.offload_maptypes.20, ptr [[TMP45]], align 8
-// CHECK-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 6
-// CHECK-NEXT: store ptr null, ptr [[TMP46]], align 8
-// CHECK-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 7
-// CHECK-NEXT: store ptr null, ptr [[TMP47]], align 8
-// CHECK-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 8
-// CHECK-NEXT: store i64 0, ptr [[TMP48]], align 8
-// CHECK-NEXT: [[TMP49:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_S0__l127.region_id, ptr [[KERNEL_ARGS10]])
-// CHECK-NEXT: [[TMP50:%.*]] = icmp ne i32 [[TMP49]], 0
-// CHECK-NEXT: br i1 [[TMP50]], label [[OMP_OFFLOAD_FAILED11:%.*]], label [[OMP_OFFLOAD_CONT12:%.*]]
+// CHECK-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 0
+// CHECK-NEXT: store i32 2, ptr [[TMP48]], align 4
+// CHECK-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 1
+// CHECK-NEXT: store i32 1, ptr [[TMP49]], align 4
+// CHECK-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 2
+// CHECK-NEXT: store ptr [[TMP46]], ptr [[TMP50]], align 8
+// CHECK-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 3
+// CHECK-NEXT: store ptr [[TMP47]], ptr [[TMP51]], align 8
+// CHECK-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 4
+// CHECK-NEXT: store ptr @.offload_sizes.19, ptr [[TMP52]], align 8
+// CHECK-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 5
+// CHECK-NEXT: store ptr @.offload_maptypes.20, ptr [[TMP53]], align 8
+// CHECK-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 6
+// CHECK-NEXT: store ptr null, ptr [[TMP54]], align 8
+// CHECK-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 7
+// CHECK-NEXT: store ptr null, ptr [[TMP55]], align 8
+// CHECK-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 8
+// CHECK-NEXT: store i64 0, ptr [[TMP56]], align 8
+// CHECK-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 9
+// CHECK-NEXT: store i64 0, ptr [[TMP57]], align 8
+// CHECK-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 10
+// CHECK-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP58]], align 4
+// CHECK-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 11
+// CHECK-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP59]], align 4
+// CHECK-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 12
+// CHECK-NEXT: store i32 0, ptr [[TMP60]], align 4
+// CHECK-NEXT: [[TMP61:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_S0__l127.region_id, ptr [[KERNEL_ARGS10]])
+// CHECK-NEXT: [[TMP62:%.*]] = icmp ne i32 [[TMP61]], 0
+// CHECK-NEXT: br i1 [[TMP62]], label [[OMP_OFFLOAD_FAILED11:%.*]], label [[OMP_OFFLOAD_CONT12:%.*]]
// CHECK: omp_offload.failed11:
// CHECK-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_S0__l127(ptr [[AA]]) #[[ATTR5]]
// CHECK-NEXT: br label [[OMP_OFFLOAD_CONT12]]
// CHECK: omp_offload.cont12:
-// CHECK-NEXT: [[TMP51:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 0
-// CHECK-NEXT: store ptr [[H]], ptr [[TMP51]], align 8
-// CHECK-NEXT: [[TMP52:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 0
-// CHECK-NEXT: store ptr [[H]], ptr [[TMP52]], align 8
-// CHECK-NEXT: [[TMP53:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 0
-// CHECK-NEXT: store ptr null, ptr [[TMP53]], align 8
-// CHECK-NEXT: [[TMP54:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 0
-// CHECK-NEXT: [[TMP55:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 0
+// CHECK-NEXT: [[TMP63:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 0
+// CHECK-NEXT: store ptr [[H]], ptr [[TMP63]], align 8
+// CHECK-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 0
+// CHECK-NEXT: store ptr [[H]], ptr [[TMP64]], align 8
+// CHECK-NEXT: [[TMP65:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 0
+// CHECK-NEXT: store ptr null, ptr [[TMP65]], align 8
+// CHECK-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 0
+// CHECK-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 0
// CHECK-NEXT: [[KERNEL_ARGS16:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 0
-// CHECK-NEXT: store i32 1, ptr [[TMP56]], align 4
-// CHECK-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 1
-// CHECK-NEXT: store i32 1, ptr [[TMP57]], align 4
-// CHECK-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 2
-// CHECK-NEXT: store ptr [[TMP54]], ptr [[TMP58]], align 8
-// CHECK-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 3
-// CHECK-NEXT: store ptr [[TMP55]], ptr [[TMP59]], align 8
-// CHECK-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 4
-// CHECK-NEXT: store ptr @.offload_sizes.21, ptr [[TMP60]], align 8
-// CHECK-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 5
-// CHECK-NEXT: store ptr @.offload_maptypes.22, ptr [[TMP61]], align 8
-// CHECK-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 6
-// CHECK-NEXT: store ptr null, ptr [[TMP62]], align 8
-// CHECK-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 7
-// CHECK-NEXT: store ptr null, ptr [[TMP63]], align 8
-// CHECK-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 8
-// CHECK-NEXT: store i64 0, ptr [[TMP64]], align 8
-// CHECK-NEXT: [[TMP65:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_S0__l129.region_id, ptr [[KERNEL_ARGS16]])
-// CHECK-NEXT: [[TMP66:%.*]] = icmp ne i32 [[TMP65]], 0
-// CHECK-NEXT: br i1 [[TMP66]], label [[OMP_OFFLOAD_FAILED17:%.*]], label [[OMP_OFFLOAD_CONT18:%.*]]
+// CHECK-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 0
+// CHECK-NEXT: store i32 2, ptr [[TMP68]], align 4
+// CHECK-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 1
+// CHECK-NEXT: store i32 1, ptr [[TMP69]], align 4
+// CHECK-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 2
+// CHECK-NEXT: store ptr [[TMP66]], ptr [[TMP70]], align 8
+// CHECK-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 3
+// CHECK-NEXT: store ptr [[TMP67]], ptr [[TMP71]], align 8
+// CHECK-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 4
+// CHECK-NEXT: store ptr @.offload_sizes.21, ptr [[TMP72]], align 8
+// CHECK-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 5
+// CHECK-NEXT: store ptr @.offload_maptypes.22, ptr [[TMP73]], align 8
+// CHECK-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 6
+// CHECK-NEXT: store ptr null, ptr [[TMP74]], align 8
+// CHECK-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 7
+// CHECK-NEXT: store ptr null, ptr [[TMP75]], align 8
+// CHECK-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 8
+// CHECK-NEXT: store i64 0, ptr [[TMP76]], align 8
+// CHECK-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 9
+// CHECK-NEXT: store i64 0, ptr [[TMP77]], align 8
+// CHECK-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 10
+// CHECK-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP78]], align 4
+// CHECK-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 11
+// CHECK-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP79]], align 4
+// CHECK-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 12
+// CHECK-NEXT: store i32 0, ptr [[TMP80]], align 4
+// CHECK-NEXT: [[TMP81:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_S0__l129.region_id, ptr [[KERNEL_ARGS16]])
+// CHECK-NEXT: [[TMP82:%.*]] = icmp ne i32 [[TMP81]], 0
+// CHECK-NEXT: br i1 [[TMP82]], label [[OMP_OFFLOAD_FAILED17:%.*]], label [[OMP_OFFLOAD_CONT18:%.*]]
// CHECK: omp_offload.failed17:
// CHECK-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_S0__l129(ptr [[H]]) #[[ATTR5]]
// CHECK-NEXT: br label [[OMP_OFFLOAD_CONT18]]
@@ -828,7 +908,7 @@ void use_template() {
// CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK-NEXT: store i32 1, ptr [[TMP6]], align 4
+// CHECK-NEXT: store i32 2, ptr [[TMP6]], align 4
// CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK-NEXT: store i32 1, ptr [[TMP7]], align 4
// CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -845,114 +925,146 @@ void use_template() {
// CHECK-NEXT: store ptr null, ptr [[TMP13]], align 8
// CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK-NEXT: store i64 0, ptr [[TMP14]], align 8
-// CHECK-NEXT: [[TMP15:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIPiET_S1__l123.region_id, ptr [[KERNEL_ARGS]])
-// CHECK-NEXT: [[TMP16:%.*]] = icmp ne i32 [[TMP15]], 0
-// CHECK-NEXT: br i1 [[TMP16]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK-NEXT: store i64 0, ptr [[TMP15]], align 8
+// CHECK-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP16]], align 4
+// CHECK-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK-NEXT: store i32 0, ptr [[TMP18]], align 4
+// CHECK-NEXT: [[TMP19:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIPiET_S1__l123.region_id, ptr [[KERNEL_ARGS]])
+// CHECK-NEXT: [[TMP20:%.*]] = icmp ne i32 [[TMP19]], 0
+// CHECK-NEXT: br i1 [[TMP20]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK: omp_offload.failed:
// CHECK-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIPiET_S1__l123(ptr [[K]]) #[[ATTR5]]
// CHECK-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK: omp_offload.cont:
-// CHECK-NEXT: [[TMP17:%.*]] = load ptr, ptr [[Z]], align 8
-// CHECK-NEXT: store ptr [[TMP17]], ptr [[TMP]], align 8
-// CHECK-NEXT: [[TMP18:%.*]] = load ptr, ptr [[TMP]], align 8
-// CHECK-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK-NEXT: store ptr [[TMP18]], ptr [[TMP19]], align 8
-// CHECK-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK-NEXT: store ptr [[TMP18]], ptr [[TMP20]], align 8
-// CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
-// CHECK-NEXT: store ptr null, ptr [[TMP21]], align 8
-// CHECK-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK-NEXT: [[TMP21:%.*]] = load ptr, ptr [[Z]], align 8
+// CHECK-NEXT: store ptr [[TMP21]], ptr [[TMP]], align 8
+// CHECK-NEXT: [[TMP22:%.*]] = load ptr, ptr [[TMP]], align 8
+// CHECK-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK-NEXT: store ptr [[TMP22]], ptr [[TMP23]], align 8
+// CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK-NEXT: store ptr [[TMP22]], ptr [[TMP24]], align 8
+// CHECK-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
+// CHECK-NEXT: store ptr null, ptr [[TMP25]], align 8
+// CHECK-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK-NEXT: [[KERNEL_ARGS4:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 0
-// CHECK-NEXT: store i32 1, ptr [[TMP24]], align 4
-// CHECK-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 1
-// CHECK-NEXT: store i32 1, ptr [[TMP25]], align 4
-// CHECK-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 2
-// CHECK-NEXT: store ptr [[TMP22]], ptr [[TMP26]], align 8
-// CHECK-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 3
-// CHECK-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
-// CHECK-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 4
-// CHECK-NEXT: store ptr @.offload_sizes.25, ptr [[TMP28]], align 8
-// CHECK-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 5
-// CHECK-NEXT: store ptr @.offload_maptypes.26, ptr [[TMP29]], align 8
-// CHECK-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 6
-// CHECK-NEXT: store ptr null, ptr [[TMP30]], align 8
-// CHECK-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 7
-// CHECK-NEXT: store ptr null, ptr [[TMP31]], align 8
-// CHECK-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 8
-// CHECK-NEXT: store i64 0, ptr [[TMP32]], align 8
-// CHECK-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIPiET_S1__l125.region_id, ptr [[KERNEL_ARGS4]])
-// CHECK-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
-// CHECK-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED5:%.*]], label [[OMP_OFFLOAD_CONT6:%.*]]
+// CHECK-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 0
+// CHECK-NEXT: store i32 2, ptr [[TMP28]], align 4
+// CHECK-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 1
+// CHECK-NEXT: store i32 1, ptr [[TMP29]], align 4
+// CHECK-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 2
+// CHECK-NEXT: store ptr [[TMP26]], ptr [[TMP30]], align 8
+// CHECK-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 3
+// CHECK-NEXT: store ptr [[TMP27]], ptr [[TMP31]], align 8
+// CHECK-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 4
+// CHECK-NEXT: store ptr @.offload_sizes.25, ptr [[TMP32]], align 8
+// CHECK-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 5
+// CHECK-NEXT: store ptr @.offload_maptypes.26, ptr [[TMP33]], align 8
+// CHECK-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 6
+// CHECK-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 7
+// CHECK-NEXT: store ptr null, ptr [[TMP35]], align 8
+// CHECK-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 8
+// CHECK-NEXT: store i64 0, ptr [[TMP36]], align 8
+// CHECK-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 9
+// CHECK-NEXT: store i64 0, ptr [[TMP37]], align 8
+// CHECK-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 10
+// CHECK-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP38]], align 4
+// CHECK-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 11
+// CHECK-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP39]], align 4
+// CHECK-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 12
+// CHECK-NEXT: store i32 0, ptr [[TMP40]], align 4
+// CHECK-NEXT: [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIPiET_S1__l125.region_id, ptr [[KERNEL_ARGS4]])
+// CHECK-NEXT: [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
+// CHECK-NEXT: br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED5:%.*]], label [[OMP_OFFLOAD_CONT6:%.*]]
// CHECK: omp_offload.failed5:
-// CHECK-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIPiET_S1__l125(ptr [[TMP18]]) #[[ATTR5]]
+// CHECK-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIPiET_S1__l125(ptr [[TMP22]]) #[[ATTR5]]
// CHECK-NEXT: br label [[OMP_OFFLOAD_CONT6]]
// CHECK: omp_offload.cont6:
-// CHECK-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 0
-// CHECK-NEXT: store ptr [[AA]], ptr [[TMP35]], align 8
-// CHECK-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 0
-// CHECK-NEXT: store ptr [[AA]], ptr [[TMP36]], align 8
-// CHECK-NEXT: [[TMP37:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i64 0, i64 0
-// CHECK-NEXT: store ptr null, ptr [[TMP37]], align 8
-// CHECK-NEXT: [[TMP38:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 0
-// CHECK-NEXT: [[TMP39:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 0
+// CHECK-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 0
+// CHECK-NEXT: store ptr [[AA]], ptr [[TMP43]], align 8
+// CHECK-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 0
+// CHECK-NEXT: store ptr [[AA]], ptr [[TMP44]], align 8
+// CHECK-NEXT: [[TMP45:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS9]], i64 0, i64 0
+// CHECK-NEXT: store ptr null, ptr [[TMP45]], align 8
+// CHECK-NEXT: [[TMP46:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS7]], i32 0, i32 0
+// CHECK-NEXT: [[TMP47:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS8]], i32 0, i32 0
// CHECK-NEXT: [[KERNEL_ARGS10:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 0
-// CHECK-NEXT: store i32 1, ptr [[TMP40]], align 4
-// CHECK-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 1
-// CHECK-NEXT: store i32 1, ptr [[TMP41]], align 4
-// CHECK-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 2
-// CHECK-NEXT: store ptr [[TMP38]], ptr [[TMP42]], align 8
-// CHECK-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 3
-// CHECK-NEXT: store ptr [[TMP39]], ptr [[TMP43]], align 8
-// CHECK-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 4
-// CHECK-NEXT: store ptr @.offload_sizes.27, ptr [[TMP44]], align 8
-// CHECK-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 5
-// CHECK-NEXT: store ptr @.offload_maptypes.28, ptr [[TMP45]], align 8
-// CHECK-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 6
-// CHECK-NEXT: store ptr null, ptr [[TMP46]], align 8
-// CHECK-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 7
-// CHECK-NEXT: store ptr null, ptr [[TMP47]], align 8
-// CHECK-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 8
-// CHECK-NEXT: store i64 0, ptr [[TMP48]], align 8
-// CHECK-NEXT: [[TMP49:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIPiET_S1__l127.region_id, ptr [[KERNEL_ARGS10]])
-// CHECK-NEXT: [[TMP50:%.*]] = icmp ne i32 [[TMP49]], 0
-// CHECK-NEXT: br i1 [[TMP50]], label [[OMP_OFFLOAD_FAILED11:%.*]], label [[OMP_OFFLOAD_CONT12:%.*]]
+// CHECK-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 0
+// CHECK-NEXT: store i32 2, ptr [[TMP48]], align 4
+// CHECK-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 1
+// CHECK-NEXT: store i32 1, ptr [[TMP49]], align 4
+// CHECK-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 2
+// CHECK-NEXT: store ptr [[TMP46]], ptr [[TMP50]], align 8
+// CHECK-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 3
+// CHECK-NEXT: store ptr [[TMP47]], ptr [[TMP51]], align 8
+// CHECK-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 4
+// CHECK-NEXT: store ptr @.offload_sizes.27, ptr [[TMP52]], align 8
+// CHECK-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 5
+// CHECK-NEXT: store ptr @.offload_maptypes.28, ptr [[TMP53]], align 8
+// CHECK-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 6
+// CHECK-NEXT: store ptr null, ptr [[TMP54]], align 8
+// CHECK-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 7
+// CHECK-NEXT: store ptr null, ptr [[TMP55]], align 8
+// CHECK-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 8
+// CHECK-NEXT: store i64 0, ptr [[TMP56]], align 8
+// CHECK-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 9
+// CHECK-NEXT: store i64 0, ptr [[TMP57]], align 8
+// CHECK-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 10
+// CHECK-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP58]], align 4
+// CHECK-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 11
+// CHECK-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP59]], align 4
+// CHECK-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS10]], i32 0, i32 12
+// CHECK-NEXT: store i32 0, ptr [[TMP60]], align 4
+// CHECK-NEXT: [[TMP61:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIPiET_S1__l127.region_id, ptr [[KERNEL_ARGS10]])
+// CHECK-NEXT: [[TMP62:%.*]] = icmp ne i32 [[TMP61]], 0
+// CHECK-NEXT: br i1 [[TMP62]], label [[OMP_OFFLOAD_FAILED11:%.*]], label [[OMP_OFFLOAD_CONT12:%.*]]
// CHECK: omp_offload.failed11:
// CHECK-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIPiET_S1__l127(ptr [[AA]]) #[[ATTR5]]
// CHECK-NEXT: br label [[OMP_OFFLOAD_CONT12]]
// CHECK: omp_offload.cont12:
-// CHECK-NEXT: [[TMP51:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 0
-// CHECK-NEXT: store ptr [[H]], ptr [[TMP51]], align 8
-// CHECK-NEXT: [[TMP52:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 0
-// CHECK-NEXT: store ptr [[H]], ptr [[TMP52]], align 8
-// CHECK-NEXT: [[TMP53:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 0
-// CHECK-NEXT: store ptr null, ptr [[TMP53]], align 8
-// CHECK-NEXT: [[TMP54:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 0
-// CHECK-NEXT: [[TMP55:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 0
+// CHECK-NEXT: [[TMP63:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 0
+// CHECK-NEXT: store ptr [[H]], ptr [[TMP63]], align 8
+// CHECK-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 0
+// CHECK-NEXT: store ptr [[H]], ptr [[TMP64]], align 8
+// CHECK-NEXT: [[TMP65:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 0
+// CHECK-NEXT: store ptr null, ptr [[TMP65]], align 8
+// CHECK-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 0
+// CHECK-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 0
// CHECK-NEXT: [[KERNEL_ARGS16:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 0
-// CHECK-NEXT: store i32 1, ptr [[TMP56]], align 4
-// CHECK-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 1
-// CHECK-NEXT: store i32 1, ptr [[TMP57]], align 4
-// CHECK-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 2
-// CHECK-NEXT: store ptr [[TMP54]], ptr [[TMP58]], align 8
-// CHECK-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 3
-// CHECK-NEXT: store ptr [[TMP55]], ptr [[TMP59]], align 8
-// CHECK-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 4
-// CHECK-NEXT: store ptr @.offload_sizes.29, ptr [[TMP60]], align 8
-// CHECK-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 5
-// CHECK-NEXT: store ptr @.offload_maptypes.30, ptr [[TMP61]], align 8
-// CHECK-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 6
-// CHECK-NEXT: store ptr null, ptr [[TMP62]], align 8
-// CHECK-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 7
-// CHECK-NEXT: store ptr null, ptr [[TMP63]], align 8
-// CHECK-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 8
-// CHECK-NEXT: store i64 0, ptr [[TMP64]], align 8
-// CHECK-NEXT: [[TMP65:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIPiET_S1__l129.region_id, ptr [[KERNEL_ARGS16]])
-// CHECK-NEXT: [[TMP66:%.*]] = icmp ne i32 [[TMP65]], 0
-// CHECK-NEXT: br i1 [[TMP66]], label [[OMP_OFFLOAD_FAILED17:%.*]], label [[OMP_OFFLOAD_CONT18:%.*]]
+// CHECK-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 0
+// CHECK-NEXT: store i32 2, ptr [[TMP68]], align 4
+// CHECK-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 1
+// CHECK-NEXT: store i32 1, ptr [[TMP69]], align 4
+// CHECK-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 2
+// CHECK-NEXT: store ptr [[TMP66]], ptr [[TMP70]], align 8
+// CHECK-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 3
+// CHECK-NEXT: store ptr [[TMP67]], ptr [[TMP71]], align 8
+// CHECK-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 4
+// CHECK-NEXT: store ptr @.offload_sizes.29, ptr [[TMP72]], align 8
+// CHECK-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 5
+// CHECK-NEXT: store ptr @.offload_maptypes.30, ptr [[TMP73]], align 8
+// CHECK-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 6
+// CHECK-NEXT: store ptr null, ptr [[TMP74]], align 8
+// CHECK-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 7
+// CHECK-NEXT: store ptr null, ptr [[TMP75]], align 8
+// CHECK-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 8
+// CHECK-NEXT: store i64 0, ptr [[TMP76]], align 8
+// CHECK-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 9
+// CHECK-NEXT: store i64 0, ptr [[TMP77]], align 8
+// CHECK-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 10
+// CHECK-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP78]], align 4
+// CHECK-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 11
+// CHECK-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP79]], align 4
+// CHECK-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 12
+// CHECK-NEXT: store i32 0, ptr [[TMP80]], align 4
+// CHECK-NEXT: [[TMP81:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIPiET_S1__l129.region_id, ptr [[KERNEL_ARGS16]])
+// CHECK-NEXT: [[TMP82:%.*]] = icmp ne i32 [[TMP81]], 0
+// CHECK-NEXT: br i1 [[TMP82]], label [[OMP_OFFLOAD_FAILED17:%.*]], label [[OMP_OFFLOAD_CONT18:%.*]]
// CHECK: omp_offload.failed17:
// CHECK-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIPiET_S1__l129(ptr [[H]]) #[[ATTR5]]
// CHECK-NEXT: br label [[OMP_OFFLOAD_CONT18]]
@@ -1127,7 +1239,7 @@ void use_template() {
// CHECK-NEXT: [[TMP20:%.*]] = sext i32 [[TMP19]] to i64
// CHECK-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK-NEXT: store i32 3, ptr [[TMP22]], align 4
// CHECK-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1144,9 +1256,17 @@ void use_template() {
// CHECK-NEXT: store ptr null, ptr [[TMP28]], align 8
// CHECK-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK-NEXT: store i64 0, ptr [[TMP29]], align 8
-// CHECK-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 [[TMP20]], i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN10SomeKernel5applyILj32EEEvv_l168.region_id, ptr [[KERNEL_ARGS]])
-// CHECK-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP31]], align 4
+// CHECK-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 [[TMP20]], i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN10SomeKernel5applyILj32EEEvv_l168.region_id, ptr [[KERNEL_ARGS]])
+// CHECK-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK: omp_offload.failed:
// CHECK-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN10SomeKernel5applyILj32EEEvv_l168(ptr [[THIS1]]) #[[ATTR5]]
// CHECK-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1196,7 +1316,7 @@ void use_template() {
//
//
// CHECK-LABEL: define {{[^@]+}}@_ZTW1h
-// CHECK-SAME: () #[[ATTR10:[0-9]+]] comdat {
+// CHECK-SAME: () #[[ATTR0]] comdat {
// CHECK-NEXT: call void @_ZTH1h()
// CHECK-NEXT: [[TMP1:%.*]] = call align 4 ptr @llvm.threadlocal.address.p0(ptr align 4 @h)
// CHECK-NEXT: ret ptr [[TMP1]]
diff --git a/clang/test/OpenMP/target_has_device_addr_codegen_01.cpp b/clang/test/OpenMP/target_has_device_addr_codegen_01.cpp
index e55b8a4ac521f..25c60fa688e74 100644
--- a/clang/test/OpenMP/target_has_device_addr_codegen_01.cpp
+++ b/clang/test/OpenMP/target_has_device_addr_codegen_01.cpp
@@ -102,7 +102,7 @@ int main() {
// CHECK-NEXT: [[TMP25:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK-NEXT: store i32 2, ptr [[TMP26]], align 4
// CHECK-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK-NEXT: store i32 6, ptr [[TMP27]], align 4
// CHECK-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -119,20 +119,28 @@ int main() {
// CHECK-NEXT: store ptr null, ptr [[TMP33]], align 8
// CHECK-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK-NEXT: store i64 0, ptr [[TMP34]], align 8
-// CHECK-NEXT: [[TMP35:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l27.region_id, ptr [[KERNEL_ARGS]])
-// CHECK-NEXT: [[TMP36:%.*]] = icmp ne i32 [[TMP35]], 0
-// CHECK-NEXT: br i1 [[TMP36]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK-NEXT: store i64 0, ptr [[TMP35]], align 8
+// CHECK-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP36]], align 4
+// CHECK-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP37]], align 4
+// CHECK-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK-NEXT: store i32 0, ptr [[TMP38]], align 4
+// CHECK-NEXT: [[TMP39:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l27.region_id, ptr [[KERNEL_ARGS]])
+// CHECK-NEXT: [[TMP40:%.*]] = icmp ne i32 [[TMP39]], 0
+// CHECK-NEXT: br i1 [[TMP40]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK: omp_offload.failed:
// CHECK-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l27(ptr [[A]], ptr [[TMP4]], ptr [[TMP5]], ptr [[ARR]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR5:[0-9]+]]
// CHECK-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK: omp_offload.cont:
-// CHECK-NEXT: [[TMP37:%.*]] = load float, ptr [[A]], align 4
-// CHECK-NEXT: [[CONV1:%.*]] = fptosi float [[TMP37]] to i32
+// CHECK-NEXT: [[TMP41:%.*]] = load float, ptr [[A]], align 4
+// CHECK-NEXT: [[CONV1:%.*]] = fptosi float [[TMP41]] to i32
// CHECK-NEXT: store i32 [[CONV1]], ptr [[RETVAL]], align 4
-// CHECK-NEXT: [[TMP38:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK-NEXT: call void @llvm.stackrestore(ptr [[TMP38]])
-// CHECK-NEXT: [[TMP39:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK-NEXT: ret i32 [[TMP39]]
+// CHECK-NEXT: [[TMP42:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK-NEXT: call void @llvm.stackrestore(ptr [[TMP42]])
+// CHECK-NEXT: [[TMP43:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK-NEXT: ret i32 [[TMP43]]
//
//
// CHECK-LABEL: define {{[^@]+}}@_ZN1SC1Ev
@@ -203,7 +211,7 @@ int main() {
// CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds [5 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK-NEXT: store i32 1, ptr [[TMP25]], align 4
+// CHECK-NEXT: store i32 2, ptr [[TMP25]], align 4
// CHECK-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK-NEXT: store i32 5, ptr [[TMP26]], align 4
// CHECK-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -220,9 +228,17 @@ int main() {
// CHECK-NEXT: store ptr null, ptr [[TMP32]], align 8
// CHECK-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK-NEXT: store i64 0, ptr [[TMP33]], align 8
-// CHECK-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN1S3fooEv_l14.region_id, ptr [[KERNEL_ARGS]])
-// CHECK-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
-// CHECK-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP35]], align 4
+// CHECK-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN1S3fooEv_l14.region_id, ptr [[KERNEL_ARGS]])
+// CHECK-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK: omp_offload.failed:
// CHECK-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN1S3fooEv_l14(ptr [[THIS1]]) #[[ATTR5]]
// CHECK-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/target_map_codegen_03.cpp b/clang/test/OpenMP/target_map_codegen_03.cpp
index 71254c01429d5..552cc794a39bf 100644
--- a/clang/test/OpenMP/target_map_codegen_03.cpp
+++ b/clang/test/OpenMP/target_map_codegen_03.cpp
@@ -96,7 +96,7 @@ void implicit_maps_nested_integer (int a){
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP8]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP8]], align 4
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP9]], align 4
// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -113,9 +113,17 @@ void implicit_maps_nested_integer (int a){
// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z28implicit_maps_nested_integeri_l48.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP18:%.*]] = icmp ne i32 [[TMP17]], 0
-// CHECK1-NEXT: br i1 [[TMP18]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP17]], align 8
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP18]], align 4
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP19]], align 4
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP20]], align 4
+// CHECK1-NEXT: [[TMP21:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z28implicit_maps_nested_integeri_l48.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP22:%.*]] = icmp ne i32 [[TMP21]], 0
+// CHECK1-NEXT: br i1 [[TMP22]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z28implicit_maps_nested_integeri_l48(i64 [[TMP2]]) #[[ATTR3:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -194,7 +202,7 @@ void implicit_maps_nested_integer (int a){
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP8]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP8]], align 4
// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP9]], align 4
// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -211,9 +219,17 @@ void implicit_maps_nested_integer (int a){
// CHECK3-NEXT: store ptr null, ptr [[TMP15]], align 4
// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP16]], align 8
-// CHECK3-NEXT: [[TMP17:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z28implicit_maps_nested_integeri_l48.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP18:%.*]] = icmp ne i32 [[TMP17]], 0
-// CHECK3-NEXT: br i1 [[TMP18]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP17]], align 8
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP18]], align 4
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP19]], align 4
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP20]], align 4
+// CHECK3-NEXT: [[TMP21:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z28implicit_maps_nested_integeri_l48.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP22:%.*]] = icmp ne i32 [[TMP21]], 0
+// CHECK3-NEXT: br i1 [[TMP22]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z28implicit_maps_nested_integeri_l48(i32 [[TMP2]]) #[[ATTR3:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/target_map_codegen_hold.cpp b/clang/test/OpenMP/target_map_codegen_hold.cpp
index 020c7f6881e6c..a5bd05536357f 100644
--- a/clang/test/OpenMP/target_map_codegen_hold.cpp
+++ b/clang/test/OpenMP/target_map_codegen_hold.cpp
@@ -1,4 +1,4 @@
-// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --check-globals --replace-value-regex "__omp_offloading_[0-9a-z]+_[0-9a-z]+" --prefix-filecheck-ir-name _ --global-value-regex ".offload_maptypes.*" ".offload_sizes.*" --global-hex-value-regex ".offload_maptypes.*"
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature --check-globals --include-generated-funcs --replace-value-regex "__omp_offloading_[0-9a-z]+_[0-9a-z]+" --prefix-filecheck-ir-name _ --global-value-regex ".offload_maptypes.*" ".offload_sizes.*" --global-hex-value-regex ".offload_maptypes.*"
// expected-no-diagnostics
#ifndef HEADER
#define HEADER
@@ -72,7 +72,6 @@
// RUN: FileCheck %s --check-prefixes=CHECK-NOUSE-I386
// Map flags used in @.offload_maptypes* below:
-//
// TO = 0x1
// FROM = 0x2
// ALWAYS = 0x4
@@ -117,7 +116,54 @@ struct ST {
void test_present_members();
};
-// CHECK-USE-PPC64LE-LABEL: @_Z20explicit_maps_singlei(
+void explicit_maps_single(int ii) {
+
+ // Map of a scalar.
+ int a = ii;
+
+ struct ST st1;
+ struct ST st2;
+
+ // Make sure the struct picks up ompx_hold even if another element of the
+ // struct doesn't have ompx_hold.
+ #pragma omp target map(tofrom : st1.i) \
+ map(ompx_hold, tofrom : a, st1.j, st2.i) \
+ map(tofrom : st2.j)
+ {
+#ifdef USE
+ st1.i++;
+ a++;
+ st1.j++;
+ st2.i++;
+ st2.j++;
+#endif
+ }
+
+ // Always Close Hold.
+ #pragma omp target map(always close ompx_hold tofrom: a)
+ {
+#ifdef USE
+ a++;
+#endif
+ }
+}
+
+void ST::test_present_members() {
+ // Make sure the struct picks up ompx_hold even if another element of the
+ // struct doesn't have ompx_hold.
+ #pragma omp target map(tofrom : i) map(ompx_hold, tofrom : j)
+ {
+#ifdef USE
+ i++;
+ j++;
+#endif
+ }
+}
+
+#endif
+//.
+// CHECK-USE-PPC64LE-LABEL: define {{[^@]+}}@_Z20explicit_maps_singlei
+// CHECK-USE-PPC64LE-SAME: (i32 noundef signext [[II:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK-USE-PPC64LE-NEXT: entry:
// CHECK-USE-PPC64LE-NEXT: [[II_ADDR:%.*]] = alloca i32, align 4
// CHECK-USE-PPC64LE-NEXT: [[A:%.*]] = alloca i32, align 4
@@ -130,7 +176,7 @@ struct ST {
// CHECK-USE-PPC64LE-NEXT: [[DOTOFFLOAD_BASEPTRS3:%.*]] = alloca [1 x i8*], align 8
// CHECK-USE-PPC64LE-NEXT: [[DOTOFFLOAD_PTRS4:%.*]] = alloca [1 x i8*], align 8
// CHECK-USE-PPC64LE-NEXT: [[DOTOFFLOAD_MAPPERS5:%.*]] = alloca [1 x i8*], align 8
-// CHECK-USE-PPC64LE-NEXT: store i32 [[II:%.*]], i32* [[II_ADDR]], align 4
+// CHECK-USE-PPC64LE-NEXT: store i32 [[II]], i32* [[II_ADDR]], align 4
// CHECK-USE-PPC64LE-NEXT: [[TMP0:%.*]] = load i32, i32* [[II_ADDR]], align 4
// CHECK-USE-PPC64LE-NEXT: store i32 [[TMP0]], i32* [[A]], align 4
// CHECK-USE-PPC64LE-NEXT: [[I:%.*]] = getelementptr inbounds [[STRUCT_ST]], %struct.ST* [[ST1]], i32 0, i32 0
@@ -218,7 +264,7 @@ struct ST {
// CHECK-USE-PPC64LE-NEXT: [[TMP55:%.*]] = getelementptr inbounds [7 x i64], [7 x i64]* [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK-USE-PPC64LE-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK-USE-PPC64LE-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK-USE-PPC64LE-NEXT: store i32 1, i32* [[TMP56]], align 4
+// CHECK-USE-PPC64LE-NEXT: store i32 2, i32* [[TMP56]], align 4
// CHECK-USE-PPC64LE-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 1
// CHECK-USE-PPC64LE-NEXT: store i32 7, i32* [[TMP57]], align 4
// CHECK-USE-PPC64LE-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 2
@@ -235,52 +281,227 @@ struct ST {
// CHECK-USE-PPC64LE-NEXT: store i8** null, i8*** [[TMP63]], align 8
// CHECK-USE-PPC64LE-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 8
// CHECK-USE-PPC64LE-NEXT: store i64 0, i64* [[TMP64]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP65:%.*]] = call i32 @__tgt_target_kernel(%struct.ident_t* @[[GLOB1:[0-9]+]], i64 -1, i32 -1, i32 0, i8* @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l782.region_id, %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]])
-// CHECK-USE-PPC64LE-NEXT: [[TMP66:%.*]] = icmp ne i32 [[TMP65]], 0
-// CHECK-USE-PPC64LE-NEXT: br i1 [[TMP66]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK-USE-PPC64LE-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK-USE-PPC64LE-NEXT: store i64 0, i64* [[TMP65]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK-USE-PPC64LE-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], [3 x i32]* [[TMP66]], align 4
+// CHECK-USE-PPC64LE-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK-USE-PPC64LE-NEXT: store [3 x i32] zeroinitializer, [3 x i32]* [[TMP67]], align 4
+// CHECK-USE-PPC64LE-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK-USE-PPC64LE-NEXT: store i32 0, i32* [[TMP68]], align 4
+// CHECK-USE-PPC64LE-NEXT: [[TMP69:%.*]] = call i32 @__tgt_target_kernel(%struct.ident_t* @[[GLOB1:[0-9]+]], i64 -1, i32 -1, i32 0, i8* @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l{{[0-9]*}}.region_id, %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]])
+// CHECK-USE-PPC64LE-NEXT: [[TMP70:%.*]] = icmp ne i32 [[TMP69]], 0
+// CHECK-USE-PPC64LE-NEXT: br i1 [[TMP70]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK-USE-PPC64LE: omp_offload.failed:
-// CHECK-USE-PPC64LE-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l782(%struct.ST* [[ST1]], i32* [[A]], %struct.ST* [[ST2]]) #[[ATTR3:[0-9]+]]
+// CHECK-USE-PPC64LE-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l{{[0-9]*}}(%struct.ST* [[ST1]], i32* [[A]], %struct.ST* [[ST2]]) #[[ATTR3:[0-9]+]]
// CHECK-USE-PPC64LE-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK-USE-PPC64LE: omp_offload.cont:
-// CHECK-USE-PPC64LE-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK-USE-PPC64LE-NEXT: [[TMP68:%.*]] = bitcast i8** [[TMP67]] to i32**
-// CHECK-USE-PPC64LE-NEXT: store i32* [[A]], i32** [[TMP68]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP69:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK-USE-PPC64LE-NEXT: [[TMP70:%.*]] = bitcast i8** [[TMP69]] to i32**
-// CHECK-USE-PPC64LE-NEXT: store i32* [[A]], i32** [[TMP70]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP71:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
-// CHECK-USE-PPC64LE-NEXT: store i8* null, i8** [[TMP71]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP72:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK-USE-PPC64LE-NEXT: [[TMP71:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK-USE-PPC64LE-NEXT: [[TMP72:%.*]] = bitcast i8** [[TMP71]] to i32**
+// CHECK-USE-PPC64LE-NEXT: store i32* [[A]], i32** [[TMP72]], align 8
// CHECK-USE-PPC64LE-NEXT: [[TMP73:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK-USE-PPC64LE-NEXT: [[TMP74:%.*]] = bitcast i8** [[TMP73]] to i32**
+// CHECK-USE-PPC64LE-NEXT: store i32* [[A]], i32** [[TMP74]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP75:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
+// CHECK-USE-PPC64LE-NEXT: store i8* null, i8** [[TMP75]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP76:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK-USE-PPC64LE-NEXT: [[TMP77:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK-USE-PPC64LE-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK-USE-PPC64LE-NEXT: store i32 1, i32* [[TMP74]], align 4
-// CHECK-USE-PPC64LE-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK-USE-PPC64LE-NEXT: store i32 1, i32* [[TMP75]], align 4
-// CHECK-USE-PPC64LE-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK-USE-PPC64LE-NEXT: store i8** [[TMP72]], i8*** [[TMP76]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK-USE-PPC64LE-NEXT: store i8** [[TMP73]], i8*** [[TMP77]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK-USE-PPC64LE-NEXT: store i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_sizes.1, i32 0, i32 0), i64** [[TMP78]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK-USE-PPC64LE-NEXT: store i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_maptypes.2, i32 0, i32 0), i64** [[TMP79]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK-USE-PPC64LE-NEXT: store i8** null, i8*** [[TMP80]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK-USE-PPC64LE-NEXT: store i8** null, i8*** [[TMP81]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK-USE-PPC64LE-NEXT: store i64 0, i64* [[TMP82]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP83:%.*]] = call i32 @__tgt_target_kernel(%struct.ident_t* @[[GLOB1]], i64 -1, i32 -1, i32 0, i8* @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l796.region_id, %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]])
-// CHECK-USE-PPC64LE-NEXT: [[TMP84:%.*]] = icmp ne i32 [[TMP83]], 0
-// CHECK-USE-PPC64LE-NEXT: br i1 [[TMP84]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK-USE-PPC64LE-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK-USE-PPC64LE-NEXT: store i32 2, i32* [[TMP78]], align 4
+// CHECK-USE-PPC64LE-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK-USE-PPC64LE-NEXT: store i32 1, i32* [[TMP79]], align 4
+// CHECK-USE-PPC64LE-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK-USE-PPC64LE-NEXT: store i8** [[TMP76]], i8*** [[TMP80]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK-USE-PPC64LE-NEXT: store i8** [[TMP77]], i8*** [[TMP81]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK-USE-PPC64LE-NEXT: store i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_sizes.1, i32 0, i32 0), i64** [[TMP82]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK-USE-PPC64LE-NEXT: store i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_maptypes.2, i32 0, i32 0), i64** [[TMP83]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK-USE-PPC64LE-NEXT: store i8** null, i8*** [[TMP84]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK-USE-PPC64LE-NEXT: store i8** null, i8*** [[TMP85]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK-USE-PPC64LE-NEXT: store i64 0, i64* [[TMP86]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK-USE-PPC64LE-NEXT: store i64 0, i64* [[TMP87]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK-USE-PPC64LE-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], [3 x i32]* [[TMP88]], align 4
+// CHECK-USE-PPC64LE-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK-USE-PPC64LE-NEXT: store [3 x i32] zeroinitializer, [3 x i32]* [[TMP89]], align 4
+// CHECK-USE-PPC64LE-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK-USE-PPC64LE-NEXT: store i32 0, i32* [[TMP90]], align 4
+// CHECK-USE-PPC64LE-NEXT: [[TMP91:%.*]] = call i32 @__tgt_target_kernel(%struct.ident_t* @[[GLOB1]], i64 -1, i32 -1, i32 0, i8* @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l{{[0-9]*}}.region_id, %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]])
+// CHECK-USE-PPC64LE-NEXT: [[TMP92:%.*]] = icmp ne i32 [[TMP91]], 0
+// CHECK-USE-PPC64LE-NEXT: br i1 [[TMP92]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK-USE-PPC64LE: omp_offload.failed7:
-// CHECK-USE-PPC64LE-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l796(i32* [[A]]) #[[ATTR3]]
+// CHECK-USE-PPC64LE-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l{{[0-9]*}}(i32* [[A]]) #[[ATTR3]]
// CHECK-USE-PPC64LE-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK-USE-PPC64LE: omp_offload.cont8:
// CHECK-USE-PPC64LE-NEXT: ret void
//
-// CHECK-USE-I386-LABEL: @_Z20explicit_maps_singlei(
+//
+// CHECK-USE-PPC64LE-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l{{[0-9]*}}
+// CHECK-USE-PPC64LE-SAME: (%struct.ST* noundef nonnull align 4 dereferenceable(8) [[ST1:%.*]], i32* noundef nonnull align 4 dereferenceable(4) [[A:%.*]], %struct.ST* noundef nonnull align 4 dereferenceable(8) [[ST2:%.*]]) #[[ATTR1:[0-9]+]] {
+// CHECK-USE-PPC64LE-NEXT: entry:
+// CHECK-USE-PPC64LE-NEXT: [[ST1_ADDR:%.*]] = alloca %struct.ST*, align 8
+// CHECK-USE-PPC64LE-NEXT: [[A_ADDR:%.*]] = alloca i32*, align 8
+// CHECK-USE-PPC64LE-NEXT: [[ST2_ADDR:%.*]] = alloca %struct.ST*, align 8
+// CHECK-USE-PPC64LE-NEXT: store %struct.ST* [[ST1]], %struct.ST** [[ST1_ADDR]], align 8
+// CHECK-USE-PPC64LE-NEXT: store i32* [[A]], i32** [[A_ADDR]], align 8
+// CHECK-USE-PPC64LE-NEXT: store %struct.ST* [[ST2]], %struct.ST** [[ST2_ADDR]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP0:%.*]] = load %struct.ST*, %struct.ST** [[ST1_ADDR]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP1:%.*]] = load i32*, i32** [[A_ADDR]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP2:%.*]] = load %struct.ST*, %struct.ST** [[ST2_ADDR]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[I:%.*]] = getelementptr inbounds [[STRUCT_ST:%.*]], %struct.ST* [[TMP0]], i32 0, i32 0
+// CHECK-USE-PPC64LE-NEXT: [[TMP3:%.*]] = load i32, i32* [[I]], align 4
+// CHECK-USE-PPC64LE-NEXT: [[INC:%.*]] = add nsw i32 [[TMP3]], 1
+// CHECK-USE-PPC64LE-NEXT: store i32 [[INC]], i32* [[I]], align 4
+// CHECK-USE-PPC64LE-NEXT: [[TMP4:%.*]] = load i32, i32* [[TMP1]], align 4
+// CHECK-USE-PPC64LE-NEXT: [[INC1:%.*]] = add nsw i32 [[TMP4]], 1
+// CHECK-USE-PPC64LE-NEXT: store i32 [[INC1]], i32* [[TMP1]], align 4
+// CHECK-USE-PPC64LE-NEXT: [[J:%.*]] = getelementptr inbounds [[STRUCT_ST]], %struct.ST* [[TMP0]], i32 0, i32 1
+// CHECK-USE-PPC64LE-NEXT: [[TMP5:%.*]] = load i32, i32* [[J]], align 4
+// CHECK-USE-PPC64LE-NEXT: [[INC2:%.*]] = add nsw i32 [[TMP5]], 1
+// CHECK-USE-PPC64LE-NEXT: store i32 [[INC2]], i32* [[J]], align 4
+// CHECK-USE-PPC64LE-NEXT: [[I3:%.*]] = getelementptr inbounds [[STRUCT_ST]], %struct.ST* [[TMP2]], i32 0, i32 0
+// CHECK-USE-PPC64LE-NEXT: [[TMP6:%.*]] = load i32, i32* [[I3]], align 4
+// CHECK-USE-PPC64LE-NEXT: [[INC4:%.*]] = add nsw i32 [[TMP6]], 1
+// CHECK-USE-PPC64LE-NEXT: store i32 [[INC4]], i32* [[I3]], align 4
+// CHECK-USE-PPC64LE-NEXT: [[J5:%.*]] = getelementptr inbounds [[STRUCT_ST]], %struct.ST* [[TMP2]], i32 0, i32 1
+// CHECK-USE-PPC64LE-NEXT: [[TMP7:%.*]] = load i32, i32* [[J5]], align 4
+// CHECK-USE-PPC64LE-NEXT: [[INC6:%.*]] = add nsw i32 [[TMP7]], 1
+// CHECK-USE-PPC64LE-NEXT: store i32 [[INC6]], i32* [[J5]], align 4
+// CHECK-USE-PPC64LE-NEXT: ret void
+//
+//
+// CHECK-USE-PPC64LE-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l{{[0-9]*}}
+// CHECK-USE-PPC64LE-SAME: (i32* noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR1]] {
+// CHECK-USE-PPC64LE-NEXT: entry:
+// CHECK-USE-PPC64LE-NEXT: [[A_ADDR:%.*]] = alloca i32*, align 8
+// CHECK-USE-PPC64LE-NEXT: store i32* [[A]], i32** [[A_ADDR]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP0:%.*]] = load i32*, i32** [[A_ADDR]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP1:%.*]] = load i32, i32* [[TMP0]], align 4
+// CHECK-USE-PPC64LE-NEXT: [[INC:%.*]] = add nsw i32 [[TMP1]], 1
+// CHECK-USE-PPC64LE-NEXT: store i32 [[INC]], i32* [[TMP0]], align 4
+// CHECK-USE-PPC64LE-NEXT: ret void
+//
+//
+// CHECK-USE-PPC64LE-LABEL: define {{[^@]+}}@_ZN2ST20test_present_membersEv
+// CHECK-USE-PPC64LE-SAME: (%struct.ST* noundef nonnull align 4 dereferenceable(8) [[THIS:%.*]]) #[[ATTR0]] align 2 {
+// CHECK-USE-PPC64LE-NEXT: entry:
+// CHECK-USE-PPC64LE-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.ST*, align 8
+// CHECK-USE-PPC64LE-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [3 x i8*], align 8
+// CHECK-USE-PPC64LE-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [3 x i8*], align 8
+// CHECK-USE-PPC64LE-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [3 x i8*], align 8
+// CHECK-USE-PPC64LE-NEXT: [[DOTOFFLOAD_SIZES:%.*]] = alloca [3 x i64], align 8
+// CHECK-USE-PPC64LE-NEXT: store %struct.ST* [[THIS]], %struct.ST** [[THIS_ADDR]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[THIS1:%.*]] = load %struct.ST*, %struct.ST** [[THIS_ADDR]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[I:%.*]] = getelementptr inbounds [[STRUCT_ST:%.*]], %struct.ST* [[THIS1]], i32 0, i32 0
+// CHECK-USE-PPC64LE-NEXT: [[J:%.*]] = getelementptr inbounds [[STRUCT_ST]], %struct.ST* [[THIS1]], i32 0, i32 1
+// CHECK-USE-PPC64LE-NEXT: [[TMP0:%.*]] = getelementptr i32, i32* [[J]], i32 1
+// CHECK-USE-PPC64LE-NEXT: [[TMP1:%.*]] = bitcast i32* [[I]] to i8*
+// CHECK-USE-PPC64LE-NEXT: [[TMP2:%.*]] = bitcast i32* [[TMP0]] to i8*
+// CHECK-USE-PPC64LE-NEXT: [[TMP3:%.*]] = ptrtoint i8* [[TMP2]] to i64
+// CHECK-USE-PPC64LE-NEXT: [[TMP4:%.*]] = ptrtoint i8* [[TMP1]] to i64
+// CHECK-USE-PPC64LE-NEXT: [[TMP5:%.*]] = sub i64 [[TMP3]], [[TMP4]]
+// CHECK-USE-PPC64LE-NEXT: [[TMP6:%.*]] = sdiv exact i64 [[TMP5]], ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)
+// CHECK-USE-PPC64LE-NEXT: [[TMP7:%.*]] = bitcast [3 x i64]* [[DOTOFFLOAD_SIZES]] to i8*
+// CHECK-USE-PPC64LE-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[TMP7]], i8* align 8 bitcast ([3 x i64]* @.offload_sizes.3 to i8*), i64 24, i1 false)
+// CHECK-USE-PPC64LE-NEXT: [[TMP8:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK-USE-PPC64LE-NEXT: [[TMP9:%.*]] = bitcast i8** [[TMP8]] to %struct.ST**
+// CHECK-USE-PPC64LE-NEXT: store %struct.ST* [[THIS1]], %struct.ST** [[TMP9]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP10:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK-USE-PPC64LE-NEXT: [[TMP11:%.*]] = bitcast i8** [[TMP10]] to i32**
+// CHECK-USE-PPC64LE-NEXT: store i32* [[I]], i32** [[TMP11]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP12:%.*]] = getelementptr inbounds [3 x i64], [3 x i64]* [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK-USE-PPC64LE-NEXT: store i64 [[TMP6]], i64* [[TMP12]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK-USE-PPC64LE-NEXT: store i8* null, i8** [[TMP13]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP14:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
+// CHECK-USE-PPC64LE-NEXT: [[TMP15:%.*]] = bitcast i8** [[TMP14]] to %struct.ST**
+// CHECK-USE-PPC64LE-NEXT: store %struct.ST* [[THIS1]], %struct.ST** [[TMP15]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_PTRS]], i32 0, i32 1
+// CHECK-USE-PPC64LE-NEXT: [[TMP17:%.*]] = bitcast i8** [[TMP16]] to i32**
+// CHECK-USE-PPC64LE-NEXT: store i32* [[I]], i32** [[TMP17]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP18:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_MAPPERS]], i64 0, i64 1
+// CHECK-USE-PPC64LE-NEXT: store i8* null, i8** [[TMP18]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP19:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
+// CHECK-USE-PPC64LE-NEXT: [[TMP20:%.*]] = bitcast i8** [[TMP19]] to %struct.ST**
+// CHECK-USE-PPC64LE-NEXT: store %struct.ST* [[THIS1]], %struct.ST** [[TMP20]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP21:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_PTRS]], i32 0, i32 2
+// CHECK-USE-PPC64LE-NEXT: [[TMP22:%.*]] = bitcast i8** [[TMP21]] to i32**
+// CHECK-USE-PPC64LE-NEXT: store i32* [[J]], i32** [[TMP22]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_MAPPERS]], i64 0, i64 2
+// CHECK-USE-PPC64LE-NEXT: store i8* null, i8** [[TMP23]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP24:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK-USE-PPC64LE-NEXT: [[TMP25:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK-USE-PPC64LE-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x i64], [3 x i64]* [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK-USE-PPC64LE-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK-USE-PPC64LE-NEXT: store i32 2, i32* [[TMP27]], align 4
+// CHECK-USE-PPC64LE-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK-USE-PPC64LE-NEXT: store i32 3, i32* [[TMP28]], align 4
+// CHECK-USE-PPC64LE-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK-USE-PPC64LE-NEXT: store i8** [[TMP24]], i8*** [[TMP29]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK-USE-PPC64LE-NEXT: store i8** [[TMP25]], i8*** [[TMP30]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK-USE-PPC64LE-NEXT: store i64* [[TMP26]], i64** [[TMP31]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK-USE-PPC64LE-NEXT: store i64* getelementptr inbounds ([3 x i64], [3 x i64]* @.offload_maptypes.4, i32 0, i32 0), i64** [[TMP32]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 6
+// CHECK-USE-PPC64LE-NEXT: store i8** null, i8*** [[TMP33]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK-USE-PPC64LE-NEXT: store i8** null, i8*** [[TMP34]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK-USE-PPC64LE-NEXT: store i64 0, i64* [[TMP35]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK-USE-PPC64LE-NEXT: store i64 0, i64* [[TMP36]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK-USE-PPC64LE-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], [3 x i32]* [[TMP37]], align 4
+// CHECK-USE-PPC64LE-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK-USE-PPC64LE-NEXT: store [3 x i32] zeroinitializer, [3 x i32]* [[TMP38]], align 4
+// CHECK-USE-PPC64LE-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK-USE-PPC64LE-NEXT: store i32 0, i32* [[TMP39]], align 4
+// CHECK-USE-PPC64LE-NEXT: [[TMP40:%.*]] = call i32 @__tgt_target_kernel(%struct.ident_t* @[[GLOB1]], i64 -1, i32 -1, i32 0, i8* @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2ST20test_present_membersEv_l{{[0-9]*}}.region_id, %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]])
+// CHECK-USE-PPC64LE-NEXT: [[TMP41:%.*]] = icmp ne i32 [[TMP40]], 0
+// CHECK-USE-PPC64LE-NEXT: br i1 [[TMP41]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK-USE-PPC64LE: omp_offload.failed:
+// CHECK-USE-PPC64LE-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2ST20test_present_membersEv_l{{[0-9]*}}(%struct.ST* [[THIS1]]) #[[ATTR3]]
+// CHECK-USE-PPC64LE-NEXT: br label [[OMP_OFFLOAD_CONT]]
+// CHECK-USE-PPC64LE: omp_offload.cont:
+// CHECK-USE-PPC64LE-NEXT: ret void
+//
+//
+// CHECK-USE-PPC64LE-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2ST20test_present_membersEv_l{{[0-9]*}}
+// CHECK-USE-PPC64LE-SAME: (%struct.ST* noundef [[THIS:%.*]]) #[[ATTR1]] {
+// CHECK-USE-PPC64LE-NEXT: entry:
+// CHECK-USE-PPC64LE-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.ST*, align 8
+// CHECK-USE-PPC64LE-NEXT: store %struct.ST* [[THIS]], %struct.ST** [[THIS_ADDR]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[TMP0:%.*]] = load %struct.ST*, %struct.ST** [[THIS_ADDR]], align 8
+// CHECK-USE-PPC64LE-NEXT: [[I:%.*]] = getelementptr inbounds [[STRUCT_ST:%.*]], %struct.ST* [[TMP0]], i32 0, i32 0
+// CHECK-USE-PPC64LE-NEXT: [[TMP1:%.*]] = load i32, i32* [[I]], align 4
+// CHECK-USE-PPC64LE-NEXT: [[INC:%.*]] = add nsw i32 [[TMP1]], 1
+// CHECK-USE-PPC64LE-NEXT: store i32 [[INC]], i32* [[I]], align 4
+// CHECK-USE-PPC64LE-NEXT: [[J:%.*]] = getelementptr inbounds [[STRUCT_ST]], %struct.ST* [[TMP0]], i32 0, i32 1
+// CHECK-USE-PPC64LE-NEXT: [[TMP2:%.*]] = load i32, i32* [[J]], align 4
+// CHECK-USE-PPC64LE-NEXT: [[INC1:%.*]] = add nsw i32 [[TMP2]], 1
+// CHECK-USE-PPC64LE-NEXT: store i32 [[INC1]], i32* [[J]], align 4
+// CHECK-USE-PPC64LE-NEXT: ret void
+//
+//
+// CHECK-USE-PPC64LE-LABEL: define {{[^@]+}}@.omp_offloading.requires_reg
+// CHECK-USE-PPC64LE-SAME: () #[[ATTR4:[0-9]+]] {
+// CHECK-USE-PPC64LE-NEXT: entry:
+// CHECK-USE-PPC64LE-NEXT: call void @__tgt_register_requires(i64 1)
+// CHECK-USE-PPC64LE-NEXT: ret void
+//
+//
+// CHECK-USE-I386-LABEL: define {{[^@]+}}@_Z20explicit_maps_singlei
+// CHECK-USE-I386-SAME: (i32 noundef [[II:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK-USE-I386-NEXT: entry:
// CHECK-USE-I386-NEXT: [[II_ADDR:%.*]] = alloca i32, align 4
// CHECK-USE-I386-NEXT: [[A:%.*]] = alloca i32, align 4
@@ -293,7 +514,7 @@ struct ST {
// CHECK-USE-I386-NEXT: [[DOTOFFLOAD_BASEPTRS3:%.*]] = alloca [1 x i8*], align 4
// CHECK-USE-I386-NEXT: [[DOTOFFLOAD_PTRS4:%.*]] = alloca [1 x i8*], align 4
// CHECK-USE-I386-NEXT: [[DOTOFFLOAD_MAPPERS5:%.*]] = alloca [1 x i8*], align 4
-// CHECK-USE-I386-NEXT: store i32 [[II:%.*]], i32* [[II_ADDR]], align 4
+// CHECK-USE-I386-NEXT: store i32 [[II]], i32* [[II_ADDR]], align 4
// CHECK-USE-I386-NEXT: [[TMP0:%.*]] = load i32, i32* [[II_ADDR]], align 4
// CHECK-USE-I386-NEXT: store i32 [[TMP0]], i32* [[A]], align 4
// CHECK-USE-I386-NEXT: [[I:%.*]] = getelementptr inbounds [[STRUCT_ST]], %struct.ST* [[ST1]], i32 0, i32 0
@@ -381,7 +602,7 @@ struct ST {
// CHECK-USE-I386-NEXT: [[TMP55:%.*]] = getelementptr inbounds [7 x i64], [7 x i64]* [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK-USE-I386-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK-USE-I386-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK-USE-I386-NEXT: store i32 1, i32* [[TMP56]], align 4
+// CHECK-USE-I386-NEXT: store i32 2, i32* [[TMP56]], align 4
// CHECK-USE-I386-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 1
// CHECK-USE-I386-NEXT: store i32 7, i32* [[TMP57]], align 4
// CHECK-USE-I386-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 2
@@ -398,52 +619,227 @@ struct ST {
// CHECK-USE-I386-NEXT: store i8** null, i8*** [[TMP63]], align 4
// CHECK-USE-I386-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 8
// CHECK-USE-I386-NEXT: store i64 0, i64* [[TMP64]], align 8
-// CHECK-USE-I386-NEXT: [[TMP65:%.*]] = call i32 @__tgt_target_kernel(%struct.ident_t* @[[GLOB1:[0-9]+]], i64 -1, i32 -1, i32 0, i8* @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l782.region_id, %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]])
-// CHECK-USE-I386-NEXT: [[TMP66:%.*]] = icmp ne i32 [[TMP65]], 0
-// CHECK-USE-I386-NEXT: br i1 [[TMP66]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK-USE-I386-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK-USE-I386-NEXT: store i64 0, i64* [[TMP65]], align 8
+// CHECK-USE-I386-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK-USE-I386-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], [3 x i32]* [[TMP66]], align 4
+// CHECK-USE-I386-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK-USE-I386-NEXT: store [3 x i32] zeroinitializer, [3 x i32]* [[TMP67]], align 4
+// CHECK-USE-I386-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK-USE-I386-NEXT: store i32 0, i32* [[TMP68]], align 4
+// CHECK-USE-I386-NEXT: [[TMP69:%.*]] = call i32 @__tgt_target_kernel(%struct.ident_t* @[[GLOB1:[0-9]+]], i64 -1, i32 -1, i32 0, i8* @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l{{[0-9]*}}.region_id, %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]])
+// CHECK-USE-I386-NEXT: [[TMP70:%.*]] = icmp ne i32 [[TMP69]], 0
+// CHECK-USE-I386-NEXT: br i1 [[TMP70]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK-USE-I386: omp_offload.failed:
-// CHECK-USE-I386-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l782(%struct.ST* [[ST1]], i32* [[A]], %struct.ST* [[ST2]]) #[[ATTR3:[0-9]+]]
+// CHECK-USE-I386-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l{{[0-9]*}}(%struct.ST* [[ST1]], i32* [[A]], %struct.ST* [[ST2]]) #[[ATTR3:[0-9]+]]
// CHECK-USE-I386-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK-USE-I386: omp_offload.cont:
-// CHECK-USE-I386-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK-USE-I386-NEXT: [[TMP68:%.*]] = bitcast i8** [[TMP67]] to i32**
-// CHECK-USE-I386-NEXT: store i32* [[A]], i32** [[TMP68]], align 4
-// CHECK-USE-I386-NEXT: [[TMP69:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK-USE-I386-NEXT: [[TMP70:%.*]] = bitcast i8** [[TMP69]] to i32**
-// CHECK-USE-I386-NEXT: store i32* [[A]], i32** [[TMP70]], align 4
-// CHECK-USE-I386-NEXT: [[TMP71:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
-// CHECK-USE-I386-NEXT: store i8* null, i8** [[TMP71]], align 4
-// CHECK-USE-I386-NEXT: [[TMP72:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK-USE-I386-NEXT: [[TMP71:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK-USE-I386-NEXT: [[TMP72:%.*]] = bitcast i8** [[TMP71]] to i32**
+// CHECK-USE-I386-NEXT: store i32* [[A]], i32** [[TMP72]], align 4
// CHECK-USE-I386-NEXT: [[TMP73:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK-USE-I386-NEXT: [[TMP74:%.*]] = bitcast i8** [[TMP73]] to i32**
+// CHECK-USE-I386-NEXT: store i32* [[A]], i32** [[TMP74]], align 4
+// CHECK-USE-I386-NEXT: [[TMP75:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
+// CHECK-USE-I386-NEXT: store i8* null, i8** [[TMP75]], align 4
+// CHECK-USE-I386-NEXT: [[TMP76:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK-USE-I386-NEXT: [[TMP77:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK-USE-I386-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK-USE-I386-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK-USE-I386-NEXT: store i32 1, i32* [[TMP74]], align 4
-// CHECK-USE-I386-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK-USE-I386-NEXT: store i32 1, i32* [[TMP75]], align 4
-// CHECK-USE-I386-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK-USE-I386-NEXT: store i8** [[TMP72]], i8*** [[TMP76]], align 4
-// CHECK-USE-I386-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK-USE-I386-NEXT: store i8** [[TMP73]], i8*** [[TMP77]], align 4
-// CHECK-USE-I386-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK-USE-I386-NEXT: store i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_sizes.1, i32 0, i32 0), i64** [[TMP78]], align 4
-// CHECK-USE-I386-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK-USE-I386-NEXT: store i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_maptypes.2, i32 0, i32 0), i64** [[TMP79]], align 4
-// CHECK-USE-I386-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK-USE-I386-NEXT: store i8** null, i8*** [[TMP80]], align 4
-// CHECK-USE-I386-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK-USE-I386-NEXT: store i8** null, i8*** [[TMP81]], align 4
-// CHECK-USE-I386-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK-USE-I386-NEXT: store i64 0, i64* [[TMP82]], align 8
-// CHECK-USE-I386-NEXT: [[TMP83:%.*]] = call i32 @__tgt_target_kernel(%struct.ident_t* @[[GLOB1]], i64 -1, i32 -1, i32 0, i8* @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l796.region_id, %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]])
-// CHECK-USE-I386-NEXT: [[TMP84:%.*]] = icmp ne i32 [[TMP83]], 0
-// CHECK-USE-I386-NEXT: br i1 [[TMP84]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK-USE-I386-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK-USE-I386-NEXT: store i32 2, i32* [[TMP78]], align 4
+// CHECK-USE-I386-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK-USE-I386-NEXT: store i32 1, i32* [[TMP79]], align 4
+// CHECK-USE-I386-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK-USE-I386-NEXT: store i8** [[TMP76]], i8*** [[TMP80]], align 4
+// CHECK-USE-I386-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK-USE-I386-NEXT: store i8** [[TMP77]], i8*** [[TMP81]], align 4
+// CHECK-USE-I386-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK-USE-I386-NEXT: store i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_sizes.1, i32 0, i32 0), i64** [[TMP82]], align 4
+// CHECK-USE-I386-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK-USE-I386-NEXT: store i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_maptypes.2, i32 0, i32 0), i64** [[TMP83]], align 4
+// CHECK-USE-I386-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK-USE-I386-NEXT: store i8** null, i8*** [[TMP84]], align 4
+// CHECK-USE-I386-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK-USE-I386-NEXT: store i8** null, i8*** [[TMP85]], align 4
+// CHECK-USE-I386-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK-USE-I386-NEXT: store i64 0, i64* [[TMP86]], align 8
+// CHECK-USE-I386-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK-USE-I386-NEXT: store i64 0, i64* [[TMP87]], align 8
+// CHECK-USE-I386-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK-USE-I386-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], [3 x i32]* [[TMP88]], align 4
+// CHECK-USE-I386-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK-USE-I386-NEXT: store [3 x i32] zeroinitializer, [3 x i32]* [[TMP89]], align 4
+// CHECK-USE-I386-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK-USE-I386-NEXT: store i32 0, i32* [[TMP90]], align 4
+// CHECK-USE-I386-NEXT: [[TMP91:%.*]] = call i32 @__tgt_target_kernel(%struct.ident_t* @[[GLOB1]], i64 -1, i32 -1, i32 0, i8* @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l{{[0-9]*}}.region_id, %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]])
+// CHECK-USE-I386-NEXT: [[TMP92:%.*]] = icmp ne i32 [[TMP91]], 0
+// CHECK-USE-I386-NEXT: br i1 [[TMP92]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK-USE-I386: omp_offload.failed7:
-// CHECK-USE-I386-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l796(i32* [[A]]) #[[ATTR3]]
+// CHECK-USE-I386-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l{{[0-9]*}}(i32* [[A]]) #[[ATTR3]]
// CHECK-USE-I386-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK-USE-I386: omp_offload.cont8:
// CHECK-USE-I386-NEXT: ret void
//
-// CHECK-NOUSE-PPC64LE-LABEL: @_Z20explicit_maps_singlei(
+//
+// CHECK-USE-I386-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l{{[0-9]*}}
+// CHECK-USE-I386-SAME: (%struct.ST* noundef nonnull align 4 dereferenceable(8) [[ST1:%.*]], i32* noundef nonnull align 4 dereferenceable(4) [[A:%.*]], %struct.ST* noundef nonnull align 4 dereferenceable(8) [[ST2:%.*]]) #[[ATTR1:[0-9]+]] {
+// CHECK-USE-I386-NEXT: entry:
+// CHECK-USE-I386-NEXT: [[ST1_ADDR:%.*]] = alloca %struct.ST*, align 4
+// CHECK-USE-I386-NEXT: [[A_ADDR:%.*]] = alloca i32*, align 4
+// CHECK-USE-I386-NEXT: [[ST2_ADDR:%.*]] = alloca %struct.ST*, align 4
+// CHECK-USE-I386-NEXT: store %struct.ST* [[ST1]], %struct.ST** [[ST1_ADDR]], align 4
+// CHECK-USE-I386-NEXT: store i32* [[A]], i32** [[A_ADDR]], align 4
+// CHECK-USE-I386-NEXT: store %struct.ST* [[ST2]], %struct.ST** [[ST2_ADDR]], align 4
+// CHECK-USE-I386-NEXT: [[TMP0:%.*]] = load %struct.ST*, %struct.ST** [[ST1_ADDR]], align 4
+// CHECK-USE-I386-NEXT: [[TMP1:%.*]] = load i32*, i32** [[A_ADDR]], align 4
+// CHECK-USE-I386-NEXT: [[TMP2:%.*]] = load %struct.ST*, %struct.ST** [[ST2_ADDR]], align 4
+// CHECK-USE-I386-NEXT: [[I:%.*]] = getelementptr inbounds [[STRUCT_ST:%.*]], %struct.ST* [[TMP0]], i32 0, i32 0
+// CHECK-USE-I386-NEXT: [[TMP3:%.*]] = load i32, i32* [[I]], align 4
+// CHECK-USE-I386-NEXT: [[INC:%.*]] = add nsw i32 [[TMP3]], 1
+// CHECK-USE-I386-NEXT: store i32 [[INC]], i32* [[I]], align 4
+// CHECK-USE-I386-NEXT: [[TMP4:%.*]] = load i32, i32* [[TMP1]], align 4
+// CHECK-USE-I386-NEXT: [[INC1:%.*]] = add nsw i32 [[TMP4]], 1
+// CHECK-USE-I386-NEXT: store i32 [[INC1]], i32* [[TMP1]], align 4
+// CHECK-USE-I386-NEXT: [[J:%.*]] = getelementptr inbounds [[STRUCT_ST]], %struct.ST* [[TMP0]], i32 0, i32 1
+// CHECK-USE-I386-NEXT: [[TMP5:%.*]] = load i32, i32* [[J]], align 4
+// CHECK-USE-I386-NEXT: [[INC2:%.*]] = add nsw i32 [[TMP5]], 1
+// CHECK-USE-I386-NEXT: store i32 [[INC2]], i32* [[J]], align 4
+// CHECK-USE-I386-NEXT: [[I3:%.*]] = getelementptr inbounds [[STRUCT_ST]], %struct.ST* [[TMP2]], i32 0, i32 0
+// CHECK-USE-I386-NEXT: [[TMP6:%.*]] = load i32, i32* [[I3]], align 4
+// CHECK-USE-I386-NEXT: [[INC4:%.*]] = add nsw i32 [[TMP6]], 1
+// CHECK-USE-I386-NEXT: store i32 [[INC4]], i32* [[I3]], align 4
+// CHECK-USE-I386-NEXT: [[J5:%.*]] = getelementptr inbounds [[STRUCT_ST]], %struct.ST* [[TMP2]], i32 0, i32 1
+// CHECK-USE-I386-NEXT: [[TMP7:%.*]] = load i32, i32* [[J5]], align 4
+// CHECK-USE-I386-NEXT: [[INC6:%.*]] = add nsw i32 [[TMP7]], 1
+// CHECK-USE-I386-NEXT: store i32 [[INC6]], i32* [[J5]], align 4
+// CHECK-USE-I386-NEXT: ret void
+//
+//
+// CHECK-USE-I386-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l{{[0-9]*}}
+// CHECK-USE-I386-SAME: (i32* noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR1]] {
+// CHECK-USE-I386-NEXT: entry:
+// CHECK-USE-I386-NEXT: [[A_ADDR:%.*]] = alloca i32*, align 4
+// CHECK-USE-I386-NEXT: store i32* [[A]], i32** [[A_ADDR]], align 4
+// CHECK-USE-I386-NEXT: [[TMP0:%.*]] = load i32*, i32** [[A_ADDR]], align 4
+// CHECK-USE-I386-NEXT: [[TMP1:%.*]] = load i32, i32* [[TMP0]], align 4
+// CHECK-USE-I386-NEXT: [[INC:%.*]] = add nsw i32 [[TMP1]], 1
+// CHECK-USE-I386-NEXT: store i32 [[INC]], i32* [[TMP0]], align 4
+// CHECK-USE-I386-NEXT: ret void
+//
+//
+// CHECK-USE-I386-LABEL: define {{[^@]+}}@_ZN2ST20test_present_membersEv
+// CHECK-USE-I386-SAME: (%struct.ST* noundef nonnull align 4 dereferenceable(8) [[THIS:%.*]]) #[[ATTR0]] align 2 {
+// CHECK-USE-I386-NEXT: entry:
+// CHECK-USE-I386-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.ST*, align 4
+// CHECK-USE-I386-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [3 x i8*], align 4
+// CHECK-USE-I386-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [3 x i8*], align 4
+// CHECK-USE-I386-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [3 x i8*], align 4
+// CHECK-USE-I386-NEXT: [[DOTOFFLOAD_SIZES:%.*]] = alloca [3 x i64], align 4
+// CHECK-USE-I386-NEXT: store %struct.ST* [[THIS]], %struct.ST** [[THIS_ADDR]], align 4
+// CHECK-USE-I386-NEXT: [[THIS1:%.*]] = load %struct.ST*, %struct.ST** [[THIS_ADDR]], align 4
+// CHECK-USE-I386-NEXT: [[I:%.*]] = getelementptr inbounds [[STRUCT_ST:%.*]], %struct.ST* [[THIS1]], i32 0, i32 0
+// CHECK-USE-I386-NEXT: [[J:%.*]] = getelementptr inbounds [[STRUCT_ST]], %struct.ST* [[THIS1]], i32 0, i32 1
+// CHECK-USE-I386-NEXT: [[TMP0:%.*]] = getelementptr i32, i32* [[J]], i32 1
+// CHECK-USE-I386-NEXT: [[TMP1:%.*]] = bitcast i32* [[I]] to i8*
+// CHECK-USE-I386-NEXT: [[TMP2:%.*]] = bitcast i32* [[TMP0]] to i8*
+// CHECK-USE-I386-NEXT: [[TMP3:%.*]] = ptrtoint i8* [[TMP2]] to i64
+// CHECK-USE-I386-NEXT: [[TMP4:%.*]] = ptrtoint i8* [[TMP1]] to i64
+// CHECK-USE-I386-NEXT: [[TMP5:%.*]] = sub i64 [[TMP3]], [[TMP4]]
+// CHECK-USE-I386-NEXT: [[TMP6:%.*]] = sdiv exact i64 [[TMP5]], ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)
+// CHECK-USE-I386-NEXT: [[TMP7:%.*]] = bitcast [3 x i64]* [[DOTOFFLOAD_SIZES]] to i8*
+// CHECK-USE-I386-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP7]], i8* align 4 bitcast ([3 x i64]* @.offload_sizes.3 to i8*), i32 24, i1 false)
+// CHECK-USE-I386-NEXT: [[TMP8:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK-USE-I386-NEXT: [[TMP9:%.*]] = bitcast i8** [[TMP8]] to %struct.ST**
+// CHECK-USE-I386-NEXT: store %struct.ST* [[THIS1]], %struct.ST** [[TMP9]], align 4
+// CHECK-USE-I386-NEXT: [[TMP10:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK-USE-I386-NEXT: [[TMP11:%.*]] = bitcast i8** [[TMP10]] to i32**
+// CHECK-USE-I386-NEXT: store i32* [[I]], i32** [[TMP11]], align 4
+// CHECK-USE-I386-NEXT: [[TMP12:%.*]] = getelementptr inbounds [3 x i64], [3 x i64]* [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK-USE-I386-NEXT: store i64 [[TMP6]], i64* [[TMP12]], align 4
+// CHECK-USE-I386-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
+// CHECK-USE-I386-NEXT: store i8* null, i8** [[TMP13]], align 4
+// CHECK-USE-I386-NEXT: [[TMP14:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
+// CHECK-USE-I386-NEXT: [[TMP15:%.*]] = bitcast i8** [[TMP14]] to %struct.ST**
+// CHECK-USE-I386-NEXT: store %struct.ST* [[THIS1]], %struct.ST** [[TMP15]], align 4
+// CHECK-USE-I386-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_PTRS]], i32 0, i32 1
+// CHECK-USE-I386-NEXT: [[TMP17:%.*]] = bitcast i8** [[TMP16]] to i32**
+// CHECK-USE-I386-NEXT: store i32* [[I]], i32** [[TMP17]], align 4
+// CHECK-USE-I386-NEXT: [[TMP18:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_MAPPERS]], i32 0, i32 1
+// CHECK-USE-I386-NEXT: store i8* null, i8** [[TMP18]], align 4
+// CHECK-USE-I386-NEXT: [[TMP19:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
+// CHECK-USE-I386-NEXT: [[TMP20:%.*]] = bitcast i8** [[TMP19]] to %struct.ST**
+// CHECK-USE-I386-NEXT: store %struct.ST* [[THIS1]], %struct.ST** [[TMP20]], align 4
+// CHECK-USE-I386-NEXT: [[TMP21:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_PTRS]], i32 0, i32 2
+// CHECK-USE-I386-NEXT: [[TMP22:%.*]] = bitcast i8** [[TMP21]] to i32**
+// CHECK-USE-I386-NEXT: store i32* [[J]], i32** [[TMP22]], align 4
+// CHECK-USE-I386-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_MAPPERS]], i32 0, i32 2
+// CHECK-USE-I386-NEXT: store i8* null, i8** [[TMP23]], align 4
+// CHECK-USE-I386-NEXT: [[TMP24:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK-USE-I386-NEXT: [[TMP25:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK-USE-I386-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x i64], [3 x i64]* [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK-USE-I386-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
+// CHECK-USE-I386-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK-USE-I386-NEXT: store i32 2, i32* [[TMP27]], align 4
+// CHECK-USE-I386-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK-USE-I386-NEXT: store i32 3, i32* [[TMP28]], align 4
+// CHECK-USE-I386-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK-USE-I386-NEXT: store i8** [[TMP24]], i8*** [[TMP29]], align 4
+// CHECK-USE-I386-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK-USE-I386-NEXT: store i8** [[TMP25]], i8*** [[TMP30]], align 4
+// CHECK-USE-I386-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK-USE-I386-NEXT: store i64* [[TMP26]], i64** [[TMP31]], align 4
+// CHECK-USE-I386-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK-USE-I386-NEXT: store i64* getelementptr inbounds ([3 x i64], [3 x i64]* @.offload_maptypes.4, i32 0, i32 0), i64** [[TMP32]], align 4
+// CHECK-USE-I386-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 6
+// CHECK-USE-I386-NEXT: store i8** null, i8*** [[TMP33]], align 4
+// CHECK-USE-I386-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK-USE-I386-NEXT: store i8** null, i8*** [[TMP34]], align 4
+// CHECK-USE-I386-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK-USE-I386-NEXT: store i64 0, i64* [[TMP35]], align 8
+// CHECK-USE-I386-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK-USE-I386-NEXT: store i64 0, i64* [[TMP36]], align 8
+// CHECK-USE-I386-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK-USE-I386-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], [3 x i32]* [[TMP37]], align 4
+// CHECK-USE-I386-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK-USE-I386-NEXT: store [3 x i32] zeroinitializer, [3 x i32]* [[TMP38]], align 4
+// CHECK-USE-I386-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK-USE-I386-NEXT: store i32 0, i32* [[TMP39]], align 4
+// CHECK-USE-I386-NEXT: [[TMP40:%.*]] = call i32 @__tgt_target_kernel(%struct.ident_t* @[[GLOB1]], i64 -1, i32 -1, i32 0, i8* @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2ST20test_present_membersEv_l{{[0-9]*}}.region_id, %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]])
+// CHECK-USE-I386-NEXT: [[TMP41:%.*]] = icmp ne i32 [[TMP40]], 0
+// CHECK-USE-I386-NEXT: br i1 [[TMP41]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK-USE-I386: omp_offload.failed:
+// CHECK-USE-I386-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2ST20test_present_membersEv_l{{[0-9]*}}(%struct.ST* [[THIS1]]) #[[ATTR3]]
+// CHECK-USE-I386-NEXT: br label [[OMP_OFFLOAD_CONT]]
+// CHECK-USE-I386: omp_offload.cont:
+// CHECK-USE-I386-NEXT: ret void
+//
+//
+// CHECK-USE-I386-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2ST20test_present_membersEv_l{{[0-9]*}}
+// CHECK-USE-I386-SAME: (%struct.ST* noundef [[THIS:%.*]]) #[[ATTR1]] {
+// CHECK-USE-I386-NEXT: entry:
+// CHECK-USE-I386-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.ST*, align 4
+// CHECK-USE-I386-NEXT: store %struct.ST* [[THIS]], %struct.ST** [[THIS_ADDR]], align 4
+// CHECK-USE-I386-NEXT: [[TMP0:%.*]] = load %struct.ST*, %struct.ST** [[THIS_ADDR]], align 4
+// CHECK-USE-I386-NEXT: [[I:%.*]] = getelementptr inbounds [[STRUCT_ST:%.*]], %struct.ST* [[TMP0]], i32 0, i32 0
+// CHECK-USE-I386-NEXT: [[TMP1:%.*]] = load i32, i32* [[I]], align 4
+// CHECK-USE-I386-NEXT: [[INC:%.*]] = add nsw i32 [[TMP1]], 1
+// CHECK-USE-I386-NEXT: store i32 [[INC]], i32* [[I]], align 4
+// CHECK-USE-I386-NEXT: [[J:%.*]] = getelementptr inbounds [[STRUCT_ST]], %struct.ST* [[TMP0]], i32 0, i32 1
+// CHECK-USE-I386-NEXT: [[TMP2:%.*]] = load i32, i32* [[J]], align 4
+// CHECK-USE-I386-NEXT: [[INC1:%.*]] = add nsw i32 [[TMP2]], 1
+// CHECK-USE-I386-NEXT: store i32 [[INC1]], i32* [[J]], align 4
+// CHECK-USE-I386-NEXT: ret void
+//
+//
+// CHECK-USE-I386-LABEL: define {{[^@]+}}@.omp_offloading.requires_reg
+// CHECK-USE-I386-SAME: () #[[ATTR4:[0-9]+]] {
+// CHECK-USE-I386-NEXT: entry:
+// CHECK-USE-I386-NEXT: call void @__tgt_register_requires(i64 1)
+// CHECK-USE-I386-NEXT: ret void
+//
+//
+// CHECK-NOUSE-PPC64LE-LABEL: define {{[^@]+}}@_Z20explicit_maps_singlei
+// CHECK-NOUSE-PPC64LE-SAME: (i32 noundef signext [[II:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK-NOUSE-PPC64LE-NEXT: entry:
// CHECK-NOUSE-PPC64LE-NEXT: [[II_ADDR:%.*]] = alloca i32, align 4
// CHECK-NOUSE-PPC64LE-NEXT: [[A:%.*]] = alloca i32, align 4
@@ -456,7 +852,7 @@ struct ST {
// CHECK-NOUSE-PPC64LE-NEXT: [[DOTOFFLOAD_BASEPTRS3:%.*]] = alloca [1 x i8*], align 8
// CHECK-NOUSE-PPC64LE-NEXT: [[DOTOFFLOAD_PTRS4:%.*]] = alloca [1 x i8*], align 8
// CHECK-NOUSE-PPC64LE-NEXT: [[DOTOFFLOAD_MAPPERS5:%.*]] = alloca [1 x i8*], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: store i32 [[II:%.*]], i32* [[II_ADDR]], align 4
+// CHECK-NOUSE-PPC64LE-NEXT: store i32 [[II]], i32* [[II_ADDR]], align 4
// CHECK-NOUSE-PPC64LE-NEXT: [[TMP0:%.*]] = load i32, i32* [[II_ADDR]], align 4
// CHECK-NOUSE-PPC64LE-NEXT: store i32 [[TMP0]], i32* [[A]], align 4
// CHECK-NOUSE-PPC64LE-NEXT: [[I:%.*]] = getelementptr inbounds [[STRUCT_ST]], %struct.ST* [[ST1]], i32 0, i32 0
@@ -544,7 +940,7 @@ struct ST {
// CHECK-NOUSE-PPC64LE-NEXT: [[TMP55:%.*]] = getelementptr inbounds [7 x i64], [7 x i64]* [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK-NOUSE-PPC64LE-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK-NOUSE-PPC64LE-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK-NOUSE-PPC64LE-NEXT: store i32 1, i32* [[TMP56]], align 4
+// CHECK-NOUSE-PPC64LE-NEXT: store i32 2, i32* [[TMP56]], align 4
// CHECK-NOUSE-PPC64LE-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 1
// CHECK-NOUSE-PPC64LE-NEXT: store i32 7, i32* [[TMP57]], align 4
// CHECK-NOUSE-PPC64LE-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 2
@@ -561,84 +957,214 @@ struct ST {
// CHECK-NOUSE-PPC64LE-NEXT: store i8** null, i8*** [[TMP63]], align 8
// CHECK-NOUSE-PPC64LE-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 8
// CHECK-NOUSE-PPC64LE-NEXT: store i64 0, i64* [[TMP64]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP65:%.*]] = call i32 @__tgt_target_kernel(%struct.ident_t* @[[GLOB1:[0-9]+]], i64 -1, i32 -1, i32 0, i8* @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l782.region_id, %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]])
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP66:%.*]] = icmp ne i32 [[TMP65]], 0
-// CHECK-NOUSE-PPC64LE-NEXT: br i1 [[TMP66]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK-NOUSE-PPC64LE-NEXT: store i64 0, i64* [[TMP65]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK-NOUSE-PPC64LE-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], [3 x i32]* [[TMP66]], align 4
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK-NOUSE-PPC64LE-NEXT: store [3 x i32] zeroinitializer, [3 x i32]* [[TMP67]], align 4
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK-NOUSE-PPC64LE-NEXT: store i32 0, i32* [[TMP68]], align 4
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP69:%.*]] = call i32 @__tgt_target_kernel(%struct.ident_t* @[[GLOB1:[0-9]+]], i64 -1, i32 -1, i32 0, i8* @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l{{[0-9]*}}.region_id, %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]])
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP70:%.*]] = icmp ne i32 [[TMP69]], 0
+// CHECK-NOUSE-PPC64LE-NEXT: br i1 [[TMP70]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK-NOUSE-PPC64LE: omp_offload.failed:
-// CHECK-NOUSE-PPC64LE-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l782() #[[ATTR3:[0-9]+]]
+// CHECK-NOUSE-PPC64LE-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l{{[0-9]*}}() #[[ATTR3:[0-9]+]]
// CHECK-NOUSE-PPC64LE-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK-NOUSE-PPC64LE: omp_offload.cont:
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP68:%.*]] = bitcast i8** [[TMP67]] to i32**
-// CHECK-NOUSE-PPC64LE-NEXT: store i32* [[A]], i32** [[TMP68]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP69:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP70:%.*]] = bitcast i8** [[TMP69]] to i32**
-// CHECK-NOUSE-PPC64LE-NEXT: store i32* [[A]], i32** [[TMP70]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP71:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
-// CHECK-NOUSE-PPC64LE-NEXT: store i8* null, i8** [[TMP71]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP72:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP71:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP72:%.*]] = bitcast i8** [[TMP71]] to i32**
+// CHECK-NOUSE-PPC64LE-NEXT: store i32* [[A]], i32** [[TMP72]], align 8
// CHECK-NOUSE-PPC64LE-NEXT: [[TMP73:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP74:%.*]] = bitcast i8** [[TMP73]] to i32**
+// CHECK-NOUSE-PPC64LE-NEXT: store i32* [[A]], i32** [[TMP74]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP75:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
+// CHECK-NOUSE-PPC64LE-NEXT: store i8* null, i8** [[TMP75]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP76:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP77:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK-NOUSE-PPC64LE-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK-NOUSE-PPC64LE-NEXT: store i32 1, i32* [[TMP74]], align 4
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK-NOUSE-PPC64LE-NEXT: store i32 1, i32* [[TMP75]], align 4
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK-NOUSE-PPC64LE-NEXT: store i8** [[TMP72]], i8*** [[TMP76]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK-NOUSE-PPC64LE-NEXT: store i8** [[TMP73]], i8*** [[TMP77]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK-NOUSE-PPC64LE-NEXT: store i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_sizes.1, i32 0, i32 0), i64** [[TMP78]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK-NOUSE-PPC64LE-NEXT: store i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_maptypes.2, i32 0, i32 0), i64** [[TMP79]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK-NOUSE-PPC64LE-NEXT: store i8** null, i8*** [[TMP80]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK-NOUSE-PPC64LE-NEXT: store i8** null, i8*** [[TMP81]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK-NOUSE-PPC64LE-NEXT: store i64 0, i64* [[TMP82]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP83:%.*]] = call i32 @__tgt_target_kernel(%struct.ident_t* @[[GLOB1]], i64 -1, i32 -1, i32 0, i8* @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l796.region_id, %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]])
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP84:%.*]] = icmp ne i32 [[TMP83]], 0
-// CHECK-NOUSE-PPC64LE-NEXT: br i1 [[TMP84]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK-NOUSE-PPC64LE-NEXT: store i32 2, i32* [[TMP78]], align 4
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK-NOUSE-PPC64LE-NEXT: store i32 1, i32* [[TMP79]], align 4
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK-NOUSE-PPC64LE-NEXT: store i8** [[TMP76]], i8*** [[TMP80]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK-NOUSE-PPC64LE-NEXT: store i8** [[TMP77]], i8*** [[TMP81]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK-NOUSE-PPC64LE-NEXT: store i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_sizes.1, i32 0, i32 0), i64** [[TMP82]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK-NOUSE-PPC64LE-NEXT: store i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_maptypes.2, i32 0, i32 0), i64** [[TMP83]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK-NOUSE-PPC64LE-NEXT: store i8** null, i8*** [[TMP84]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK-NOUSE-PPC64LE-NEXT: store i8** null, i8*** [[TMP85]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK-NOUSE-PPC64LE-NEXT: store i64 0, i64* [[TMP86]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK-NOUSE-PPC64LE-NEXT: store i64 0, i64* [[TMP87]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK-NOUSE-PPC64LE-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], [3 x i32]* [[TMP88]], align 4
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK-NOUSE-PPC64LE-NEXT: store [3 x i32] zeroinitializer, [3 x i32]* [[TMP89]], align 4
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK-NOUSE-PPC64LE-NEXT: store i32 0, i32* [[TMP90]], align 4
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP91:%.*]] = call i32 @__tgt_target_kernel(%struct.ident_t* @[[GLOB1]], i64 -1, i32 -1, i32 0, i8* @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l{{[0-9]*}}.region_id, %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]])
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP92:%.*]] = icmp ne i32 [[TMP91]], 0
+// CHECK-NOUSE-PPC64LE-NEXT: br i1 [[TMP92]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK-NOUSE-PPC64LE: omp_offload.failed7:
-// CHECK-NOUSE-PPC64LE-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l796() #[[ATTR3]]
+// CHECK-NOUSE-PPC64LE-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l{{[0-9]*}}() #[[ATTR3]]
// CHECK-NOUSE-PPC64LE-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK-NOUSE-PPC64LE: omp_offload.cont8:
// CHECK-NOUSE-PPC64LE-NEXT: ret void
//
-// CHECK-NOUSE-I386-LABEL: @_Z20explicit_maps_singlei(
-// CHECK-NOUSE-I386-NEXT: entry:
-// CHECK-NOUSE-I386-NEXT: [[II_ADDR:%.*]] = alloca i32, align 4
-// CHECK-NOUSE-I386-NEXT: [[A:%.*]] = alloca i32, align 4
-// CHECK-NOUSE-I386-NEXT: [[ST1:%.*]] = alloca [[STRUCT_ST:%.*]], align 4
-// CHECK-NOUSE-I386-NEXT: [[ST2:%.*]] = alloca [[STRUCT_ST]], align 4
-// CHECK-NOUSE-I386-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [7 x i8*], align 4
-// CHECK-NOUSE-I386-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [7 x i8*], align 4
-// CHECK-NOUSE-I386-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [7 x i8*], align 4
-// CHECK-NOUSE-I386-NEXT: [[DOTOFFLOAD_SIZES:%.*]] = alloca [7 x i64], align 4
-// CHECK-NOUSE-I386-NEXT: [[DOTOFFLOAD_BASEPTRS3:%.*]] = alloca [1 x i8*], align 4
-// CHECK-NOUSE-I386-NEXT: [[DOTOFFLOAD_PTRS4:%.*]] = alloca [1 x i8*], align 4
-// CHECK-NOUSE-I386-NEXT: [[DOTOFFLOAD_MAPPERS5:%.*]] = alloca [1 x i8*], align 4
-// CHECK-NOUSE-I386-NEXT: store i32 [[II:%.*]], i32* [[II_ADDR]], align 4
-// CHECK-NOUSE-I386-NEXT: [[TMP0:%.*]] = load i32, i32* [[II_ADDR]], align 4
-// CHECK-NOUSE-I386-NEXT: store i32 [[TMP0]], i32* [[A]], align 4
-// CHECK-NOUSE-I386-NEXT: [[I:%.*]] = getelementptr inbounds [[STRUCT_ST]], %struct.ST* [[ST1]], i32 0, i32 0
-// CHECK-NOUSE-I386-NEXT: [[J:%.*]] = getelementptr inbounds [[STRUCT_ST]], %struct.ST* [[ST1]], i32 0, i32 1
-// CHECK-NOUSE-I386-NEXT: [[TMP1:%.*]] = getelementptr i32, i32* [[J]], i32 1
-// CHECK-NOUSE-I386-NEXT: [[TMP2:%.*]] = bitcast i32* [[I]] to i8*
-// CHECK-NOUSE-I386-NEXT: [[TMP3:%.*]] = bitcast i32* [[TMP1]] to i8*
-// CHECK-NOUSE-I386-NEXT: [[TMP4:%.*]] = ptrtoint i8* [[TMP3]] to i64
-// CHECK-NOUSE-I386-NEXT: [[TMP5:%.*]] = ptrtoint i8* [[TMP2]] to i64
-// CHECK-NOUSE-I386-NEXT: [[TMP6:%.*]] = sub i64 [[TMP4]], [[TMP5]]
-// CHECK-NOUSE-I386-NEXT: [[TMP7:%.*]] = sdiv exact i64 [[TMP6]], ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)
-// CHECK-NOUSE-I386-NEXT: [[I1:%.*]] = getelementptr inbounds [[STRUCT_ST]], %struct.ST* [[ST2]], i32 0, i32 0
-// CHECK-NOUSE-I386-NEXT: [[J2:%.*]] = getelementptr inbounds [[STRUCT_ST]], %struct.ST* [[ST2]], i32 0, i32 1
-// CHECK-NOUSE-I386-NEXT: [[TMP8:%.*]] = getelementptr i32, i32* [[J2]], i32 1
-// CHECK-NOUSE-I386-NEXT: [[TMP9:%.*]] = bitcast i32* [[I1]] to i8*
-// CHECK-NOUSE-I386-NEXT: [[TMP10:%.*]] = bitcast i32* [[TMP8]] to i8*
-// CHECK-NOUSE-I386-NEXT: [[TMP11:%.*]] = ptrtoint i8* [[TMP10]] to i64
-// CHECK-NOUSE-I386-NEXT: [[TMP12:%.*]] = ptrtoint i8* [[TMP9]] to i64
-// CHECK-NOUSE-I386-NEXT: [[TMP13:%.*]] = sub i64 [[TMP11]], [[TMP12]]
+//
+// CHECK-NOUSE-PPC64LE-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l{{[0-9]*}}
+// CHECK-NOUSE-PPC64LE-SAME: () #[[ATTR1:[0-9]+]] {
+// CHECK-NOUSE-PPC64LE-NEXT: entry:
+// CHECK-NOUSE-PPC64LE-NEXT: ret void
+//
+//
+// CHECK-NOUSE-PPC64LE-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l{{[0-9]*}}
+// CHECK-NOUSE-PPC64LE-SAME: () #[[ATTR1]] {
+// CHECK-NOUSE-PPC64LE-NEXT: entry:
+// CHECK-NOUSE-PPC64LE-NEXT: ret void
+//
+//
+// CHECK-NOUSE-PPC64LE-LABEL: define {{[^@]+}}@_ZN2ST20test_present_membersEv
+// CHECK-NOUSE-PPC64LE-SAME: (%struct.ST* noundef nonnull align 4 dereferenceable(8) [[THIS:%.*]]) #[[ATTR0]] align 2 {
+// CHECK-NOUSE-PPC64LE-NEXT: entry:
+// CHECK-NOUSE-PPC64LE-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.ST*, align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [3 x i8*], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [3 x i8*], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [3 x i8*], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[DOTOFFLOAD_SIZES:%.*]] = alloca [3 x i64], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: store %struct.ST* [[THIS]], %struct.ST** [[THIS_ADDR]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[THIS1:%.*]] = load %struct.ST*, %struct.ST** [[THIS_ADDR]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[I:%.*]] = getelementptr inbounds [[STRUCT_ST:%.*]], %struct.ST* [[THIS1]], i32 0, i32 0
+// CHECK-NOUSE-PPC64LE-NEXT: [[J:%.*]] = getelementptr inbounds [[STRUCT_ST]], %struct.ST* [[THIS1]], i32 0, i32 1
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP0:%.*]] = getelementptr i32, i32* [[J]], i32 1
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP1:%.*]] = bitcast i32* [[I]] to i8*
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP2:%.*]] = bitcast i32* [[TMP0]] to i8*
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP3:%.*]] = ptrtoint i8* [[TMP2]] to i64
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP4:%.*]] = ptrtoint i8* [[TMP1]] to i64
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP5:%.*]] = sub i64 [[TMP3]], [[TMP4]]
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP6:%.*]] = sdiv exact i64 [[TMP5]], ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP7:%.*]] = bitcast [3 x i64]* [[DOTOFFLOAD_SIZES]] to i8*
+// CHECK-NOUSE-PPC64LE-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[TMP7]], i8* align 8 bitcast ([3 x i64]* @.offload_sizes.3 to i8*), i64 24, i1 false)
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP8:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP9:%.*]] = bitcast i8** [[TMP8]] to %struct.ST**
+// CHECK-NOUSE-PPC64LE-NEXT: store %struct.ST* [[THIS1]], %struct.ST** [[TMP9]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP10:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP11:%.*]] = bitcast i8** [[TMP10]] to i32**
+// CHECK-NOUSE-PPC64LE-NEXT: store i32* [[I]], i32** [[TMP11]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP12:%.*]] = getelementptr inbounds [3 x i64], [3 x i64]* [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK-NOUSE-PPC64LE-NEXT: store i64 [[TMP6]], i64* [[TMP12]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK-NOUSE-PPC64LE-NEXT: store i8* null, i8** [[TMP13]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP14:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP15:%.*]] = bitcast i8** [[TMP14]] to %struct.ST**
+// CHECK-NOUSE-PPC64LE-NEXT: store %struct.ST* [[THIS1]], %struct.ST** [[TMP15]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_PTRS]], i32 0, i32 1
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP17:%.*]] = bitcast i8** [[TMP16]] to i32**
+// CHECK-NOUSE-PPC64LE-NEXT: store i32* [[I]], i32** [[TMP17]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP18:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_MAPPERS]], i64 0, i64 1
+// CHECK-NOUSE-PPC64LE-NEXT: store i8* null, i8** [[TMP18]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP19:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP20:%.*]] = bitcast i8** [[TMP19]] to %struct.ST**
+// CHECK-NOUSE-PPC64LE-NEXT: store %struct.ST* [[THIS1]], %struct.ST** [[TMP20]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP21:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_PTRS]], i32 0, i32 2
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP22:%.*]] = bitcast i8** [[TMP21]] to i32**
+// CHECK-NOUSE-PPC64LE-NEXT: store i32* [[J]], i32** [[TMP22]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_MAPPERS]], i64 0, i64 2
+// CHECK-NOUSE-PPC64LE-NEXT: store i8* null, i8** [[TMP23]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP24:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP25:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x i64], [3 x i64]* [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK-NOUSE-PPC64LE-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK-NOUSE-PPC64LE-NEXT: store i32 2, i32* [[TMP27]], align 4
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK-NOUSE-PPC64LE-NEXT: store i32 3, i32* [[TMP28]], align 4
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK-NOUSE-PPC64LE-NEXT: store i8** [[TMP24]], i8*** [[TMP29]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK-NOUSE-PPC64LE-NEXT: store i8** [[TMP25]], i8*** [[TMP30]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK-NOUSE-PPC64LE-NEXT: store i64* [[TMP26]], i64** [[TMP31]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK-NOUSE-PPC64LE-NEXT: store i64* getelementptr inbounds ([3 x i64], [3 x i64]* @.offload_maptypes.4, i32 0, i32 0), i64** [[TMP32]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 6
+// CHECK-NOUSE-PPC64LE-NEXT: store i8** null, i8*** [[TMP33]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK-NOUSE-PPC64LE-NEXT: store i8** null, i8*** [[TMP34]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK-NOUSE-PPC64LE-NEXT: store i64 0, i64* [[TMP35]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK-NOUSE-PPC64LE-NEXT: store i64 0, i64* [[TMP36]], align 8
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK-NOUSE-PPC64LE-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], [3 x i32]* [[TMP37]], align 4
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK-NOUSE-PPC64LE-NEXT: store [3 x i32] zeroinitializer, [3 x i32]* [[TMP38]], align 4
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK-NOUSE-PPC64LE-NEXT: store i32 0, i32* [[TMP39]], align 4
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP40:%.*]] = call i32 @__tgt_target_kernel(%struct.ident_t* @[[GLOB1]], i64 -1, i32 -1, i32 0, i8* @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2ST20test_present_membersEv_l{{[0-9]*}}.region_id, %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]])
+// CHECK-NOUSE-PPC64LE-NEXT: [[TMP41:%.*]] = icmp ne i32 [[TMP40]], 0
+// CHECK-NOUSE-PPC64LE-NEXT: br i1 [[TMP41]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK-NOUSE-PPC64LE: omp_offload.failed:
+// CHECK-NOUSE-PPC64LE-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2ST20test_present_membersEv_l{{[0-9]*}}() #[[ATTR3]]
+// CHECK-NOUSE-PPC64LE-NEXT: br label [[OMP_OFFLOAD_CONT]]
+// CHECK-NOUSE-PPC64LE: omp_offload.cont:
+// CHECK-NOUSE-PPC64LE-NEXT: ret void
+//
+//
+// CHECK-NOUSE-PPC64LE-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2ST20test_present_membersEv_l{{[0-9]*}}
+// CHECK-NOUSE-PPC64LE-SAME: () #[[ATTR1]] {
+// CHECK-NOUSE-PPC64LE-NEXT: entry:
+// CHECK-NOUSE-PPC64LE-NEXT: ret void
+//
+//
+// CHECK-NOUSE-PPC64LE-LABEL: define {{[^@]+}}@.omp_offloading.requires_reg
+// CHECK-NOUSE-PPC64LE-SAME: () #[[ATTR4:[0-9]+]] {
+// CHECK-NOUSE-PPC64LE-NEXT: entry:
+// CHECK-NOUSE-PPC64LE-NEXT: call void @__tgt_register_requires(i64 1)
+// CHECK-NOUSE-PPC64LE-NEXT: ret void
+//
+//
+// CHECK-NOUSE-I386-LABEL: define {{[^@]+}}@_Z20explicit_maps_singlei
+// CHECK-NOUSE-I386-SAME: (i32 noundef [[II:%.*]]) #[[ATTR0:[0-9]+]] {
+// CHECK-NOUSE-I386-NEXT: entry:
+// CHECK-NOUSE-I386-NEXT: [[II_ADDR:%.*]] = alloca i32, align 4
+// CHECK-NOUSE-I386-NEXT: [[A:%.*]] = alloca i32, align 4
+// CHECK-NOUSE-I386-NEXT: [[ST1:%.*]] = alloca [[STRUCT_ST:%.*]], align 4
+// CHECK-NOUSE-I386-NEXT: [[ST2:%.*]] = alloca [[STRUCT_ST]], align 4
+// CHECK-NOUSE-I386-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [7 x i8*], align 4
+// CHECK-NOUSE-I386-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [7 x i8*], align 4
+// CHECK-NOUSE-I386-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [7 x i8*], align 4
+// CHECK-NOUSE-I386-NEXT: [[DOTOFFLOAD_SIZES:%.*]] = alloca [7 x i64], align 4
+// CHECK-NOUSE-I386-NEXT: [[DOTOFFLOAD_BASEPTRS3:%.*]] = alloca [1 x i8*], align 4
+// CHECK-NOUSE-I386-NEXT: [[DOTOFFLOAD_PTRS4:%.*]] = alloca [1 x i8*], align 4
+// CHECK-NOUSE-I386-NEXT: [[DOTOFFLOAD_MAPPERS5:%.*]] = alloca [1 x i8*], align 4
+// CHECK-NOUSE-I386-NEXT: store i32 [[II]], i32* [[II_ADDR]], align 4
+// CHECK-NOUSE-I386-NEXT: [[TMP0:%.*]] = load i32, i32* [[II_ADDR]], align 4
+// CHECK-NOUSE-I386-NEXT: store i32 [[TMP0]], i32* [[A]], align 4
+// CHECK-NOUSE-I386-NEXT: [[I:%.*]] = getelementptr inbounds [[STRUCT_ST]], %struct.ST* [[ST1]], i32 0, i32 0
+// CHECK-NOUSE-I386-NEXT: [[J:%.*]] = getelementptr inbounds [[STRUCT_ST]], %struct.ST* [[ST1]], i32 0, i32 1
+// CHECK-NOUSE-I386-NEXT: [[TMP1:%.*]] = getelementptr i32, i32* [[J]], i32 1
+// CHECK-NOUSE-I386-NEXT: [[TMP2:%.*]] = bitcast i32* [[I]] to i8*
+// CHECK-NOUSE-I386-NEXT: [[TMP3:%.*]] = bitcast i32* [[TMP1]] to i8*
+// CHECK-NOUSE-I386-NEXT: [[TMP4:%.*]] = ptrtoint i8* [[TMP3]] to i64
+// CHECK-NOUSE-I386-NEXT: [[TMP5:%.*]] = ptrtoint i8* [[TMP2]] to i64
+// CHECK-NOUSE-I386-NEXT: [[TMP6:%.*]] = sub i64 [[TMP4]], [[TMP5]]
+// CHECK-NOUSE-I386-NEXT: [[TMP7:%.*]] = sdiv exact i64 [[TMP6]], ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)
+// CHECK-NOUSE-I386-NEXT: [[I1:%.*]] = getelementptr inbounds [[STRUCT_ST]], %struct.ST* [[ST2]], i32 0, i32 0
+// CHECK-NOUSE-I386-NEXT: [[J2:%.*]] = getelementptr inbounds [[STRUCT_ST]], %struct.ST* [[ST2]], i32 0, i32 1
+// CHECK-NOUSE-I386-NEXT: [[TMP8:%.*]] = getelementptr i32, i32* [[J2]], i32 1
+// CHECK-NOUSE-I386-NEXT: [[TMP9:%.*]] = bitcast i32* [[I1]] to i8*
+// CHECK-NOUSE-I386-NEXT: [[TMP10:%.*]] = bitcast i32* [[TMP8]] to i8*
+// CHECK-NOUSE-I386-NEXT: [[TMP11:%.*]] = ptrtoint i8* [[TMP10]] to i64
+// CHECK-NOUSE-I386-NEXT: [[TMP12:%.*]] = ptrtoint i8* [[TMP9]] to i64
+// CHECK-NOUSE-I386-NEXT: [[TMP13:%.*]] = sub i64 [[TMP11]], [[TMP12]]
// CHECK-NOUSE-I386-NEXT: [[TMP14:%.*]] = sdiv exact i64 [[TMP13]], ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)
// CHECK-NOUSE-I386-NEXT: [[TMP15:%.*]] = bitcast [7 x i64]* [[DOTOFFLOAD_SIZES]] to i8*
// CHECK-NOUSE-I386-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP15]], i8* align 4 bitcast ([7 x i64]* @.offload_sizes to i8*), i32 56, i1 false)
@@ -707,7 +1233,7 @@ struct ST {
// CHECK-NOUSE-I386-NEXT: [[TMP55:%.*]] = getelementptr inbounds [7 x i64], [7 x i64]* [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK-NOUSE-I386-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK-NOUSE-I386-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK-NOUSE-I386-NEXT: store i32 1, i32* [[TMP56]], align 4
+// CHECK-NOUSE-I386-NEXT: store i32 2, i32* [[TMP56]], align 4
// CHECK-NOUSE-I386-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 1
// CHECK-NOUSE-I386-NEXT: store i32 7, i32* [[TMP57]], align 4
// CHECK-NOUSE-I386-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 2
@@ -724,322 +1250,89 @@ struct ST {
// CHECK-NOUSE-I386-NEXT: store i8** null, i8*** [[TMP63]], align 4
// CHECK-NOUSE-I386-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 8
// CHECK-NOUSE-I386-NEXT: store i64 0, i64* [[TMP64]], align 8
-// CHECK-NOUSE-I386-NEXT: [[TMP65:%.*]] = call i32 @__tgt_target_kernel(%struct.ident_t* @[[GLOB1:[0-9]+]], i64 -1, i32 -1, i32 0, i8* @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l782.region_id, %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]])
-// CHECK-NOUSE-I386-NEXT: [[TMP66:%.*]] = icmp ne i32 [[TMP65]], 0
-// CHECK-NOUSE-I386-NEXT: br i1 [[TMP66]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK-NOUSE-I386-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK-NOUSE-I386-NEXT: store i64 0, i64* [[TMP65]], align 8
+// CHECK-NOUSE-I386-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK-NOUSE-I386-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], [3 x i32]* [[TMP66]], align 4
+// CHECK-NOUSE-I386-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK-NOUSE-I386-NEXT: store [3 x i32] zeroinitializer, [3 x i32]* [[TMP67]], align 4
+// CHECK-NOUSE-I386-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK-NOUSE-I386-NEXT: store i32 0, i32* [[TMP68]], align 4
+// CHECK-NOUSE-I386-NEXT: [[TMP69:%.*]] = call i32 @__tgt_target_kernel(%struct.ident_t* @[[GLOB1:[0-9]+]], i64 -1, i32 -1, i32 0, i8* @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l{{[0-9]*}}.region_id, %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]])
+// CHECK-NOUSE-I386-NEXT: [[TMP70:%.*]] = icmp ne i32 [[TMP69]], 0
+// CHECK-NOUSE-I386-NEXT: br i1 [[TMP70]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK-NOUSE-I386: omp_offload.failed:
-// CHECK-NOUSE-I386-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l782() #[[ATTR3:[0-9]+]]
+// CHECK-NOUSE-I386-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l{{[0-9]*}}() #[[ATTR3:[0-9]+]]
// CHECK-NOUSE-I386-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK-NOUSE-I386: omp_offload.cont:
-// CHECK-NOUSE-I386-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK-NOUSE-I386-NEXT: [[TMP68:%.*]] = bitcast i8** [[TMP67]] to i32**
-// CHECK-NOUSE-I386-NEXT: store i32* [[A]], i32** [[TMP68]], align 4
-// CHECK-NOUSE-I386-NEXT: [[TMP69:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK-NOUSE-I386-NEXT: [[TMP70:%.*]] = bitcast i8** [[TMP69]] to i32**
-// CHECK-NOUSE-I386-NEXT: store i32* [[A]], i32** [[TMP70]], align 4
-// CHECK-NOUSE-I386-NEXT: [[TMP71:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
-// CHECK-NOUSE-I386-NEXT: store i8* null, i8** [[TMP71]], align 4
-// CHECK-NOUSE-I386-NEXT: [[TMP72:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK-NOUSE-I386-NEXT: [[TMP71:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK-NOUSE-I386-NEXT: [[TMP72:%.*]] = bitcast i8** [[TMP71]] to i32**
+// CHECK-NOUSE-I386-NEXT: store i32* [[A]], i32** [[TMP72]], align 4
// CHECK-NOUSE-I386-NEXT: [[TMP73:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK-NOUSE-I386-NEXT: [[TMP74:%.*]] = bitcast i8** [[TMP73]] to i32**
+// CHECK-NOUSE-I386-NEXT: store i32* [[A]], i32** [[TMP74]], align 4
+// CHECK-NOUSE-I386-NEXT: [[TMP75:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
+// CHECK-NOUSE-I386-NEXT: store i8* null, i8** [[TMP75]], align 4
+// CHECK-NOUSE-I386-NEXT: [[TMP76:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK-NOUSE-I386-NEXT: [[TMP77:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK-NOUSE-I386-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK-NOUSE-I386-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK-NOUSE-I386-NEXT: store i32 1, i32* [[TMP74]], align 4
-// CHECK-NOUSE-I386-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK-NOUSE-I386-NEXT: store i32 1, i32* [[TMP75]], align 4
-// CHECK-NOUSE-I386-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK-NOUSE-I386-NEXT: store i8** [[TMP72]], i8*** [[TMP76]], align 4
-// CHECK-NOUSE-I386-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK-NOUSE-I386-NEXT: store i8** [[TMP73]], i8*** [[TMP77]], align 4
-// CHECK-NOUSE-I386-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK-NOUSE-I386-NEXT: store i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_sizes.1, i32 0, i32 0), i64** [[TMP78]], align 4
-// CHECK-NOUSE-I386-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK-NOUSE-I386-NEXT: store i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_maptypes.2, i32 0, i32 0), i64** [[TMP79]], align 4
-// CHECK-NOUSE-I386-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK-NOUSE-I386-NEXT: store i8** null, i8*** [[TMP80]], align 4
-// CHECK-NOUSE-I386-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK-NOUSE-I386-NEXT: store i8** null, i8*** [[TMP81]], align 4
-// CHECK-NOUSE-I386-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK-NOUSE-I386-NEXT: store i64 0, i64* [[TMP82]], align 8
-// CHECK-NOUSE-I386-NEXT: [[TMP83:%.*]] = call i32 @__tgt_target_kernel(%struct.ident_t* @[[GLOB1]], i64 -1, i32 -1, i32 0, i8* @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l796.region_id, %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]])
-// CHECK-NOUSE-I386-NEXT: [[TMP84:%.*]] = icmp ne i32 [[TMP83]], 0
-// CHECK-NOUSE-I386-NEXT: br i1 [[TMP84]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK-NOUSE-I386-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK-NOUSE-I386-NEXT: store i32 2, i32* [[TMP78]], align 4
+// CHECK-NOUSE-I386-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK-NOUSE-I386-NEXT: store i32 1, i32* [[TMP79]], align 4
+// CHECK-NOUSE-I386-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK-NOUSE-I386-NEXT: store i8** [[TMP76]], i8*** [[TMP80]], align 4
+// CHECK-NOUSE-I386-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK-NOUSE-I386-NEXT: store i8** [[TMP77]], i8*** [[TMP81]], align 4
+// CHECK-NOUSE-I386-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK-NOUSE-I386-NEXT: store i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_sizes.1, i32 0, i32 0), i64** [[TMP82]], align 4
+// CHECK-NOUSE-I386-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK-NOUSE-I386-NEXT: store i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_maptypes.2, i32 0, i32 0), i64** [[TMP83]], align 4
+// CHECK-NOUSE-I386-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK-NOUSE-I386-NEXT: store i8** null, i8*** [[TMP84]], align 4
+// CHECK-NOUSE-I386-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK-NOUSE-I386-NEXT: store i8** null, i8*** [[TMP85]], align 4
+// CHECK-NOUSE-I386-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK-NOUSE-I386-NEXT: store i64 0, i64* [[TMP86]], align 8
+// CHECK-NOUSE-I386-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK-NOUSE-I386-NEXT: store i64 0, i64* [[TMP87]], align 8
+// CHECK-NOUSE-I386-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK-NOUSE-I386-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], [3 x i32]* [[TMP88]], align 4
+// CHECK-NOUSE-I386-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK-NOUSE-I386-NEXT: store [3 x i32] zeroinitializer, [3 x i32]* [[TMP89]], align 4
+// CHECK-NOUSE-I386-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK-NOUSE-I386-NEXT: store i32 0, i32* [[TMP90]], align 4
+// CHECK-NOUSE-I386-NEXT: [[TMP91:%.*]] = call i32 @__tgt_target_kernel(%struct.ident_t* @[[GLOB1]], i64 -1, i32 -1, i32 0, i8* @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l{{[0-9]*}}.region_id, %struct.__tgt_kernel_arguments* [[KERNEL_ARGS6]])
+// CHECK-NOUSE-I386-NEXT: [[TMP92:%.*]] = icmp ne i32 [[TMP91]], 0
+// CHECK-NOUSE-I386-NEXT: br i1 [[TMP92]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK-NOUSE-I386: omp_offload.failed7:
-// CHECK-NOUSE-I386-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l796() #[[ATTR3]]
+// CHECK-NOUSE-I386-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l{{[0-9]*}}() #[[ATTR3]]
// CHECK-NOUSE-I386-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK-NOUSE-I386: omp_offload.cont8:
// CHECK-NOUSE-I386-NEXT: ret void
//
-void explicit_maps_single(int ii) {
-
- // Map of a scalar.
- int a = ii;
-
- struct ST st1;
- struct ST st2;
-
- // Make sure the struct picks up ompx_hold even if another element of the
- // struct doesn't have ompx_hold.
- #pragma omp target map(tofrom : st1.i) \
- map(ompx_hold, tofrom : a, st1.j, st2.i) \
- map(tofrom : st2.j)
- {
-#ifdef USE
- st1.i++;
- a++;
- st1.j++;
- st2.i++;
- st2.j++;
-#endif
- }
-
- // Always Close Hold.
- #pragma omp target map(always close ompx_hold tofrom: a)
- {
-#ifdef USE
- a++;
-#endif
- }
-}
-
-// CHECK-USE-PPC64LE-LABEL: @_ZN2ST20test_present_membersEv(
-// CHECK-USE-PPC64LE-NEXT: entry:
-// CHECK-USE-PPC64LE-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.ST*, align 8
-// CHECK-USE-PPC64LE-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [3 x i8*], align 8
-// CHECK-USE-PPC64LE-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [3 x i8*], align 8
-// CHECK-USE-PPC64LE-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [3 x i8*], align 8
-// CHECK-USE-PPC64LE-NEXT: [[DOTOFFLOAD_SIZES:%.*]] = alloca [3 x i64], align 8
-// CHECK-USE-PPC64LE-NEXT: store %struct.ST* [[THIS:%.*]], %struct.ST** [[THIS_ADDR]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[THIS1:%.*]] = load %struct.ST*, %struct.ST** [[THIS_ADDR]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[I:%.*]] = getelementptr inbounds [[STRUCT_ST:%.*]], %struct.ST* [[THIS1]], i32 0, i32 0
-// CHECK-USE-PPC64LE-NEXT: [[J:%.*]] = getelementptr inbounds [[STRUCT_ST]], %struct.ST* [[THIS1]], i32 0, i32 1
-// CHECK-USE-PPC64LE-NEXT: [[TMP0:%.*]] = getelementptr i32, i32* [[J]], i32 1
-// CHECK-USE-PPC64LE-NEXT: [[TMP1:%.*]] = bitcast i32* [[I]] to i8*
-// CHECK-USE-PPC64LE-NEXT: [[TMP2:%.*]] = bitcast i32* [[TMP0]] to i8*
-// CHECK-USE-PPC64LE-NEXT: [[TMP3:%.*]] = ptrtoint i8* [[TMP2]] to i64
-// CHECK-USE-PPC64LE-NEXT: [[TMP4:%.*]] = ptrtoint i8* [[TMP1]] to i64
-// CHECK-USE-PPC64LE-NEXT: [[TMP5:%.*]] = sub i64 [[TMP3]], [[TMP4]]
-// CHECK-USE-PPC64LE-NEXT: [[TMP6:%.*]] = sdiv exact i64 [[TMP5]], ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)
-// CHECK-USE-PPC64LE-NEXT: [[TMP7:%.*]] = bitcast [3 x i64]* [[DOTOFFLOAD_SIZES]] to i8*
-// CHECK-USE-PPC64LE-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[TMP7]], i8* align 8 bitcast ([3 x i64]* @.offload_sizes.3 to i8*), i64 24, i1 false)
-// CHECK-USE-PPC64LE-NEXT: [[TMP8:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK-USE-PPC64LE-NEXT: [[TMP9:%.*]] = bitcast i8** [[TMP8]] to %struct.ST**
-// CHECK-USE-PPC64LE-NEXT: store %struct.ST* [[THIS1]], %struct.ST** [[TMP9]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP10:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK-USE-PPC64LE-NEXT: [[TMP11:%.*]] = bitcast i8** [[TMP10]] to i32**
-// CHECK-USE-PPC64LE-NEXT: store i32* [[I]], i32** [[TMP11]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP12:%.*]] = getelementptr inbounds [3 x i64], [3 x i64]* [[DOTOFFLOAD_SIZES]], i32 0, i32 0
-// CHECK-USE-PPC64LE-NEXT: store i64 [[TMP6]], i64* [[TMP12]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK-USE-PPC64LE-NEXT: store i8* null, i8** [[TMP13]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP14:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
-// CHECK-USE-PPC64LE-NEXT: [[TMP15:%.*]] = bitcast i8** [[TMP14]] to %struct.ST**
-// CHECK-USE-PPC64LE-NEXT: store %struct.ST* [[THIS1]], %struct.ST** [[TMP15]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_PTRS]], i32 0, i32 1
-// CHECK-USE-PPC64LE-NEXT: [[TMP17:%.*]] = bitcast i8** [[TMP16]] to i32**
-// CHECK-USE-PPC64LE-NEXT: store i32* [[I]], i32** [[TMP17]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP18:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_MAPPERS]], i64 0, i64 1
-// CHECK-USE-PPC64LE-NEXT: store i8* null, i8** [[TMP18]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP19:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
-// CHECK-USE-PPC64LE-NEXT: [[TMP20:%.*]] = bitcast i8** [[TMP19]] to %struct.ST**
-// CHECK-USE-PPC64LE-NEXT: store %struct.ST* [[THIS1]], %struct.ST** [[TMP20]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP21:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_PTRS]], i32 0, i32 2
-// CHECK-USE-PPC64LE-NEXT: [[TMP22:%.*]] = bitcast i8** [[TMP21]] to i32**
-// CHECK-USE-PPC64LE-NEXT: store i32* [[J]], i32** [[TMP22]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_MAPPERS]], i64 0, i64 2
-// CHECK-USE-PPC64LE-NEXT: store i8* null, i8** [[TMP23]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP24:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK-USE-PPC64LE-NEXT: [[TMP25:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK-USE-PPC64LE-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x i64], [3 x i64]* [[DOTOFFLOAD_SIZES]], i32 0, i32 0
-// CHECK-USE-PPC64LE-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK-USE-PPC64LE-NEXT: store i32 1, i32* [[TMP27]], align 4
-// CHECK-USE-PPC64LE-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK-USE-PPC64LE-NEXT: store i32 3, i32* [[TMP28]], align 4
-// CHECK-USE-PPC64LE-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK-USE-PPC64LE-NEXT: store i8** [[TMP24]], i8*** [[TMP29]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK-USE-PPC64LE-NEXT: store i8** [[TMP25]], i8*** [[TMP30]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK-USE-PPC64LE-NEXT: store i64* [[TMP26]], i64** [[TMP31]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK-USE-PPC64LE-NEXT: store i64* getelementptr inbounds ([3 x i64], [3 x i64]* @.offload_maptypes.4, i32 0, i32 0), i64** [[TMP32]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK-USE-PPC64LE-NEXT: store i8** null, i8*** [[TMP33]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 7
-// CHECK-USE-PPC64LE-NEXT: store i8** null, i8*** [[TMP34]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK-USE-PPC64LE-NEXT: store i64 0, i64* [[TMP35]], align 8
-// CHECK-USE-PPC64LE-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(%struct.ident_t* @[[GLOB1]], i64 -1, i32 -1, i32 0, i8* @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2ST20test_present_membersEv_l1115.region_id, %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]])
-// CHECK-USE-PPC64LE-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
-// CHECK-USE-PPC64LE-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
-// CHECK-USE-PPC64LE: omp_offload.failed:
-// CHECK-USE-PPC64LE-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2ST20test_present_membersEv_l1115(%struct.ST* [[THIS1]]) #[[ATTR3]]
-// CHECK-USE-PPC64LE-NEXT: br label [[OMP_OFFLOAD_CONT]]
-// CHECK-USE-PPC64LE: omp_offload.cont:
-// CHECK-USE-PPC64LE-NEXT: ret void
//
-// CHECK-USE-I386-LABEL: @_ZN2ST20test_present_membersEv(
-// CHECK-USE-I386-NEXT: entry:
-// CHECK-USE-I386-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.ST*, align 4
-// CHECK-USE-I386-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [3 x i8*], align 4
-// CHECK-USE-I386-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [3 x i8*], align 4
-// CHECK-USE-I386-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [3 x i8*], align 4
-// CHECK-USE-I386-NEXT: [[DOTOFFLOAD_SIZES:%.*]] = alloca [3 x i64], align 4
-// CHECK-USE-I386-NEXT: store %struct.ST* [[THIS:%.*]], %struct.ST** [[THIS_ADDR]], align 4
-// CHECK-USE-I386-NEXT: [[THIS1:%.*]] = load %struct.ST*, %struct.ST** [[THIS_ADDR]], align 4
-// CHECK-USE-I386-NEXT: [[I:%.*]] = getelementptr inbounds [[STRUCT_ST:%.*]], %struct.ST* [[THIS1]], i32 0, i32 0
-// CHECK-USE-I386-NEXT: [[J:%.*]] = getelementptr inbounds [[STRUCT_ST]], %struct.ST* [[THIS1]], i32 0, i32 1
-// CHECK-USE-I386-NEXT: [[TMP0:%.*]] = getelementptr i32, i32* [[J]], i32 1
-// CHECK-USE-I386-NEXT: [[TMP1:%.*]] = bitcast i32* [[I]] to i8*
-// CHECK-USE-I386-NEXT: [[TMP2:%.*]] = bitcast i32* [[TMP0]] to i8*
-// CHECK-USE-I386-NEXT: [[TMP3:%.*]] = ptrtoint i8* [[TMP2]] to i64
-// CHECK-USE-I386-NEXT: [[TMP4:%.*]] = ptrtoint i8* [[TMP1]] to i64
-// CHECK-USE-I386-NEXT: [[TMP5:%.*]] = sub i64 [[TMP3]], [[TMP4]]
-// CHECK-USE-I386-NEXT: [[TMP6:%.*]] = sdiv exact i64 [[TMP5]], ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)
-// CHECK-USE-I386-NEXT: [[TMP7:%.*]] = bitcast [3 x i64]* [[DOTOFFLOAD_SIZES]] to i8*
-// CHECK-USE-I386-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP7]], i8* align 4 bitcast ([3 x i64]* @.offload_sizes.3 to i8*), i32 24, i1 false)
-// CHECK-USE-I386-NEXT: [[TMP8:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK-USE-I386-NEXT: [[TMP9:%.*]] = bitcast i8** [[TMP8]] to %struct.ST**
-// CHECK-USE-I386-NEXT: store %struct.ST* [[THIS1]], %struct.ST** [[TMP9]], align 4
-// CHECK-USE-I386-NEXT: [[TMP10:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK-USE-I386-NEXT: [[TMP11:%.*]] = bitcast i8** [[TMP10]] to i32**
-// CHECK-USE-I386-NEXT: store i32* [[I]], i32** [[TMP11]], align 4
-// CHECK-USE-I386-NEXT: [[TMP12:%.*]] = getelementptr inbounds [3 x i64], [3 x i64]* [[DOTOFFLOAD_SIZES]], i32 0, i32 0
-// CHECK-USE-I386-NEXT: store i64 [[TMP6]], i64* [[TMP12]], align 4
-// CHECK-USE-I386-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
-// CHECK-USE-I386-NEXT: store i8* null, i8** [[TMP13]], align 4
-// CHECK-USE-I386-NEXT: [[TMP14:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
-// CHECK-USE-I386-NEXT: [[TMP15:%.*]] = bitcast i8** [[TMP14]] to %struct.ST**
-// CHECK-USE-I386-NEXT: store %struct.ST* [[THIS1]], %struct.ST** [[TMP15]], align 4
-// CHECK-USE-I386-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_PTRS]], i32 0, i32 1
-// CHECK-USE-I386-NEXT: [[TMP17:%.*]] = bitcast i8** [[TMP16]] to i32**
-// CHECK-USE-I386-NEXT: store i32* [[I]], i32** [[TMP17]], align 4
-// CHECK-USE-I386-NEXT: [[TMP18:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_MAPPERS]], i32 0, i32 1
-// CHECK-USE-I386-NEXT: store i8* null, i8** [[TMP18]], align 4
-// CHECK-USE-I386-NEXT: [[TMP19:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
-// CHECK-USE-I386-NEXT: [[TMP20:%.*]] = bitcast i8** [[TMP19]] to %struct.ST**
-// CHECK-USE-I386-NEXT: store %struct.ST* [[THIS1]], %struct.ST** [[TMP20]], align 4
-// CHECK-USE-I386-NEXT: [[TMP21:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_PTRS]], i32 0, i32 2
-// CHECK-USE-I386-NEXT: [[TMP22:%.*]] = bitcast i8** [[TMP21]] to i32**
-// CHECK-USE-I386-NEXT: store i32* [[J]], i32** [[TMP22]], align 4
-// CHECK-USE-I386-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_MAPPERS]], i32 0, i32 2
-// CHECK-USE-I386-NEXT: store i8* null, i8** [[TMP23]], align 4
-// CHECK-USE-I386-NEXT: [[TMP24:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK-USE-I386-NEXT: [[TMP25:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK-USE-I386-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x i64], [3 x i64]* [[DOTOFFLOAD_SIZES]], i32 0, i32 0
-// CHECK-USE-I386-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK-USE-I386-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK-USE-I386-NEXT: store i32 1, i32* [[TMP27]], align 4
-// CHECK-USE-I386-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK-USE-I386-NEXT: store i32 3, i32* [[TMP28]], align 4
-// CHECK-USE-I386-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK-USE-I386-NEXT: store i8** [[TMP24]], i8*** [[TMP29]], align 4
-// CHECK-USE-I386-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK-USE-I386-NEXT: store i8** [[TMP25]], i8*** [[TMP30]], align 4
-// CHECK-USE-I386-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK-USE-I386-NEXT: store i64* [[TMP26]], i64** [[TMP31]], align 4
-// CHECK-USE-I386-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK-USE-I386-NEXT: store i64* getelementptr inbounds ([3 x i64], [3 x i64]* @.offload_maptypes.4, i32 0, i32 0), i64** [[TMP32]], align 4
-// CHECK-USE-I386-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK-USE-I386-NEXT: store i8** null, i8*** [[TMP33]], align 4
-// CHECK-USE-I386-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 7
-// CHECK-USE-I386-NEXT: store i8** null, i8*** [[TMP34]], align 4
-// CHECK-USE-I386-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK-USE-I386-NEXT: store i64 0, i64* [[TMP35]], align 8
-// CHECK-USE-I386-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(%struct.ident_t* @[[GLOB1]], i64 -1, i32 -1, i32 0, i8* @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2ST20test_present_membersEv_l1115.region_id, %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]])
-// CHECK-USE-I386-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
-// CHECK-USE-I386-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
-// CHECK-USE-I386: omp_offload.failed:
-// CHECK-USE-I386-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2ST20test_present_membersEv_l1115(%struct.ST* [[THIS1]]) #[[ATTR3]]
-// CHECK-USE-I386-NEXT: br label [[OMP_OFFLOAD_CONT]]
-// CHECK-USE-I386: omp_offload.cont:
-// CHECK-USE-I386-NEXT: ret void
+// CHECK-NOUSE-I386-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l{{[0-9]*}}
+// CHECK-NOUSE-I386-SAME: () #[[ATTR1:[0-9]+]] {
+// CHECK-NOUSE-I386-NEXT: entry:
+// CHECK-NOUSE-I386-NEXT: ret void
//
-// CHECK-NOUSE-PPC64LE-LABEL: @_ZN2ST20test_present_membersEv(
-// CHECK-NOUSE-PPC64LE-NEXT: entry:
-// CHECK-NOUSE-PPC64LE-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.ST*, align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [3 x i8*], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [3 x i8*], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [3 x i8*], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[DOTOFFLOAD_SIZES:%.*]] = alloca [3 x i64], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: store %struct.ST* [[THIS:%.*]], %struct.ST** [[THIS_ADDR]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[THIS1:%.*]] = load %struct.ST*, %struct.ST** [[THIS_ADDR]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[I:%.*]] = getelementptr inbounds [[STRUCT_ST:%.*]], %struct.ST* [[THIS1]], i32 0, i32 0
-// CHECK-NOUSE-PPC64LE-NEXT: [[J:%.*]] = getelementptr inbounds [[STRUCT_ST]], %struct.ST* [[THIS1]], i32 0, i32 1
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP0:%.*]] = getelementptr i32, i32* [[J]], i32 1
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP1:%.*]] = bitcast i32* [[I]] to i8*
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP2:%.*]] = bitcast i32* [[TMP0]] to i8*
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP3:%.*]] = ptrtoint i8* [[TMP2]] to i64
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP4:%.*]] = ptrtoint i8* [[TMP1]] to i64
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP5:%.*]] = sub i64 [[TMP3]], [[TMP4]]
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP6:%.*]] = sdiv exact i64 [[TMP5]], ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP7:%.*]] = bitcast [3 x i64]* [[DOTOFFLOAD_SIZES]] to i8*
-// CHECK-NOUSE-PPC64LE-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[TMP7]], i8* align 8 bitcast ([3 x i64]* @.offload_sizes.3 to i8*), i64 24, i1 false)
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP8:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP9:%.*]] = bitcast i8** [[TMP8]] to %struct.ST**
-// CHECK-NOUSE-PPC64LE-NEXT: store %struct.ST* [[THIS1]], %struct.ST** [[TMP9]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP10:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP11:%.*]] = bitcast i8** [[TMP10]] to i32**
-// CHECK-NOUSE-PPC64LE-NEXT: store i32* [[I]], i32** [[TMP11]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP12:%.*]] = getelementptr inbounds [3 x i64], [3 x i64]* [[DOTOFFLOAD_SIZES]], i32 0, i32 0
-// CHECK-NOUSE-PPC64LE-NEXT: store i64 [[TMP6]], i64* [[TMP12]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK-NOUSE-PPC64LE-NEXT: store i8* null, i8** [[TMP13]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP14:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP15:%.*]] = bitcast i8** [[TMP14]] to %struct.ST**
-// CHECK-NOUSE-PPC64LE-NEXT: store %struct.ST* [[THIS1]], %struct.ST** [[TMP15]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_PTRS]], i32 0, i32 1
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP17:%.*]] = bitcast i8** [[TMP16]] to i32**
-// CHECK-NOUSE-PPC64LE-NEXT: store i32* [[I]], i32** [[TMP17]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP18:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_MAPPERS]], i64 0, i64 1
-// CHECK-NOUSE-PPC64LE-NEXT: store i8* null, i8** [[TMP18]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP19:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP20:%.*]] = bitcast i8** [[TMP19]] to %struct.ST**
-// CHECK-NOUSE-PPC64LE-NEXT: store %struct.ST* [[THIS1]], %struct.ST** [[TMP20]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP21:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_PTRS]], i32 0, i32 2
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP22:%.*]] = bitcast i8** [[TMP21]] to i32**
-// CHECK-NOUSE-PPC64LE-NEXT: store i32* [[J]], i32** [[TMP22]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_MAPPERS]], i64 0, i64 2
-// CHECK-NOUSE-PPC64LE-NEXT: store i8* null, i8** [[TMP23]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP24:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP25:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x i64], [3 x i64]* [[DOTOFFLOAD_SIZES]], i32 0, i32 0
-// CHECK-NOUSE-PPC64LE-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK-NOUSE-PPC64LE-NEXT: store i32 1, i32* [[TMP27]], align 4
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK-NOUSE-PPC64LE-NEXT: store i32 3, i32* [[TMP28]], align 4
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK-NOUSE-PPC64LE-NEXT: store i8** [[TMP24]], i8*** [[TMP29]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK-NOUSE-PPC64LE-NEXT: store i8** [[TMP25]], i8*** [[TMP30]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK-NOUSE-PPC64LE-NEXT: store i64* [[TMP26]], i64** [[TMP31]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK-NOUSE-PPC64LE-NEXT: store i64* getelementptr inbounds ([3 x i64], [3 x i64]* @.offload_maptypes.4, i32 0, i32 0), i64** [[TMP32]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK-NOUSE-PPC64LE-NEXT: store i8** null, i8*** [[TMP33]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 7
-// CHECK-NOUSE-PPC64LE-NEXT: store i8** null, i8*** [[TMP34]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK-NOUSE-PPC64LE-NEXT: store i64 0, i64* [[TMP35]], align 8
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(%struct.ident_t* @[[GLOB1]], i64 -1, i32 -1, i32 0, i8* @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2ST20test_present_membersEv_l1115.region_id, %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]])
-// CHECK-NOUSE-PPC64LE-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
-// CHECK-NOUSE-PPC64LE-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
-// CHECK-NOUSE-PPC64LE: omp_offload.failed:
-// CHECK-NOUSE-PPC64LE-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2ST20test_present_membersEv_l1115() #[[ATTR3]]
-// CHECK-NOUSE-PPC64LE-NEXT: br label [[OMP_OFFLOAD_CONT]]
-// CHECK-NOUSE-PPC64LE: omp_offload.cont:
-// CHECK-NOUSE-PPC64LE-NEXT: ret void
//
-// CHECK-NOUSE-I386-LABEL: @_ZN2ST20test_present_membersEv(
+// CHECK-NOUSE-I386-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z20explicit_maps_singlei_l{{[0-9]*}}
+// CHECK-NOUSE-I386-SAME: () #[[ATTR1]] {
+// CHECK-NOUSE-I386-NEXT: entry:
+// CHECK-NOUSE-I386-NEXT: ret void
+//
+//
+// CHECK-NOUSE-I386-LABEL: define {{[^@]+}}@_ZN2ST20test_present_membersEv
+// CHECK-NOUSE-I386-SAME: (%struct.ST* noundef nonnull align 4 dereferenceable(8) [[THIS:%.*]]) #[[ATTR0]] align 2 {
// CHECK-NOUSE-I386-NEXT: entry:
// CHECK-NOUSE-I386-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.ST*, align 4
// CHECK-NOUSE-I386-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [3 x i8*], align 4
// CHECK-NOUSE-I386-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [3 x i8*], align 4
// CHECK-NOUSE-I386-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [3 x i8*], align 4
// CHECK-NOUSE-I386-NEXT: [[DOTOFFLOAD_SIZES:%.*]] = alloca [3 x i64], align 4
-// CHECK-NOUSE-I386-NEXT: store %struct.ST* [[THIS:%.*]], %struct.ST** [[THIS_ADDR]], align 4
+// CHECK-NOUSE-I386-NEXT: store %struct.ST* [[THIS]], %struct.ST** [[THIS_ADDR]], align 4
// CHECK-NOUSE-I386-NEXT: [[THIS1:%.*]] = load %struct.ST*, %struct.ST** [[THIS_ADDR]], align 4
// CHECK-NOUSE-I386-NEXT: [[I:%.*]] = getelementptr inbounds [[STRUCT_ST:%.*]], %struct.ST* [[THIS1]], i32 0, i32 0
// CHECK-NOUSE-I386-NEXT: [[J:%.*]] = getelementptr inbounds [[STRUCT_ST]], %struct.ST* [[THIS1]], i32 0, i32 1
@@ -1083,7 +1376,7 @@ void explicit_maps_single(int ii) {
// CHECK-NOUSE-I386-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x i64], [3 x i64]* [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK-NOUSE-I386-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK-NOUSE-I386-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK-NOUSE-I386-NEXT: store i32 1, i32* [[TMP27]], align 4
+// CHECK-NOUSE-I386-NEXT: store i32 2, i32* [[TMP27]], align 4
// CHECK-NOUSE-I386-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 1
// CHECK-NOUSE-I386-NEXT: store i32 3, i32* [[TMP28]], align 4
// CHECK-NOUSE-I386-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 2
@@ -1100,25 +1393,32 @@ void explicit_maps_single(int ii) {
// CHECK-NOUSE-I386-NEXT: store i8** null, i8*** [[TMP34]], align 4
// CHECK-NOUSE-I386-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 8
// CHECK-NOUSE-I386-NEXT: store i64 0, i64* [[TMP35]], align 8
-// CHECK-NOUSE-I386-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(%struct.ident_t* @[[GLOB1]], i64 -1, i32 -1, i32 0, i8* @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2ST20test_present_membersEv_l1115.region_id, %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]])
-// CHECK-NOUSE-I386-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
-// CHECK-NOUSE-I386-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK-NOUSE-I386-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK-NOUSE-I386-NEXT: store i64 0, i64* [[TMP36]], align 8
+// CHECK-NOUSE-I386-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK-NOUSE-I386-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], [3 x i32]* [[TMP37]], align 4
+// CHECK-NOUSE-I386-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK-NOUSE-I386-NEXT: store [3 x i32] zeroinitializer, [3 x i32]* [[TMP38]], align 4
+// CHECK-NOUSE-I386-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK-NOUSE-I386-NEXT: store i32 0, i32* [[TMP39]], align 4
+// CHECK-NOUSE-I386-NEXT: [[TMP40:%.*]] = call i32 @__tgt_target_kernel(%struct.ident_t* @[[GLOB1]], i64 -1, i32 -1, i32 0, i8* @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2ST20test_present_membersEv_l{{[0-9]*}}.region_id, %struct.__tgt_kernel_arguments* [[KERNEL_ARGS]])
+// CHECK-NOUSE-I386-NEXT: [[TMP41:%.*]] = icmp ne i32 [[TMP40]], 0
+// CHECK-NOUSE-I386-NEXT: br i1 [[TMP41]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK-NOUSE-I386: omp_offload.failed:
-// CHECK-NOUSE-I386-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2ST20test_present_membersEv_l1115() #[[ATTR3]]
+// CHECK-NOUSE-I386-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2ST20test_present_membersEv_l{{[0-9]*}}() #[[ATTR3]]
// CHECK-NOUSE-I386-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK-NOUSE-I386: omp_offload.cont:
// CHECK-NOUSE-I386-NEXT: ret void
//
-void ST::test_present_members() {
- // Make sure the struct picks up ompx_hold even if another element of the
- // struct doesn't have ompx_hold.
- #pragma omp target map(tofrom : i) map(ompx_hold, tofrom : j)
- {
-#ifdef USE
- i++;
- j++;
-#endif
- }
-}
-
-#endif
+//
+// CHECK-NOUSE-I386-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2ST20test_present_membersEv_l{{[0-9]*}}
+// CHECK-NOUSE-I386-SAME: () #[[ATTR1]] {
+// CHECK-NOUSE-I386-NEXT: entry:
+// CHECK-NOUSE-I386-NEXT: ret void
+//
+//
+// CHECK-NOUSE-I386-LABEL: define {{[^@]+}}@.omp_offloading.requires_reg
+// CHECK-NOUSE-I386-SAME: () #[[ATTR4:[0-9]+]] {
+// CHECK-NOUSE-I386-NEXT: entry:
+// CHECK-NOUSE-I386-NEXT: call void @__tgt_register_requires(i64 1)
+// CHECK-NOUSE-I386-NEXT: ret void
diff --git a/clang/test/OpenMP/target_map_member_expr_codegen.cpp b/clang/test/OpenMP/target_map_member_expr_codegen.cpp
index 2d9777197a791..c4dace141fc4d 100644
--- a/clang/test/OpenMP/target_map_member_expr_codegen.cpp
+++ b/clang/test/OpenMP/target_map_member_expr_codegen.cpp
@@ -52,7 +52,6 @@ class B : public A {
// CHECK-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [4 x ptr], align 8
// CHECK-NEXT: store ptr [[THIS:%.*]], ptr [[THIS_ADDR]], align 8
// CHECK-NEXT: [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
-// CHECK-NOT: sdiv exact i64 {{.*}}, ptrtoint
// CHECK-NEXT: [[RES:%.*]] = getelementptr inbounds [[CLASS_B:%.*]], ptr [[THIS1]], i32 0, i32 1
// CHECK-NEXT: [[X:%.*]] = getelementptr inbounds [[CLASS_A:%.*]], ptr [[THIS1]], i32 0, i32 0
// CHECK-NEXT: [[Y:%.*]] = getelementptr inbounds [[CLASS_A]], ptr [[THIS1]], i32 0, i32 1
@@ -84,7 +83,7 @@ class B : public A {
// CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK-NEXT: store i32 1, ptr [[TMP14]], align 4
+// CHECK-NEXT: store i32 2, ptr [[TMP14]], align 4
// CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK-NEXT: store i32 4, ptr [[TMP15]], align 4
// CHECK-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -101,9 +100,17 @@ class B : public A {
// CHECK-NEXT: store ptr null, ptr [[TMP21]], align 8
// CHECK-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK-NEXT: store i64 0, ptr [[TMP22]], align 8
-// CHECK-NEXT: [[TMP23:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 -1, i32 0, ptr @.__omp_offloading_{{.*}}__ZN1B3runEv_{{.*}}.region_id, ptr [[KERNEL_ARGS]])
-// CHECK-NEXT: [[TMP24:%.*]] = icmp ne i32 [[TMP23]], 0
-// CHECK-NEXT: br i1 [[TMP24]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK-NEXT: store i64 0, ptr [[TMP23]], align 8
+// CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP24]], align 4
+// CHECK-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK-NEXT: store i32 0, ptr [[TMP26]], align 4
+// CHECK-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 -1, i32 0, ptr @.__omp_offloading_{{.*}}__ZN1B3runEv_{{.*}}.region_id, ptr [[KERNEL_ARGS]])
+// CHECK-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
+// CHECK-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK: omp_offload.failed:
// CHECK-NEXT: call void @__omp_offloading_{{.*}}__ZN1B3runEv_{{.*}}(ptr [[THIS1]]) #[[ATTR3:[0-9]+]]
// CHECK-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/target_offload_mandatory_codegen.cpp b/clang/test/OpenMP/target_offload_mandatory_codegen.cpp
index 7eee911c5ab40..2062e2e390fd4 100644
--- a/clang/test/OpenMP/target_offload_mandatory_codegen.cpp
+++ b/clang/test/OpenMP/target_offload_mandatory_codegen.cpp
@@ -33,7 +33,7 @@ void host_dev(int device) {
// MANDATORY-NEXT: entry:
// MANDATORY-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// MANDATORY-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// MANDATORY-NEXT: store i32 1, ptr [[TMP0]], align 4
+// MANDATORY-NEXT: store i32 2, ptr [[TMP0]], align 4
// MANDATORY-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// MANDATORY-NEXT: store i32 0, ptr [[TMP1]], align 4
// MANDATORY-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -50,9 +50,17 @@ void host_dev(int device) {
// MANDATORY-NEXT: store ptr null, ptr [[TMP7]], align 8
// MANDATORY-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// MANDATORY-NEXT: store i64 0, ptr [[TMP8]], align 8
-// MANDATORY-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z4hostv_l12.region_id, ptr [[KERNEL_ARGS]])
-// MANDATORY-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// MANDATORY-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// MANDATORY-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// MANDATORY-NEXT: store i64 0, ptr [[TMP9]], align 8
+// MANDATORY-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// MANDATORY-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP10]], align 4
+// MANDATORY-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// MANDATORY-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// MANDATORY-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// MANDATORY-NEXT: store i32 0, ptr [[TMP12]], align 4
+// MANDATORY-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z4hostv_l12.region_id, ptr [[KERNEL_ARGS]])
+// MANDATORY-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// MANDATORY-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// MANDATORY: omp_offload.failed:
// MANDATORY-NEXT: unreachable
// MANDATORY: omp_offload.cont:
@@ -71,7 +79,7 @@ void host_dev(int device) {
// MANDATORY: omp_if.then:
// MANDATORY-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// MANDATORY-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// MANDATORY-NEXT: store i32 1, ptr [[TMP1]], align 4
+// MANDATORY-NEXT: store i32 2, ptr [[TMP1]], align 4
// MANDATORY-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// MANDATORY-NEXT: store i32 0, ptr [[TMP2]], align 4
// MANDATORY-NEXT: [[TMP3:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -88,9 +96,17 @@ void host_dev(int device) {
// MANDATORY-NEXT: store ptr null, ptr [[TMP8]], align 8
// MANDATORY-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// MANDATORY-NEXT: store i64 0, ptr [[TMP9]], align 8
-// MANDATORY-NEXT: [[TMP10:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z7host_ifb_l17.region_id, ptr [[KERNEL_ARGS]])
-// MANDATORY-NEXT: [[TMP11:%.*]] = icmp ne i32 [[TMP10]], 0
-// MANDATORY-NEXT: br i1 [[TMP11]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// MANDATORY-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// MANDATORY-NEXT: store i64 0, ptr [[TMP10]], align 8
+// MANDATORY-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// MANDATORY-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP11]], align 4
+// MANDATORY-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// MANDATORY-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP12]], align 4
+// MANDATORY-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// MANDATORY-NEXT: store i32 0, ptr [[TMP13]], align 4
+// MANDATORY-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z7host_ifb_l17.region_id, ptr [[KERNEL_ARGS]])
+// MANDATORY-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
+// MANDATORY-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// MANDATORY: omp_offload.failed:
// MANDATORY-NEXT: unreachable
// MANDATORY: omp_offload.cont:
@@ -113,7 +129,7 @@ void host_dev(int device) {
// MANDATORY-NEXT: [[TMP2:%.*]] = sext i32 [[TMP1]] to i64
// MANDATORY-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// MANDATORY-NEXT: [[TMP3:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// MANDATORY-NEXT: store i32 1, ptr [[TMP3]], align 4
+// MANDATORY-NEXT: store i32 2, ptr [[TMP3]], align 4
// MANDATORY-NEXT: [[TMP4:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// MANDATORY-NEXT: store i32 0, ptr [[TMP4]], align 4
// MANDATORY-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -130,9 +146,17 @@ void host_dev(int device) {
// MANDATORY-NEXT: store ptr null, ptr [[TMP10]], align 8
// MANDATORY-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// MANDATORY-NEXT: store i64 0, ptr [[TMP11]], align 8
-// MANDATORY-NEXT: [[TMP12:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 [[TMP2]], i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z8host_devi_l22.region_id, ptr [[KERNEL_ARGS]])
-// MANDATORY-NEXT: [[TMP13:%.*]] = icmp ne i32 [[TMP12]], 0
-// MANDATORY-NEXT: br i1 [[TMP13]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// MANDATORY-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// MANDATORY-NEXT: store i64 0, ptr [[TMP12]], align 8
+// MANDATORY-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// MANDATORY-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP13]], align 4
+// MANDATORY-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// MANDATORY-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP14]], align 4
+// MANDATORY-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// MANDATORY-NEXT: store i32 0, ptr [[TMP15]], align 4
+// MANDATORY-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 [[TMP2]], i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z8host_devi_l22.region_id, ptr [[KERNEL_ARGS]])
+// MANDATORY-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
+// MANDATORY-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// MANDATORY: omp_offload.failed:
// MANDATORY-NEXT: unreachable
// MANDATORY: omp_offload.cont:
diff --git a/clang/test/OpenMP/target_parallel_codegen.cpp b/clang/test/OpenMP/target_parallel_codegen.cpp
index f4c15f6c5da0e..02788b69e2680 100644
--- a/clang/test/OpenMP/target_parallel_codegen.cpp
+++ b/clang/test/OpenMP/target_parallel_codegen.cpp
@@ -354,7 +354,7 @@ int bar(int n){
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP19]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP19]], align 4
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP20]], align 4
// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -371,175 +371,199 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP26]], align 8
// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP27]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
-// CHECK1-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP29]], align 4
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP31]], align 4
+// CHECK1-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
+// CHECK1-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110(i64 [[TMP13]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
-// CHECK1-NEXT: [[TMP30:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: store i32 [[TMP30]], ptr [[A_CASTED2]], align 4
-// CHECK1-NEXT: [[TMP31:%.*]] = load i64, ptr [[A_CASTED2]], align 8
-// CHECK1-NEXT: [[TMP32:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK1-NEXT: store i16 [[TMP32]], ptr [[AA_CASTED3]], align 2
-// CHECK1-NEXT: [[TMP33:%.*]] = load i64, ptr [[AA_CASTED3]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP34]], 10
+// CHECK1-NEXT: [[TMP34:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: store i32 [[TMP34]], ptr [[A_CASTED2]], align 4
+// CHECK1-NEXT: [[TMP35:%.*]] = load i64, ptr [[A_CASTED2]], align 8
+// CHECK1-NEXT: [[TMP36:%.*]] = load i16, ptr [[AA]], align 2
+// CHECK1-NEXT: store i16 [[TMP36]], ptr [[AA_CASTED3]], align 2
+// CHECK1-NEXT: [[TMP37:%.*]] = load i64, ptr [[AA_CASTED3]], align 8
+// CHECK1-NEXT: [[TMP38:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP38]], 10
// CHECK1-NEXT: br i1 [[CMP]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK1: omp_if.then:
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP31]], ptr [[TMP35]], align 8
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP31]], ptr [[TMP36]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP33]], ptr [[TMP38]], align 8
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP33]], ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP35]], ptr [[TMP39]], align 8
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP35]], ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP37]], ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP37]], ptr [[TMP43]], align 8
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP44]], align 8
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP43]], align 4
-// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK1-NEXT: store i32 2, ptr [[TMP44]], align 4
-// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP41]], ptr [[TMP45]], align 8
-// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP42]], ptr [[TMP46]], align 8
-// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.5, ptr [[TMP47]], align 8
-// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP48]], align 8
-// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP49]], align 8
-// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP50]], align 8
-// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP51]], align 8
-// CHECK1-NEXT: [[TMP52:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK1-NEXT: [[TMP53:%.*]] = icmp ne i32 [[TMP52]], 0
-// CHECK1-NEXT: br i1 [[TMP53]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP47]], align 4
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK1-NEXT: store i32 2, ptr [[TMP48]], align 4
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP45]], ptr [[TMP49]], align 8
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP46]], ptr [[TMP50]], align 8
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.5, ptr [[TMP51]], align 8
+// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP52]], align 8
+// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP53]], align 8
+// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP54]], align 8
+// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP55]], align 8
+// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP56]], align 8
+// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP57]], align 4
+// CHECK1-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP58]], align 4
+// CHECK1-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP59]], align 4
+// CHECK1-NEXT: [[TMP60:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK1-NEXT: [[TMP61:%.*]] = icmp ne i32 [[TMP60]], 0
+// CHECK1-NEXT: br i1 [[TMP61]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK1: omp_offload.failed8:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119(i64 [[TMP31]], i64 [[TMP33]]) #[[ATTR4]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119(i64 [[TMP35]], i64 [[TMP37]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK1: omp_offload.cont9:
// CHECK1-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK1: omp_if.else:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119(i64 [[TMP31]], i64 [[TMP33]]) #[[ATTR4]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119(i64 [[TMP35]], i64 [[TMP37]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
-// CHECK1-NEXT: [[TMP54:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: store i32 [[TMP54]], ptr [[A_CASTED10]], align 4
-// CHECK1-NEXT: [[TMP55:%.*]] = load i64, ptr [[A_CASTED10]], align 8
-// CHECK1-NEXT: [[TMP56:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: [[CMP11:%.*]] = icmp sgt i32 [[TMP56]], 20
+// CHECK1-NEXT: [[TMP62:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: store i32 [[TMP62]], ptr [[A_CASTED10]], align 4
+// CHECK1-NEXT: [[TMP63:%.*]] = load i64, ptr [[A_CASTED10]], align 8
+// CHECK1-NEXT: [[TMP64:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: [[CMP11:%.*]] = icmp sgt i32 [[TMP64]], 20
// CHECK1-NEXT: br i1 [[CMP11]], label [[OMP_IF_THEN12:%.*]], label [[OMP_IF_ELSE19:%.*]]
// CHECK1: omp_if.then12:
-// CHECK1-NEXT: [[TMP57:%.*]] = mul nuw i64 [[TMP2]], 4
-// CHECK1-NEXT: [[TMP58:%.*]] = mul nuw i64 5, [[TMP5]]
-// CHECK1-NEXT: [[TMP59:%.*]] = mul nuw i64 [[TMP58]], 8
+// CHECK1-NEXT: [[TMP65:%.*]] = mul nuw i64 [[TMP2]], 4
+// CHECK1-NEXT: [[TMP66:%.*]] = mul nuw i64 5, [[TMP5]]
+// CHECK1-NEXT: [[TMP67:%.*]] = mul nuw i64 [[TMP66]], 8
// CHECK1-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES]], ptr align 8 @.offload_sizes.8, i64 72, i1 false)
-// CHECK1-NEXT: [[TMP60:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP55]], ptr [[TMP60]], align 8
-// CHECK1-NEXT: [[TMP61:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP55]], ptr [[TMP61]], align 8
-// CHECK1-NEXT: [[TMP62:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP62]], align 8
-// CHECK1-NEXT: [[TMP63:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 1
-// CHECK1-NEXT: store ptr [[B]], ptr [[TMP63]], align 8
-// CHECK1-NEXT: [[TMP64:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 1
-// CHECK1-NEXT: store ptr [[B]], ptr [[TMP64]], align 8
-// CHECK1-NEXT: [[TMP65:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP65]], align 8
-// CHECK1-NEXT: [[TMP66:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 2
-// CHECK1-NEXT: store i64 [[TMP2]], ptr [[TMP66]], align 8
-// CHECK1-NEXT: [[TMP67:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 2
-// CHECK1-NEXT: store i64 [[TMP2]], ptr [[TMP67]], align 8
-// CHECK1-NEXT: [[TMP68:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP68]], align 8
-// CHECK1-NEXT: [[TMP69:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[VLA]], ptr [[TMP69]], align 8
-// CHECK1-NEXT: [[TMP70:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[VLA]], ptr [[TMP70]], align 8
-// CHECK1-NEXT: [[TMP71:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
-// CHECK1-NEXT: store i64 [[TMP57]], ptr [[TMP71]], align 8
-// CHECK1-NEXT: [[TMP72:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP72]], align 8
-// CHECK1-NEXT: [[TMP73:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 4
-// CHECK1-NEXT: store ptr [[C]], ptr [[TMP73]], align 8
-// CHECK1-NEXT: [[TMP74:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 4
-// CHECK1-NEXT: store ptr [[C]], ptr [[TMP74]], align 8
-// CHECK1-NEXT: [[TMP75:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP75]], align 8
-// CHECK1-NEXT: [[TMP76:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 5
-// CHECK1-NEXT: store i64 5, ptr [[TMP76]], align 8
-// CHECK1-NEXT: [[TMP77:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 5
-// CHECK1-NEXT: store i64 5, ptr [[TMP77]], align 8
-// CHECK1-NEXT: [[TMP78:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP78]], align 8
-// CHECK1-NEXT: [[TMP79:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 6
-// CHECK1-NEXT: store i64 [[TMP5]], ptr [[TMP79]], align 8
-// CHECK1-NEXT: [[TMP80:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 6
-// CHECK1-NEXT: store i64 [[TMP5]], ptr [[TMP80]], align 8
-// CHECK1-NEXT: [[TMP81:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP81]], align 8
-// CHECK1-NEXT: [[TMP82:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 7
-// CHECK1-NEXT: store ptr [[VLA1]], ptr [[TMP82]], align 8
-// CHECK1-NEXT: [[TMP83:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 7
-// CHECK1-NEXT: store ptr [[VLA1]], ptr [[TMP83]], align 8
-// CHECK1-NEXT: [[TMP84:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
-// CHECK1-NEXT: store i64 [[TMP59]], ptr [[TMP84]], align 8
-// CHECK1-NEXT: [[TMP85:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP85]], align 8
-// CHECK1-NEXT: [[TMP86:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 8
-// CHECK1-NEXT: store ptr [[D]], ptr [[TMP86]], align 8
-// CHECK1-NEXT: [[TMP87:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 8
-// CHECK1-NEXT: store ptr [[D]], ptr [[TMP87]], align 8
-// CHECK1-NEXT: [[TMP88:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 8
-// CHECK1-NEXT: store ptr null, ptr [[TMP88]], align 8
-// CHECK1-NEXT: [[TMP89:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP90:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP91:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP68:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP63]], ptr [[TMP68]], align 8
+// CHECK1-NEXT: [[TMP69:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP63]], ptr [[TMP69]], align 8
+// CHECK1-NEXT: [[TMP70:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP70]], align 8
+// CHECK1-NEXT: [[TMP71:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 1
+// CHECK1-NEXT: store ptr [[B]], ptr [[TMP71]], align 8
+// CHECK1-NEXT: [[TMP72:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 1
+// CHECK1-NEXT: store ptr [[B]], ptr [[TMP72]], align 8
+// CHECK1-NEXT: [[TMP73:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP73]], align 8
+// CHECK1-NEXT: [[TMP74:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 2
+// CHECK1-NEXT: store i64 [[TMP2]], ptr [[TMP74]], align 8
+// CHECK1-NEXT: [[TMP75:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 2
+// CHECK1-NEXT: store i64 [[TMP2]], ptr [[TMP75]], align 8
+// CHECK1-NEXT: [[TMP76:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 2
+// CHECK1-NEXT: store ptr null, ptr [[TMP76]], align 8
+// CHECK1-NEXT: [[TMP77:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[VLA]], ptr [[TMP77]], align 8
+// CHECK1-NEXT: [[TMP78:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[VLA]], ptr [[TMP78]], align 8
+// CHECK1-NEXT: [[TMP79:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
+// CHECK1-NEXT: store i64 [[TMP65]], ptr [[TMP79]], align 8
+// CHECK1-NEXT: [[TMP80:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 3
+// CHECK1-NEXT: store ptr null, ptr [[TMP80]], align 8
+// CHECK1-NEXT: [[TMP81:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 4
+// CHECK1-NEXT: store ptr [[C]], ptr [[TMP81]], align 8
+// CHECK1-NEXT: [[TMP82:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 4
+// CHECK1-NEXT: store ptr [[C]], ptr [[TMP82]], align 8
+// CHECK1-NEXT: [[TMP83:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP83]], align 8
+// CHECK1-NEXT: [[TMP84:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 5
+// CHECK1-NEXT: store i64 5, ptr [[TMP84]], align 8
+// CHECK1-NEXT: [[TMP85:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 5
+// CHECK1-NEXT: store i64 5, ptr [[TMP85]], align 8
+// CHECK1-NEXT: [[TMP86:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP86]], align 8
+// CHECK1-NEXT: [[TMP87:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 6
+// CHECK1-NEXT: store i64 [[TMP5]], ptr [[TMP87]], align 8
+// CHECK1-NEXT: [[TMP88:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 6
+// CHECK1-NEXT: store i64 [[TMP5]], ptr [[TMP88]], align 8
+// CHECK1-NEXT: [[TMP89:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP89]], align 8
+// CHECK1-NEXT: [[TMP90:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 7
+// CHECK1-NEXT: store ptr [[VLA1]], ptr [[TMP90]], align 8
+// CHECK1-NEXT: [[TMP91:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 7
+// CHECK1-NEXT: store ptr [[VLA1]], ptr [[TMP91]], align 8
+// CHECK1-NEXT: [[TMP92:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
+// CHECK1-NEXT: store i64 [[TMP67]], ptr [[TMP92]], align 8
+// CHECK1-NEXT: [[TMP93:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP93]], align 8
+// CHECK1-NEXT: [[TMP94:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 8
+// CHECK1-NEXT: store ptr [[D]], ptr [[TMP94]], align 8
+// CHECK1-NEXT: [[TMP95:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 8
+// CHECK1-NEXT: store ptr [[D]], ptr [[TMP95]], align 8
+// CHECK1-NEXT: [[TMP96:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 8
+// CHECK1-NEXT: store ptr null, ptr [[TMP96]], align 8
+// CHECK1-NEXT: [[TMP97:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP98:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP99:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS16:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP92]], align 4
-// CHECK1-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 1
-// CHECK1-NEXT: store i32 9, ptr [[TMP93]], align 4
-// CHECK1-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP89]], ptr [[TMP94]], align 8
-// CHECK1-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP90]], ptr [[TMP95]], align 8
-// CHECK1-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 4
-// CHECK1-NEXT: store ptr [[TMP91]], ptr [[TMP96]], align 8
-// CHECK1-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP97]], align 8
-// CHECK1-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP98]], align 8
-// CHECK1-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP99]], align 8
-// CHECK1-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP100]], align 8
-// CHECK1-NEXT: [[TMP101:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144.region_id, ptr [[KERNEL_ARGS16]])
-// CHECK1-NEXT: [[TMP102:%.*]] = icmp ne i32 [[TMP101]], 0
-// CHECK1-NEXT: br i1 [[TMP102]], label [[OMP_OFFLOAD_FAILED17:%.*]], label [[OMP_OFFLOAD_CONT18:%.*]]
+// CHECK1-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP100]], align 4
+// CHECK1-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 1
+// CHECK1-NEXT: store i32 9, ptr [[TMP101]], align 4
+// CHECK1-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP97]], ptr [[TMP102]], align 8
+// CHECK1-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP98]], ptr [[TMP103]], align 8
+// CHECK1-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 4
+// CHECK1-NEXT: store ptr [[TMP99]], ptr [[TMP104]], align 8
+// CHECK1-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP105]], align 8
+// CHECK1-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP106]], align 8
+// CHECK1-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP107]], align 8
+// CHECK1-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP108]], align 8
+// CHECK1-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP109]], align 8
+// CHECK1-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP110]], align 4
+// CHECK1-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP111]], align 4
+// CHECK1-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP112]], align 4
+// CHECK1-NEXT: [[TMP113:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144.region_id, ptr [[KERNEL_ARGS16]])
+// CHECK1-NEXT: [[TMP114:%.*]] = icmp ne i32 [[TMP113]], 0
+// CHECK1-NEXT: br i1 [[TMP114]], label [[OMP_OFFLOAD_FAILED17:%.*]], label [[OMP_OFFLOAD_CONT18:%.*]]
// CHECK1: omp_offload.failed17:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144(i64 [[TMP55]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]]) #[[ATTR4]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144(i64 [[TMP63]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT18]]
// CHECK1: omp_offload.cont18:
// CHECK1-NEXT: br label [[OMP_IF_END20:%.*]]
// CHECK1: omp_if.else19:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144(i64 [[TMP55]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]]) #[[ATTR4]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144(i64 [[TMP63]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_IF_END20]]
// CHECK1: omp_if.end20:
-// CHECK1-NEXT: [[TMP103:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: [[TMP104:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK1-NEXT: call void @llvm.stackrestore(ptr [[TMP104]])
-// CHECK1-NEXT: ret i32 [[TMP103]]
+// CHECK1-NEXT: [[TMP115:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: [[TMP116:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK1-NEXT: call void @llvm.stackrestore(ptr [[TMP116]])
+// CHECK1-NEXT: ret i32 [[TMP115]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l100
@@ -590,7 +614,7 @@ int bar(int n){
// CHECK1-NEXT: store ptr [[TMP3]], ptr [[DOTTASK_T__ADDR_I]], align 8, !noalias !21
// CHECK1-NEXT: store ptr [[TMP7]], ptr [[__CONTEXT_ADDR_I]], align 8, !noalias !21
// CHECK1-NEXT: [[TMP8:%.*]] = load ptr, ptr [[__CONTEXT_ADDR_I]], align 8, !noalias !21
-// CHECK1-NEXT: store i32 1, ptr [[KERNEL_ARGS_I]], align 4, !noalias !21
+// CHECK1-NEXT: store i32 2, ptr [[KERNEL_ARGS_I]], align 4, !noalias !21
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP9]], align 4, !noalias !21
// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 2
@@ -607,9 +631,17 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8, !noalias !21
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP16]], align 8, !noalias !21
-// CHECK1-NEXT: [[TMP17:%.*]] = call i32 @__tgt_target_kernel_nowait(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l100.region_id, ptr [[KERNEL_ARGS_I]], i32 0, ptr null, i32 0, ptr null)
-// CHECK1-NEXT: [[TMP18:%.*]] = icmp ne i32 [[TMP17]], 0
-// CHECK1-NEXT: br i1 [[TMP18]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__1_EXIT:%.*]]
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 9
+// CHECK1-NEXT: store i64 1, ptr [[TMP17]], align 8, !noalias !21
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP18]], align 4, !noalias !21
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP19]], align 4, !noalias !21
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP20]], align 4, !noalias !21
+// CHECK1-NEXT: [[TMP21:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l100.region_id, ptr [[KERNEL_ARGS_I]])
+// CHECK1-NEXT: [[TMP22:%.*]] = icmp ne i32 [[TMP21]], 0
+// CHECK1-NEXT: br i1 [[TMP22]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__1_EXIT:%.*]]
// CHECK1: omp_offload.failed.i:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l100() #[[ATTR4]]
// CHECK1-NEXT: br label [[DOTOMP_OUTLINED__1_EXIT]]
@@ -940,7 +972,7 @@ int bar(int n){
// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [5 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP29]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP29]], align 4
// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 5, ptr [[TMP30]], align 4
// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -957,9 +989,17 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP36]], align 8
// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK1-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP38]], align 8
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP39]], align 4
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP40]], align 4
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP41]], align 4
+// CHECK1-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
+// CHECK1-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216(ptr [[THIS1]], i64 [[TMP6]], i64 2, i64 [[TMP2]], ptr [[VLA]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -969,15 +1009,15 @@ int bar(int n){
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216(ptr [[THIS1]], i64 [[TMP6]], i64 2, i64 [[TMP2]], ptr [[VLA]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
-// CHECK1-NEXT: [[TMP40:%.*]] = mul nsw i64 1, [[TMP2]]
-// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i64 [[TMP40]]
+// CHECK1-NEXT: [[TMP44:%.*]] = mul nsw i64 1, [[TMP2]]
+// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i64 [[TMP44]]
// CHECK1-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i16, ptr [[ARRAYIDX]], i64 1
-// CHECK1-NEXT: [[TMP41:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
-// CHECK1-NEXT: [[CONV:%.*]] = sext i16 [[TMP41]] to i32
-// CHECK1-NEXT: [[TMP42:%.*]] = load i32, ptr [[B]], align 4
-// CHECK1-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP42]]
-// CHECK1-NEXT: [[TMP43:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK1-NEXT: call void @llvm.stackrestore(ptr [[TMP43]])
+// CHECK1-NEXT: [[TMP45:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
+// CHECK1-NEXT: [[CONV:%.*]] = sext i16 [[TMP45]] to i32
+// CHECK1-NEXT: [[TMP46:%.*]] = load i32, ptr [[B]], align 4
+// CHECK1-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP46]]
+// CHECK1-NEXT: [[TMP47:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK1-NEXT: call void @llvm.stackrestore(ptr [[TMP47]])
// CHECK1-NEXT: ret i32 [[ADD3]]
//
//
@@ -1040,7 +1080,7 @@ int bar(int n){
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 4, ptr [[TMP22]], align 4
// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1057,9 +1097,17 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l198.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP31]], align 4
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK1-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l198.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK1-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l198(i64 [[TMP1]], i64 [[TMP3]], i64 [[TMP5]], ptr [[B]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1069,8 +1117,8 @@ int bar(int n){
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l198(i64 [[TMP1]], i64 [[TMP3]], i64 [[TMP5]], ptr [[B]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
-// CHECK1-NEXT: [[TMP32:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: ret i32 [[TMP32]]
+// CHECK1-NEXT: [[TMP36:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: ret i32 [[TMP36]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_Z9ftemplateIiET_i
@@ -1120,7 +1168,7 @@ int bar(int n){
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP16]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP16]], align 4
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 3, ptr [[TMP17]], align 4
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1137,9 +1185,17 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP23]], align 8
// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l181.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK1-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP25]], align 8
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP28]], align 4
+// CHECK1-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l181.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK1-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l181(i64 [[TMP1]], i64 [[TMP3]], ptr [[B]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1149,8 +1205,8 @@ int bar(int n){
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l181(i64 [[TMP1]], i64 [[TMP3]], ptr [[B]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
-// CHECK1-NEXT: [[TMP27:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: ret i32 [[TMP27]]
+// CHECK1-NEXT: [[TMP31:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: ret i32 [[TMP31]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216
@@ -1399,7 +1455,7 @@ int bar(int n){
// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP17]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP17]], align 4
// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP18]], align 4
// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1416,177 +1472,201 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP24]], align 4
// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP25]], align 8
-// CHECK3-NEXT: [[TMP26:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP27:%.*]] = icmp ne i32 [[TMP26]], 0
-// CHECK3-NEXT: br i1 [[TMP27]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP26]], align 8
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
+// CHECK3-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110(i32 [[TMP11]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
-// CHECK3-NEXT: [[TMP28:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: store i32 [[TMP28]], ptr [[A_CASTED2]], align 4
-// CHECK3-NEXT: [[TMP29:%.*]] = load i32, ptr [[A_CASTED2]], align 4
-// CHECK3-NEXT: [[TMP30:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK3-NEXT: store i16 [[TMP30]], ptr [[AA_CASTED3]], align 2
-// CHECK3-NEXT: [[TMP31:%.*]] = load i32, ptr [[AA_CASTED3]], align 4
-// CHECK3-NEXT: [[TMP32:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP32]], 10
+// CHECK3-NEXT: [[TMP32:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: store i32 [[TMP32]], ptr [[A_CASTED2]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = load i32, ptr [[A_CASTED2]], align 4
+// CHECK3-NEXT: [[TMP34:%.*]] = load i16, ptr [[AA]], align 2
+// CHECK3-NEXT: store i16 [[TMP34]], ptr [[AA_CASTED3]], align 2
+// CHECK3-NEXT: [[TMP35:%.*]] = load i32, ptr [[AA_CASTED3]], align 4
+// CHECK3-NEXT: [[TMP36:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP36]], 10
// CHECK3-NEXT: br i1 [[CMP]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK3: omp_if.then:
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP29]], ptr [[TMP33]], align 4
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP29]], ptr [[TMP34]], align 4
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP35]], align 4
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP31]], ptr [[TMP36]], align 4
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP31]], ptr [[TMP37]], align 4
-// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
-// CHECK3-NEXT: store ptr null, ptr [[TMP38]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP33]], ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP33]], ptr [[TMP38]], align 4
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP39]], align 4
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP35]], ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP35]], ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
+// CHECK3-NEXT: store ptr null, ptr [[TMP42]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP41]], align 4
-// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK3-NEXT: store i32 2, ptr [[TMP42]], align 4
-// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP39]], ptr [[TMP43]], align 4
-// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP40]], ptr [[TMP44]], align 4
-// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.5, ptr [[TMP45]], align 4
-// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP46]], align 4
-// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP47]], align 4
-// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP48]], align 4
-// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP49]], align 8
-// CHECK3-NEXT: [[TMP50:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK3-NEXT: [[TMP51:%.*]] = icmp ne i32 [[TMP50]], 0
-// CHECK3-NEXT: br i1 [[TMP51]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK3-NEXT: store i32 2, ptr [[TMP46]], align 4
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 4
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 4
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.5, ptr [[TMP49]], align 4
+// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP50]], align 4
+// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP52]], align 4
+// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP53]], align 8
+// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP55]], align 4
+// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK3-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK3-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK3-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK3: omp_offload.failed8:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119(i32 [[TMP29]], i32 [[TMP31]]) #[[ATTR4]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119(i32 [[TMP33]], i32 [[TMP35]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK3: omp_offload.cont9:
// CHECK3-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK3: omp_if.else:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119(i32 [[TMP29]], i32 [[TMP31]]) #[[ATTR4]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119(i32 [[TMP33]], i32 [[TMP35]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
-// CHECK3-NEXT: [[TMP52:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: store i32 [[TMP52]], ptr [[A_CASTED10]], align 4
-// CHECK3-NEXT: [[TMP53:%.*]] = load i32, ptr [[A_CASTED10]], align 4
-// CHECK3-NEXT: [[TMP54:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: [[CMP11:%.*]] = icmp sgt i32 [[TMP54]], 20
+// CHECK3-NEXT: [[TMP60:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: store i32 [[TMP60]], ptr [[A_CASTED10]], align 4
+// CHECK3-NEXT: [[TMP61:%.*]] = load i32, ptr [[A_CASTED10]], align 4
+// CHECK3-NEXT: [[TMP62:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: [[CMP11:%.*]] = icmp sgt i32 [[TMP62]], 20
// CHECK3-NEXT: br i1 [[CMP11]], label [[OMP_IF_THEN12:%.*]], label [[OMP_IF_ELSE19:%.*]]
// CHECK3: omp_if.then12:
-// CHECK3-NEXT: [[TMP55:%.*]] = mul nuw i32 [[TMP1]], 4
-// CHECK3-NEXT: [[TMP56:%.*]] = sext i32 [[TMP55]] to i64
-// CHECK3-NEXT: [[TMP57:%.*]] = mul nuw i32 5, [[TMP3]]
-// CHECK3-NEXT: [[TMP58:%.*]] = mul nuw i32 [[TMP57]], 8
-// CHECK3-NEXT: [[TMP59:%.*]] = sext i32 [[TMP58]] to i64
+// CHECK3-NEXT: [[TMP63:%.*]] = mul nuw i32 [[TMP1]], 4
+// CHECK3-NEXT: [[TMP64:%.*]] = sext i32 [[TMP63]] to i64
+// CHECK3-NEXT: [[TMP65:%.*]] = mul nuw i32 5, [[TMP3]]
+// CHECK3-NEXT: [[TMP66:%.*]] = mul nuw i32 [[TMP65]], 8
+// CHECK3-NEXT: [[TMP67:%.*]] = sext i32 [[TMP66]] to i64
// CHECK3-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES]], ptr align 4 @.offload_sizes.8, i32 72, i1 false)
-// CHECK3-NEXT: [[TMP60:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP53]], ptr [[TMP60]], align 4
-// CHECK3-NEXT: [[TMP61:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP53]], ptr [[TMP61]], align 4
-// CHECK3-NEXT: [[TMP62:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP62]], align 4
-// CHECK3-NEXT: [[TMP63:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 1
-// CHECK3-NEXT: store ptr [[B]], ptr [[TMP63]], align 4
-// CHECK3-NEXT: [[TMP64:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 1
-// CHECK3-NEXT: store ptr [[B]], ptr [[TMP64]], align 4
-// CHECK3-NEXT: [[TMP65:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i32 0, i32 1
-// CHECK3-NEXT: store ptr null, ptr [[TMP65]], align 4
-// CHECK3-NEXT: [[TMP66:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 2
-// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP66]], align 4
-// CHECK3-NEXT: [[TMP67:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 2
-// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP67]], align 4
-// CHECK3-NEXT: [[TMP68:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i32 0, i32 2
-// CHECK3-NEXT: store ptr null, ptr [[TMP68]], align 4
-// CHECK3-NEXT: [[TMP69:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[VLA]], ptr [[TMP69]], align 4
-// CHECK3-NEXT: [[TMP70:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[VLA]], ptr [[TMP70]], align 4
-// CHECK3-NEXT: [[TMP71:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
-// CHECK3-NEXT: store i64 [[TMP56]], ptr [[TMP71]], align 4
-// CHECK3-NEXT: [[TMP72:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i32 0, i32 3
-// CHECK3-NEXT: store ptr null, ptr [[TMP72]], align 4
-// CHECK3-NEXT: [[TMP73:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 4
-// CHECK3-NEXT: store ptr [[C]], ptr [[TMP73]], align 4
-// CHECK3-NEXT: [[TMP74:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 4
-// CHECK3-NEXT: store ptr [[C]], ptr [[TMP74]], align 4
-// CHECK3-NEXT: [[TMP75:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i32 0, i32 4
-// CHECK3-NEXT: store ptr null, ptr [[TMP75]], align 4
-// CHECK3-NEXT: [[TMP76:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 5
-// CHECK3-NEXT: store i32 5, ptr [[TMP76]], align 4
-// CHECK3-NEXT: [[TMP77:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 5
-// CHECK3-NEXT: store i32 5, ptr [[TMP77]], align 4
-// CHECK3-NEXT: [[TMP78:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i32 0, i32 5
-// CHECK3-NEXT: store ptr null, ptr [[TMP78]], align 4
-// CHECK3-NEXT: [[TMP79:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 6
-// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP79]], align 4
-// CHECK3-NEXT: [[TMP80:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 6
-// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP80]], align 4
-// CHECK3-NEXT: [[TMP81:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP81]], align 4
-// CHECK3-NEXT: [[TMP82:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 7
-// CHECK3-NEXT: store ptr [[VLA1]], ptr [[TMP82]], align 4
-// CHECK3-NEXT: [[TMP83:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 7
-// CHECK3-NEXT: store ptr [[VLA1]], ptr [[TMP83]], align 4
-// CHECK3-NEXT: [[TMP84:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
-// CHECK3-NEXT: store i64 [[TMP59]], ptr [[TMP84]], align 4
-// CHECK3-NEXT: [[TMP85:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP85]], align 4
-// CHECK3-NEXT: [[TMP86:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 8
-// CHECK3-NEXT: store ptr [[D]], ptr [[TMP86]], align 4
-// CHECK3-NEXT: [[TMP87:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 8
-// CHECK3-NEXT: store ptr [[D]], ptr [[TMP87]], align 4
-// CHECK3-NEXT: [[TMP88:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i32 0, i32 8
-// CHECK3-NEXT: store ptr null, ptr [[TMP88]], align 4
-// CHECK3-NEXT: [[TMP89:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP90:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP91:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP68:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP61]], ptr [[TMP68]], align 4
+// CHECK3-NEXT: [[TMP69:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP61]], ptr [[TMP69]], align 4
+// CHECK3-NEXT: [[TMP70:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP70]], align 4
+// CHECK3-NEXT: [[TMP71:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 1
+// CHECK3-NEXT: store ptr [[B]], ptr [[TMP71]], align 4
+// CHECK3-NEXT: [[TMP72:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 1
+// CHECK3-NEXT: store ptr [[B]], ptr [[TMP72]], align 4
+// CHECK3-NEXT: [[TMP73:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i32 0, i32 1
+// CHECK3-NEXT: store ptr null, ptr [[TMP73]], align 4
+// CHECK3-NEXT: [[TMP74:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 2
+// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP74]], align 4
+// CHECK3-NEXT: [[TMP75:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 2
+// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP75]], align 4
+// CHECK3-NEXT: [[TMP76:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i32 0, i32 2
+// CHECK3-NEXT: store ptr null, ptr [[TMP76]], align 4
+// CHECK3-NEXT: [[TMP77:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[VLA]], ptr [[TMP77]], align 4
+// CHECK3-NEXT: [[TMP78:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[VLA]], ptr [[TMP78]], align 4
+// CHECK3-NEXT: [[TMP79:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
+// CHECK3-NEXT: store i64 [[TMP64]], ptr [[TMP79]], align 4
+// CHECK3-NEXT: [[TMP80:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i32 0, i32 3
+// CHECK3-NEXT: store ptr null, ptr [[TMP80]], align 4
+// CHECK3-NEXT: [[TMP81:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 4
+// CHECK3-NEXT: store ptr [[C]], ptr [[TMP81]], align 4
+// CHECK3-NEXT: [[TMP82:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 4
+// CHECK3-NEXT: store ptr [[C]], ptr [[TMP82]], align 4
+// CHECK3-NEXT: [[TMP83:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i32 0, i32 4
+// CHECK3-NEXT: store ptr null, ptr [[TMP83]], align 4
+// CHECK3-NEXT: [[TMP84:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 5
+// CHECK3-NEXT: store i32 5, ptr [[TMP84]], align 4
+// CHECK3-NEXT: [[TMP85:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 5
+// CHECK3-NEXT: store i32 5, ptr [[TMP85]], align 4
+// CHECK3-NEXT: [[TMP86:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i32 0, i32 5
+// CHECK3-NEXT: store ptr null, ptr [[TMP86]], align 4
+// CHECK3-NEXT: [[TMP87:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 6
+// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP87]], align 4
+// CHECK3-NEXT: [[TMP88:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 6
+// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP88]], align 4
+// CHECK3-NEXT: [[TMP89:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP89]], align 4
+// CHECK3-NEXT: [[TMP90:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 7
+// CHECK3-NEXT: store ptr [[VLA1]], ptr [[TMP90]], align 4
+// CHECK3-NEXT: [[TMP91:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 7
+// CHECK3-NEXT: store ptr [[VLA1]], ptr [[TMP91]], align 4
+// CHECK3-NEXT: [[TMP92:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
+// CHECK3-NEXT: store i64 [[TMP67]], ptr [[TMP92]], align 4
+// CHECK3-NEXT: [[TMP93:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP93]], align 4
+// CHECK3-NEXT: [[TMP94:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 8
+// CHECK3-NEXT: store ptr [[D]], ptr [[TMP94]], align 4
+// CHECK3-NEXT: [[TMP95:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 8
+// CHECK3-NEXT: store ptr [[D]], ptr [[TMP95]], align 4
+// CHECK3-NEXT: [[TMP96:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i32 0, i32 8
+// CHECK3-NEXT: store ptr null, ptr [[TMP96]], align 4
+// CHECK3-NEXT: [[TMP97:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP98:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP99:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS16:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP92]], align 4
-// CHECK3-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 1
-// CHECK3-NEXT: store i32 9, ptr [[TMP93]], align 4
-// CHECK3-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP89]], ptr [[TMP94]], align 4
-// CHECK3-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP90]], ptr [[TMP95]], align 4
-// CHECK3-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 4
-// CHECK3-NEXT: store ptr [[TMP91]], ptr [[TMP96]], align 4
-// CHECK3-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP97]], align 4
-// CHECK3-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP98]], align 4
-// CHECK3-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP99]], align 4
-// CHECK3-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP100]], align 8
-// CHECK3-NEXT: [[TMP101:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144.region_id, ptr [[KERNEL_ARGS16]])
-// CHECK3-NEXT: [[TMP102:%.*]] = icmp ne i32 [[TMP101]], 0
-// CHECK3-NEXT: br i1 [[TMP102]], label [[OMP_OFFLOAD_FAILED17:%.*]], label [[OMP_OFFLOAD_CONT18:%.*]]
+// CHECK3-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP100]], align 4
+// CHECK3-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 1
+// CHECK3-NEXT: store i32 9, ptr [[TMP101]], align 4
+// CHECK3-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP97]], ptr [[TMP102]], align 4
+// CHECK3-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP98]], ptr [[TMP103]], align 4
+// CHECK3-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 4
+// CHECK3-NEXT: store ptr [[TMP99]], ptr [[TMP104]], align 4
+// CHECK3-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP105]], align 4
+// CHECK3-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP106]], align 4
+// CHECK3-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP107]], align 4
+// CHECK3-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP108]], align 8
+// CHECK3-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP109]], align 8
+// CHECK3-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP110]], align 4
+// CHECK3-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP111]], align 4
+// CHECK3-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP112]], align 4
+// CHECK3-NEXT: [[TMP113:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144.region_id, ptr [[KERNEL_ARGS16]])
+// CHECK3-NEXT: [[TMP114:%.*]] = icmp ne i32 [[TMP113]], 0
+// CHECK3-NEXT: br i1 [[TMP114]], label [[OMP_OFFLOAD_FAILED17:%.*]], label [[OMP_OFFLOAD_CONT18:%.*]]
// CHECK3: omp_offload.failed17:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144(i32 [[TMP53]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]]) #[[ATTR4]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144(i32 [[TMP61]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT18]]
// CHECK3: omp_offload.cont18:
// CHECK3-NEXT: br label [[OMP_IF_END20:%.*]]
// CHECK3: omp_if.else19:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144(i32 [[TMP53]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]]) #[[ATTR4]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144(i32 [[TMP61]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_IF_END20]]
// CHECK3: omp_if.end20:
-// CHECK3-NEXT: [[TMP103:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: [[TMP104:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK3-NEXT: call void @llvm.stackrestore(ptr [[TMP104]])
-// CHECK3-NEXT: ret i32 [[TMP103]]
+// CHECK3-NEXT: [[TMP115:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: [[TMP116:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK3-NEXT: call void @llvm.stackrestore(ptr [[TMP116]])
+// CHECK3-NEXT: ret i32 [[TMP115]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l100
@@ -1637,7 +1717,7 @@ int bar(int n){
// CHECK3-NEXT: store ptr [[TMP3]], ptr [[DOTTASK_T__ADDR_I]], align 4, !noalias !22
// CHECK3-NEXT: store ptr [[TMP7]], ptr [[__CONTEXT_ADDR_I]], align 4, !noalias !22
// CHECK3-NEXT: [[TMP8:%.*]] = load ptr, ptr [[__CONTEXT_ADDR_I]], align 4, !noalias !22
-// CHECK3-NEXT: store i32 1, ptr [[KERNEL_ARGS_I]], align 4, !noalias !22
+// CHECK3-NEXT: store i32 2, ptr [[KERNEL_ARGS_I]], align 4, !noalias !22
// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP9]], align 4, !noalias !22
// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 2
@@ -1654,9 +1734,17 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP15]], align 4, !noalias !22
// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP16]], align 8, !noalias !22
-// CHECK3-NEXT: [[TMP17:%.*]] = call i32 @__tgt_target_kernel_nowait(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l100.region_id, ptr [[KERNEL_ARGS_I]], i32 0, ptr null, i32 0, ptr null)
-// CHECK3-NEXT: [[TMP18:%.*]] = icmp ne i32 [[TMP17]], 0
-// CHECK3-NEXT: br i1 [[TMP18]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__1_EXIT:%.*]]
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 9
+// CHECK3-NEXT: store i64 1, ptr [[TMP17]], align 8, !noalias !22
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP18]], align 4, !noalias !22
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP19]], align 4, !noalias !22
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP20]], align 4, !noalias !22
+// CHECK3-NEXT: [[TMP21:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l100.region_id, ptr [[KERNEL_ARGS_I]])
+// CHECK3-NEXT: [[TMP22:%.*]] = icmp ne i32 [[TMP21]], 0
+// CHECK3-NEXT: br i1 [[TMP22]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__1_EXIT:%.*]]
// CHECK3: omp_offload.failed.i:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l100() #[[ATTR4]]
// CHECK3-NEXT: br label [[DOTOMP_OUTLINED__1_EXIT]]
@@ -1987,7 +2075,7 @@ int bar(int n){
// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [5 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP29]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP29]], align 4
// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 5, ptr [[TMP30]], align 4
// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2004,9 +2092,17 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP36]], align 4
// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP37]], align 8
-// CHECK3-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK3-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP38]], align 8
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP39]], align 4
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
+// CHECK3-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216(ptr [[THIS1]], i32 [[TMP5]], i32 2, i32 [[TMP1]], ptr [[VLA]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -2016,15 +2112,15 @@ int bar(int n){
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216(ptr [[THIS1]], i32 [[TMP5]], i32 2, i32 [[TMP1]], ptr [[VLA]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
-// CHECK3-NEXT: [[TMP40:%.*]] = mul nsw i32 1, [[TMP1]]
-// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i32 [[TMP40]]
+// CHECK3-NEXT: [[TMP44:%.*]] = mul nsw i32 1, [[TMP1]]
+// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i32 [[TMP44]]
// CHECK3-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i16, ptr [[ARRAYIDX]], i32 1
-// CHECK3-NEXT: [[TMP41:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
-// CHECK3-NEXT: [[CONV:%.*]] = sext i16 [[TMP41]] to i32
-// CHECK3-NEXT: [[TMP42:%.*]] = load i32, ptr [[B]], align 4
-// CHECK3-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP42]]
-// CHECK3-NEXT: [[TMP43:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK3-NEXT: call void @llvm.stackrestore(ptr [[TMP43]])
+// CHECK3-NEXT: [[TMP45:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
+// CHECK3-NEXT: [[CONV:%.*]] = sext i16 [[TMP45]] to i32
+// CHECK3-NEXT: [[TMP46:%.*]] = load i32, ptr [[B]], align 4
+// CHECK3-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP46]]
+// CHECK3-NEXT: [[TMP47:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK3-NEXT: call void @llvm.stackrestore(ptr [[TMP47]])
// CHECK3-NEXT: ret i32 [[ADD3]]
//
//
@@ -2087,7 +2183,7 @@ int bar(int n){
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 4, ptr [[TMP22]], align 4
// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2104,9 +2200,17 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4
// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP29]], align 8
-// CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l198.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK3-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK3-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l198.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK3-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l198(i32 [[TMP1]], i32 [[TMP3]], i32 [[TMP5]], ptr [[B]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -2116,8 +2220,8 @@ int bar(int n){
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l198(i32 [[TMP1]], i32 [[TMP3]], i32 [[TMP5]], ptr [[B]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
-// CHECK3-NEXT: [[TMP32:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: ret i32 [[TMP32]]
+// CHECK3-NEXT: [[TMP36:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: ret i32 [[TMP36]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_Z9ftemplateIiET_i
@@ -2167,7 +2271,7 @@ int bar(int n){
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP16]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP16]], align 4
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 3, ptr [[TMP17]], align 4
// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2184,9 +2288,17 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP23]], align 4
// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP24]], align 8
-// CHECK3-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l181.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK3-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP25]], align 8
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l181.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK3-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l181(i32 [[TMP1]], i32 [[TMP3]], ptr [[B]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -2196,8 +2308,8 @@ int bar(int n){
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l181(i32 [[TMP1]], i32 [[TMP3]], ptr [[B]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
-// CHECK3-NEXT: [[TMP27:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: ret i32 [[TMP27]]
+// CHECK3-NEXT: [[TMP31:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: ret i32 [[TMP31]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216
diff --git a/clang/test/OpenMP/target_parallel_depend_codegen.cpp b/clang/test/OpenMP/target_parallel_depend_codegen.cpp
index 49cb76f61dc03..e7992e76f4ea2 100644
--- a/clang/test/OpenMP/target_parallel_depend_codegen.cpp
+++ b/clang/test/OpenMP/target_parallel_depend_codegen.cpp
@@ -205,7 +205,7 @@ int foo(int n) {
// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2
// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]],
// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64
-// CHECK: [[RET:%.+]] = call i32 @__tgt_target_kernel_nowait(%struct.ident_t* @{{.+}}, i64 [[DEVICE]], i32 1, i32 0, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* %{{.+}})
+// CHECK: [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 [[DEVICE]], i32 1, i32 0, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* %{{.+}})
// CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]]
diff --git a/clang/test/OpenMP/target_parallel_for_codegen.cpp b/clang/test/OpenMP/target_parallel_for_codegen.cpp
index 8f14afb70c674..2d14c963b0577 100644
--- a/clang/test/OpenMP/target_parallel_for_codegen.cpp
+++ b/clang/test/OpenMP/target_parallel_for_codegen.cpp
@@ -364,7 +364,7 @@ int bar(int n){
// CHECK1-NEXT: store i64 [[TMP5]], ptr [[__VLA_EXPR1]], align 8
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP8]], align 4
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -381,250 +381,274 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l103.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK1-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l103.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l103() #[[ATTR4:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[CALL:%.*]] = call noundef i64 @_Z7get_valv()
// CHECK1-NEXT: store i64 [[CALL]], ptr [[K]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: store i32 [[TMP18]], ptr [[A_CASTED]], align 4
-// CHECK1-NEXT: [[TMP19:%.*]] = load i64, ptr [[A_CASTED]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = load i64, ptr [[K]], align 8
-// CHECK1-NEXT: store i64 [[TMP20]], ptr [[K_CASTED]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = load i64, ptr [[K_CASTED]], align 8
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110(i64 [[TMP19]], i64 [[TMP21]]) #[[ATTR4]]
+// CHECK1-NEXT: [[TMP22:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: store i32 [[TMP22]], ptr [[A_CASTED]], align 4
+// CHECK1-NEXT: [[TMP23:%.*]] = load i64, ptr [[A_CASTED]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = load i64, ptr [[K]], align 8
+// CHECK1-NEXT: store i64 [[TMP24]], ptr [[K_CASTED]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = load i64, ptr [[K_CASTED]], align 8
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110(i64 [[TMP23]], i64 [[TMP25]]) #[[ATTR4]]
// CHECK1-NEXT: store i32 12, ptr [[LIN]], align 4
-// CHECK1-NEXT: [[TMP22:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK1-NEXT: store i16 [[TMP22]], ptr [[AA_CASTED]], align 2
-// CHECK1-NEXT: [[TMP23:%.*]] = load i64, ptr [[AA_CASTED]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = load i32, ptr [[LIN]], align 4
-// CHECK1-NEXT: store i32 [[TMP24]], ptr [[LIN_CASTED]], align 4
-// CHECK1-NEXT: [[TMP25:%.*]] = load i64, ptr [[LIN_CASTED]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: store i32 [[TMP26]], ptr [[A_CASTED2]], align 4
-// CHECK1-NEXT: [[TMP27:%.*]] = load i64, ptr [[A_CASTED2]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP23]], ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP23]], ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP30]], align 8
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP25]], ptr [[TMP31]], align 8
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP25]], ptr [[TMP32]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
-// CHECK1-NEXT: store i64 [[TMP27]], ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
-// CHECK1-NEXT: store i64 [[TMP27]], ptr [[TMP35]], align 8
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP36]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP40:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK1-NEXT: store i16 [[TMP40]], ptr [[TMP39]], align 4
-// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 1
-// CHECK1-NEXT: [[TMP42:%.*]] = load i32, ptr [[LIN]], align 4
-// CHECK1-NEXT: store i32 [[TMP42]], ptr [[TMP41]], align 4
-// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 2
-// CHECK1-NEXT: [[TMP44:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: store i32 [[TMP44]], ptr [[TMP43]], align 4
-// CHECK1-NEXT: [[TMP45:%.*]] = call ptr @__kmpc_omp_target_task_alloc(ptr @[[GLOB2]], i32 [[TMP0]], i32 1, i64 120, i64 12, ptr @.omp_task_entry., i64 -1)
-// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], ptr [[TMP45]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T:%.*]], ptr [[TMP46]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP48:%.*]] = load ptr, ptr [[TMP47]], align 8
-// CHECK1-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 4 [[TMP48]], ptr align 4 [[AGG_CAPTURED]], i64 12, i1 false)
-// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES]], ptr [[TMP45]], i32 0, i32 1
-// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T:%.*]], ptr [[TMP49]], i32 0, i32 0
-// CHECK1-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[TMP50]], ptr align 8 [[TMP37]], i64 24, i1 false)
-// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T]], ptr [[TMP49]], i32 0, i32 1
-// CHECK1-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[TMP51]], ptr align 8 [[TMP38]], i64 24, i1 false)
-// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T]], ptr [[TMP49]], i32 0, i32 2
-// CHECK1-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[TMP52]], ptr align 8 @.offload_sizes, i64 24, i1 false)
-// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T]], ptr [[TMP49]], i32 0, i32 3
-// CHECK1-NEXT: [[TMP54:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK1-NEXT: store i16 [[TMP54]], ptr [[TMP53]], align 8
-// CHECK1-NEXT: [[TMP55:%.*]] = call i32 @__kmpc_omp_task(ptr @[[GLOB2]], i32 [[TMP0]], ptr [[TMP45]])
-// CHECK1-NEXT: [[TMP56:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: store i32 [[TMP56]], ptr [[A_CASTED3]], align 4
-// CHECK1-NEXT: [[TMP57:%.*]] = load i64, ptr [[A_CASTED3]], align 8
+// CHECK1-NEXT: [[TMP26:%.*]] = load i16, ptr [[AA]], align 2
+// CHECK1-NEXT: store i16 [[TMP26]], ptr [[AA_CASTED]], align 2
+// CHECK1-NEXT: [[TMP27:%.*]] = load i64, ptr [[AA_CASTED]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = load i32, ptr [[LIN]], align 4
+// CHECK1-NEXT: store i32 [[TMP28]], ptr [[LIN_CASTED]], align 4
+// CHECK1-NEXT: [[TMP29:%.*]] = load i64, ptr [[LIN_CASTED]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: store i32 [[TMP30]], ptr [[A_CASTED2]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = load i64, ptr [[A_CASTED2]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP27]], ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP27]], ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP29]], ptr [[TMP35]], align 8
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP29]], ptr [[TMP36]], align 8
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP37]], align 8
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
+// CHECK1-NEXT: store i64 [[TMP31]], ptr [[TMP38]], align 8
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
+// CHECK1-NEXT: store i64 [[TMP31]], ptr [[TMP39]], align 8
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 2
+// CHECK1-NEXT: store ptr null, ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP44:%.*]] = load i16, ptr [[AA]], align 2
+// CHECK1-NEXT: store i16 [[TMP44]], ptr [[TMP43]], align 4
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 1
+// CHECK1-NEXT: [[TMP46:%.*]] = load i32, ptr [[LIN]], align 4
+// CHECK1-NEXT: store i32 [[TMP46]], ptr [[TMP45]], align 4
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 2
+// CHECK1-NEXT: [[TMP48:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: store i32 [[TMP48]], ptr [[TMP47]], align 4
+// CHECK1-NEXT: [[TMP49:%.*]] = call ptr @__kmpc_omp_target_task_alloc(ptr @[[GLOB2]], i32 [[TMP0]], i32 1, i64 120, i64 12, ptr @.omp_task_entry., i64 -1)
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], ptr [[TMP49]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T:%.*]], ptr [[TMP50]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP52:%.*]] = load ptr, ptr [[TMP51]], align 8
+// CHECK1-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 4 [[TMP52]], ptr align 4 [[AGG_CAPTURED]], i64 12, i1 false)
+// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES]], ptr [[TMP49]], i32 0, i32 1
+// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T:%.*]], ptr [[TMP53]], i32 0, i32 0
+// CHECK1-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[TMP54]], ptr align 8 [[TMP41]], i64 24, i1 false)
+// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T]], ptr [[TMP53]], i32 0, i32 1
+// CHECK1-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[TMP55]], ptr align 8 [[TMP42]], i64 24, i1 false)
+// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T]], ptr [[TMP53]], i32 0, i32 2
+// CHECK1-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[TMP56]], ptr align 8 @.offload_sizes, i64 24, i1 false)
+// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T]], ptr [[TMP53]], i32 0, i32 3
// CHECK1-NEXT: [[TMP58:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK1-NEXT: store i16 [[TMP58]], ptr [[AA_CASTED4]], align 2
-// CHECK1-NEXT: [[TMP59:%.*]] = load i64, ptr [[AA_CASTED4]], align 8
-// CHECK1-NEXT: [[TMP60:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP60]], 10
+// CHECK1-NEXT: store i16 [[TMP58]], ptr [[TMP57]], align 8
+// CHECK1-NEXT: [[TMP59:%.*]] = call i32 @__kmpc_omp_task(ptr @[[GLOB2]], i32 [[TMP0]], ptr [[TMP49]])
+// CHECK1-NEXT: [[TMP60:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: store i32 [[TMP60]], ptr [[A_CASTED3]], align 4
+// CHECK1-NEXT: [[TMP61:%.*]] = load i64, ptr [[A_CASTED3]], align 8
+// CHECK1-NEXT: [[TMP62:%.*]] = load i16, ptr [[AA]], align 2
+// CHECK1-NEXT: store i16 [[TMP62]], ptr [[AA_CASTED4]], align 2
+// CHECK1-NEXT: [[TMP63:%.*]] = load i64, ptr [[AA_CASTED4]], align 8
+// CHECK1-NEXT: [[TMP64:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP64]], 10
// CHECK1-NEXT: br i1 [[CMP]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK1: omp_if.then:
-// CHECK1-NEXT: [[TMP61:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP57]], ptr [[TMP61]], align 8
-// CHECK1-NEXT: [[TMP62:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP57]], ptr [[TMP62]], align 8
-// CHECK1-NEXT: [[TMP63:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP63]], align 8
-// CHECK1-NEXT: [[TMP64:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP59]], ptr [[TMP64]], align 8
-// CHECK1-NEXT: [[TMP65:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP59]], ptr [[TMP65]], align 8
-// CHECK1-NEXT: [[TMP66:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP66]], align 8
-// CHECK1-NEXT: [[TMP67:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP68:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP65:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP61]], ptr [[TMP65]], align 8
+// CHECK1-NEXT: [[TMP66:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP61]], ptr [[TMP66]], align 8
+// CHECK1-NEXT: [[TMP67:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP67]], align 8
+// CHECK1-NEXT: [[TMP68:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP63]], ptr [[TMP68]], align 8
+// CHECK1-NEXT: [[TMP69:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP63]], ptr [[TMP69]], align 8
+// CHECK1-NEXT: [[TMP70:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP70]], align 8
+// CHECK1-NEXT: [[TMP71:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP72:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS8:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP69]], align 4
-// CHECK1-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 1
-// CHECK1-NEXT: store i32 2, ptr [[TMP70]], align 4
-// CHECK1-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP67]], ptr [[TMP71]], align 8
-// CHECK1-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 8
-// CHECK1-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.5, ptr [[TMP73]], align 8
-// CHECK1-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP74]], align 8
-// CHECK1-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP75]], align 8
-// CHECK1-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP76]], align 8
-// CHECK1-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP77]], align 8
-// CHECK1-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l136.region_id, ptr [[KERNEL_ARGS8]])
-// CHECK1-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
-// CHECK1-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED9:%.*]], label [[OMP_OFFLOAD_CONT10:%.*]]
+// CHECK1-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP73]], align 4
+// CHECK1-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 1
+// CHECK1-NEXT: store i32 2, ptr [[TMP74]], align 4
+// CHECK1-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP71]], ptr [[TMP75]], align 8
+// CHECK1-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP72]], ptr [[TMP76]], align 8
+// CHECK1-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.5, ptr [[TMP77]], align 8
+// CHECK1-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP78]], align 8
+// CHECK1-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP79]], align 8
+// CHECK1-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP80]], align 8
+// CHECK1-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP81]], align 8
+// CHECK1-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP82]], align 8
+// CHECK1-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP83]], align 4
+// CHECK1-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP84]], align 4
+// CHECK1-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP85]], align 4
+// CHECK1-NEXT: [[TMP86:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l136.region_id, ptr [[KERNEL_ARGS8]])
+// CHECK1-NEXT: [[TMP87:%.*]] = icmp ne i32 [[TMP86]], 0
+// CHECK1-NEXT: br i1 [[TMP87]], label [[OMP_OFFLOAD_FAILED9:%.*]], label [[OMP_OFFLOAD_CONT10:%.*]]
// CHECK1: omp_offload.failed9:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l136(i64 [[TMP57]], i64 [[TMP59]]) #[[ATTR4]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l136(i64 [[TMP61]], i64 [[TMP63]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT10]]
// CHECK1: omp_offload.cont10:
// CHECK1-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK1: omp_if.else:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l136(i64 [[TMP57]], i64 [[TMP59]]) #[[ATTR4]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l136(i64 [[TMP61]], i64 [[TMP63]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
-// CHECK1-NEXT: [[TMP80:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: store i32 [[TMP80]], ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK1-NEXT: [[TMP81:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: store i32 [[TMP81]], ptr [[A_CASTED11]], align 4
-// CHECK1-NEXT: [[TMP82:%.*]] = load i64, ptr [[A_CASTED11]], align 8
-// CHECK1-NEXT: [[TMP83:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK1-NEXT: store i32 [[TMP83]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK1-NEXT: [[TMP84:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK1-NEXT: [[TMP85:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: [[CMP12:%.*]] = icmp sgt i32 [[TMP85]], 20
+// CHECK1-NEXT: [[TMP88:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: store i32 [[TMP88]], ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK1-NEXT: [[TMP89:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: store i32 [[TMP89]], ptr [[A_CASTED11]], align 4
+// CHECK1-NEXT: [[TMP90:%.*]] = load i64, ptr [[A_CASTED11]], align 8
+// CHECK1-NEXT: [[TMP91:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK1-NEXT: store i32 [[TMP91]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK1-NEXT: [[TMP92:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK1-NEXT: [[TMP93:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: [[CMP12:%.*]] = icmp sgt i32 [[TMP93]], 20
// CHECK1-NEXT: br i1 [[CMP12]], label [[OMP_IF_THEN13:%.*]], label [[OMP_IF_ELSE20:%.*]]
// CHECK1: omp_if.then13:
-// CHECK1-NEXT: [[TMP86:%.*]] = mul nuw i64 [[TMP2]], 4
-// CHECK1-NEXT: [[TMP87:%.*]] = mul nuw i64 5, [[TMP5]]
-// CHECK1-NEXT: [[TMP88:%.*]] = mul nuw i64 [[TMP87]], 8
+// CHECK1-NEXT: [[TMP94:%.*]] = mul nuw i64 [[TMP2]], 4
+// CHECK1-NEXT: [[TMP95:%.*]] = mul nuw i64 5, [[TMP5]]
+// CHECK1-NEXT: [[TMP96:%.*]] = mul nuw i64 [[TMP95]], 8
// CHECK1-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES]], ptr align 8 @.offload_sizes.8, i64 80, i1 false)
-// CHECK1-NEXT: [[TMP89:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP82]], ptr [[TMP89]], align 8
-// CHECK1-NEXT: [[TMP90:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP82]], ptr [[TMP90]], align 8
-// CHECK1-NEXT: [[TMP91:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP91]], align 8
-// CHECK1-NEXT: [[TMP92:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 1
-// CHECK1-NEXT: store ptr [[B]], ptr [[TMP92]], align 8
-// CHECK1-NEXT: [[TMP93:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 1
-// CHECK1-NEXT: store ptr [[B]], ptr [[TMP93]], align 8
-// CHECK1-NEXT: [[TMP94:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP94]], align 8
-// CHECK1-NEXT: [[TMP95:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 2
-// CHECK1-NEXT: store i64 [[TMP2]], ptr [[TMP95]], align 8
-// CHECK1-NEXT: [[TMP96:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 2
-// CHECK1-NEXT: store i64 [[TMP2]], ptr [[TMP96]], align 8
-// CHECK1-NEXT: [[TMP97:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP97]], align 8
-// CHECK1-NEXT: [[TMP98:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[VLA]], ptr [[TMP98]], align 8
-// CHECK1-NEXT: [[TMP99:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[VLA]], ptr [[TMP99]], align 8
-// CHECK1-NEXT: [[TMP100:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
-// CHECK1-NEXT: store i64 [[TMP86]], ptr [[TMP100]], align 8
-// CHECK1-NEXT: [[TMP101:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP101]], align 8
-// CHECK1-NEXT: [[TMP102:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 4
-// CHECK1-NEXT: store ptr [[C]], ptr [[TMP102]], align 8
-// CHECK1-NEXT: [[TMP103:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 4
-// CHECK1-NEXT: store ptr [[C]], ptr [[TMP103]], align 8
-// CHECK1-NEXT: [[TMP104:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP104]], align 8
-// CHECK1-NEXT: [[TMP105:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 5
-// CHECK1-NEXT: store i64 5, ptr [[TMP105]], align 8
-// CHECK1-NEXT: [[TMP106:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 5
-// CHECK1-NEXT: store i64 5, ptr [[TMP106]], align 8
-// CHECK1-NEXT: [[TMP107:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP107]], align 8
-// CHECK1-NEXT: [[TMP108:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 6
-// CHECK1-NEXT: store i64 [[TMP5]], ptr [[TMP108]], align 8
-// CHECK1-NEXT: [[TMP109:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 6
-// CHECK1-NEXT: store i64 [[TMP5]], ptr [[TMP109]], align 8
-// CHECK1-NEXT: [[TMP110:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP110]], align 8
-// CHECK1-NEXT: [[TMP111:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 7
-// CHECK1-NEXT: store ptr [[VLA1]], ptr [[TMP111]], align 8
-// CHECK1-NEXT: [[TMP112:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 7
-// CHECK1-NEXT: store ptr [[VLA1]], ptr [[TMP112]], align 8
-// CHECK1-NEXT: [[TMP113:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
-// CHECK1-NEXT: store i64 [[TMP88]], ptr [[TMP113]], align 8
-// CHECK1-NEXT: [[TMP114:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP114]], align 8
-// CHECK1-NEXT: [[TMP115:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 8
-// CHECK1-NEXT: store ptr [[D]], ptr [[TMP115]], align 8
-// CHECK1-NEXT: [[TMP116:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 8
-// CHECK1-NEXT: store ptr [[D]], ptr [[TMP116]], align 8
-// CHECK1-NEXT: [[TMP117:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 8
-// CHECK1-NEXT: store ptr null, ptr [[TMP117]], align 8
-// CHECK1-NEXT: [[TMP118:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 9
-// CHECK1-NEXT: store i64 [[TMP84]], ptr [[TMP118]], align 8
-// CHECK1-NEXT: [[TMP119:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 9
-// CHECK1-NEXT: store i64 [[TMP84]], ptr [[TMP119]], align 8
-// CHECK1-NEXT: [[TMP120:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 9
-// CHECK1-NEXT: store ptr null, ptr [[TMP120]], align 8
-// CHECK1-NEXT: [[TMP121:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP122:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP123:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP97:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP90]], ptr [[TMP97]], align 8
+// CHECK1-NEXT: [[TMP98:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP90]], ptr [[TMP98]], align 8
+// CHECK1-NEXT: [[TMP99:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP99]], align 8
+// CHECK1-NEXT: [[TMP100:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 1
+// CHECK1-NEXT: store ptr [[B]], ptr [[TMP100]], align 8
+// CHECK1-NEXT: [[TMP101:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 1
+// CHECK1-NEXT: store ptr [[B]], ptr [[TMP101]], align 8
+// CHECK1-NEXT: [[TMP102:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP102]], align 8
+// CHECK1-NEXT: [[TMP103:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 2
+// CHECK1-NEXT: store i64 [[TMP2]], ptr [[TMP103]], align 8
+// CHECK1-NEXT: [[TMP104:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 2
+// CHECK1-NEXT: store i64 [[TMP2]], ptr [[TMP104]], align 8
+// CHECK1-NEXT: [[TMP105:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 2
+// CHECK1-NEXT: store ptr null, ptr [[TMP105]], align 8
+// CHECK1-NEXT: [[TMP106:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[VLA]], ptr [[TMP106]], align 8
+// CHECK1-NEXT: [[TMP107:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[VLA]], ptr [[TMP107]], align 8
+// CHECK1-NEXT: [[TMP108:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
+// CHECK1-NEXT: store i64 [[TMP94]], ptr [[TMP108]], align 8
+// CHECK1-NEXT: [[TMP109:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 3
+// CHECK1-NEXT: store ptr null, ptr [[TMP109]], align 8
+// CHECK1-NEXT: [[TMP110:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 4
+// CHECK1-NEXT: store ptr [[C]], ptr [[TMP110]], align 8
+// CHECK1-NEXT: [[TMP111:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 4
+// CHECK1-NEXT: store ptr [[C]], ptr [[TMP111]], align 8
+// CHECK1-NEXT: [[TMP112:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP112]], align 8
+// CHECK1-NEXT: [[TMP113:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 5
+// CHECK1-NEXT: store i64 5, ptr [[TMP113]], align 8
+// CHECK1-NEXT: [[TMP114:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 5
+// CHECK1-NEXT: store i64 5, ptr [[TMP114]], align 8
+// CHECK1-NEXT: [[TMP115:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP115]], align 8
+// CHECK1-NEXT: [[TMP116:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 6
+// CHECK1-NEXT: store i64 [[TMP5]], ptr [[TMP116]], align 8
+// CHECK1-NEXT: [[TMP117:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 6
+// CHECK1-NEXT: store i64 [[TMP5]], ptr [[TMP117]], align 8
+// CHECK1-NEXT: [[TMP118:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP118]], align 8
+// CHECK1-NEXT: [[TMP119:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 7
+// CHECK1-NEXT: store ptr [[VLA1]], ptr [[TMP119]], align 8
+// CHECK1-NEXT: [[TMP120:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 7
+// CHECK1-NEXT: store ptr [[VLA1]], ptr [[TMP120]], align 8
+// CHECK1-NEXT: [[TMP121:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
+// CHECK1-NEXT: store i64 [[TMP96]], ptr [[TMP121]], align 8
+// CHECK1-NEXT: [[TMP122:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP122]], align 8
+// CHECK1-NEXT: [[TMP123:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 8
+// CHECK1-NEXT: store ptr [[D]], ptr [[TMP123]], align 8
+// CHECK1-NEXT: [[TMP124:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 8
+// CHECK1-NEXT: store ptr [[D]], ptr [[TMP124]], align 8
+// CHECK1-NEXT: [[TMP125:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 8
+// CHECK1-NEXT: store ptr null, ptr [[TMP125]], align 8
+// CHECK1-NEXT: [[TMP126:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 9
+// CHECK1-NEXT: store i64 [[TMP92]], ptr [[TMP126]], align 8
+// CHECK1-NEXT: [[TMP127:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 9
+// CHECK1-NEXT: store i64 [[TMP92]], ptr [[TMP127]], align 8
+// CHECK1-NEXT: [[TMP128:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 9
+// CHECK1-NEXT: store ptr null, ptr [[TMP128]], align 8
+// CHECK1-NEXT: [[TMP129:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP130:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP131:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS17:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP124:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP124]], align 4
-// CHECK1-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 1
-// CHECK1-NEXT: store i32 10, ptr [[TMP125]], align 4
-// CHECK1-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP121]], ptr [[TMP126]], align 8
-// CHECK1-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP122]], ptr [[TMP127]], align 8
-// CHECK1-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 4
-// CHECK1-NEXT: store ptr [[TMP123]], ptr [[TMP128]], align 8
-// CHECK1-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP129]], align 8
-// CHECK1-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP130]], align 8
-// CHECK1-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP131]], align 8
-// CHECK1-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP132]], align 8
-// CHECK1-NEXT: [[TMP133:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l160.region_id, ptr [[KERNEL_ARGS17]])
-// CHECK1-NEXT: [[TMP134:%.*]] = icmp ne i32 [[TMP133]], 0
-// CHECK1-NEXT: br i1 [[TMP134]], label [[OMP_OFFLOAD_FAILED18:%.*]], label [[OMP_OFFLOAD_CONT19:%.*]]
+// CHECK1-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP132]], align 4
+// CHECK1-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 1
+// CHECK1-NEXT: store i32 10, ptr [[TMP133]], align 4
+// CHECK1-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP129]], ptr [[TMP134]], align 8
+// CHECK1-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP130]], ptr [[TMP135]], align 8
+// CHECK1-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 4
+// CHECK1-NEXT: store ptr [[TMP131]], ptr [[TMP136]], align 8
+// CHECK1-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP137]], align 8
+// CHECK1-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP138]], align 8
+// CHECK1-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP139]], align 8
+// CHECK1-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP140]], align 8
+// CHECK1-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP141]], align 8
+// CHECK1-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP142]], align 4
+// CHECK1-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP143]], align 4
+// CHECK1-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP144]], align 4
+// CHECK1-NEXT: [[TMP145:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l160.region_id, ptr [[KERNEL_ARGS17]])
+// CHECK1-NEXT: [[TMP146:%.*]] = icmp ne i32 [[TMP145]], 0
+// CHECK1-NEXT: br i1 [[TMP146]], label [[OMP_OFFLOAD_FAILED18:%.*]], label [[OMP_OFFLOAD_CONT19:%.*]]
// CHECK1: omp_offload.failed18:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l160(i64 [[TMP82]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]], i64 [[TMP84]]) #[[ATTR4]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l160(i64 [[TMP90]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]], i64 [[TMP92]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT19]]
// CHECK1: omp_offload.cont19:
// CHECK1-NEXT: br label [[OMP_IF_END21:%.*]]
// CHECK1: omp_if.else20:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l160(i64 [[TMP82]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]], i64 [[TMP84]]) #[[ATTR4]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l160(i64 [[TMP90]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]], i64 [[TMP92]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_IF_END21]]
// CHECK1: omp_if.end21:
-// CHECK1-NEXT: [[TMP135:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: [[TMP136:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK1-NEXT: call void @llvm.stackrestore(ptr [[TMP136]])
-// CHECK1-NEXT: ret i32 [[TMP135]]
+// CHECK1-NEXT: [[TMP147:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: [[TMP148:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK1-NEXT: call void @llvm.stackrestore(ptr [[TMP148]])
+// CHECK1-NEXT: ret i32 [[TMP147]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l103
@@ -1023,7 +1047,7 @@ int bar(int n){
// CHECK1-NEXT: [[TMP15:%.*]] = load ptr, ptr [[DOTFIRSTPRIV_PTR_ADDR3_I]], align 8, !noalias !24
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT_ANON:%.*]], ptr [[TMP9]], i32 0, i32 1
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[TMP9]], i32 0, i32 2
-// CHECK1-NEXT: store i32 1, ptr [[KERNEL_ARGS_I]], align 4, !noalias !24
+// CHECK1-NEXT: store i32 2, ptr [[KERNEL_ARGS_I]], align 4, !noalias !24
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 1
// CHECK1-NEXT: store i32 3, ptr [[TMP18]], align 4, !noalias !24
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 2
@@ -1040,20 +1064,28 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP24]], align 8, !noalias !24
// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP25]], align 8, !noalias !24
-// CHECK1-NEXT: [[TMP26:%.*]] = call i32 @__tgt_target_kernel_nowait(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l128.region_id, ptr [[KERNEL_ARGS_I]], i32 0, ptr null, i32 0, ptr null)
-// CHECK1-NEXT: [[TMP27:%.*]] = icmp ne i32 [[TMP26]], 0
-// CHECK1-NEXT: br i1 [[TMP27]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__3_EXIT:%.*]]
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 9
+// CHECK1-NEXT: store i64 1, ptr [[TMP26]], align 8, !noalias !24
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP27]], align 4, !noalias !24
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4, !noalias !24
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP29]], align 4, !noalias !24
+// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l128.region_id, ptr [[KERNEL_ARGS_I]])
+// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
+// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__3_EXIT:%.*]]
// CHECK1: omp_offload.failed.i:
-// CHECK1-NEXT: [[TMP28:%.*]] = load i16, ptr [[TMP12]], align 2
-// CHECK1-NEXT: store i16 [[TMP28]], ptr [[AA_CASTED_I]], align 2, !noalias !24
-// CHECK1-NEXT: [[TMP29:%.*]] = load i64, ptr [[AA_CASTED_I]], align 8, !noalias !24
-// CHECK1-NEXT: [[TMP30:%.*]] = load i32, ptr [[TMP16]], align 4
-// CHECK1-NEXT: store i32 [[TMP30]], ptr [[LIN_CASTED_I]], align 4, !noalias !24
-// CHECK1-NEXT: [[TMP31:%.*]] = load i64, ptr [[LIN_CASTED_I]], align 8, !noalias !24
-// CHECK1-NEXT: [[TMP32:%.*]] = load i32, ptr [[TMP17]], align 4
-// CHECK1-NEXT: store i32 [[TMP32]], ptr [[A_CASTED_I]], align 4, !noalias !24
-// CHECK1-NEXT: [[TMP33:%.*]] = load i64, ptr [[A_CASTED_I]], align 8, !noalias !24
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l128(i64 [[TMP29]], i64 [[TMP31]], i64 [[TMP33]]) #[[ATTR4]]
+// CHECK1-NEXT: [[TMP32:%.*]] = load i16, ptr [[TMP12]], align 2
+// CHECK1-NEXT: store i16 [[TMP32]], ptr [[AA_CASTED_I]], align 2, !noalias !24
+// CHECK1-NEXT: [[TMP33:%.*]] = load i64, ptr [[AA_CASTED_I]], align 8, !noalias !24
+// CHECK1-NEXT: [[TMP34:%.*]] = load i32, ptr [[TMP16]], align 4
+// CHECK1-NEXT: store i32 [[TMP34]], ptr [[LIN_CASTED_I]], align 4, !noalias !24
+// CHECK1-NEXT: [[TMP35:%.*]] = load i64, ptr [[LIN_CASTED_I]], align 8, !noalias !24
+// CHECK1-NEXT: [[TMP36:%.*]] = load i32, ptr [[TMP17]], align 4
+// CHECK1-NEXT: store i32 [[TMP36]], ptr [[A_CASTED_I]], align 4, !noalias !24
+// CHECK1-NEXT: [[TMP37:%.*]] = load i64, ptr [[A_CASTED_I]], align 8, !noalias !24
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l128(i64 [[TMP33]], i64 [[TMP35]], i64 [[TMP37]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[DOTOMP_OUTLINED__3_EXIT]]
// CHECK1: .omp_outlined..3.exit:
// CHECK1-NEXT: ret i32 0
@@ -1443,7 +1475,7 @@ int bar(int n){
// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [5 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP29]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP29]], align 4
// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 5, ptr [[TMP30]], align 4
// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1460,9 +1492,17 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP36]], align 8
// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l232.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK1-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP38]], align 8
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP39]], align 4
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP40]], align 4
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP41]], align 4
+// CHECK1-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l232.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
+// CHECK1-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l232(ptr [[THIS1]], i64 [[TMP6]], i64 2, i64 [[TMP2]], ptr [[VLA]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1472,15 +1512,15 @@ int bar(int n){
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l232(ptr [[THIS1]], i64 [[TMP6]], i64 2, i64 [[TMP2]], ptr [[VLA]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
-// CHECK1-NEXT: [[TMP40:%.*]] = mul nsw i64 1, [[TMP2]]
-// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i64 [[TMP40]]
+// CHECK1-NEXT: [[TMP44:%.*]] = mul nsw i64 1, [[TMP2]]
+// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i64 [[TMP44]]
// CHECK1-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i16, ptr [[ARRAYIDX]], i64 1
-// CHECK1-NEXT: [[TMP41:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
-// CHECK1-NEXT: [[CONV:%.*]] = sext i16 [[TMP41]] to i32
-// CHECK1-NEXT: [[TMP42:%.*]] = load i32, ptr [[B]], align 4
-// CHECK1-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP42]]
-// CHECK1-NEXT: [[TMP43:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK1-NEXT: call void @llvm.stackrestore(ptr [[TMP43]])
+// CHECK1-NEXT: [[TMP45:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
+// CHECK1-NEXT: [[CONV:%.*]] = sext i16 [[TMP45]] to i32
+// CHECK1-NEXT: [[TMP46:%.*]] = load i32, ptr [[B]], align 4
+// CHECK1-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP46]]
+// CHECK1-NEXT: [[TMP47:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK1-NEXT: call void @llvm.stackrestore(ptr [[TMP47]])
// CHECK1-NEXT: ret i32 [[ADD3]]
//
//
@@ -1543,7 +1583,7 @@ int bar(int n){
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 4, ptr [[TMP22]], align 4
// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1560,9 +1600,17 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l214.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP31]], align 4
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK1-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l214.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK1-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l214(i64 [[TMP1]], i64 [[TMP3]], i64 [[TMP5]], ptr [[B]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1572,8 +1620,8 @@ int bar(int n){
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l214(i64 [[TMP1]], i64 [[TMP3]], i64 [[TMP5]], ptr [[B]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
-// CHECK1-NEXT: [[TMP32:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: ret i32 [[TMP32]]
+// CHECK1-NEXT: [[TMP36:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: ret i32 [[TMP36]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_Z9ftemplateIiET_i
@@ -1623,7 +1671,7 @@ int bar(int n){
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP16]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP16]], align 4
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 3, ptr [[TMP17]], align 4
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1640,9 +1688,17 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP23]], align 8
// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l197.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK1-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP25]], align 8
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP28]], align 4
+// CHECK1-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l197.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK1-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l197(i64 [[TMP1]], i64 [[TMP3]], ptr [[B]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1652,8 +1708,8 @@ int bar(int n){
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l197(i64 [[TMP1]], i64 [[TMP3]], ptr [[B]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
-// CHECK1-NEXT: [[TMP27:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: ret i32 [[TMP27]]
+// CHECK1-NEXT: [[TMP31:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: ret i32 [[TMP31]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l232
@@ -1981,7 +2037,7 @@ int bar(int n){
// CHECK3-NEXT: store i32 [[TMP3]], ptr [[__VLA_EXPR1]], align 4
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1998,249 +2054,273 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l103.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l103.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l103() #[[ATTR4:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: [[CALL:%.*]] = call noundef i64 @_Z7get_valv()
// CHECK3-NEXT: store i64 [[CALL]], ptr [[K]], align 8
-// CHECK3-NEXT: [[TMP16:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: store i32 [[TMP16]], ptr [[A_CASTED]], align 4
-// CHECK3-NEXT: [[TMP17:%.*]] = load i32, ptr [[A_CASTED]], align 4
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110(i32 [[TMP17]], ptr [[K]]) #[[ATTR4]]
+// CHECK3-NEXT: [[TMP20:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: store i32 [[TMP20]], ptr [[A_CASTED]], align 4
+// CHECK3-NEXT: [[TMP21:%.*]] = load i32, ptr [[A_CASTED]], align 4
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110(i32 [[TMP21]], ptr [[K]]) #[[ATTR4]]
// CHECK3-NEXT: store i32 12, ptr [[LIN]], align 4
-// CHECK3-NEXT: [[TMP18:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK3-NEXT: store i16 [[TMP18]], ptr [[AA_CASTED]], align 2
-// CHECK3-NEXT: [[TMP19:%.*]] = load i32, ptr [[AA_CASTED]], align 4
-// CHECK3-NEXT: [[TMP20:%.*]] = load i32, ptr [[LIN]], align 4
-// CHECK3-NEXT: store i32 [[TMP20]], ptr [[LIN_CASTED]], align 4
-// CHECK3-NEXT: [[TMP21:%.*]] = load i32, ptr [[LIN_CASTED]], align 4
-// CHECK3-NEXT: [[TMP22:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: store i32 [[TMP22]], ptr [[A_CASTED2]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = load i32, ptr [[A_CASTED2]], align 4
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP19]], ptr [[TMP24]], align 4
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP19]], ptr [[TMP25]], align 4
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP26]], align 4
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP21]], ptr [[TMP27]], align 4
-// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP21]], ptr [[TMP28]], align 4
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 1
-// CHECK3-NEXT: store ptr null, ptr [[TMP29]], align 4
-// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
-// CHECK3-NEXT: store i32 [[TMP23]], ptr [[TMP30]], align 4
-// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
-// CHECK3-NEXT: store i32 [[TMP23]], ptr [[TMP31]], align 4
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 2
-// CHECK3-NEXT: store ptr null, ptr [[TMP32]], align 4
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP36:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK3-NEXT: store i16 [[TMP36]], ptr [[TMP35]], align 4
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 1
-// CHECK3-NEXT: [[TMP38:%.*]] = load i32, ptr [[LIN]], align 4
-// CHECK3-NEXT: store i32 [[TMP38]], ptr [[TMP37]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 2
-// CHECK3-NEXT: [[TMP40:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: store i32 [[TMP40]], ptr [[TMP39]], align 4
-// CHECK3-NEXT: [[TMP41:%.*]] = call ptr @__kmpc_omp_target_task_alloc(ptr @[[GLOB2]], i32 [[TMP0]], i32 1, i32 72, i32 12, ptr @.omp_task_entry., i64 -1)
-// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], ptr [[TMP41]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T:%.*]], ptr [[TMP42]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP44:%.*]] = load ptr, ptr [[TMP43]], align 4
-// CHECK3-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[TMP44]], ptr align 4 [[AGG_CAPTURED]], i32 12, i1 false)
-// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES]], ptr [[TMP41]], i32 0, i32 1
-// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T:%.*]], ptr [[TMP45]], i32 0, i32 0
-// CHECK3-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[TMP46]], ptr align 4 @.offload_sizes, i32 24, i1 false)
-// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T]], ptr [[TMP45]], i32 0, i32 1
-// CHECK3-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[TMP47]], ptr align 4 [[TMP33]], i32 12, i1 false)
-// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T]], ptr [[TMP45]], i32 0, i32 2
-// CHECK3-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[TMP48]], ptr align 4 [[TMP34]], i32 12, i1 false)
-// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T]], ptr [[TMP45]], i32 0, i32 3
-// CHECK3-NEXT: [[TMP50:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK3-NEXT: store i16 [[TMP50]], ptr [[TMP49]], align 4
-// CHECK3-NEXT: [[TMP51:%.*]] = call i32 @__kmpc_omp_task(ptr @[[GLOB2]], i32 [[TMP0]], ptr [[TMP41]])
-// CHECK3-NEXT: [[TMP52:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: store i32 [[TMP52]], ptr [[A_CASTED3]], align 4
-// CHECK3-NEXT: [[TMP53:%.*]] = load i32, ptr [[A_CASTED3]], align 4
+// CHECK3-NEXT: [[TMP22:%.*]] = load i16, ptr [[AA]], align 2
+// CHECK3-NEXT: store i16 [[TMP22]], ptr [[AA_CASTED]], align 2
+// CHECK3-NEXT: [[TMP23:%.*]] = load i32, ptr [[AA_CASTED]], align 4
+// CHECK3-NEXT: [[TMP24:%.*]] = load i32, ptr [[LIN]], align 4
+// CHECK3-NEXT: store i32 [[TMP24]], ptr [[LIN_CASTED]], align 4
+// CHECK3-NEXT: [[TMP25:%.*]] = load i32, ptr [[LIN_CASTED]], align 4
+// CHECK3-NEXT: [[TMP26:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: store i32 [[TMP26]], ptr [[A_CASTED2]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = load i32, ptr [[A_CASTED2]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP23]], ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP23]], ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP25]], ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP25]], ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 1
+// CHECK3-NEXT: store ptr null, ptr [[TMP33]], align 4
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
+// CHECK3-NEXT: store i32 [[TMP27]], ptr [[TMP34]], align 4
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
+// CHECK3-NEXT: store i32 [[TMP27]], ptr [[TMP35]], align 4
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 2
+// CHECK3-NEXT: store ptr null, ptr [[TMP36]], align 4
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP40:%.*]] = load i16, ptr [[AA]], align 2
+// CHECK3-NEXT: store i16 [[TMP40]], ptr [[TMP39]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 1
+// CHECK3-NEXT: [[TMP42:%.*]] = load i32, ptr [[LIN]], align 4
+// CHECK3-NEXT: store i32 [[TMP42]], ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 2
+// CHECK3-NEXT: [[TMP44:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: store i32 [[TMP44]], ptr [[TMP43]], align 4
+// CHECK3-NEXT: [[TMP45:%.*]] = call ptr @__kmpc_omp_target_task_alloc(ptr @[[GLOB2]], i32 [[TMP0]], i32 1, i32 72, i32 12, ptr @.omp_task_entry., i64 -1)
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], ptr [[TMP45]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T:%.*]], ptr [[TMP46]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP48:%.*]] = load ptr, ptr [[TMP47]], align 4
+// CHECK3-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[TMP48]], ptr align 4 [[AGG_CAPTURED]], i32 12, i1 false)
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES]], ptr [[TMP45]], i32 0, i32 1
+// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T:%.*]], ptr [[TMP49]], i32 0, i32 0
+// CHECK3-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[TMP50]], ptr align 4 @.offload_sizes, i32 24, i1 false)
+// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T]], ptr [[TMP49]], i32 0, i32 1
+// CHECK3-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[TMP51]], ptr align 4 [[TMP37]], i32 12, i1 false)
+// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T]], ptr [[TMP49]], i32 0, i32 2
+// CHECK3-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[TMP52]], ptr align 4 [[TMP38]], i32 12, i1 false)
+// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T]], ptr [[TMP49]], i32 0, i32 3
// CHECK3-NEXT: [[TMP54:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK3-NEXT: store i16 [[TMP54]], ptr [[AA_CASTED4]], align 2
-// CHECK3-NEXT: [[TMP55:%.*]] = load i32, ptr [[AA_CASTED4]], align 4
-// CHECK3-NEXT: [[TMP56:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP56]], 10
+// CHECK3-NEXT: store i16 [[TMP54]], ptr [[TMP53]], align 4
+// CHECK3-NEXT: [[TMP55:%.*]] = call i32 @__kmpc_omp_task(ptr @[[GLOB2]], i32 [[TMP0]], ptr [[TMP45]])
+// CHECK3-NEXT: [[TMP56:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: store i32 [[TMP56]], ptr [[A_CASTED3]], align 4
+// CHECK3-NEXT: [[TMP57:%.*]] = load i32, ptr [[A_CASTED3]], align 4
+// CHECK3-NEXT: [[TMP58:%.*]] = load i16, ptr [[AA]], align 2
+// CHECK3-NEXT: store i16 [[TMP58]], ptr [[AA_CASTED4]], align 2
+// CHECK3-NEXT: [[TMP59:%.*]] = load i32, ptr [[AA_CASTED4]], align 4
+// CHECK3-NEXT: [[TMP60:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP60]], 10
// CHECK3-NEXT: br i1 [[CMP]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK3: omp_if.then:
-// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP53]], ptr [[TMP57]], align 4
-// CHECK3-NEXT: [[TMP58:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP53]], ptr [[TMP58]], align 4
-// CHECK3-NEXT: [[TMP59:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP59]], align 4
-// CHECK3-NEXT: [[TMP60:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP55]], ptr [[TMP60]], align 4
-// CHECK3-NEXT: [[TMP61:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP55]], ptr [[TMP61]], align 4
-// CHECK3-NEXT: [[TMP62:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 1
-// CHECK3-NEXT: store ptr null, ptr [[TMP62]], align 4
-// CHECK3-NEXT: [[TMP63:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP64:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP61:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP57]], ptr [[TMP61]], align 4
+// CHECK3-NEXT: [[TMP62:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP57]], ptr [[TMP62]], align 4
+// CHECK3-NEXT: [[TMP63:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP63]], align 4
+// CHECK3-NEXT: [[TMP64:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP59]], ptr [[TMP64]], align 4
+// CHECK3-NEXT: [[TMP65:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP59]], ptr [[TMP65]], align 4
+// CHECK3-NEXT: [[TMP66:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 1
+// CHECK3-NEXT: store ptr null, ptr [[TMP66]], align 4
+// CHECK3-NEXT: [[TMP67:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP68:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS8:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP65]], align 4
-// CHECK3-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 1
-// CHECK3-NEXT: store i32 2, ptr [[TMP66]], align 4
-// CHECK3-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP63]], ptr [[TMP67]], align 4
-// CHECK3-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP64]], ptr [[TMP68]], align 4
-// CHECK3-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.5, ptr [[TMP69]], align 4
-// CHECK3-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP70]], align 4
-// CHECK3-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP71]], align 4
-// CHECK3-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP72]], align 4
-// CHECK3-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP73]], align 8
-// CHECK3-NEXT: [[TMP74:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l136.region_id, ptr [[KERNEL_ARGS8]])
-// CHECK3-NEXT: [[TMP75:%.*]] = icmp ne i32 [[TMP74]], 0
-// CHECK3-NEXT: br i1 [[TMP75]], label [[OMP_OFFLOAD_FAILED9:%.*]], label [[OMP_OFFLOAD_CONT10:%.*]]
+// CHECK3-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP69]], align 4
+// CHECK3-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 1
+// CHECK3-NEXT: store i32 2, ptr [[TMP70]], align 4
+// CHECK3-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP67]], ptr [[TMP71]], align 4
+// CHECK3-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 4
+// CHECK3-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.5, ptr [[TMP73]], align 4
+// CHECK3-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP74]], align 4
+// CHECK3-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP75]], align 4
+// CHECK3-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP76]], align 4
+// CHECK3-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP77]], align 8
+// CHECK3-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP78]], align 8
+// CHECK3-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP79]], align 4
+// CHECK3-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP80]], align 4
+// CHECK3-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP81]], align 4
+// CHECK3-NEXT: [[TMP82:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l136.region_id, ptr [[KERNEL_ARGS8]])
+// CHECK3-NEXT: [[TMP83:%.*]] = icmp ne i32 [[TMP82]], 0
+// CHECK3-NEXT: br i1 [[TMP83]], label [[OMP_OFFLOAD_FAILED9:%.*]], label [[OMP_OFFLOAD_CONT10:%.*]]
// CHECK3: omp_offload.failed9:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l136(i32 [[TMP53]], i32 [[TMP55]]) #[[ATTR4]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l136(i32 [[TMP57]], i32 [[TMP59]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT10]]
// CHECK3: omp_offload.cont10:
// CHECK3-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK3: omp_if.else:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l136(i32 [[TMP53]], i32 [[TMP55]]) #[[ATTR4]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l136(i32 [[TMP57]], i32 [[TMP59]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
-// CHECK3-NEXT: [[TMP76:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: store i32 [[TMP76]], ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK3-NEXT: [[TMP77:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: store i32 [[TMP77]], ptr [[A_CASTED11]], align 4
-// CHECK3-NEXT: [[TMP78:%.*]] = load i32, ptr [[A_CASTED11]], align 4
-// CHECK3-NEXT: [[TMP79:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK3-NEXT: store i32 [[TMP79]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK3-NEXT: [[TMP80:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK3-NEXT: [[TMP81:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: [[CMP12:%.*]] = icmp sgt i32 [[TMP81]], 20
+// CHECK3-NEXT: [[TMP84:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: store i32 [[TMP84]], ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK3-NEXT: [[TMP85:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: store i32 [[TMP85]], ptr [[A_CASTED11]], align 4
+// CHECK3-NEXT: [[TMP86:%.*]] = load i32, ptr [[A_CASTED11]], align 4
+// CHECK3-NEXT: [[TMP87:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK3-NEXT: store i32 [[TMP87]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK3-NEXT: [[TMP88:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK3-NEXT: [[TMP89:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: [[CMP12:%.*]] = icmp sgt i32 [[TMP89]], 20
// CHECK3-NEXT: br i1 [[CMP12]], label [[OMP_IF_THEN13:%.*]], label [[OMP_IF_ELSE20:%.*]]
// CHECK3: omp_if.then13:
-// CHECK3-NEXT: [[TMP82:%.*]] = mul nuw i32 [[TMP1]], 4
-// CHECK3-NEXT: [[TMP83:%.*]] = sext i32 [[TMP82]] to i64
-// CHECK3-NEXT: [[TMP84:%.*]] = mul nuw i32 5, [[TMP3]]
-// CHECK3-NEXT: [[TMP85:%.*]] = mul nuw i32 [[TMP84]], 8
-// CHECK3-NEXT: [[TMP86:%.*]] = sext i32 [[TMP85]] to i64
+// CHECK3-NEXT: [[TMP90:%.*]] = mul nuw i32 [[TMP1]], 4
+// CHECK3-NEXT: [[TMP91:%.*]] = sext i32 [[TMP90]] to i64
+// CHECK3-NEXT: [[TMP92:%.*]] = mul nuw i32 5, [[TMP3]]
+// CHECK3-NEXT: [[TMP93:%.*]] = mul nuw i32 [[TMP92]], 8
+// CHECK3-NEXT: [[TMP94:%.*]] = sext i32 [[TMP93]] to i64
// CHECK3-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES]], ptr align 4 @.offload_sizes.8, i32 80, i1 false)
-// CHECK3-NEXT: [[TMP87:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP78]], ptr [[TMP87]], align 4
-// CHECK3-NEXT: [[TMP88:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP78]], ptr [[TMP88]], align 4
-// CHECK3-NEXT: [[TMP89:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP89]], align 4
-// CHECK3-NEXT: [[TMP90:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 1
-// CHECK3-NEXT: store ptr [[B]], ptr [[TMP90]], align 4
-// CHECK3-NEXT: [[TMP91:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 1
-// CHECK3-NEXT: store ptr [[B]], ptr [[TMP91]], align 4
-// CHECK3-NEXT: [[TMP92:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 1
-// CHECK3-NEXT: store ptr null, ptr [[TMP92]], align 4
-// CHECK3-NEXT: [[TMP93:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 2
-// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP93]], align 4
-// CHECK3-NEXT: [[TMP94:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 2
-// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP94]], align 4
-// CHECK3-NEXT: [[TMP95:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 2
-// CHECK3-NEXT: store ptr null, ptr [[TMP95]], align 4
-// CHECK3-NEXT: [[TMP96:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[VLA]], ptr [[TMP96]], align 4
-// CHECK3-NEXT: [[TMP97:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[VLA]], ptr [[TMP97]], align 4
-// CHECK3-NEXT: [[TMP98:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
-// CHECK3-NEXT: store i64 [[TMP83]], ptr [[TMP98]], align 4
-// CHECK3-NEXT: [[TMP99:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 3
-// CHECK3-NEXT: store ptr null, ptr [[TMP99]], align 4
-// CHECK3-NEXT: [[TMP100:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 4
-// CHECK3-NEXT: store ptr [[C]], ptr [[TMP100]], align 4
-// CHECK3-NEXT: [[TMP101:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 4
-// CHECK3-NEXT: store ptr [[C]], ptr [[TMP101]], align 4
-// CHECK3-NEXT: [[TMP102:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 4
-// CHECK3-NEXT: store ptr null, ptr [[TMP102]], align 4
-// CHECK3-NEXT: [[TMP103:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 5
-// CHECK3-NEXT: store i32 5, ptr [[TMP103]], align 4
-// CHECK3-NEXT: [[TMP104:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 5
-// CHECK3-NEXT: store i32 5, ptr [[TMP104]], align 4
-// CHECK3-NEXT: [[TMP105:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 5
-// CHECK3-NEXT: store ptr null, ptr [[TMP105]], align 4
-// CHECK3-NEXT: [[TMP106:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 6
-// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP106]], align 4
-// CHECK3-NEXT: [[TMP107:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 6
-// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP107]], align 4
-// CHECK3-NEXT: [[TMP108:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP108]], align 4
-// CHECK3-NEXT: [[TMP109:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 7
-// CHECK3-NEXT: store ptr [[VLA1]], ptr [[TMP109]], align 4
-// CHECK3-NEXT: [[TMP110:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 7
-// CHECK3-NEXT: store ptr [[VLA1]], ptr [[TMP110]], align 4
-// CHECK3-NEXT: [[TMP111:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
-// CHECK3-NEXT: store i64 [[TMP86]], ptr [[TMP111]], align 4
-// CHECK3-NEXT: [[TMP112:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP112]], align 4
-// CHECK3-NEXT: [[TMP113:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 8
-// CHECK3-NEXT: store ptr [[D]], ptr [[TMP113]], align 4
-// CHECK3-NEXT: [[TMP114:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 8
-// CHECK3-NEXT: store ptr [[D]], ptr [[TMP114]], align 4
-// CHECK3-NEXT: [[TMP115:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 8
-// CHECK3-NEXT: store ptr null, ptr [[TMP115]], align 4
-// CHECK3-NEXT: [[TMP116:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 9
-// CHECK3-NEXT: store i32 [[TMP80]], ptr [[TMP116]], align 4
-// CHECK3-NEXT: [[TMP117:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 9
-// CHECK3-NEXT: store i32 [[TMP80]], ptr [[TMP117]], align 4
-// CHECK3-NEXT: [[TMP118:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 9
-// CHECK3-NEXT: store ptr null, ptr [[TMP118]], align 4
-// CHECK3-NEXT: [[TMP119:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP120:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP121:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP95:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP86]], ptr [[TMP95]], align 4
+// CHECK3-NEXT: [[TMP96:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP86]], ptr [[TMP96]], align 4
+// CHECK3-NEXT: [[TMP97:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP97]], align 4
+// CHECK3-NEXT: [[TMP98:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 1
+// CHECK3-NEXT: store ptr [[B]], ptr [[TMP98]], align 4
+// CHECK3-NEXT: [[TMP99:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 1
+// CHECK3-NEXT: store ptr [[B]], ptr [[TMP99]], align 4
+// CHECK3-NEXT: [[TMP100:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 1
+// CHECK3-NEXT: store ptr null, ptr [[TMP100]], align 4
+// CHECK3-NEXT: [[TMP101:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 2
+// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP101]], align 4
+// CHECK3-NEXT: [[TMP102:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 2
+// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP102]], align 4
+// CHECK3-NEXT: [[TMP103:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 2
+// CHECK3-NEXT: store ptr null, ptr [[TMP103]], align 4
+// CHECK3-NEXT: [[TMP104:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[VLA]], ptr [[TMP104]], align 4
+// CHECK3-NEXT: [[TMP105:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[VLA]], ptr [[TMP105]], align 4
+// CHECK3-NEXT: [[TMP106:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
+// CHECK3-NEXT: store i64 [[TMP91]], ptr [[TMP106]], align 4
+// CHECK3-NEXT: [[TMP107:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 3
+// CHECK3-NEXT: store ptr null, ptr [[TMP107]], align 4
+// CHECK3-NEXT: [[TMP108:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 4
+// CHECK3-NEXT: store ptr [[C]], ptr [[TMP108]], align 4
+// CHECK3-NEXT: [[TMP109:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 4
+// CHECK3-NEXT: store ptr [[C]], ptr [[TMP109]], align 4
+// CHECK3-NEXT: [[TMP110:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 4
+// CHECK3-NEXT: store ptr null, ptr [[TMP110]], align 4
+// CHECK3-NEXT: [[TMP111:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 5
+// CHECK3-NEXT: store i32 5, ptr [[TMP111]], align 4
+// CHECK3-NEXT: [[TMP112:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 5
+// CHECK3-NEXT: store i32 5, ptr [[TMP112]], align 4
+// CHECK3-NEXT: [[TMP113:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 5
+// CHECK3-NEXT: store ptr null, ptr [[TMP113]], align 4
+// CHECK3-NEXT: [[TMP114:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 6
+// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP114]], align 4
+// CHECK3-NEXT: [[TMP115:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 6
+// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP115]], align 4
+// CHECK3-NEXT: [[TMP116:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP116]], align 4
+// CHECK3-NEXT: [[TMP117:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 7
+// CHECK3-NEXT: store ptr [[VLA1]], ptr [[TMP117]], align 4
+// CHECK3-NEXT: [[TMP118:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 7
+// CHECK3-NEXT: store ptr [[VLA1]], ptr [[TMP118]], align 4
+// CHECK3-NEXT: [[TMP119:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
+// CHECK3-NEXT: store i64 [[TMP94]], ptr [[TMP119]], align 4
+// CHECK3-NEXT: [[TMP120:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP120]], align 4
+// CHECK3-NEXT: [[TMP121:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 8
+// CHECK3-NEXT: store ptr [[D]], ptr [[TMP121]], align 4
+// CHECK3-NEXT: [[TMP122:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 8
+// CHECK3-NEXT: store ptr [[D]], ptr [[TMP122]], align 4
+// CHECK3-NEXT: [[TMP123:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 8
+// CHECK3-NEXT: store ptr null, ptr [[TMP123]], align 4
+// CHECK3-NEXT: [[TMP124:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 9
+// CHECK3-NEXT: store i32 [[TMP88]], ptr [[TMP124]], align 4
+// CHECK3-NEXT: [[TMP125:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 9
+// CHECK3-NEXT: store i32 [[TMP88]], ptr [[TMP125]], align 4
+// CHECK3-NEXT: [[TMP126:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 9
+// CHECK3-NEXT: store ptr null, ptr [[TMP126]], align 4
+// CHECK3-NEXT: [[TMP127:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP128:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP129:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS17:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP122:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP122]], align 4
-// CHECK3-NEXT: [[TMP123:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 1
-// CHECK3-NEXT: store i32 10, ptr [[TMP123]], align 4
-// CHECK3-NEXT: [[TMP124:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP119]], ptr [[TMP124]], align 4
-// CHECK3-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP120]], ptr [[TMP125]], align 4
-// CHECK3-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 4
-// CHECK3-NEXT: store ptr [[TMP121]], ptr [[TMP126]], align 4
-// CHECK3-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP127]], align 4
-// CHECK3-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP128]], align 4
-// CHECK3-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP129]], align 4
-// CHECK3-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP130]], align 8
-// CHECK3-NEXT: [[TMP131:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l160.region_id, ptr [[KERNEL_ARGS17]])
-// CHECK3-NEXT: [[TMP132:%.*]] = icmp ne i32 [[TMP131]], 0
-// CHECK3-NEXT: br i1 [[TMP132]], label [[OMP_OFFLOAD_FAILED18:%.*]], label [[OMP_OFFLOAD_CONT19:%.*]]
+// CHECK3-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP130]], align 4
+// CHECK3-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 1
+// CHECK3-NEXT: store i32 10, ptr [[TMP131]], align 4
+// CHECK3-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP127]], ptr [[TMP132]], align 4
+// CHECK3-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP128]], ptr [[TMP133]], align 4
+// CHECK3-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 4
+// CHECK3-NEXT: store ptr [[TMP129]], ptr [[TMP134]], align 4
+// CHECK3-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP135]], align 4
+// CHECK3-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP136]], align 4
+// CHECK3-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP137]], align 4
+// CHECK3-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP138]], align 8
+// CHECK3-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP139]], align 8
+// CHECK3-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP140]], align 4
+// CHECK3-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP141]], align 4
+// CHECK3-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP142]], align 4
+// CHECK3-NEXT: [[TMP143:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l160.region_id, ptr [[KERNEL_ARGS17]])
+// CHECK3-NEXT: [[TMP144:%.*]] = icmp ne i32 [[TMP143]], 0
+// CHECK3-NEXT: br i1 [[TMP144]], label [[OMP_OFFLOAD_FAILED18:%.*]], label [[OMP_OFFLOAD_CONT19:%.*]]
// CHECK3: omp_offload.failed18:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l160(i32 [[TMP78]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]], i32 [[TMP80]]) #[[ATTR4]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l160(i32 [[TMP86]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]], i32 [[TMP88]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT19]]
// CHECK3: omp_offload.cont19:
// CHECK3-NEXT: br label [[OMP_IF_END21:%.*]]
// CHECK3: omp_if.else20:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l160(i32 [[TMP78]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]], i32 [[TMP80]]) #[[ATTR4]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l160(i32 [[TMP86]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]], i32 [[TMP88]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_IF_END21]]
// CHECK3: omp_if.end21:
-// CHECK3-NEXT: [[TMP133:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: [[TMP134:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK3-NEXT: call void @llvm.stackrestore(ptr [[TMP134]])
-// CHECK3-NEXT: ret i32 [[TMP133]]
+// CHECK3-NEXT: [[TMP145:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: [[TMP146:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK3-NEXT: call void @llvm.stackrestore(ptr [[TMP146]])
+// CHECK3-NEXT: ret i32 [[TMP145]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l103
@@ -2637,7 +2717,7 @@ int bar(int n){
// CHECK3-NEXT: [[TMP15:%.*]] = load ptr, ptr [[DOTFIRSTPRIV_PTR_ADDR3_I]], align 4, !noalias !25
// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT_ANON:%.*]], ptr [[TMP9]], i32 0, i32 1
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[TMP9]], i32 0, i32 2
-// CHECK3-NEXT: store i32 1, ptr [[KERNEL_ARGS_I]], align 4, !noalias !25
+// CHECK3-NEXT: store i32 2, ptr [[KERNEL_ARGS_I]], align 4, !noalias !25
// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 1
// CHECK3-NEXT: store i32 3, ptr [[TMP18]], align 4, !noalias !25
// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 2
@@ -2654,20 +2734,28 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP24]], align 4, !noalias !25
// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP25]], align 8, !noalias !25
-// CHECK3-NEXT: [[TMP26:%.*]] = call i32 @__tgt_target_kernel_nowait(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l128.region_id, ptr [[KERNEL_ARGS_I]], i32 0, ptr null, i32 0, ptr null)
-// CHECK3-NEXT: [[TMP27:%.*]] = icmp ne i32 [[TMP26]], 0
-// CHECK3-NEXT: br i1 [[TMP27]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__3_EXIT:%.*]]
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 9
+// CHECK3-NEXT: store i64 1, ptr [[TMP26]], align 8, !noalias !25
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP27]], align 4, !noalias !25
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4, !noalias !25
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP29]], align 4, !noalias !25
+// CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l128.region_id, ptr [[KERNEL_ARGS_I]])
+// CHECK3-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
+// CHECK3-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__3_EXIT:%.*]]
// CHECK3: omp_offload.failed.i:
-// CHECK3-NEXT: [[TMP28:%.*]] = load i16, ptr [[TMP12]], align 2
-// CHECK3-NEXT: store i16 [[TMP28]], ptr [[AA_CASTED_I]], align 2, !noalias !25
-// CHECK3-NEXT: [[TMP29:%.*]] = load i32, ptr [[AA_CASTED_I]], align 4, !noalias !25
-// CHECK3-NEXT: [[TMP30:%.*]] = load i32, ptr [[TMP16]], align 4
-// CHECK3-NEXT: store i32 [[TMP30]], ptr [[LIN_CASTED_I]], align 4, !noalias !25
-// CHECK3-NEXT: [[TMP31:%.*]] = load i32, ptr [[LIN_CASTED_I]], align 4, !noalias !25
-// CHECK3-NEXT: [[TMP32:%.*]] = load i32, ptr [[TMP17]], align 4
-// CHECK3-NEXT: store i32 [[TMP32]], ptr [[A_CASTED_I]], align 4, !noalias !25
-// CHECK3-NEXT: [[TMP33:%.*]] = load i32, ptr [[A_CASTED_I]], align 4, !noalias !25
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l128(i32 [[TMP29]], i32 [[TMP31]], i32 [[TMP33]]) #[[ATTR4]]
+// CHECK3-NEXT: [[TMP32:%.*]] = load i16, ptr [[TMP12]], align 2
+// CHECK3-NEXT: store i16 [[TMP32]], ptr [[AA_CASTED_I]], align 2, !noalias !25
+// CHECK3-NEXT: [[TMP33:%.*]] = load i32, ptr [[AA_CASTED_I]], align 4, !noalias !25
+// CHECK3-NEXT: [[TMP34:%.*]] = load i32, ptr [[TMP16]], align 4
+// CHECK3-NEXT: store i32 [[TMP34]], ptr [[LIN_CASTED_I]], align 4, !noalias !25
+// CHECK3-NEXT: [[TMP35:%.*]] = load i32, ptr [[LIN_CASTED_I]], align 4, !noalias !25
+// CHECK3-NEXT: [[TMP36:%.*]] = load i32, ptr [[TMP17]], align 4
+// CHECK3-NEXT: store i32 [[TMP36]], ptr [[A_CASTED_I]], align 4, !noalias !25
+// CHECK3-NEXT: [[TMP37:%.*]] = load i32, ptr [[A_CASTED_I]], align 4, !noalias !25
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l128(i32 [[TMP33]], i32 [[TMP35]], i32 [[TMP37]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[DOTOMP_OUTLINED__3_EXIT]]
// CHECK3: .omp_outlined..3.exit:
// CHECK3-NEXT: ret i32 0
@@ -3057,7 +3145,7 @@ int bar(int n){
// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [5 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP29]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP29]], align 4
// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 5, ptr [[TMP30]], align 4
// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3074,9 +3162,17 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP36]], align 4
// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP37]], align 8
-// CHECK3-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l232.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK3-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP38]], align 8
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP39]], align 4
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l232.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
+// CHECK3-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l232(ptr [[THIS1]], i32 [[TMP5]], i32 2, i32 [[TMP1]], ptr [[VLA]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -3086,15 +3182,15 @@ int bar(int n){
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l232(ptr [[THIS1]], i32 [[TMP5]], i32 2, i32 [[TMP1]], ptr [[VLA]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
-// CHECK3-NEXT: [[TMP40:%.*]] = mul nsw i32 1, [[TMP1]]
-// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i32 [[TMP40]]
+// CHECK3-NEXT: [[TMP44:%.*]] = mul nsw i32 1, [[TMP1]]
+// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i32 [[TMP44]]
// CHECK3-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i16, ptr [[ARRAYIDX]], i32 1
-// CHECK3-NEXT: [[TMP41:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
-// CHECK3-NEXT: [[CONV:%.*]] = sext i16 [[TMP41]] to i32
-// CHECK3-NEXT: [[TMP42:%.*]] = load i32, ptr [[B]], align 4
-// CHECK3-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP42]]
-// CHECK3-NEXT: [[TMP43:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK3-NEXT: call void @llvm.stackrestore(ptr [[TMP43]])
+// CHECK3-NEXT: [[TMP45:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
+// CHECK3-NEXT: [[CONV:%.*]] = sext i16 [[TMP45]] to i32
+// CHECK3-NEXT: [[TMP46:%.*]] = load i32, ptr [[B]], align 4
+// CHECK3-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP46]]
+// CHECK3-NEXT: [[TMP47:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK3-NEXT: call void @llvm.stackrestore(ptr [[TMP47]])
// CHECK3-NEXT: ret i32 [[ADD3]]
//
//
@@ -3157,7 +3253,7 @@ int bar(int n){
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 4, ptr [[TMP22]], align 4
// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3174,9 +3270,17 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4
// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP29]], align 8
-// CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l214.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK3-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK3-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l214.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK3-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l214(i32 [[TMP1]], i32 [[TMP3]], i32 [[TMP5]], ptr [[B]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -3186,8 +3290,8 @@ int bar(int n){
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l214(i32 [[TMP1]], i32 [[TMP3]], i32 [[TMP5]], ptr [[B]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
-// CHECK3-NEXT: [[TMP32:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: ret i32 [[TMP32]]
+// CHECK3-NEXT: [[TMP36:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: ret i32 [[TMP36]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_Z9ftemplateIiET_i
@@ -3237,7 +3341,7 @@ int bar(int n){
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP16]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP16]], align 4
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 3, ptr [[TMP17]], align 4
// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3254,9 +3358,17 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP23]], align 4
// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP24]], align 8
-// CHECK3-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l197.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK3-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP25]], align 8
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l197.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK3-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l197(i32 [[TMP1]], i32 [[TMP3]], ptr [[B]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -3266,8 +3378,8 @@ int bar(int n){
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l197(i32 [[TMP1]], i32 [[TMP3]], ptr [[B]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
-// CHECK3-NEXT: [[TMP27:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: ret i32 [[TMP27]]
+// CHECK3-NEXT: [[TMP31:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: ret i32 [[TMP31]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l232
@@ -5128,7 +5240,7 @@ int bar(int n){
// CHECK17-NEXT: store i64 [[TMP5]], ptr [[__VLA_EXPR1]], align 8
// CHECK17-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK17-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK17-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK17-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK17-NEXT: store i32 0, ptr [[TMP8]], align 4
// CHECK17-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -5145,250 +5257,274 @@ int bar(int n){
// CHECK17-NEXT: store ptr null, ptr [[TMP14]], align 8
// CHECK17-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK17-NEXT: store i64 0, ptr [[TMP15]], align 8
-// CHECK17-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l103.region_id, ptr [[KERNEL_ARGS]])
-// CHECK17-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK17-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK17-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK17-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP17]], align 4
+// CHECK17-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK17-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK17-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l103.region_id, ptr [[KERNEL_ARGS]])
+// CHECK17-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK17-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK17: omp_offload.failed:
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l103() #[[ATTR4:[0-9]+]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK17: omp_offload.cont:
// CHECK17-NEXT: [[CALL:%.*]] = call noundef i64 @_Z7get_valv()
// CHECK17-NEXT: store i64 [[CALL]], ptr [[K]], align 8
-// CHECK17-NEXT: [[TMP18:%.*]] = load i32, ptr [[A]], align 4
-// CHECK17-NEXT: store i32 [[TMP18]], ptr [[A_CASTED]], align 4
-// CHECK17-NEXT: [[TMP19:%.*]] = load i64, ptr [[A_CASTED]], align 8
-// CHECK17-NEXT: [[TMP20:%.*]] = load i64, ptr [[K]], align 8
-// CHECK17-NEXT: store i64 [[TMP20]], ptr [[K_CASTED]], align 8
-// CHECK17-NEXT: [[TMP21:%.*]] = load i64, ptr [[K_CASTED]], align 8
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110(i64 [[TMP19]], i64 [[TMP21]]) #[[ATTR4]]
+// CHECK17-NEXT: [[TMP22:%.*]] = load i32, ptr [[A]], align 4
+// CHECK17-NEXT: store i32 [[TMP22]], ptr [[A_CASTED]], align 4
+// CHECK17-NEXT: [[TMP23:%.*]] = load i64, ptr [[A_CASTED]], align 8
+// CHECK17-NEXT: [[TMP24:%.*]] = load i64, ptr [[K]], align 8
+// CHECK17-NEXT: store i64 [[TMP24]], ptr [[K_CASTED]], align 8
+// CHECK17-NEXT: [[TMP25:%.*]] = load i64, ptr [[K_CASTED]], align 8
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110(i64 [[TMP23]], i64 [[TMP25]]) #[[ATTR4]]
// CHECK17-NEXT: store i32 12, ptr [[LIN]], align 4
-// CHECK17-NEXT: [[TMP22:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK17-NEXT: store i16 [[TMP22]], ptr [[AA_CASTED]], align 2
-// CHECK17-NEXT: [[TMP23:%.*]] = load i64, ptr [[AA_CASTED]], align 8
-// CHECK17-NEXT: [[TMP24:%.*]] = load i32, ptr [[LIN]], align 4
-// CHECK17-NEXT: store i32 [[TMP24]], ptr [[LIN_CASTED]], align 4
-// CHECK17-NEXT: [[TMP25:%.*]] = load i64, ptr [[LIN_CASTED]], align 8
-// CHECK17-NEXT: [[TMP26:%.*]] = load i32, ptr [[A]], align 4
-// CHECK17-NEXT: store i32 [[TMP26]], ptr [[A_CASTED2]], align 4
-// CHECK17-NEXT: [[TMP27:%.*]] = load i64, ptr [[A_CASTED2]], align 8
-// CHECK17-NEXT: [[TMP28:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP23]], ptr [[TMP28]], align 8
-// CHECK17-NEXT: [[TMP29:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP23]], ptr [[TMP29]], align 8
-// CHECK17-NEXT: [[TMP30:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP30]], align 8
-// CHECK17-NEXT: [[TMP31:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP25]], ptr [[TMP31]], align 8
-// CHECK17-NEXT: [[TMP32:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP25]], ptr [[TMP32]], align 8
-// CHECK17-NEXT: [[TMP33:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 1
-// CHECK17-NEXT: store ptr null, ptr [[TMP33]], align 8
-// CHECK17-NEXT: [[TMP34:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
-// CHECK17-NEXT: store i64 [[TMP27]], ptr [[TMP34]], align 8
-// CHECK17-NEXT: [[TMP35:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
-// CHECK17-NEXT: store i64 [[TMP27]], ptr [[TMP35]], align 8
-// CHECK17-NEXT: [[TMP36:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 2
-// CHECK17-NEXT: store ptr null, ptr [[TMP36]], align 8
-// CHECK17-NEXT: [[TMP37:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP40:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK17-NEXT: store i16 [[TMP40]], ptr [[TMP39]], align 4
-// CHECK17-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 1
-// CHECK17-NEXT: [[TMP42:%.*]] = load i32, ptr [[LIN]], align 4
-// CHECK17-NEXT: store i32 [[TMP42]], ptr [[TMP41]], align 4
-// CHECK17-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 2
-// CHECK17-NEXT: [[TMP44:%.*]] = load i32, ptr [[A]], align 4
-// CHECK17-NEXT: store i32 [[TMP44]], ptr [[TMP43]], align 4
-// CHECK17-NEXT: [[TMP45:%.*]] = call ptr @__kmpc_omp_target_task_alloc(ptr @[[GLOB2]], i32 [[TMP0]], i32 1, i64 120, i64 12, ptr @.omp_task_entry., i64 -1)
-// CHECK17-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], ptr [[TMP45]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T:%.*]], ptr [[TMP46]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP48:%.*]] = load ptr, ptr [[TMP47]], align 8
-// CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 4 [[TMP48]], ptr align 4 [[AGG_CAPTURED]], i64 12, i1 false)
-// CHECK17-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES]], ptr [[TMP45]], i32 0, i32 1
-// CHECK17-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T:%.*]], ptr [[TMP49]], i32 0, i32 0
-// CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[TMP50]], ptr align 8 [[TMP37]], i64 24, i1 false)
-// CHECK17-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T]], ptr [[TMP49]], i32 0, i32 1
-// CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[TMP51]], ptr align 8 [[TMP38]], i64 24, i1 false)
-// CHECK17-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T]], ptr [[TMP49]], i32 0, i32 2
-// CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[TMP52]], ptr align 8 @.offload_sizes, i64 24, i1 false)
-// CHECK17-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T]], ptr [[TMP49]], i32 0, i32 3
-// CHECK17-NEXT: [[TMP54:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK17-NEXT: store i16 [[TMP54]], ptr [[TMP53]], align 8
-// CHECK17-NEXT: [[TMP55:%.*]] = call i32 @__kmpc_omp_task(ptr @[[GLOB2]], i32 [[TMP0]], ptr [[TMP45]])
-// CHECK17-NEXT: [[TMP56:%.*]] = load i32, ptr [[A]], align 4
-// CHECK17-NEXT: store i32 [[TMP56]], ptr [[A_CASTED3]], align 4
-// CHECK17-NEXT: [[TMP57:%.*]] = load i64, ptr [[A_CASTED3]], align 8
+// CHECK17-NEXT: [[TMP26:%.*]] = load i16, ptr [[AA]], align 2
+// CHECK17-NEXT: store i16 [[TMP26]], ptr [[AA_CASTED]], align 2
+// CHECK17-NEXT: [[TMP27:%.*]] = load i64, ptr [[AA_CASTED]], align 8
+// CHECK17-NEXT: [[TMP28:%.*]] = load i32, ptr [[LIN]], align 4
+// CHECK17-NEXT: store i32 [[TMP28]], ptr [[LIN_CASTED]], align 4
+// CHECK17-NEXT: [[TMP29:%.*]] = load i64, ptr [[LIN_CASTED]], align 8
+// CHECK17-NEXT: [[TMP30:%.*]] = load i32, ptr [[A]], align 4
+// CHECK17-NEXT: store i32 [[TMP30]], ptr [[A_CASTED2]], align 4
+// CHECK17-NEXT: [[TMP31:%.*]] = load i64, ptr [[A_CASTED2]], align 8
+// CHECK17-NEXT: [[TMP32:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP27]], ptr [[TMP32]], align 8
+// CHECK17-NEXT: [[TMP33:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP27]], ptr [[TMP33]], align 8
+// CHECK17-NEXT: [[TMP34:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK17-NEXT: [[TMP35:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP29]], ptr [[TMP35]], align 8
+// CHECK17-NEXT: [[TMP36:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP29]], ptr [[TMP36]], align 8
+// CHECK17-NEXT: [[TMP37:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 1
+// CHECK17-NEXT: store ptr null, ptr [[TMP37]], align 8
+// CHECK17-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
+// CHECK17-NEXT: store i64 [[TMP31]], ptr [[TMP38]], align 8
+// CHECK17-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
+// CHECK17-NEXT: store i64 [[TMP31]], ptr [[TMP39]], align 8
+// CHECK17-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 2
+// CHECK17-NEXT: store ptr null, ptr [[TMP40]], align 8
+// CHECK17-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP44:%.*]] = load i16, ptr [[AA]], align 2
+// CHECK17-NEXT: store i16 [[TMP44]], ptr [[TMP43]], align 4
+// CHECK17-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 1
+// CHECK17-NEXT: [[TMP46:%.*]] = load i32, ptr [[LIN]], align 4
+// CHECK17-NEXT: store i32 [[TMP46]], ptr [[TMP45]], align 4
+// CHECK17-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 2
+// CHECK17-NEXT: [[TMP48:%.*]] = load i32, ptr [[A]], align 4
+// CHECK17-NEXT: store i32 [[TMP48]], ptr [[TMP47]], align 4
+// CHECK17-NEXT: [[TMP49:%.*]] = call ptr @__kmpc_omp_target_task_alloc(ptr @[[GLOB2]], i32 [[TMP0]], i32 1, i64 120, i64 12, ptr @.omp_task_entry., i64 -1)
+// CHECK17-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], ptr [[TMP49]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T:%.*]], ptr [[TMP50]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP52:%.*]] = load ptr, ptr [[TMP51]], align 8
+// CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 4 [[TMP52]], ptr align 4 [[AGG_CAPTURED]], i64 12, i1 false)
+// CHECK17-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES]], ptr [[TMP49]], i32 0, i32 1
+// CHECK17-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T:%.*]], ptr [[TMP53]], i32 0, i32 0
+// CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[TMP54]], ptr align 8 [[TMP41]], i64 24, i1 false)
+// CHECK17-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T]], ptr [[TMP53]], i32 0, i32 1
+// CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[TMP55]], ptr align 8 [[TMP42]], i64 24, i1 false)
+// CHECK17-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T]], ptr [[TMP53]], i32 0, i32 2
+// CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[TMP56]], ptr align 8 @.offload_sizes, i64 24, i1 false)
+// CHECK17-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T]], ptr [[TMP53]], i32 0, i32 3
// CHECK17-NEXT: [[TMP58:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK17-NEXT: store i16 [[TMP58]], ptr [[AA_CASTED4]], align 2
-// CHECK17-NEXT: [[TMP59:%.*]] = load i64, ptr [[AA_CASTED4]], align 8
-// CHECK17-NEXT: [[TMP60:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK17-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP60]], 10
+// CHECK17-NEXT: store i16 [[TMP58]], ptr [[TMP57]], align 8
+// CHECK17-NEXT: [[TMP59:%.*]] = call i32 @__kmpc_omp_task(ptr @[[GLOB2]], i32 [[TMP0]], ptr [[TMP49]])
+// CHECK17-NEXT: [[TMP60:%.*]] = load i32, ptr [[A]], align 4
+// CHECK17-NEXT: store i32 [[TMP60]], ptr [[A_CASTED3]], align 4
+// CHECK17-NEXT: [[TMP61:%.*]] = load i64, ptr [[A_CASTED3]], align 8
+// CHECK17-NEXT: [[TMP62:%.*]] = load i16, ptr [[AA]], align 2
+// CHECK17-NEXT: store i16 [[TMP62]], ptr [[AA_CASTED4]], align 2
+// CHECK17-NEXT: [[TMP63:%.*]] = load i64, ptr [[AA_CASTED4]], align 8
+// CHECK17-NEXT: [[TMP64:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK17-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP64]], 10
// CHECK17-NEXT: br i1 [[CMP]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK17: omp_if.then:
-// CHECK17-NEXT: [[TMP61:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP57]], ptr [[TMP61]], align 8
-// CHECK17-NEXT: [[TMP62:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP57]], ptr [[TMP62]], align 8
-// CHECK17-NEXT: [[TMP63:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP63]], align 8
-// CHECK17-NEXT: [[TMP64:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP59]], ptr [[TMP64]], align 8
-// CHECK17-NEXT: [[TMP65:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP59]], ptr [[TMP65]], align 8
-// CHECK17-NEXT: [[TMP66:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 1
-// CHECK17-NEXT: store ptr null, ptr [[TMP66]], align 8
-// CHECK17-NEXT: [[TMP67:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP68:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP65:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP61]], ptr [[TMP65]], align 8
+// CHECK17-NEXT: [[TMP66:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP61]], ptr [[TMP66]], align 8
+// CHECK17-NEXT: [[TMP67:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP67]], align 8
+// CHECK17-NEXT: [[TMP68:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP63]], ptr [[TMP68]], align 8
+// CHECK17-NEXT: [[TMP69:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP63]], ptr [[TMP69]], align 8
+// CHECK17-NEXT: [[TMP70:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 1
+// CHECK17-NEXT: store ptr null, ptr [[TMP70]], align 8
+// CHECK17-NEXT: [[TMP71:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP72:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS8:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP69]], align 4
-// CHECK17-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 1
-// CHECK17-NEXT: store i32 2, ptr [[TMP70]], align 4
-// CHECK17-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP67]], ptr [[TMP71]], align 8
-// CHECK17-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 8
-// CHECK17-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 4
-// CHECK17-NEXT: store ptr @.offload_sizes.5, ptr [[TMP73]], align 8
-// CHECK17-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP74]], align 8
-// CHECK17-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP75]], align 8
-// CHECK17-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP76]], align 8
-// CHECK17-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 8
-// CHECK17-NEXT: store i64 0, ptr [[TMP77]], align 8
-// CHECK17-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l136.region_id, ptr [[KERNEL_ARGS8]])
-// CHECK17-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
-// CHECK17-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED9:%.*]], label [[OMP_OFFLOAD_CONT10:%.*]]
+// CHECK17-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP73]], align 4
+// CHECK17-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 1
+// CHECK17-NEXT: store i32 2, ptr [[TMP74]], align 4
+// CHECK17-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP71]], ptr [[TMP75]], align 8
+// CHECK17-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP72]], ptr [[TMP76]], align 8
+// CHECK17-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 4
+// CHECK17-NEXT: store ptr @.offload_sizes.5, ptr [[TMP77]], align 8
+// CHECK17-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP78]], align 8
+// CHECK17-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP79]], align 8
+// CHECK17-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP80]], align 8
+// CHECK17-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 8
+// CHECK17-NEXT: store i64 0, ptr [[TMP81]], align 8
+// CHECK17-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP82]], align 8
+// CHECK17-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP83]], align 4
+// CHECK17-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP84]], align 4
+// CHECK17-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP85]], align 4
+// CHECK17-NEXT: [[TMP86:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l136.region_id, ptr [[KERNEL_ARGS8]])
+// CHECK17-NEXT: [[TMP87:%.*]] = icmp ne i32 [[TMP86]], 0
+// CHECK17-NEXT: br i1 [[TMP87]], label [[OMP_OFFLOAD_FAILED9:%.*]], label [[OMP_OFFLOAD_CONT10:%.*]]
// CHECK17: omp_offload.failed9:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l136(i64 [[TMP57]], i64 [[TMP59]]) #[[ATTR4]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l136(i64 [[TMP61]], i64 [[TMP63]]) #[[ATTR4]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT10]]
// CHECK17: omp_offload.cont10:
// CHECK17-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK17: omp_if.else:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l136(i64 [[TMP57]], i64 [[TMP59]]) #[[ATTR4]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l136(i64 [[TMP61]], i64 [[TMP63]]) #[[ATTR4]]
// CHECK17-NEXT: br label [[OMP_IF_END]]
// CHECK17: omp_if.end:
-// CHECK17-NEXT: [[TMP80:%.*]] = load i32, ptr [[A]], align 4
-// CHECK17-NEXT: store i32 [[TMP80]], ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK17-NEXT: [[TMP81:%.*]] = load i32, ptr [[A]], align 4
-// CHECK17-NEXT: store i32 [[TMP81]], ptr [[A_CASTED11]], align 4
-// CHECK17-NEXT: [[TMP82:%.*]] = load i64, ptr [[A_CASTED11]], align 8
-// CHECK17-NEXT: [[TMP83:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK17-NEXT: store i32 [[TMP83]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK17-NEXT: [[TMP84:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK17-NEXT: [[TMP85:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK17-NEXT: [[CMP12:%.*]] = icmp sgt i32 [[TMP85]], 20
+// CHECK17-NEXT: [[TMP88:%.*]] = load i32, ptr [[A]], align 4
+// CHECK17-NEXT: store i32 [[TMP88]], ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK17-NEXT: [[TMP89:%.*]] = load i32, ptr [[A]], align 4
+// CHECK17-NEXT: store i32 [[TMP89]], ptr [[A_CASTED11]], align 4
+// CHECK17-NEXT: [[TMP90:%.*]] = load i64, ptr [[A_CASTED11]], align 8
+// CHECK17-NEXT: [[TMP91:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK17-NEXT: store i32 [[TMP91]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK17-NEXT: [[TMP92:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK17-NEXT: [[TMP93:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK17-NEXT: [[CMP12:%.*]] = icmp sgt i32 [[TMP93]], 20
// CHECK17-NEXT: br i1 [[CMP12]], label [[OMP_IF_THEN13:%.*]], label [[OMP_IF_ELSE20:%.*]]
// CHECK17: omp_if.then13:
-// CHECK17-NEXT: [[TMP86:%.*]] = mul nuw i64 [[TMP2]], 4
-// CHECK17-NEXT: [[TMP87:%.*]] = mul nuw i64 5, [[TMP5]]
-// CHECK17-NEXT: [[TMP88:%.*]] = mul nuw i64 [[TMP87]], 8
+// CHECK17-NEXT: [[TMP94:%.*]] = mul nuw i64 [[TMP2]], 4
+// CHECK17-NEXT: [[TMP95:%.*]] = mul nuw i64 5, [[TMP5]]
+// CHECK17-NEXT: [[TMP96:%.*]] = mul nuw i64 [[TMP95]], 8
// CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES]], ptr align 8 @.offload_sizes.8, i64 80, i1 false)
-// CHECK17-NEXT: [[TMP89:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP82]], ptr [[TMP89]], align 8
-// CHECK17-NEXT: [[TMP90:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP82]], ptr [[TMP90]], align 8
-// CHECK17-NEXT: [[TMP91:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP91]], align 8
-// CHECK17-NEXT: [[TMP92:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 1
-// CHECK17-NEXT: store ptr [[B]], ptr [[TMP92]], align 8
-// CHECK17-NEXT: [[TMP93:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 1
-// CHECK17-NEXT: store ptr [[B]], ptr [[TMP93]], align 8
-// CHECK17-NEXT: [[TMP94:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 1
-// CHECK17-NEXT: store ptr null, ptr [[TMP94]], align 8
-// CHECK17-NEXT: [[TMP95:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 2
-// CHECK17-NEXT: store i64 [[TMP2]], ptr [[TMP95]], align 8
-// CHECK17-NEXT: [[TMP96:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 2
-// CHECK17-NEXT: store i64 [[TMP2]], ptr [[TMP96]], align 8
-// CHECK17-NEXT: [[TMP97:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 2
-// CHECK17-NEXT: store ptr null, ptr [[TMP97]], align 8
-// CHECK17-NEXT: [[TMP98:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP98]], align 8
-// CHECK17-NEXT: [[TMP99:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP99]], align 8
-// CHECK17-NEXT: [[TMP100:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
-// CHECK17-NEXT: store i64 [[TMP86]], ptr [[TMP100]], align 8
-// CHECK17-NEXT: [[TMP101:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 3
-// CHECK17-NEXT: store ptr null, ptr [[TMP101]], align 8
-// CHECK17-NEXT: [[TMP102:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 4
-// CHECK17-NEXT: store ptr [[C]], ptr [[TMP102]], align 8
-// CHECK17-NEXT: [[TMP103:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 4
-// CHECK17-NEXT: store ptr [[C]], ptr [[TMP103]], align 8
-// CHECK17-NEXT: [[TMP104:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 4
-// CHECK17-NEXT: store ptr null, ptr [[TMP104]], align 8
-// CHECK17-NEXT: [[TMP105:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 5
-// CHECK17-NEXT: store i64 5, ptr [[TMP105]], align 8
-// CHECK17-NEXT: [[TMP106:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 5
-// CHECK17-NEXT: store i64 5, ptr [[TMP106]], align 8
-// CHECK17-NEXT: [[TMP107:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 5
-// CHECK17-NEXT: store ptr null, ptr [[TMP107]], align 8
-// CHECK17-NEXT: [[TMP108:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 6
-// CHECK17-NEXT: store i64 [[TMP5]], ptr [[TMP108]], align 8
-// CHECK17-NEXT: [[TMP109:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 6
-// CHECK17-NEXT: store i64 [[TMP5]], ptr [[TMP109]], align 8
-// CHECK17-NEXT: [[TMP110:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP110]], align 8
-// CHECK17-NEXT: [[TMP111:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 7
-// CHECK17-NEXT: store ptr [[VLA1]], ptr [[TMP111]], align 8
-// CHECK17-NEXT: [[TMP112:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 7
-// CHECK17-NEXT: store ptr [[VLA1]], ptr [[TMP112]], align 8
-// CHECK17-NEXT: [[TMP113:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
-// CHECK17-NEXT: store i64 [[TMP88]], ptr [[TMP113]], align 8
-// CHECK17-NEXT: [[TMP114:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP114]], align 8
-// CHECK17-NEXT: [[TMP115:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 8
-// CHECK17-NEXT: store ptr [[D]], ptr [[TMP115]], align 8
-// CHECK17-NEXT: [[TMP116:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 8
-// CHECK17-NEXT: store ptr [[D]], ptr [[TMP116]], align 8
-// CHECK17-NEXT: [[TMP117:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 8
-// CHECK17-NEXT: store ptr null, ptr [[TMP117]], align 8
-// CHECK17-NEXT: [[TMP118:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 9
-// CHECK17-NEXT: store i64 [[TMP84]], ptr [[TMP118]], align 8
-// CHECK17-NEXT: [[TMP119:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 9
-// CHECK17-NEXT: store i64 [[TMP84]], ptr [[TMP119]], align 8
-// CHECK17-NEXT: [[TMP120:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 9
-// CHECK17-NEXT: store ptr null, ptr [[TMP120]], align 8
-// CHECK17-NEXT: [[TMP121:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP122:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP123:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP97:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP90]], ptr [[TMP97]], align 8
+// CHECK17-NEXT: [[TMP98:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP90]], ptr [[TMP98]], align 8
+// CHECK17-NEXT: [[TMP99:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP99]], align 8
+// CHECK17-NEXT: [[TMP100:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 1
+// CHECK17-NEXT: store ptr [[B]], ptr [[TMP100]], align 8
+// CHECK17-NEXT: [[TMP101:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 1
+// CHECK17-NEXT: store ptr [[B]], ptr [[TMP101]], align 8
+// CHECK17-NEXT: [[TMP102:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 1
+// CHECK17-NEXT: store ptr null, ptr [[TMP102]], align 8
+// CHECK17-NEXT: [[TMP103:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 2
+// CHECK17-NEXT: store i64 [[TMP2]], ptr [[TMP103]], align 8
+// CHECK17-NEXT: [[TMP104:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 2
+// CHECK17-NEXT: store i64 [[TMP2]], ptr [[TMP104]], align 8
+// CHECK17-NEXT: [[TMP105:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 2
+// CHECK17-NEXT: store ptr null, ptr [[TMP105]], align 8
+// CHECK17-NEXT: [[TMP106:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP106]], align 8
+// CHECK17-NEXT: [[TMP107:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP107]], align 8
+// CHECK17-NEXT: [[TMP108:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
+// CHECK17-NEXT: store i64 [[TMP94]], ptr [[TMP108]], align 8
+// CHECK17-NEXT: [[TMP109:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 3
+// CHECK17-NEXT: store ptr null, ptr [[TMP109]], align 8
+// CHECK17-NEXT: [[TMP110:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 4
+// CHECK17-NEXT: store ptr [[C]], ptr [[TMP110]], align 8
+// CHECK17-NEXT: [[TMP111:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 4
+// CHECK17-NEXT: store ptr [[C]], ptr [[TMP111]], align 8
+// CHECK17-NEXT: [[TMP112:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 4
+// CHECK17-NEXT: store ptr null, ptr [[TMP112]], align 8
+// CHECK17-NEXT: [[TMP113:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 5
+// CHECK17-NEXT: store i64 5, ptr [[TMP113]], align 8
+// CHECK17-NEXT: [[TMP114:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 5
+// CHECK17-NEXT: store i64 5, ptr [[TMP114]], align 8
+// CHECK17-NEXT: [[TMP115:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 5
+// CHECK17-NEXT: store ptr null, ptr [[TMP115]], align 8
+// CHECK17-NEXT: [[TMP116:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 6
+// CHECK17-NEXT: store i64 [[TMP5]], ptr [[TMP116]], align 8
+// CHECK17-NEXT: [[TMP117:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 6
+// CHECK17-NEXT: store i64 [[TMP5]], ptr [[TMP117]], align 8
+// CHECK17-NEXT: [[TMP118:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP118]], align 8
+// CHECK17-NEXT: [[TMP119:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 7
+// CHECK17-NEXT: store ptr [[VLA1]], ptr [[TMP119]], align 8
+// CHECK17-NEXT: [[TMP120:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 7
+// CHECK17-NEXT: store ptr [[VLA1]], ptr [[TMP120]], align 8
+// CHECK17-NEXT: [[TMP121:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
+// CHECK17-NEXT: store i64 [[TMP96]], ptr [[TMP121]], align 8
+// CHECK17-NEXT: [[TMP122:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP122]], align 8
+// CHECK17-NEXT: [[TMP123:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 8
+// CHECK17-NEXT: store ptr [[D]], ptr [[TMP123]], align 8
+// CHECK17-NEXT: [[TMP124:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 8
+// CHECK17-NEXT: store ptr [[D]], ptr [[TMP124]], align 8
+// CHECK17-NEXT: [[TMP125:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 8
+// CHECK17-NEXT: store ptr null, ptr [[TMP125]], align 8
+// CHECK17-NEXT: [[TMP126:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 9
+// CHECK17-NEXT: store i64 [[TMP92]], ptr [[TMP126]], align 8
+// CHECK17-NEXT: [[TMP127:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 9
+// CHECK17-NEXT: store i64 [[TMP92]], ptr [[TMP127]], align 8
+// CHECK17-NEXT: [[TMP128:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 9
+// CHECK17-NEXT: store ptr null, ptr [[TMP128]], align 8
+// CHECK17-NEXT: [[TMP129:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP130:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP131:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS17:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP124:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP124]], align 4
-// CHECK17-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 1
-// CHECK17-NEXT: store i32 10, ptr [[TMP125]], align 4
-// CHECK17-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP121]], ptr [[TMP126]], align 8
-// CHECK17-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP122]], ptr [[TMP127]], align 8
-// CHECK17-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 4
-// CHECK17-NEXT: store ptr [[TMP123]], ptr [[TMP128]], align 8
-// CHECK17-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP129]], align 8
-// CHECK17-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP130]], align 8
-// CHECK17-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP131]], align 8
-// CHECK17-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 8
-// CHECK17-NEXT: store i64 0, ptr [[TMP132]], align 8
-// CHECK17-NEXT: [[TMP133:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l160.region_id, ptr [[KERNEL_ARGS17]])
-// CHECK17-NEXT: [[TMP134:%.*]] = icmp ne i32 [[TMP133]], 0
-// CHECK17-NEXT: br i1 [[TMP134]], label [[OMP_OFFLOAD_FAILED18:%.*]], label [[OMP_OFFLOAD_CONT19:%.*]]
+// CHECK17-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP132]], align 4
+// CHECK17-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 1
+// CHECK17-NEXT: store i32 10, ptr [[TMP133]], align 4
+// CHECK17-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP129]], ptr [[TMP134]], align 8
+// CHECK17-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP130]], ptr [[TMP135]], align 8
+// CHECK17-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 4
+// CHECK17-NEXT: store ptr [[TMP131]], ptr [[TMP136]], align 8
+// CHECK17-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP137]], align 8
+// CHECK17-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP138]], align 8
+// CHECK17-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP139]], align 8
+// CHECK17-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 8
+// CHECK17-NEXT: store i64 0, ptr [[TMP140]], align 8
+// CHECK17-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP141]], align 8
+// CHECK17-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP142]], align 4
+// CHECK17-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP143]], align 4
+// CHECK17-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP144]], align 4
+// CHECK17-NEXT: [[TMP145:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l160.region_id, ptr [[KERNEL_ARGS17]])
+// CHECK17-NEXT: [[TMP146:%.*]] = icmp ne i32 [[TMP145]], 0
+// CHECK17-NEXT: br i1 [[TMP146]], label [[OMP_OFFLOAD_FAILED18:%.*]], label [[OMP_OFFLOAD_CONT19:%.*]]
// CHECK17: omp_offload.failed18:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l160(i64 [[TMP82]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]], i64 [[TMP84]]) #[[ATTR4]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l160(i64 [[TMP90]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]], i64 [[TMP92]]) #[[ATTR4]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT19]]
// CHECK17: omp_offload.cont19:
// CHECK17-NEXT: br label [[OMP_IF_END21:%.*]]
// CHECK17: omp_if.else20:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l160(i64 [[TMP82]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]], i64 [[TMP84]]) #[[ATTR4]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l160(i64 [[TMP90]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]], i64 [[TMP92]]) #[[ATTR4]]
// CHECK17-NEXT: br label [[OMP_IF_END21]]
// CHECK17: omp_if.end21:
-// CHECK17-NEXT: [[TMP135:%.*]] = load i32, ptr [[A]], align 4
-// CHECK17-NEXT: [[TMP136:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK17-NEXT: call void @llvm.stackrestore(ptr [[TMP136]])
-// CHECK17-NEXT: ret i32 [[TMP135]]
+// CHECK17-NEXT: [[TMP147:%.*]] = load i32, ptr [[A]], align 4
+// CHECK17-NEXT: [[TMP148:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK17-NEXT: call void @llvm.stackrestore(ptr [[TMP148]])
+// CHECK17-NEXT: ret i32 [[TMP147]]
//
//
// CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l103
@@ -5787,7 +5923,7 @@ int bar(int n){
// CHECK17-NEXT: [[TMP15:%.*]] = load ptr, ptr [[DOTFIRSTPRIV_PTR_ADDR3_I]], align 8, !noalias !24
// CHECK17-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT_ANON:%.*]], ptr [[TMP9]], i32 0, i32 1
// CHECK17-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[TMP9]], i32 0, i32 2
-// CHECK17-NEXT: store i32 1, ptr [[KERNEL_ARGS_I]], align 4, !noalias !24
+// CHECK17-NEXT: store i32 2, ptr [[KERNEL_ARGS_I]], align 4, !noalias !24
// CHECK17-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 1
// CHECK17-NEXT: store i32 3, ptr [[TMP18]], align 4, !noalias !24
// CHECK17-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 2
@@ -5804,20 +5940,28 @@ int bar(int n){
// CHECK17-NEXT: store ptr null, ptr [[TMP24]], align 8, !noalias !24
// CHECK17-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 8
// CHECK17-NEXT: store i64 0, ptr [[TMP25]], align 8, !noalias !24
-// CHECK17-NEXT: [[TMP26:%.*]] = call i32 @__tgt_target_kernel_nowait(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l128.region_id, ptr [[KERNEL_ARGS_I]], i32 0, ptr null, i32 0, ptr null)
-// CHECK17-NEXT: [[TMP27:%.*]] = icmp ne i32 [[TMP26]], 0
-// CHECK17-NEXT: br i1 [[TMP27]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__3_EXIT:%.*]]
+// CHECK17-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 9
+// CHECK17-NEXT: store i64 1, ptr [[TMP26]], align 8, !noalias !24
+// CHECK17-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP27]], align 4, !noalias !24
+// CHECK17-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4, !noalias !24
+// CHECK17-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP29]], align 4, !noalias !24
+// CHECK17-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l128.region_id, ptr [[KERNEL_ARGS_I]])
+// CHECK17-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
+// CHECK17-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__3_EXIT:%.*]]
// CHECK17: omp_offload.failed.i:
-// CHECK17-NEXT: [[TMP28:%.*]] = load i16, ptr [[TMP12]], align 2
-// CHECK17-NEXT: store i16 [[TMP28]], ptr [[AA_CASTED_I]], align 2, !noalias !24
-// CHECK17-NEXT: [[TMP29:%.*]] = load i64, ptr [[AA_CASTED_I]], align 8, !noalias !24
-// CHECK17-NEXT: [[TMP30:%.*]] = load i32, ptr [[TMP16]], align 4
-// CHECK17-NEXT: store i32 [[TMP30]], ptr [[LIN_CASTED_I]], align 4, !noalias !24
-// CHECK17-NEXT: [[TMP31:%.*]] = load i64, ptr [[LIN_CASTED_I]], align 8, !noalias !24
-// CHECK17-NEXT: [[TMP32:%.*]] = load i32, ptr [[TMP17]], align 4
-// CHECK17-NEXT: store i32 [[TMP32]], ptr [[A_CASTED_I]], align 4, !noalias !24
-// CHECK17-NEXT: [[TMP33:%.*]] = load i64, ptr [[A_CASTED_I]], align 8, !noalias !24
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l128(i64 [[TMP29]], i64 [[TMP31]], i64 [[TMP33]]) #[[ATTR4]]
+// CHECK17-NEXT: [[TMP32:%.*]] = load i16, ptr [[TMP12]], align 2
+// CHECK17-NEXT: store i16 [[TMP32]], ptr [[AA_CASTED_I]], align 2, !noalias !24
+// CHECK17-NEXT: [[TMP33:%.*]] = load i64, ptr [[AA_CASTED_I]], align 8, !noalias !24
+// CHECK17-NEXT: [[TMP34:%.*]] = load i32, ptr [[TMP16]], align 4
+// CHECK17-NEXT: store i32 [[TMP34]], ptr [[LIN_CASTED_I]], align 4, !noalias !24
+// CHECK17-NEXT: [[TMP35:%.*]] = load i64, ptr [[LIN_CASTED_I]], align 8, !noalias !24
+// CHECK17-NEXT: [[TMP36:%.*]] = load i32, ptr [[TMP17]], align 4
+// CHECK17-NEXT: store i32 [[TMP36]], ptr [[A_CASTED_I]], align 4, !noalias !24
+// CHECK17-NEXT: [[TMP37:%.*]] = load i64, ptr [[A_CASTED_I]], align 8, !noalias !24
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l128(i64 [[TMP33]], i64 [[TMP35]], i64 [[TMP37]]) #[[ATTR4]]
// CHECK17-NEXT: br label [[DOTOMP_OUTLINED__3_EXIT]]
// CHECK17: .omp_outlined..3.exit:
// CHECK17-NEXT: ret i32 0
@@ -6207,7 +6351,7 @@ int bar(int n){
// CHECK17-NEXT: [[TMP28:%.*]] = getelementptr inbounds [5 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK17-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP29]], align 4
+// CHECK17-NEXT: store i32 2, ptr [[TMP29]], align 4
// CHECK17-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK17-NEXT: store i32 5, ptr [[TMP30]], align 4
// CHECK17-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -6224,9 +6368,17 @@ int bar(int n){
// CHECK17-NEXT: store ptr null, ptr [[TMP36]], align 8
// CHECK17-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK17-NEXT: store i64 0, ptr [[TMP37]], align 8
-// CHECK17-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l232.region_id, ptr [[KERNEL_ARGS]])
-// CHECK17-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK17-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK17-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP38]], align 8
+// CHECK17-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP39]], align 4
+// CHECK17-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP40]], align 4
+// CHECK17-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP41]], align 4
+// CHECK17-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l232.region_id, ptr [[KERNEL_ARGS]])
+// CHECK17-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
+// CHECK17-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK17: omp_offload.failed:
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l232(ptr [[THIS1]], i64 [[TMP6]], i64 2, i64 [[TMP2]], ptr [[VLA]]) #[[ATTR4]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -6236,15 +6388,15 @@ int bar(int n){
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l232(ptr [[THIS1]], i64 [[TMP6]], i64 2, i64 [[TMP2]], ptr [[VLA]]) #[[ATTR4]]
// CHECK17-NEXT: br label [[OMP_IF_END]]
// CHECK17: omp_if.end:
-// CHECK17-NEXT: [[TMP40:%.*]] = mul nsw i64 1, [[TMP2]]
-// CHECK17-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i64 [[TMP40]]
+// CHECK17-NEXT: [[TMP44:%.*]] = mul nsw i64 1, [[TMP2]]
+// CHECK17-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i64 [[TMP44]]
// CHECK17-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i16, ptr [[ARRAYIDX]], i64 1
-// CHECK17-NEXT: [[TMP41:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
-// CHECK17-NEXT: [[CONV:%.*]] = sext i16 [[TMP41]] to i32
-// CHECK17-NEXT: [[TMP42:%.*]] = load i32, ptr [[B]], align 4
-// CHECK17-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP42]]
-// CHECK17-NEXT: [[TMP43:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK17-NEXT: call void @llvm.stackrestore(ptr [[TMP43]])
+// CHECK17-NEXT: [[TMP45:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
+// CHECK17-NEXT: [[CONV:%.*]] = sext i16 [[TMP45]] to i32
+// CHECK17-NEXT: [[TMP46:%.*]] = load i32, ptr [[B]], align 4
+// CHECK17-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP46]]
+// CHECK17-NEXT: [[TMP47:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK17-NEXT: call void @llvm.stackrestore(ptr [[TMP47]])
// CHECK17-NEXT: ret i32 [[ADD3]]
//
//
@@ -6307,7 +6459,7 @@ int bar(int n){
// CHECK17-NEXT: [[TMP20:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK17-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK17-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK17-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK17-NEXT: store i32 4, ptr [[TMP22]], align 4
// CHECK17-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -6324,9 +6476,17 @@ int bar(int n){
// CHECK17-NEXT: store ptr null, ptr [[TMP28]], align 8
// CHECK17-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK17-NEXT: store i64 0, ptr [[TMP29]], align 8
-// CHECK17-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l214.region_id, ptr [[KERNEL_ARGS]])
-// CHECK17-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK17-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK17-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK17-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP31]], align 4
+// CHECK17-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK17-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK17-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l214.region_id, ptr [[KERNEL_ARGS]])
+// CHECK17-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK17-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK17: omp_offload.failed:
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l214(i64 [[TMP1]], i64 [[TMP3]], i64 [[TMP5]], ptr [[B]]) #[[ATTR4]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -6336,8 +6496,8 @@ int bar(int n){
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l214(i64 [[TMP1]], i64 [[TMP3]], i64 [[TMP5]], ptr [[B]]) #[[ATTR4]]
// CHECK17-NEXT: br label [[OMP_IF_END]]
// CHECK17: omp_if.end:
-// CHECK17-NEXT: [[TMP32:%.*]] = load i32, ptr [[A]], align 4
-// CHECK17-NEXT: ret i32 [[TMP32]]
+// CHECK17-NEXT: [[TMP36:%.*]] = load i32, ptr [[A]], align 4
+// CHECK17-NEXT: ret i32 [[TMP36]]
//
//
// CHECK17-LABEL: define {{[^@]+}}@_Z9ftemplateIiET_i
@@ -6387,7 +6547,7 @@ int bar(int n){
// CHECK17-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK17-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP16]], align 4
+// CHECK17-NEXT: store i32 2, ptr [[TMP16]], align 4
// CHECK17-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK17-NEXT: store i32 3, ptr [[TMP17]], align 4
// CHECK17-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -6404,9 +6564,17 @@ int bar(int n){
// CHECK17-NEXT: store ptr null, ptr [[TMP23]], align 8
// CHECK17-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK17-NEXT: store i64 0, ptr [[TMP24]], align 8
-// CHECK17-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l197.region_id, ptr [[KERNEL_ARGS]])
-// CHECK17-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK17-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK17-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP25]], align 8
+// CHECK17-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK17-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP27]], align 4
+// CHECK17-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP28]], align 4
+// CHECK17-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l197.region_id, ptr [[KERNEL_ARGS]])
+// CHECK17-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK17-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK17: omp_offload.failed:
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l197(i64 [[TMP1]], i64 [[TMP3]], ptr [[B]]) #[[ATTR4]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -6416,8 +6584,8 @@ int bar(int n){
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l197(i64 [[TMP1]], i64 [[TMP3]], ptr [[B]]) #[[ATTR4]]
// CHECK17-NEXT: br label [[OMP_IF_END]]
// CHECK17: omp_if.end:
-// CHECK17-NEXT: [[TMP27:%.*]] = load i32, ptr [[A]], align 4
-// CHECK17-NEXT: ret i32 [[TMP27]]
+// CHECK17-NEXT: [[TMP31:%.*]] = load i32, ptr [[A]], align 4
+// CHECK17-NEXT: ret i32 [[TMP31]]
//
//
// CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l232
@@ -6745,7 +6913,7 @@ int bar(int n){
// CHECK19-NEXT: store i32 [[TMP3]], ptr [[__VLA_EXPR1]], align 4
// CHECK19-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK19-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK19-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK19-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK19-NEXT: store i32 0, ptr [[TMP6]], align 4
// CHECK19-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -6762,249 +6930,273 @@ int bar(int n){
// CHECK19-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK19-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK19-NEXT: store i64 0, ptr [[TMP13]], align 8
-// CHECK19-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l103.region_id, ptr [[KERNEL_ARGS]])
-// CHECK19-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK19-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK19-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK19-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP15]], align 4
+// CHECK19-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK19-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK19-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l103.region_id, ptr [[KERNEL_ARGS]])
+// CHECK19-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK19-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK19: omp_offload.failed:
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l103() #[[ATTR4:[0-9]+]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK19: omp_offload.cont:
// CHECK19-NEXT: [[CALL:%.*]] = call noundef i64 @_Z7get_valv()
// CHECK19-NEXT: store i64 [[CALL]], ptr [[K]], align 8
-// CHECK19-NEXT: [[TMP16:%.*]] = load i32, ptr [[A]], align 4
-// CHECK19-NEXT: store i32 [[TMP16]], ptr [[A_CASTED]], align 4
-// CHECK19-NEXT: [[TMP17:%.*]] = load i32, ptr [[A_CASTED]], align 4
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110(i32 [[TMP17]], ptr [[K]]) #[[ATTR4]]
+// CHECK19-NEXT: [[TMP20:%.*]] = load i32, ptr [[A]], align 4
+// CHECK19-NEXT: store i32 [[TMP20]], ptr [[A_CASTED]], align 4
+// CHECK19-NEXT: [[TMP21:%.*]] = load i32, ptr [[A_CASTED]], align 4
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110(i32 [[TMP21]], ptr [[K]]) #[[ATTR4]]
// CHECK19-NEXT: store i32 12, ptr [[LIN]], align 4
-// CHECK19-NEXT: [[TMP18:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK19-NEXT: store i16 [[TMP18]], ptr [[AA_CASTED]], align 2
-// CHECK19-NEXT: [[TMP19:%.*]] = load i32, ptr [[AA_CASTED]], align 4
-// CHECK19-NEXT: [[TMP20:%.*]] = load i32, ptr [[LIN]], align 4
-// CHECK19-NEXT: store i32 [[TMP20]], ptr [[LIN_CASTED]], align 4
-// CHECK19-NEXT: [[TMP21:%.*]] = load i32, ptr [[LIN_CASTED]], align 4
-// CHECK19-NEXT: [[TMP22:%.*]] = load i32, ptr [[A]], align 4
-// CHECK19-NEXT: store i32 [[TMP22]], ptr [[A_CASTED2]], align 4
-// CHECK19-NEXT: [[TMP23:%.*]] = load i32, ptr [[A_CASTED2]], align 4
-// CHECK19-NEXT: [[TMP24:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP19]], ptr [[TMP24]], align 4
-// CHECK19-NEXT: [[TMP25:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP19]], ptr [[TMP25]], align 4
-// CHECK19-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP26]], align 4
-// CHECK19-NEXT: [[TMP27:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP21]], ptr [[TMP27]], align 4
-// CHECK19-NEXT: [[TMP28:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP21]], ptr [[TMP28]], align 4
-// CHECK19-NEXT: [[TMP29:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 1
-// CHECK19-NEXT: store ptr null, ptr [[TMP29]], align 4
-// CHECK19-NEXT: [[TMP30:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
-// CHECK19-NEXT: store i32 [[TMP23]], ptr [[TMP30]], align 4
-// CHECK19-NEXT: [[TMP31:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
-// CHECK19-NEXT: store i32 [[TMP23]], ptr [[TMP31]], align 4
-// CHECK19-NEXT: [[TMP32:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 2
-// CHECK19-NEXT: store ptr null, ptr [[TMP32]], align 4
-// CHECK19-NEXT: [[TMP33:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP34:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP36:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK19-NEXT: store i16 [[TMP36]], ptr [[TMP35]], align 4
-// CHECK19-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 1
-// CHECK19-NEXT: [[TMP38:%.*]] = load i32, ptr [[LIN]], align 4
-// CHECK19-NEXT: store i32 [[TMP38]], ptr [[TMP37]], align 4
-// CHECK19-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 2
-// CHECK19-NEXT: [[TMP40:%.*]] = load i32, ptr [[A]], align 4
-// CHECK19-NEXT: store i32 [[TMP40]], ptr [[TMP39]], align 4
-// CHECK19-NEXT: [[TMP41:%.*]] = call ptr @__kmpc_omp_target_task_alloc(ptr @[[GLOB2]], i32 [[TMP0]], i32 1, i32 72, i32 12, ptr @.omp_task_entry., i64 -1)
-// CHECK19-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], ptr [[TMP41]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T:%.*]], ptr [[TMP42]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP44:%.*]] = load ptr, ptr [[TMP43]], align 4
-// CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[TMP44]], ptr align 4 [[AGG_CAPTURED]], i32 12, i1 false)
-// CHECK19-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES]], ptr [[TMP41]], i32 0, i32 1
-// CHECK19-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T:%.*]], ptr [[TMP45]], i32 0, i32 0
-// CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[TMP46]], ptr align 4 @.offload_sizes, i32 24, i1 false)
-// CHECK19-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T]], ptr [[TMP45]], i32 0, i32 1
-// CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[TMP47]], ptr align 4 [[TMP33]], i32 12, i1 false)
-// CHECK19-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T]], ptr [[TMP45]], i32 0, i32 2
-// CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[TMP48]], ptr align 4 [[TMP34]], i32 12, i1 false)
-// CHECK19-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T]], ptr [[TMP45]], i32 0, i32 3
-// CHECK19-NEXT: [[TMP50:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK19-NEXT: store i16 [[TMP50]], ptr [[TMP49]], align 4
-// CHECK19-NEXT: [[TMP51:%.*]] = call i32 @__kmpc_omp_task(ptr @[[GLOB2]], i32 [[TMP0]], ptr [[TMP41]])
-// CHECK19-NEXT: [[TMP52:%.*]] = load i32, ptr [[A]], align 4
-// CHECK19-NEXT: store i32 [[TMP52]], ptr [[A_CASTED3]], align 4
-// CHECK19-NEXT: [[TMP53:%.*]] = load i32, ptr [[A_CASTED3]], align 4
+// CHECK19-NEXT: [[TMP22:%.*]] = load i16, ptr [[AA]], align 2
+// CHECK19-NEXT: store i16 [[TMP22]], ptr [[AA_CASTED]], align 2
+// CHECK19-NEXT: [[TMP23:%.*]] = load i32, ptr [[AA_CASTED]], align 4
+// CHECK19-NEXT: [[TMP24:%.*]] = load i32, ptr [[LIN]], align 4
+// CHECK19-NEXT: store i32 [[TMP24]], ptr [[LIN_CASTED]], align 4
+// CHECK19-NEXT: [[TMP25:%.*]] = load i32, ptr [[LIN_CASTED]], align 4
+// CHECK19-NEXT: [[TMP26:%.*]] = load i32, ptr [[A]], align 4
+// CHECK19-NEXT: store i32 [[TMP26]], ptr [[A_CASTED2]], align 4
+// CHECK19-NEXT: [[TMP27:%.*]] = load i32, ptr [[A_CASTED2]], align 4
+// CHECK19-NEXT: [[TMP28:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP23]], ptr [[TMP28]], align 4
+// CHECK19-NEXT: [[TMP29:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP23]], ptr [[TMP29]], align 4
+// CHECK19-NEXT: [[TMP30:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP30]], align 4
+// CHECK19-NEXT: [[TMP31:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP25]], ptr [[TMP31]], align 4
+// CHECK19-NEXT: [[TMP32:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP25]], ptr [[TMP32]], align 4
+// CHECK19-NEXT: [[TMP33:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 1
+// CHECK19-NEXT: store ptr null, ptr [[TMP33]], align 4
+// CHECK19-NEXT: [[TMP34:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
+// CHECK19-NEXT: store i32 [[TMP27]], ptr [[TMP34]], align 4
+// CHECK19-NEXT: [[TMP35:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
+// CHECK19-NEXT: store i32 [[TMP27]], ptr [[TMP35]], align 4
+// CHECK19-NEXT: [[TMP36:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 2
+// CHECK19-NEXT: store ptr null, ptr [[TMP36]], align 4
+// CHECK19-NEXT: [[TMP37:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP40:%.*]] = load i16, ptr [[AA]], align 2
+// CHECK19-NEXT: store i16 [[TMP40]], ptr [[TMP39]], align 4
+// CHECK19-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 1
+// CHECK19-NEXT: [[TMP42:%.*]] = load i32, ptr [[LIN]], align 4
+// CHECK19-NEXT: store i32 [[TMP42]], ptr [[TMP41]], align 4
+// CHECK19-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 2
+// CHECK19-NEXT: [[TMP44:%.*]] = load i32, ptr [[A]], align 4
+// CHECK19-NEXT: store i32 [[TMP44]], ptr [[TMP43]], align 4
+// CHECK19-NEXT: [[TMP45:%.*]] = call ptr @__kmpc_omp_target_task_alloc(ptr @[[GLOB2]], i32 [[TMP0]], i32 1, i32 72, i32 12, ptr @.omp_task_entry., i64 -1)
+// CHECK19-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], ptr [[TMP45]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T:%.*]], ptr [[TMP46]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP48:%.*]] = load ptr, ptr [[TMP47]], align 4
+// CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[TMP48]], ptr align 4 [[AGG_CAPTURED]], i32 12, i1 false)
+// CHECK19-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES]], ptr [[TMP45]], i32 0, i32 1
+// CHECK19-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T:%.*]], ptr [[TMP49]], i32 0, i32 0
+// CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[TMP50]], ptr align 4 @.offload_sizes, i32 24, i1 false)
+// CHECK19-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T]], ptr [[TMP49]], i32 0, i32 1
+// CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[TMP51]], ptr align 4 [[TMP37]], i32 12, i1 false)
+// CHECK19-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T]], ptr [[TMP49]], i32 0, i32 2
+// CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[TMP52]], ptr align 4 [[TMP38]], i32 12, i1 false)
+// CHECK19-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T]], ptr [[TMP49]], i32 0, i32 3
// CHECK19-NEXT: [[TMP54:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK19-NEXT: store i16 [[TMP54]], ptr [[AA_CASTED4]], align 2
-// CHECK19-NEXT: [[TMP55:%.*]] = load i32, ptr [[AA_CASTED4]], align 4
-// CHECK19-NEXT: [[TMP56:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK19-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP56]], 10
+// CHECK19-NEXT: store i16 [[TMP54]], ptr [[TMP53]], align 4
+// CHECK19-NEXT: [[TMP55:%.*]] = call i32 @__kmpc_omp_task(ptr @[[GLOB2]], i32 [[TMP0]], ptr [[TMP45]])
+// CHECK19-NEXT: [[TMP56:%.*]] = load i32, ptr [[A]], align 4
+// CHECK19-NEXT: store i32 [[TMP56]], ptr [[A_CASTED3]], align 4
+// CHECK19-NEXT: [[TMP57:%.*]] = load i32, ptr [[A_CASTED3]], align 4
+// CHECK19-NEXT: [[TMP58:%.*]] = load i16, ptr [[AA]], align 2
+// CHECK19-NEXT: store i16 [[TMP58]], ptr [[AA_CASTED4]], align 2
+// CHECK19-NEXT: [[TMP59:%.*]] = load i32, ptr [[AA_CASTED4]], align 4
+// CHECK19-NEXT: [[TMP60:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK19-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP60]], 10
// CHECK19-NEXT: br i1 [[CMP]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK19: omp_if.then:
-// CHECK19-NEXT: [[TMP57:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP53]], ptr [[TMP57]], align 4
-// CHECK19-NEXT: [[TMP58:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP53]], ptr [[TMP58]], align 4
-// CHECK19-NEXT: [[TMP59:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP59]], align 4
-// CHECK19-NEXT: [[TMP60:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP55]], ptr [[TMP60]], align 4
-// CHECK19-NEXT: [[TMP61:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP55]], ptr [[TMP61]], align 4
-// CHECK19-NEXT: [[TMP62:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 1
-// CHECK19-NEXT: store ptr null, ptr [[TMP62]], align 4
-// CHECK19-NEXT: [[TMP63:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP64:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP61:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP57]], ptr [[TMP61]], align 4
+// CHECK19-NEXT: [[TMP62:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP57]], ptr [[TMP62]], align 4
+// CHECK19-NEXT: [[TMP63:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP63]], align 4
+// CHECK19-NEXT: [[TMP64:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP59]], ptr [[TMP64]], align 4
+// CHECK19-NEXT: [[TMP65:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP59]], ptr [[TMP65]], align 4
+// CHECK19-NEXT: [[TMP66:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 1
+// CHECK19-NEXT: store ptr null, ptr [[TMP66]], align 4
+// CHECK19-NEXT: [[TMP67:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP68:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS8:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP65]], align 4
-// CHECK19-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 1
-// CHECK19-NEXT: store i32 2, ptr [[TMP66]], align 4
-// CHECK19-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP63]], ptr [[TMP67]], align 4
-// CHECK19-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP64]], ptr [[TMP68]], align 4
-// CHECK19-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 4
-// CHECK19-NEXT: store ptr @.offload_sizes.5, ptr [[TMP69]], align 4
-// CHECK19-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP70]], align 4
-// CHECK19-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP71]], align 4
-// CHECK19-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP72]], align 4
-// CHECK19-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 8
-// CHECK19-NEXT: store i64 0, ptr [[TMP73]], align 8
-// CHECK19-NEXT: [[TMP74:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l136.region_id, ptr [[KERNEL_ARGS8]])
-// CHECK19-NEXT: [[TMP75:%.*]] = icmp ne i32 [[TMP74]], 0
-// CHECK19-NEXT: br i1 [[TMP75]], label [[OMP_OFFLOAD_FAILED9:%.*]], label [[OMP_OFFLOAD_CONT10:%.*]]
+// CHECK19-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP69]], align 4
+// CHECK19-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 1
+// CHECK19-NEXT: store i32 2, ptr [[TMP70]], align 4
+// CHECK19-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP67]], ptr [[TMP71]], align 4
+// CHECK19-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 4
+// CHECK19-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 4
+// CHECK19-NEXT: store ptr @.offload_sizes.5, ptr [[TMP73]], align 4
+// CHECK19-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP74]], align 4
+// CHECK19-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP75]], align 4
+// CHECK19-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP76]], align 4
+// CHECK19-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 8
+// CHECK19-NEXT: store i64 0, ptr [[TMP77]], align 8
+// CHECK19-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP78]], align 8
+// CHECK19-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP79]], align 4
+// CHECK19-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP80]], align 4
+// CHECK19-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP81]], align 4
+// CHECK19-NEXT: [[TMP82:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l136.region_id, ptr [[KERNEL_ARGS8]])
+// CHECK19-NEXT: [[TMP83:%.*]] = icmp ne i32 [[TMP82]], 0
+// CHECK19-NEXT: br i1 [[TMP83]], label [[OMP_OFFLOAD_FAILED9:%.*]], label [[OMP_OFFLOAD_CONT10:%.*]]
// CHECK19: omp_offload.failed9:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l136(i32 [[TMP53]], i32 [[TMP55]]) #[[ATTR4]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l136(i32 [[TMP57]], i32 [[TMP59]]) #[[ATTR4]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT10]]
// CHECK19: omp_offload.cont10:
// CHECK19-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK19: omp_if.else:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l136(i32 [[TMP53]], i32 [[TMP55]]) #[[ATTR4]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l136(i32 [[TMP57]], i32 [[TMP59]]) #[[ATTR4]]
// CHECK19-NEXT: br label [[OMP_IF_END]]
// CHECK19: omp_if.end:
-// CHECK19-NEXT: [[TMP76:%.*]] = load i32, ptr [[A]], align 4
-// CHECK19-NEXT: store i32 [[TMP76]], ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK19-NEXT: [[TMP77:%.*]] = load i32, ptr [[A]], align 4
-// CHECK19-NEXT: store i32 [[TMP77]], ptr [[A_CASTED11]], align 4
-// CHECK19-NEXT: [[TMP78:%.*]] = load i32, ptr [[A_CASTED11]], align 4
-// CHECK19-NEXT: [[TMP79:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK19-NEXT: store i32 [[TMP79]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK19-NEXT: [[TMP80:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK19-NEXT: [[TMP81:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK19-NEXT: [[CMP12:%.*]] = icmp sgt i32 [[TMP81]], 20
+// CHECK19-NEXT: [[TMP84:%.*]] = load i32, ptr [[A]], align 4
+// CHECK19-NEXT: store i32 [[TMP84]], ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK19-NEXT: [[TMP85:%.*]] = load i32, ptr [[A]], align 4
+// CHECK19-NEXT: store i32 [[TMP85]], ptr [[A_CASTED11]], align 4
+// CHECK19-NEXT: [[TMP86:%.*]] = load i32, ptr [[A_CASTED11]], align 4
+// CHECK19-NEXT: [[TMP87:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK19-NEXT: store i32 [[TMP87]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK19-NEXT: [[TMP88:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK19-NEXT: [[TMP89:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK19-NEXT: [[CMP12:%.*]] = icmp sgt i32 [[TMP89]], 20
// CHECK19-NEXT: br i1 [[CMP12]], label [[OMP_IF_THEN13:%.*]], label [[OMP_IF_ELSE20:%.*]]
// CHECK19: omp_if.then13:
-// CHECK19-NEXT: [[TMP82:%.*]] = mul nuw i32 [[TMP1]], 4
-// CHECK19-NEXT: [[TMP83:%.*]] = sext i32 [[TMP82]] to i64
-// CHECK19-NEXT: [[TMP84:%.*]] = mul nuw i32 5, [[TMP3]]
-// CHECK19-NEXT: [[TMP85:%.*]] = mul nuw i32 [[TMP84]], 8
-// CHECK19-NEXT: [[TMP86:%.*]] = sext i32 [[TMP85]] to i64
+// CHECK19-NEXT: [[TMP90:%.*]] = mul nuw i32 [[TMP1]], 4
+// CHECK19-NEXT: [[TMP91:%.*]] = sext i32 [[TMP90]] to i64
+// CHECK19-NEXT: [[TMP92:%.*]] = mul nuw i32 5, [[TMP3]]
+// CHECK19-NEXT: [[TMP93:%.*]] = mul nuw i32 [[TMP92]], 8
+// CHECK19-NEXT: [[TMP94:%.*]] = sext i32 [[TMP93]] to i64
// CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES]], ptr align 4 @.offload_sizes.8, i32 80, i1 false)
-// CHECK19-NEXT: [[TMP87:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP78]], ptr [[TMP87]], align 4
-// CHECK19-NEXT: [[TMP88:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP78]], ptr [[TMP88]], align 4
-// CHECK19-NEXT: [[TMP89:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP89]], align 4
-// CHECK19-NEXT: [[TMP90:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 1
-// CHECK19-NEXT: store ptr [[B]], ptr [[TMP90]], align 4
-// CHECK19-NEXT: [[TMP91:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 1
-// CHECK19-NEXT: store ptr [[B]], ptr [[TMP91]], align 4
-// CHECK19-NEXT: [[TMP92:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 1
-// CHECK19-NEXT: store ptr null, ptr [[TMP92]], align 4
-// CHECK19-NEXT: [[TMP93:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 2
-// CHECK19-NEXT: store i32 [[TMP1]], ptr [[TMP93]], align 4
-// CHECK19-NEXT: [[TMP94:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 2
-// CHECK19-NEXT: store i32 [[TMP1]], ptr [[TMP94]], align 4
-// CHECK19-NEXT: [[TMP95:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 2
-// CHECK19-NEXT: store ptr null, ptr [[TMP95]], align 4
-// CHECK19-NEXT: [[TMP96:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP96]], align 4
-// CHECK19-NEXT: [[TMP97:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP97]], align 4
-// CHECK19-NEXT: [[TMP98:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
-// CHECK19-NEXT: store i64 [[TMP83]], ptr [[TMP98]], align 4
-// CHECK19-NEXT: [[TMP99:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 3
-// CHECK19-NEXT: store ptr null, ptr [[TMP99]], align 4
-// CHECK19-NEXT: [[TMP100:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 4
-// CHECK19-NEXT: store ptr [[C]], ptr [[TMP100]], align 4
-// CHECK19-NEXT: [[TMP101:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 4
-// CHECK19-NEXT: store ptr [[C]], ptr [[TMP101]], align 4
-// CHECK19-NEXT: [[TMP102:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 4
-// CHECK19-NEXT: store ptr null, ptr [[TMP102]], align 4
-// CHECK19-NEXT: [[TMP103:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 5
-// CHECK19-NEXT: store i32 5, ptr [[TMP103]], align 4
-// CHECK19-NEXT: [[TMP104:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 5
-// CHECK19-NEXT: store i32 5, ptr [[TMP104]], align 4
-// CHECK19-NEXT: [[TMP105:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 5
-// CHECK19-NEXT: store ptr null, ptr [[TMP105]], align 4
-// CHECK19-NEXT: [[TMP106:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 6
-// CHECK19-NEXT: store i32 [[TMP3]], ptr [[TMP106]], align 4
-// CHECK19-NEXT: [[TMP107:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 6
-// CHECK19-NEXT: store i32 [[TMP3]], ptr [[TMP107]], align 4
-// CHECK19-NEXT: [[TMP108:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP108]], align 4
-// CHECK19-NEXT: [[TMP109:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 7
-// CHECK19-NEXT: store ptr [[VLA1]], ptr [[TMP109]], align 4
-// CHECK19-NEXT: [[TMP110:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 7
-// CHECK19-NEXT: store ptr [[VLA1]], ptr [[TMP110]], align 4
-// CHECK19-NEXT: [[TMP111:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
-// CHECK19-NEXT: store i64 [[TMP86]], ptr [[TMP111]], align 4
-// CHECK19-NEXT: [[TMP112:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP112]], align 4
-// CHECK19-NEXT: [[TMP113:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 8
-// CHECK19-NEXT: store ptr [[D]], ptr [[TMP113]], align 4
-// CHECK19-NEXT: [[TMP114:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 8
-// CHECK19-NEXT: store ptr [[D]], ptr [[TMP114]], align 4
-// CHECK19-NEXT: [[TMP115:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 8
-// CHECK19-NEXT: store ptr null, ptr [[TMP115]], align 4
-// CHECK19-NEXT: [[TMP116:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 9
-// CHECK19-NEXT: store i32 [[TMP80]], ptr [[TMP116]], align 4
-// CHECK19-NEXT: [[TMP117:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 9
-// CHECK19-NEXT: store i32 [[TMP80]], ptr [[TMP117]], align 4
-// CHECK19-NEXT: [[TMP118:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 9
-// CHECK19-NEXT: store ptr null, ptr [[TMP118]], align 4
-// CHECK19-NEXT: [[TMP119:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP120:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP121:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP95:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP86]], ptr [[TMP95]], align 4
+// CHECK19-NEXT: [[TMP96:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP86]], ptr [[TMP96]], align 4
+// CHECK19-NEXT: [[TMP97:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP97]], align 4
+// CHECK19-NEXT: [[TMP98:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 1
+// CHECK19-NEXT: store ptr [[B]], ptr [[TMP98]], align 4
+// CHECK19-NEXT: [[TMP99:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 1
+// CHECK19-NEXT: store ptr [[B]], ptr [[TMP99]], align 4
+// CHECK19-NEXT: [[TMP100:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 1
+// CHECK19-NEXT: store ptr null, ptr [[TMP100]], align 4
+// CHECK19-NEXT: [[TMP101:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 2
+// CHECK19-NEXT: store i32 [[TMP1]], ptr [[TMP101]], align 4
+// CHECK19-NEXT: [[TMP102:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 2
+// CHECK19-NEXT: store i32 [[TMP1]], ptr [[TMP102]], align 4
+// CHECK19-NEXT: [[TMP103:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 2
+// CHECK19-NEXT: store ptr null, ptr [[TMP103]], align 4
+// CHECK19-NEXT: [[TMP104:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP104]], align 4
+// CHECK19-NEXT: [[TMP105:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP105]], align 4
+// CHECK19-NEXT: [[TMP106:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
+// CHECK19-NEXT: store i64 [[TMP91]], ptr [[TMP106]], align 4
+// CHECK19-NEXT: [[TMP107:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 3
+// CHECK19-NEXT: store ptr null, ptr [[TMP107]], align 4
+// CHECK19-NEXT: [[TMP108:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 4
+// CHECK19-NEXT: store ptr [[C]], ptr [[TMP108]], align 4
+// CHECK19-NEXT: [[TMP109:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 4
+// CHECK19-NEXT: store ptr [[C]], ptr [[TMP109]], align 4
+// CHECK19-NEXT: [[TMP110:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 4
+// CHECK19-NEXT: store ptr null, ptr [[TMP110]], align 4
+// CHECK19-NEXT: [[TMP111:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 5
+// CHECK19-NEXT: store i32 5, ptr [[TMP111]], align 4
+// CHECK19-NEXT: [[TMP112:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 5
+// CHECK19-NEXT: store i32 5, ptr [[TMP112]], align 4
+// CHECK19-NEXT: [[TMP113:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 5
+// CHECK19-NEXT: store ptr null, ptr [[TMP113]], align 4
+// CHECK19-NEXT: [[TMP114:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 6
+// CHECK19-NEXT: store i32 [[TMP3]], ptr [[TMP114]], align 4
+// CHECK19-NEXT: [[TMP115:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 6
+// CHECK19-NEXT: store i32 [[TMP3]], ptr [[TMP115]], align 4
+// CHECK19-NEXT: [[TMP116:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP116]], align 4
+// CHECK19-NEXT: [[TMP117:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 7
+// CHECK19-NEXT: store ptr [[VLA1]], ptr [[TMP117]], align 4
+// CHECK19-NEXT: [[TMP118:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 7
+// CHECK19-NEXT: store ptr [[VLA1]], ptr [[TMP118]], align 4
+// CHECK19-NEXT: [[TMP119:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
+// CHECK19-NEXT: store i64 [[TMP94]], ptr [[TMP119]], align 4
+// CHECK19-NEXT: [[TMP120:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP120]], align 4
+// CHECK19-NEXT: [[TMP121:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 8
+// CHECK19-NEXT: store ptr [[D]], ptr [[TMP121]], align 4
+// CHECK19-NEXT: [[TMP122:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 8
+// CHECK19-NEXT: store ptr [[D]], ptr [[TMP122]], align 4
+// CHECK19-NEXT: [[TMP123:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 8
+// CHECK19-NEXT: store ptr null, ptr [[TMP123]], align 4
+// CHECK19-NEXT: [[TMP124:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 9
+// CHECK19-NEXT: store i32 [[TMP88]], ptr [[TMP124]], align 4
+// CHECK19-NEXT: [[TMP125:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 9
+// CHECK19-NEXT: store i32 [[TMP88]], ptr [[TMP125]], align 4
+// CHECK19-NEXT: [[TMP126:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 9
+// CHECK19-NEXT: store ptr null, ptr [[TMP126]], align 4
+// CHECK19-NEXT: [[TMP127:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP128:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP129:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS17:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP122:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP122]], align 4
-// CHECK19-NEXT: [[TMP123:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 1
-// CHECK19-NEXT: store i32 10, ptr [[TMP123]], align 4
-// CHECK19-NEXT: [[TMP124:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP119]], ptr [[TMP124]], align 4
-// CHECK19-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP120]], ptr [[TMP125]], align 4
-// CHECK19-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 4
-// CHECK19-NEXT: store ptr [[TMP121]], ptr [[TMP126]], align 4
-// CHECK19-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP127]], align 4
-// CHECK19-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP128]], align 4
-// CHECK19-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP129]], align 4
-// CHECK19-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 8
-// CHECK19-NEXT: store i64 0, ptr [[TMP130]], align 8
-// CHECK19-NEXT: [[TMP131:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l160.region_id, ptr [[KERNEL_ARGS17]])
-// CHECK19-NEXT: [[TMP132:%.*]] = icmp ne i32 [[TMP131]], 0
-// CHECK19-NEXT: br i1 [[TMP132]], label [[OMP_OFFLOAD_FAILED18:%.*]], label [[OMP_OFFLOAD_CONT19:%.*]]
+// CHECK19-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP130]], align 4
+// CHECK19-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 1
+// CHECK19-NEXT: store i32 10, ptr [[TMP131]], align 4
+// CHECK19-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP127]], ptr [[TMP132]], align 4
+// CHECK19-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP128]], ptr [[TMP133]], align 4
+// CHECK19-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 4
+// CHECK19-NEXT: store ptr [[TMP129]], ptr [[TMP134]], align 4
+// CHECK19-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP135]], align 4
+// CHECK19-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP136]], align 4
+// CHECK19-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP137]], align 4
+// CHECK19-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 8
+// CHECK19-NEXT: store i64 0, ptr [[TMP138]], align 8
+// CHECK19-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP139]], align 8
+// CHECK19-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP140]], align 4
+// CHECK19-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP141]], align 4
+// CHECK19-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP142]], align 4
+// CHECK19-NEXT: [[TMP143:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l160.region_id, ptr [[KERNEL_ARGS17]])
+// CHECK19-NEXT: [[TMP144:%.*]] = icmp ne i32 [[TMP143]], 0
+// CHECK19-NEXT: br i1 [[TMP144]], label [[OMP_OFFLOAD_FAILED18:%.*]], label [[OMP_OFFLOAD_CONT19:%.*]]
// CHECK19: omp_offload.failed18:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l160(i32 [[TMP78]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]], i32 [[TMP80]]) #[[ATTR4]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l160(i32 [[TMP86]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]], i32 [[TMP88]]) #[[ATTR4]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT19]]
// CHECK19: omp_offload.cont19:
// CHECK19-NEXT: br label [[OMP_IF_END21:%.*]]
// CHECK19: omp_if.else20:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l160(i32 [[TMP78]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]], i32 [[TMP80]]) #[[ATTR4]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l160(i32 [[TMP86]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]], i32 [[TMP88]]) #[[ATTR4]]
// CHECK19-NEXT: br label [[OMP_IF_END21]]
// CHECK19: omp_if.end21:
-// CHECK19-NEXT: [[TMP133:%.*]] = load i32, ptr [[A]], align 4
-// CHECK19-NEXT: [[TMP134:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK19-NEXT: call void @llvm.stackrestore(ptr [[TMP134]])
-// CHECK19-NEXT: ret i32 [[TMP133]]
+// CHECK19-NEXT: [[TMP145:%.*]] = load i32, ptr [[A]], align 4
+// CHECK19-NEXT: [[TMP146:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK19-NEXT: call void @llvm.stackrestore(ptr [[TMP146]])
+// CHECK19-NEXT: ret i32 [[TMP145]]
//
//
// CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l103
@@ -7401,7 +7593,7 @@ int bar(int n){
// CHECK19-NEXT: [[TMP15:%.*]] = load ptr, ptr [[DOTFIRSTPRIV_PTR_ADDR3_I]], align 4, !noalias !25
// CHECK19-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT_ANON:%.*]], ptr [[TMP9]], i32 0, i32 1
// CHECK19-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[TMP9]], i32 0, i32 2
-// CHECK19-NEXT: store i32 1, ptr [[KERNEL_ARGS_I]], align 4, !noalias !25
+// CHECK19-NEXT: store i32 2, ptr [[KERNEL_ARGS_I]], align 4, !noalias !25
// CHECK19-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 1
// CHECK19-NEXT: store i32 3, ptr [[TMP18]], align 4, !noalias !25
// CHECK19-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 2
@@ -7418,20 +7610,28 @@ int bar(int n){
// CHECK19-NEXT: store ptr null, ptr [[TMP24]], align 4, !noalias !25
// CHECK19-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 8
// CHECK19-NEXT: store i64 0, ptr [[TMP25]], align 8, !noalias !25
-// CHECK19-NEXT: [[TMP26:%.*]] = call i32 @__tgt_target_kernel_nowait(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l128.region_id, ptr [[KERNEL_ARGS_I]], i32 0, ptr null, i32 0, ptr null)
-// CHECK19-NEXT: [[TMP27:%.*]] = icmp ne i32 [[TMP26]], 0
-// CHECK19-NEXT: br i1 [[TMP27]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__3_EXIT:%.*]]
+// CHECK19-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 9
+// CHECK19-NEXT: store i64 1, ptr [[TMP26]], align 8, !noalias !25
+// CHECK19-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP27]], align 4, !noalias !25
+// CHECK19-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4, !noalias !25
+// CHECK19-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP29]], align 4, !noalias !25
+// CHECK19-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l128.region_id, ptr [[KERNEL_ARGS_I]])
+// CHECK19-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
+// CHECK19-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__3_EXIT:%.*]]
// CHECK19: omp_offload.failed.i:
-// CHECK19-NEXT: [[TMP28:%.*]] = load i16, ptr [[TMP12]], align 2
-// CHECK19-NEXT: store i16 [[TMP28]], ptr [[AA_CASTED_I]], align 2, !noalias !25
-// CHECK19-NEXT: [[TMP29:%.*]] = load i32, ptr [[AA_CASTED_I]], align 4, !noalias !25
-// CHECK19-NEXT: [[TMP30:%.*]] = load i32, ptr [[TMP16]], align 4
-// CHECK19-NEXT: store i32 [[TMP30]], ptr [[LIN_CASTED_I]], align 4, !noalias !25
-// CHECK19-NEXT: [[TMP31:%.*]] = load i32, ptr [[LIN_CASTED_I]], align 4, !noalias !25
-// CHECK19-NEXT: [[TMP32:%.*]] = load i32, ptr [[TMP17]], align 4
-// CHECK19-NEXT: store i32 [[TMP32]], ptr [[A_CASTED_I]], align 4, !noalias !25
-// CHECK19-NEXT: [[TMP33:%.*]] = load i32, ptr [[A_CASTED_I]], align 4, !noalias !25
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l128(i32 [[TMP29]], i32 [[TMP31]], i32 [[TMP33]]) #[[ATTR4]]
+// CHECK19-NEXT: [[TMP32:%.*]] = load i16, ptr [[TMP12]], align 2
+// CHECK19-NEXT: store i16 [[TMP32]], ptr [[AA_CASTED_I]], align 2, !noalias !25
+// CHECK19-NEXT: [[TMP33:%.*]] = load i32, ptr [[AA_CASTED_I]], align 4, !noalias !25
+// CHECK19-NEXT: [[TMP34:%.*]] = load i32, ptr [[TMP16]], align 4
+// CHECK19-NEXT: store i32 [[TMP34]], ptr [[LIN_CASTED_I]], align 4, !noalias !25
+// CHECK19-NEXT: [[TMP35:%.*]] = load i32, ptr [[LIN_CASTED_I]], align 4, !noalias !25
+// CHECK19-NEXT: [[TMP36:%.*]] = load i32, ptr [[TMP17]], align 4
+// CHECK19-NEXT: store i32 [[TMP36]], ptr [[A_CASTED_I]], align 4, !noalias !25
+// CHECK19-NEXT: [[TMP37:%.*]] = load i32, ptr [[A_CASTED_I]], align 4, !noalias !25
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l128(i32 [[TMP33]], i32 [[TMP35]], i32 [[TMP37]]) #[[ATTR4]]
// CHECK19-NEXT: br label [[DOTOMP_OUTLINED__3_EXIT]]
// CHECK19: .omp_outlined..3.exit:
// CHECK19-NEXT: ret i32 0
@@ -7821,7 +8021,7 @@ int bar(int n){
// CHECK19-NEXT: [[TMP28:%.*]] = getelementptr inbounds [5 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK19-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP29]], align 4
+// CHECK19-NEXT: store i32 2, ptr [[TMP29]], align 4
// CHECK19-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK19-NEXT: store i32 5, ptr [[TMP30]], align 4
// CHECK19-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -7838,9 +8038,17 @@ int bar(int n){
// CHECK19-NEXT: store ptr null, ptr [[TMP36]], align 4
// CHECK19-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK19-NEXT: store i64 0, ptr [[TMP37]], align 8
-// CHECK19-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l232.region_id, ptr [[KERNEL_ARGS]])
-// CHECK19-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK19-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK19-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP38]], align 8
+// CHECK19-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP39]], align 4
+// CHECK19-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP40]], align 4
+// CHECK19-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP41]], align 4
+// CHECK19-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l232.region_id, ptr [[KERNEL_ARGS]])
+// CHECK19-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
+// CHECK19-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK19: omp_offload.failed:
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l232(ptr [[THIS1]], i32 [[TMP5]], i32 2, i32 [[TMP1]], ptr [[VLA]]) #[[ATTR4]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -7850,15 +8058,15 @@ int bar(int n){
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l232(ptr [[THIS1]], i32 [[TMP5]], i32 2, i32 [[TMP1]], ptr [[VLA]]) #[[ATTR4]]
// CHECK19-NEXT: br label [[OMP_IF_END]]
// CHECK19: omp_if.end:
-// CHECK19-NEXT: [[TMP40:%.*]] = mul nsw i32 1, [[TMP1]]
-// CHECK19-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i32 [[TMP40]]
+// CHECK19-NEXT: [[TMP44:%.*]] = mul nsw i32 1, [[TMP1]]
+// CHECK19-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i32 [[TMP44]]
// CHECK19-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i16, ptr [[ARRAYIDX]], i32 1
-// CHECK19-NEXT: [[TMP41:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
-// CHECK19-NEXT: [[CONV:%.*]] = sext i16 [[TMP41]] to i32
-// CHECK19-NEXT: [[TMP42:%.*]] = load i32, ptr [[B]], align 4
-// CHECK19-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP42]]
-// CHECK19-NEXT: [[TMP43:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK19-NEXT: call void @llvm.stackrestore(ptr [[TMP43]])
+// CHECK19-NEXT: [[TMP45:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
+// CHECK19-NEXT: [[CONV:%.*]] = sext i16 [[TMP45]] to i32
+// CHECK19-NEXT: [[TMP46:%.*]] = load i32, ptr [[B]], align 4
+// CHECK19-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP46]]
+// CHECK19-NEXT: [[TMP47:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK19-NEXT: call void @llvm.stackrestore(ptr [[TMP47]])
// CHECK19-NEXT: ret i32 [[ADD3]]
//
//
@@ -7921,7 +8129,7 @@ int bar(int n){
// CHECK19-NEXT: [[TMP20:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK19-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK19-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK19-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK19-NEXT: store i32 4, ptr [[TMP22]], align 4
// CHECK19-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -7938,9 +8146,17 @@ int bar(int n){
// CHECK19-NEXT: store ptr null, ptr [[TMP28]], align 4
// CHECK19-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK19-NEXT: store i64 0, ptr [[TMP29]], align 8
-// CHECK19-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l214.region_id, ptr [[KERNEL_ARGS]])
-// CHECK19-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK19-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK19-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK19-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP31]], align 4
+// CHECK19-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK19-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK19-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l214.region_id, ptr [[KERNEL_ARGS]])
+// CHECK19-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK19-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK19: omp_offload.failed:
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l214(i32 [[TMP1]], i32 [[TMP3]], i32 [[TMP5]], ptr [[B]]) #[[ATTR4]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -7950,8 +8166,8 @@ int bar(int n){
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l214(i32 [[TMP1]], i32 [[TMP3]], i32 [[TMP5]], ptr [[B]]) #[[ATTR4]]
// CHECK19-NEXT: br label [[OMP_IF_END]]
// CHECK19: omp_if.end:
-// CHECK19-NEXT: [[TMP32:%.*]] = load i32, ptr [[A]], align 4
-// CHECK19-NEXT: ret i32 [[TMP32]]
+// CHECK19-NEXT: [[TMP36:%.*]] = load i32, ptr [[A]], align 4
+// CHECK19-NEXT: ret i32 [[TMP36]]
//
//
// CHECK19-LABEL: define {{[^@]+}}@_Z9ftemplateIiET_i
@@ -8001,7 +8217,7 @@ int bar(int n){
// CHECK19-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK19-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP16]], align 4
+// CHECK19-NEXT: store i32 2, ptr [[TMP16]], align 4
// CHECK19-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK19-NEXT: store i32 3, ptr [[TMP17]], align 4
// CHECK19-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -8018,9 +8234,17 @@ int bar(int n){
// CHECK19-NEXT: store ptr null, ptr [[TMP23]], align 4
// CHECK19-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK19-NEXT: store i64 0, ptr [[TMP24]], align 8
-// CHECK19-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l197.region_id, ptr [[KERNEL_ARGS]])
-// CHECK19-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK19-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK19-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP25]], align 8
+// CHECK19-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK19-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP27]], align 4
+// CHECK19-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP28]], align 4
+// CHECK19-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l197.region_id, ptr [[KERNEL_ARGS]])
+// CHECK19-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK19-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK19: omp_offload.failed:
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l197(i32 [[TMP1]], i32 [[TMP3]], ptr [[B]]) #[[ATTR4]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -8030,8 +8254,8 @@ int bar(int n){
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l197(i32 [[TMP1]], i32 [[TMP3]], ptr [[B]]) #[[ATTR4]]
// CHECK19-NEXT: br label [[OMP_IF_END]]
// CHECK19: omp_if.end:
-// CHECK19-NEXT: [[TMP27:%.*]] = load i32, ptr [[A]], align 4
-// CHECK19-NEXT: ret i32 [[TMP27]]
+// CHECK19-NEXT: [[TMP31:%.*]] = load i32, ptr [[A]], align 4
+// CHECK19-NEXT: ret i32 [[TMP31]]
//
//
// CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l232
diff --git a/clang/test/OpenMP/target_parallel_for_depend_codegen.cpp b/clang/test/OpenMP/target_parallel_for_depend_codegen.cpp
index d243bbca6a77e..56f5926b99ba7 100644
--- a/clang/test/OpenMP/target_parallel_for_depend_codegen.cpp
+++ b/clang/test/OpenMP/target_parallel_for_depend_codegen.cpp
@@ -200,7 +200,7 @@ int foo(int n) {
// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2
// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]],
// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64
-// CHECK: [[RET:%.+]] = call i32 @__tgt_target_kernel_nowait(%struct.ident_t* @{{.+}}, i64 [[DEVICE]], i32 1, i32 0, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* %{{.+}})
+// CHECK: [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 [[DEVICE]], i32 1, i32 0, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* %{{.+}})
// CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]]
diff --git a/clang/test/OpenMP/target_parallel_for_simd_codegen.cpp b/clang/test/OpenMP/target_parallel_for_simd_codegen.cpp
index e9970392a1f29..387c93cd35647 100644
--- a/clang/test/OpenMP/target_parallel_for_simd_codegen.cpp
+++ b/clang/test/OpenMP/target_parallel_for_simd_codegen.cpp
@@ -393,7 +393,7 @@ int bar(int n){
// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP31]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP31]], align 4
// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 3, ptr [[TMP32]], align 4
// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -410,186 +410,210 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP38]], align 8
// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l108.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP41:%.*]] = icmp ne i32 [[TMP40]], 0
-// CHECK1-NEXT: br i1 [[TMP41]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP41]], align 4
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP42]], align 4
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP43]], align 4
+// CHECK1-NEXT: [[TMP44:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l108.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP45:%.*]] = icmp ne i32 [[TMP44]], 0
+// CHECK1-NEXT: br i1 [[TMP45]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l108(i64 [[TMP15]], i64 [[TMP17]], i64 [[TMP19]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
-// CHECK1-NEXT: [[TMP42:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: store i32 [[TMP42]], ptr [[A_CASTED3]], align 4
-// CHECK1-NEXT: [[TMP43:%.*]] = load i64, ptr [[A_CASTED3]], align 8
-// CHECK1-NEXT: [[TMP44:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK1-NEXT: store i16 [[TMP44]], ptr [[AA_CASTED4]], align 2
-// CHECK1-NEXT: [[TMP45:%.*]] = load i64, ptr [[AA_CASTED4]], align 8
-// CHECK1-NEXT: [[TMP46:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP46]], 10
+// CHECK1-NEXT: [[TMP46:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: store i32 [[TMP46]], ptr [[A_CASTED3]], align 4
+// CHECK1-NEXT: [[TMP47:%.*]] = load i64, ptr [[A_CASTED3]], align 8
+// CHECK1-NEXT: [[TMP48:%.*]] = load i16, ptr [[AA]], align 2
+// CHECK1-NEXT: store i16 [[TMP48]], ptr [[AA_CASTED4]], align 2
+// CHECK1-NEXT: [[TMP49:%.*]] = load i64, ptr [[AA_CASTED4]], align 8
+// CHECK1-NEXT: [[TMP50:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP50]], 10
// CHECK1-NEXT: br i1 [[CMP]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK1: omp_if.then:
-// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP43]], ptr [[TMP47]], align 8
-// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP43]], ptr [[TMP48]], align 8
-// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP49]], align 8
-// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP45]], ptr [[TMP50]], align 8
-// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP45]], ptr [[TMP51]], align 8
-// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP52]], align 8
-// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP47]], ptr [[TMP51]], align 8
+// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP47]], ptr [[TMP52]], align 8
+// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP53]], align 8
+// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP49]], ptr [[TMP54]], align 8
+// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP49]], ptr [[TMP55]], align 8
+// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP56]], align 8
+// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP58:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS8:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP55]], align 4
-// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 1
-// CHECK1-NEXT: store i32 2, ptr [[TMP56]], align 4
-// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP53]], ptr [[TMP57]], align 8
-// CHECK1-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP54]], ptr [[TMP58]], align 8
-// CHECK1-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.5, ptr [[TMP59]], align 8
-// CHECK1-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP60]], align 8
-// CHECK1-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP61]], align 8
-// CHECK1-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP62]], align 8
-// CHECK1-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP63]], align 8
-// CHECK1-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l116.region_id, ptr [[KERNEL_ARGS8]])
-// CHECK1-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
-// CHECK1-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED9:%.*]], label [[OMP_OFFLOAD_CONT10:%.*]]
+// CHECK1-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP59]], align 4
+// CHECK1-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 1
+// CHECK1-NEXT: store i32 2, ptr [[TMP60]], align 4
+// CHECK1-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP57]], ptr [[TMP61]], align 8
+// CHECK1-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP58]], ptr [[TMP62]], align 8
+// CHECK1-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.5, ptr [[TMP63]], align 8
+// CHECK1-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP64]], align 8
+// CHECK1-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP65]], align 8
+// CHECK1-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP66]], align 8
+// CHECK1-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP67]], align 8
+// CHECK1-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP68]], align 8
+// CHECK1-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP69]], align 4
+// CHECK1-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP70]], align 4
+// CHECK1-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP71]], align 4
+// CHECK1-NEXT: [[TMP72:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l116.region_id, ptr [[KERNEL_ARGS8]])
+// CHECK1-NEXT: [[TMP73:%.*]] = icmp ne i32 [[TMP72]], 0
+// CHECK1-NEXT: br i1 [[TMP73]], label [[OMP_OFFLOAD_FAILED9:%.*]], label [[OMP_OFFLOAD_CONT10:%.*]]
// CHECK1: omp_offload.failed9:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l116(i64 [[TMP43]], i64 [[TMP45]]) #[[ATTR4]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l116(i64 [[TMP47]], i64 [[TMP49]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT10]]
// CHECK1: omp_offload.cont10:
// CHECK1-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK1: omp_if.else:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l116(i64 [[TMP43]], i64 [[TMP45]]) #[[ATTR4]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l116(i64 [[TMP47]], i64 [[TMP49]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
-// CHECK1-NEXT: [[TMP66:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: store i32 [[TMP66]], ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK1-NEXT: [[TMP67:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: store i32 [[TMP67]], ptr [[A_CASTED11]], align 4
-// CHECK1-NEXT: [[TMP68:%.*]] = load i64, ptr [[A_CASTED11]], align 8
-// CHECK1-NEXT: [[TMP69:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK1-NEXT: store i32 [[TMP69]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK1-NEXT: [[TMP70:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK1-NEXT: [[TMP71:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: [[CMP12:%.*]] = icmp sgt i32 [[TMP71]], 20
+// CHECK1-NEXT: [[TMP74:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: store i32 [[TMP74]], ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK1-NEXT: [[TMP75:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: store i32 [[TMP75]], ptr [[A_CASTED11]], align 4
+// CHECK1-NEXT: [[TMP76:%.*]] = load i64, ptr [[A_CASTED11]], align 8
+// CHECK1-NEXT: [[TMP77:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK1-NEXT: store i32 [[TMP77]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK1-NEXT: [[TMP78:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK1-NEXT: [[TMP79:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: [[CMP12:%.*]] = icmp sgt i32 [[TMP79]], 20
// CHECK1-NEXT: br i1 [[CMP12]], label [[OMP_IF_THEN13:%.*]], label [[OMP_IF_ELSE20:%.*]]
// CHECK1: omp_if.then13:
-// CHECK1-NEXT: [[TMP72:%.*]] = mul nuw i64 [[TMP2]], 4
-// CHECK1-NEXT: [[TMP73:%.*]] = mul nuw i64 5, [[TMP5]]
-// CHECK1-NEXT: [[TMP74:%.*]] = mul nuw i64 [[TMP73]], 8
+// CHECK1-NEXT: [[TMP80:%.*]] = mul nuw i64 [[TMP2]], 4
+// CHECK1-NEXT: [[TMP81:%.*]] = mul nuw i64 5, [[TMP5]]
+// CHECK1-NEXT: [[TMP82:%.*]] = mul nuw i64 [[TMP81]], 8
// CHECK1-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES]], ptr align 8 @.offload_sizes.8, i64 80, i1 false)
-// CHECK1-NEXT: [[TMP75:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP68]], ptr [[TMP75]], align 8
-// CHECK1-NEXT: [[TMP76:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP68]], ptr [[TMP76]], align 8
-// CHECK1-NEXT: [[TMP77:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP77]], align 8
-// CHECK1-NEXT: [[TMP78:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 1
-// CHECK1-NEXT: store ptr [[B]], ptr [[TMP78]], align 8
-// CHECK1-NEXT: [[TMP79:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 1
-// CHECK1-NEXT: store ptr [[B]], ptr [[TMP79]], align 8
-// CHECK1-NEXT: [[TMP80:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP80]], align 8
-// CHECK1-NEXT: [[TMP81:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 2
-// CHECK1-NEXT: store i64 [[TMP2]], ptr [[TMP81]], align 8
-// CHECK1-NEXT: [[TMP82:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 2
-// CHECK1-NEXT: store i64 [[TMP2]], ptr [[TMP82]], align 8
-// CHECK1-NEXT: [[TMP83:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP83]], align 8
-// CHECK1-NEXT: [[TMP84:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[VLA]], ptr [[TMP84]], align 8
-// CHECK1-NEXT: [[TMP85:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[VLA]], ptr [[TMP85]], align 8
-// CHECK1-NEXT: [[TMP86:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
-// CHECK1-NEXT: store i64 [[TMP72]], ptr [[TMP86]], align 8
-// CHECK1-NEXT: [[TMP87:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP87]], align 8
-// CHECK1-NEXT: [[TMP88:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 4
-// CHECK1-NEXT: store ptr [[C]], ptr [[TMP88]], align 8
-// CHECK1-NEXT: [[TMP89:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 4
-// CHECK1-NEXT: store ptr [[C]], ptr [[TMP89]], align 8
-// CHECK1-NEXT: [[TMP90:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP90]], align 8
-// CHECK1-NEXT: [[TMP91:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 5
-// CHECK1-NEXT: store i64 5, ptr [[TMP91]], align 8
-// CHECK1-NEXT: [[TMP92:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 5
-// CHECK1-NEXT: store i64 5, ptr [[TMP92]], align 8
-// CHECK1-NEXT: [[TMP93:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP93]], align 8
-// CHECK1-NEXT: [[TMP94:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 6
-// CHECK1-NEXT: store i64 [[TMP5]], ptr [[TMP94]], align 8
-// CHECK1-NEXT: [[TMP95:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 6
-// CHECK1-NEXT: store i64 [[TMP5]], ptr [[TMP95]], align 8
-// CHECK1-NEXT: [[TMP96:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP96]], align 8
-// CHECK1-NEXT: [[TMP97:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 7
-// CHECK1-NEXT: store ptr [[VLA1]], ptr [[TMP97]], align 8
-// CHECK1-NEXT: [[TMP98:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 7
-// CHECK1-NEXT: store ptr [[VLA1]], ptr [[TMP98]], align 8
-// CHECK1-NEXT: [[TMP99:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
-// CHECK1-NEXT: store i64 [[TMP74]], ptr [[TMP99]], align 8
-// CHECK1-NEXT: [[TMP100:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP100]], align 8
-// CHECK1-NEXT: [[TMP101:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 8
-// CHECK1-NEXT: store ptr [[D]], ptr [[TMP101]], align 8
-// CHECK1-NEXT: [[TMP102:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 8
-// CHECK1-NEXT: store ptr [[D]], ptr [[TMP102]], align 8
-// CHECK1-NEXT: [[TMP103:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 8
-// CHECK1-NEXT: store ptr null, ptr [[TMP103]], align 8
-// CHECK1-NEXT: [[TMP104:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 9
-// CHECK1-NEXT: store i64 [[TMP70]], ptr [[TMP104]], align 8
-// CHECK1-NEXT: [[TMP105:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 9
-// CHECK1-NEXT: store i64 [[TMP70]], ptr [[TMP105]], align 8
-// CHECK1-NEXT: [[TMP106:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 9
-// CHECK1-NEXT: store ptr null, ptr [[TMP106]], align 8
-// CHECK1-NEXT: [[TMP107:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP108:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP109:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP83:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP76]], ptr [[TMP83]], align 8
+// CHECK1-NEXT: [[TMP84:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP76]], ptr [[TMP84]], align 8
+// CHECK1-NEXT: [[TMP85:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP85]], align 8
+// CHECK1-NEXT: [[TMP86:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 1
+// CHECK1-NEXT: store ptr [[B]], ptr [[TMP86]], align 8
+// CHECK1-NEXT: [[TMP87:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 1
+// CHECK1-NEXT: store ptr [[B]], ptr [[TMP87]], align 8
+// CHECK1-NEXT: [[TMP88:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP88]], align 8
+// CHECK1-NEXT: [[TMP89:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 2
+// CHECK1-NEXT: store i64 [[TMP2]], ptr [[TMP89]], align 8
+// CHECK1-NEXT: [[TMP90:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 2
+// CHECK1-NEXT: store i64 [[TMP2]], ptr [[TMP90]], align 8
+// CHECK1-NEXT: [[TMP91:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 2
+// CHECK1-NEXT: store ptr null, ptr [[TMP91]], align 8
+// CHECK1-NEXT: [[TMP92:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[VLA]], ptr [[TMP92]], align 8
+// CHECK1-NEXT: [[TMP93:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[VLA]], ptr [[TMP93]], align 8
+// CHECK1-NEXT: [[TMP94:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
+// CHECK1-NEXT: store i64 [[TMP80]], ptr [[TMP94]], align 8
+// CHECK1-NEXT: [[TMP95:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 3
+// CHECK1-NEXT: store ptr null, ptr [[TMP95]], align 8
+// CHECK1-NEXT: [[TMP96:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 4
+// CHECK1-NEXT: store ptr [[C]], ptr [[TMP96]], align 8
+// CHECK1-NEXT: [[TMP97:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 4
+// CHECK1-NEXT: store ptr [[C]], ptr [[TMP97]], align 8
+// CHECK1-NEXT: [[TMP98:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP98]], align 8
+// CHECK1-NEXT: [[TMP99:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 5
+// CHECK1-NEXT: store i64 5, ptr [[TMP99]], align 8
+// CHECK1-NEXT: [[TMP100:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 5
+// CHECK1-NEXT: store i64 5, ptr [[TMP100]], align 8
+// CHECK1-NEXT: [[TMP101:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP101]], align 8
+// CHECK1-NEXT: [[TMP102:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 6
+// CHECK1-NEXT: store i64 [[TMP5]], ptr [[TMP102]], align 8
+// CHECK1-NEXT: [[TMP103:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 6
+// CHECK1-NEXT: store i64 [[TMP5]], ptr [[TMP103]], align 8
+// CHECK1-NEXT: [[TMP104:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP104]], align 8
+// CHECK1-NEXT: [[TMP105:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 7
+// CHECK1-NEXT: store ptr [[VLA1]], ptr [[TMP105]], align 8
+// CHECK1-NEXT: [[TMP106:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 7
+// CHECK1-NEXT: store ptr [[VLA1]], ptr [[TMP106]], align 8
+// CHECK1-NEXT: [[TMP107:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
+// CHECK1-NEXT: store i64 [[TMP82]], ptr [[TMP107]], align 8
+// CHECK1-NEXT: [[TMP108:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP108]], align 8
+// CHECK1-NEXT: [[TMP109:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 8
+// CHECK1-NEXT: store ptr [[D]], ptr [[TMP109]], align 8
+// CHECK1-NEXT: [[TMP110:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 8
+// CHECK1-NEXT: store ptr [[D]], ptr [[TMP110]], align 8
+// CHECK1-NEXT: [[TMP111:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 8
+// CHECK1-NEXT: store ptr null, ptr [[TMP111]], align 8
+// CHECK1-NEXT: [[TMP112:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 9
+// CHECK1-NEXT: store i64 [[TMP78]], ptr [[TMP112]], align 8
+// CHECK1-NEXT: [[TMP113:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 9
+// CHECK1-NEXT: store i64 [[TMP78]], ptr [[TMP113]], align 8
+// CHECK1-NEXT: [[TMP114:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 9
+// CHECK1-NEXT: store ptr null, ptr [[TMP114]], align 8
+// CHECK1-NEXT: [[TMP115:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP116:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP117:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS17:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP110]], align 4
-// CHECK1-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 1
-// CHECK1-NEXT: store i32 10, ptr [[TMP111]], align 4
-// CHECK1-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP107]], ptr [[TMP112]], align 8
-// CHECK1-NEXT: [[TMP113:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP108]], ptr [[TMP113]], align 8
-// CHECK1-NEXT: [[TMP114:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 4
-// CHECK1-NEXT: store ptr [[TMP109]], ptr [[TMP114]], align 8
-// CHECK1-NEXT: [[TMP115:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP115]], align 8
-// CHECK1-NEXT: [[TMP116:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP116]], align 8
-// CHECK1-NEXT: [[TMP117:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP117]], align 8
-// CHECK1-NEXT: [[TMP118:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP118]], align 8
-// CHECK1-NEXT: [[TMP119:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l140.region_id, ptr [[KERNEL_ARGS17]])
-// CHECK1-NEXT: [[TMP120:%.*]] = icmp ne i32 [[TMP119]], 0
-// CHECK1-NEXT: br i1 [[TMP120]], label [[OMP_OFFLOAD_FAILED18:%.*]], label [[OMP_OFFLOAD_CONT19:%.*]]
+// CHECK1-NEXT: [[TMP118:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP118]], align 4
+// CHECK1-NEXT: [[TMP119:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 1
+// CHECK1-NEXT: store i32 10, ptr [[TMP119]], align 4
+// CHECK1-NEXT: [[TMP120:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP115]], ptr [[TMP120]], align 8
+// CHECK1-NEXT: [[TMP121:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP116]], ptr [[TMP121]], align 8
+// CHECK1-NEXT: [[TMP122:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 4
+// CHECK1-NEXT: store ptr [[TMP117]], ptr [[TMP122]], align 8
+// CHECK1-NEXT: [[TMP123:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP123]], align 8
+// CHECK1-NEXT: [[TMP124:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP124]], align 8
+// CHECK1-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP125]], align 8
+// CHECK1-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP126]], align 8
+// CHECK1-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP127]], align 8
+// CHECK1-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP128]], align 4
+// CHECK1-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP129]], align 4
+// CHECK1-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP130]], align 4
+// CHECK1-NEXT: [[TMP131:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l140.region_id, ptr [[KERNEL_ARGS17]])
+// CHECK1-NEXT: [[TMP132:%.*]] = icmp ne i32 [[TMP131]], 0
+// CHECK1-NEXT: br i1 [[TMP132]], label [[OMP_OFFLOAD_FAILED18:%.*]], label [[OMP_OFFLOAD_CONT19:%.*]]
// CHECK1: omp_offload.failed18:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l140(i64 [[TMP68]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]], i64 [[TMP70]]) #[[ATTR4]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l140(i64 [[TMP76]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]], i64 [[TMP78]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT19]]
// CHECK1: omp_offload.cont19:
// CHECK1-NEXT: br label [[OMP_IF_END21:%.*]]
// CHECK1: omp_if.else20:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l140(i64 [[TMP68]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]], i64 [[TMP70]]) #[[ATTR4]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l140(i64 [[TMP76]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]], i64 [[TMP78]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_IF_END21]]
// CHECK1: omp_if.end21:
-// CHECK1-NEXT: [[TMP121:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: [[TMP122:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK1-NEXT: call void @llvm.stackrestore(ptr [[TMP122]])
-// CHECK1-NEXT: ret i32 [[TMP121]]
+// CHECK1-NEXT: [[TMP133:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: [[TMP134:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK1-NEXT: call void @llvm.stackrestore(ptr [[TMP134]])
+// CHECK1-NEXT: ret i32 [[TMP133]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l96
@@ -697,7 +721,7 @@ int bar(int n){
// CHECK1-NEXT: store ptr [[TMP3]], ptr [[DOTTASK_T__ADDR_I]], align 8, !noalias !25
// CHECK1-NEXT: store ptr [[TMP7]], ptr [[__CONTEXT_ADDR_I]], align 8, !noalias !25
// CHECK1-NEXT: [[TMP8:%.*]] = load ptr, ptr [[__CONTEXT_ADDR_I]], align 8, !noalias !25
-// CHECK1-NEXT: store i32 1, ptr [[KERNEL_ARGS_I]], align 4, !noalias !25
+// CHECK1-NEXT: store i32 2, ptr [[KERNEL_ARGS_I]], align 4, !noalias !25
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP9]], align 4, !noalias !25
// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 2
@@ -714,9 +738,17 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8, !noalias !25
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP16]], align 8, !noalias !25
-// CHECK1-NEXT: [[TMP17:%.*]] = call i32 @__tgt_target_kernel_nowait(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l96.region_id, ptr [[KERNEL_ARGS_I]], i32 0, ptr null, i32 0, ptr null)
-// CHECK1-NEXT: [[TMP18:%.*]] = icmp ne i32 [[TMP17]], 0
-// CHECK1-NEXT: br i1 [[TMP18]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__1_EXIT:%.*]]
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 9
+// CHECK1-NEXT: store i64 1, ptr [[TMP17]], align 8, !noalias !25
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP18]], align 4, !noalias !25
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP19]], align 4, !noalias !25
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP20]], align 4, !noalias !25
+// CHECK1-NEXT: [[TMP21:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l96.region_id, ptr [[KERNEL_ARGS_I]])
+// CHECK1-NEXT: [[TMP22:%.*]] = icmp ne i32 [[TMP21]], 0
+// CHECK1-NEXT: br i1 [[TMP22]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__1_EXIT:%.*]]
// CHECK1: omp_offload.failed.i:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l96() #[[ATTR4]]
// CHECK1-NEXT: br label [[DOTOMP_OUTLINED__1_EXIT]]
@@ -1371,7 +1403,7 @@ int bar(int n){
// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [5 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP29]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP29]], align 4
// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 5, ptr [[TMP30]], align 4
// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1388,9 +1420,17 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP36]], align 8
// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK1-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP38]], align 8
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP39]], align 4
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP40]], align 4
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP41]], align 4
+// CHECK1-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
+// CHECK1-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216(ptr [[THIS1]], i64 [[TMP6]], i64 2, i64 [[TMP2]], ptr [[VLA]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1400,15 +1440,15 @@ int bar(int n){
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216(ptr [[THIS1]], i64 [[TMP6]], i64 2, i64 [[TMP2]], ptr [[VLA]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
-// CHECK1-NEXT: [[TMP40:%.*]] = mul nsw i64 1, [[TMP2]]
-// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i64 [[TMP40]]
+// CHECK1-NEXT: [[TMP44:%.*]] = mul nsw i64 1, [[TMP2]]
+// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i64 [[TMP44]]
// CHECK1-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i16, ptr [[ARRAYIDX]], i64 1
-// CHECK1-NEXT: [[TMP41:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
-// CHECK1-NEXT: [[CONV:%.*]] = sext i16 [[TMP41]] to i32
-// CHECK1-NEXT: [[TMP42:%.*]] = load i32, ptr [[B]], align 4
-// CHECK1-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP42]]
-// CHECK1-NEXT: [[TMP43:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK1-NEXT: call void @llvm.stackrestore(ptr [[TMP43]])
+// CHECK1-NEXT: [[TMP45:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
+// CHECK1-NEXT: [[CONV:%.*]] = sext i16 [[TMP45]] to i32
+// CHECK1-NEXT: [[TMP46:%.*]] = load i32, ptr [[B]], align 4
+// CHECK1-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP46]]
+// CHECK1-NEXT: [[TMP47:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK1-NEXT: call void @llvm.stackrestore(ptr [[TMP47]])
// CHECK1-NEXT: ret i32 [[ADD3]]
//
//
@@ -1471,7 +1511,7 @@ int bar(int n){
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 4, ptr [[TMP22]], align 4
// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1488,9 +1528,17 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l195.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP31]], align 4
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK1-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l195.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK1-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l195(i64 [[TMP1]], i64 [[TMP3]], i64 [[TMP5]], ptr [[B]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1500,8 +1548,8 @@ int bar(int n){
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l195(i64 [[TMP1]], i64 [[TMP3]], i64 [[TMP5]], ptr [[B]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
-// CHECK1-NEXT: [[TMP32:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: ret i32 [[TMP32]]
+// CHECK1-NEXT: [[TMP36:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: ret i32 [[TMP36]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_Z9ftemplateIiET_i
@@ -1551,7 +1599,7 @@ int bar(int n){
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP16]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP16]], align 4
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 3, ptr [[TMP17]], align 4
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1568,9 +1616,17 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP23]], align 8
// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l178.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK1-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP25]], align 8
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP28]], align 4
+// CHECK1-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l178.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK1-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l178(i64 [[TMP1]], i64 [[TMP3]], ptr [[B]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1580,8 +1636,8 @@ int bar(int n){
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l178(i64 [[TMP1]], i64 [[TMP3]], ptr [[B]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
-// CHECK1-NEXT: [[TMP27:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: ret i32 [[TMP27]]
+// CHECK1-NEXT: [[TMP31:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: ret i32 [[TMP31]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216
@@ -1962,7 +2018,7 @@ int bar(int n){
// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP27]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP27]], align 4
// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 3, ptr [[TMP28]], align 4
// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1979,188 +2035,212 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP34]], align 4
// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP35]], align 8
-// CHECK3-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l108.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
-// CHECK3-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP36]], align 8
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP38]], align 4
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP39]], align 4
+// CHECK3-NEXT: [[TMP40:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l108.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP41:%.*]] = icmp ne i32 [[TMP40]], 0
+// CHECK3-NEXT: br i1 [[TMP41]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l108(i32 [[TMP11]], i32 [[TMP13]], i32 [[TMP15]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
-// CHECK3-NEXT: [[TMP38:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: store i32 [[TMP38]], ptr [[A_CASTED3]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = load i32, ptr [[A_CASTED3]], align 4
-// CHECK3-NEXT: [[TMP40:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK3-NEXT: store i16 [[TMP40]], ptr [[AA_CASTED4]], align 2
-// CHECK3-NEXT: [[TMP41:%.*]] = load i32, ptr [[AA_CASTED4]], align 4
-// CHECK3-NEXT: [[TMP42:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP42]], 10
+// CHECK3-NEXT: [[TMP42:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: store i32 [[TMP42]], ptr [[A_CASTED3]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = load i32, ptr [[A_CASTED3]], align 4
+// CHECK3-NEXT: [[TMP44:%.*]] = load i16, ptr [[AA]], align 2
+// CHECK3-NEXT: store i16 [[TMP44]], ptr [[AA_CASTED4]], align 2
+// CHECK3-NEXT: [[TMP45:%.*]] = load i32, ptr [[AA_CASTED4]], align 4
+// CHECK3-NEXT: [[TMP46:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP46]], 10
// CHECK3-NEXT: br i1 [[CMP]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK3: omp_if.then:
-// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP39]], ptr [[TMP43]], align 4
-// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP39]], ptr [[TMP44]], align 4
-// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP45]], align 4
-// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP41]], ptr [[TMP46]], align 4
-// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP41]], ptr [[TMP47]], align 4
-// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 1
-// CHECK3-NEXT: store ptr null, ptr [[TMP48]], align 4
-// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP43]], ptr [[TMP47]], align 4
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP43]], ptr [[TMP48]], align 4
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP49]], align 4
+// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP45]], ptr [[TMP50]], align 4
+// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP45]], ptr [[TMP51]], align 4
+// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 1
+// CHECK3-NEXT: store ptr null, ptr [[TMP52]], align 4
+// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS8:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP51]], align 4
-// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 1
-// CHECK3-NEXT: store i32 2, ptr [[TMP52]], align 4
-// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 4
-// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP50]], ptr [[TMP54]], align 4
-// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.5, ptr [[TMP55]], align 4
-// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP56]], align 4
-// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP57]], align 4
-// CHECK3-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP58]], align 4
-// CHECK3-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP59]], align 8
-// CHECK3-NEXT: [[TMP60:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l116.region_id, ptr [[KERNEL_ARGS8]])
-// CHECK3-NEXT: [[TMP61:%.*]] = icmp ne i32 [[TMP60]], 0
-// CHECK3-NEXT: br i1 [[TMP61]], label [[OMP_OFFLOAD_FAILED9:%.*]], label [[OMP_OFFLOAD_CONT10:%.*]]
+// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP55]], align 4
+// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 1
+// CHECK3-NEXT: store i32 2, ptr [[TMP56]], align 4
+// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP53]], ptr [[TMP57]], align 4
+// CHECK3-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP54]], ptr [[TMP58]], align 4
+// CHECK3-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.5, ptr [[TMP59]], align 4
+// CHECK3-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP60]], align 4
+// CHECK3-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP61]], align 4
+// CHECK3-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP62]], align 4
+// CHECK3-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP63]], align 8
+// CHECK3-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP64]], align 8
+// CHECK3-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP65]], align 4
+// CHECK3-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP66]], align 4
+// CHECK3-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP67]], align 4
+// CHECK3-NEXT: [[TMP68:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l116.region_id, ptr [[KERNEL_ARGS8]])
+// CHECK3-NEXT: [[TMP69:%.*]] = icmp ne i32 [[TMP68]], 0
+// CHECK3-NEXT: br i1 [[TMP69]], label [[OMP_OFFLOAD_FAILED9:%.*]], label [[OMP_OFFLOAD_CONT10:%.*]]
// CHECK3: omp_offload.failed9:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l116(i32 [[TMP39]], i32 [[TMP41]]) #[[ATTR4]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l116(i32 [[TMP43]], i32 [[TMP45]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT10]]
// CHECK3: omp_offload.cont10:
// CHECK3-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK3: omp_if.else:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l116(i32 [[TMP39]], i32 [[TMP41]]) #[[ATTR4]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l116(i32 [[TMP43]], i32 [[TMP45]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
-// CHECK3-NEXT: [[TMP62:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: store i32 [[TMP62]], ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK3-NEXT: [[TMP63:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: store i32 [[TMP63]], ptr [[A_CASTED11]], align 4
-// CHECK3-NEXT: [[TMP64:%.*]] = load i32, ptr [[A_CASTED11]], align 4
-// CHECK3-NEXT: [[TMP65:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK3-NEXT: store i32 [[TMP65]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK3-NEXT: [[TMP66:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK3-NEXT: [[TMP67:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: [[CMP12:%.*]] = icmp sgt i32 [[TMP67]], 20
+// CHECK3-NEXT: [[TMP70:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: store i32 [[TMP70]], ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK3-NEXT: [[TMP71:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: store i32 [[TMP71]], ptr [[A_CASTED11]], align 4
+// CHECK3-NEXT: [[TMP72:%.*]] = load i32, ptr [[A_CASTED11]], align 4
+// CHECK3-NEXT: [[TMP73:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK3-NEXT: store i32 [[TMP73]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK3-NEXT: [[TMP74:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK3-NEXT: [[TMP75:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: [[CMP12:%.*]] = icmp sgt i32 [[TMP75]], 20
// CHECK3-NEXT: br i1 [[CMP12]], label [[OMP_IF_THEN13:%.*]], label [[OMP_IF_ELSE20:%.*]]
// CHECK3: omp_if.then13:
-// CHECK3-NEXT: [[TMP68:%.*]] = mul nuw i32 [[TMP1]], 4
-// CHECK3-NEXT: [[TMP69:%.*]] = sext i32 [[TMP68]] to i64
-// CHECK3-NEXT: [[TMP70:%.*]] = mul nuw i32 5, [[TMP3]]
-// CHECK3-NEXT: [[TMP71:%.*]] = mul nuw i32 [[TMP70]], 8
-// CHECK3-NEXT: [[TMP72:%.*]] = sext i32 [[TMP71]] to i64
+// CHECK3-NEXT: [[TMP76:%.*]] = mul nuw i32 [[TMP1]], 4
+// CHECK3-NEXT: [[TMP77:%.*]] = sext i32 [[TMP76]] to i64
+// CHECK3-NEXT: [[TMP78:%.*]] = mul nuw i32 5, [[TMP3]]
+// CHECK3-NEXT: [[TMP79:%.*]] = mul nuw i32 [[TMP78]], 8
+// CHECK3-NEXT: [[TMP80:%.*]] = sext i32 [[TMP79]] to i64
// CHECK3-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES]], ptr align 4 @.offload_sizes.8, i32 80, i1 false)
-// CHECK3-NEXT: [[TMP73:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP64]], ptr [[TMP73]], align 4
-// CHECK3-NEXT: [[TMP74:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP64]], ptr [[TMP74]], align 4
-// CHECK3-NEXT: [[TMP75:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP75]], align 4
-// CHECK3-NEXT: [[TMP76:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 1
-// CHECK3-NEXT: store ptr [[B]], ptr [[TMP76]], align 4
-// CHECK3-NEXT: [[TMP77:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 1
-// CHECK3-NEXT: store ptr [[B]], ptr [[TMP77]], align 4
-// CHECK3-NEXT: [[TMP78:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 1
-// CHECK3-NEXT: store ptr null, ptr [[TMP78]], align 4
-// CHECK3-NEXT: [[TMP79:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 2
-// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP79]], align 4
-// CHECK3-NEXT: [[TMP80:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 2
-// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP80]], align 4
-// CHECK3-NEXT: [[TMP81:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 2
-// CHECK3-NEXT: store ptr null, ptr [[TMP81]], align 4
-// CHECK3-NEXT: [[TMP82:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[VLA]], ptr [[TMP82]], align 4
-// CHECK3-NEXT: [[TMP83:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[VLA]], ptr [[TMP83]], align 4
-// CHECK3-NEXT: [[TMP84:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
-// CHECK3-NEXT: store i64 [[TMP69]], ptr [[TMP84]], align 4
-// CHECK3-NEXT: [[TMP85:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 3
-// CHECK3-NEXT: store ptr null, ptr [[TMP85]], align 4
-// CHECK3-NEXT: [[TMP86:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 4
-// CHECK3-NEXT: store ptr [[C]], ptr [[TMP86]], align 4
-// CHECK3-NEXT: [[TMP87:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 4
-// CHECK3-NEXT: store ptr [[C]], ptr [[TMP87]], align 4
-// CHECK3-NEXT: [[TMP88:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 4
-// CHECK3-NEXT: store ptr null, ptr [[TMP88]], align 4
-// CHECK3-NEXT: [[TMP89:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 5
-// CHECK3-NEXT: store i32 5, ptr [[TMP89]], align 4
-// CHECK3-NEXT: [[TMP90:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 5
-// CHECK3-NEXT: store i32 5, ptr [[TMP90]], align 4
-// CHECK3-NEXT: [[TMP91:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 5
-// CHECK3-NEXT: store ptr null, ptr [[TMP91]], align 4
-// CHECK3-NEXT: [[TMP92:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 6
-// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP92]], align 4
-// CHECK3-NEXT: [[TMP93:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 6
-// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP93]], align 4
-// CHECK3-NEXT: [[TMP94:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP94]], align 4
-// CHECK3-NEXT: [[TMP95:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 7
-// CHECK3-NEXT: store ptr [[VLA1]], ptr [[TMP95]], align 4
-// CHECK3-NEXT: [[TMP96:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 7
-// CHECK3-NEXT: store ptr [[VLA1]], ptr [[TMP96]], align 4
-// CHECK3-NEXT: [[TMP97:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
-// CHECK3-NEXT: store i64 [[TMP72]], ptr [[TMP97]], align 4
-// CHECK3-NEXT: [[TMP98:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP98]], align 4
-// CHECK3-NEXT: [[TMP99:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 8
-// CHECK3-NEXT: store ptr [[D]], ptr [[TMP99]], align 4
-// CHECK3-NEXT: [[TMP100:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 8
-// CHECK3-NEXT: store ptr [[D]], ptr [[TMP100]], align 4
-// CHECK3-NEXT: [[TMP101:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 8
-// CHECK3-NEXT: store ptr null, ptr [[TMP101]], align 4
-// CHECK3-NEXT: [[TMP102:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 9
-// CHECK3-NEXT: store i32 [[TMP66]], ptr [[TMP102]], align 4
-// CHECK3-NEXT: [[TMP103:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 9
-// CHECK3-NEXT: store i32 [[TMP66]], ptr [[TMP103]], align 4
-// CHECK3-NEXT: [[TMP104:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 9
-// CHECK3-NEXT: store ptr null, ptr [[TMP104]], align 4
-// CHECK3-NEXT: [[TMP105:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP106:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP107:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP81:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP72]], ptr [[TMP81]], align 4
+// CHECK3-NEXT: [[TMP82:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP72]], ptr [[TMP82]], align 4
+// CHECK3-NEXT: [[TMP83:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP83]], align 4
+// CHECK3-NEXT: [[TMP84:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 1
+// CHECK3-NEXT: store ptr [[B]], ptr [[TMP84]], align 4
+// CHECK3-NEXT: [[TMP85:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 1
+// CHECK3-NEXT: store ptr [[B]], ptr [[TMP85]], align 4
+// CHECK3-NEXT: [[TMP86:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 1
+// CHECK3-NEXT: store ptr null, ptr [[TMP86]], align 4
+// CHECK3-NEXT: [[TMP87:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 2
+// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP87]], align 4
+// CHECK3-NEXT: [[TMP88:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 2
+// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP88]], align 4
+// CHECK3-NEXT: [[TMP89:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 2
+// CHECK3-NEXT: store ptr null, ptr [[TMP89]], align 4
+// CHECK3-NEXT: [[TMP90:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[VLA]], ptr [[TMP90]], align 4
+// CHECK3-NEXT: [[TMP91:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[VLA]], ptr [[TMP91]], align 4
+// CHECK3-NEXT: [[TMP92:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
+// CHECK3-NEXT: store i64 [[TMP77]], ptr [[TMP92]], align 4
+// CHECK3-NEXT: [[TMP93:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 3
+// CHECK3-NEXT: store ptr null, ptr [[TMP93]], align 4
+// CHECK3-NEXT: [[TMP94:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 4
+// CHECK3-NEXT: store ptr [[C]], ptr [[TMP94]], align 4
+// CHECK3-NEXT: [[TMP95:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 4
+// CHECK3-NEXT: store ptr [[C]], ptr [[TMP95]], align 4
+// CHECK3-NEXT: [[TMP96:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 4
+// CHECK3-NEXT: store ptr null, ptr [[TMP96]], align 4
+// CHECK3-NEXT: [[TMP97:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 5
+// CHECK3-NEXT: store i32 5, ptr [[TMP97]], align 4
+// CHECK3-NEXT: [[TMP98:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 5
+// CHECK3-NEXT: store i32 5, ptr [[TMP98]], align 4
+// CHECK3-NEXT: [[TMP99:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 5
+// CHECK3-NEXT: store ptr null, ptr [[TMP99]], align 4
+// CHECK3-NEXT: [[TMP100:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 6
+// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP100]], align 4
+// CHECK3-NEXT: [[TMP101:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 6
+// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP101]], align 4
+// CHECK3-NEXT: [[TMP102:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP102]], align 4
+// CHECK3-NEXT: [[TMP103:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 7
+// CHECK3-NEXT: store ptr [[VLA1]], ptr [[TMP103]], align 4
+// CHECK3-NEXT: [[TMP104:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 7
+// CHECK3-NEXT: store ptr [[VLA1]], ptr [[TMP104]], align 4
+// CHECK3-NEXT: [[TMP105:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
+// CHECK3-NEXT: store i64 [[TMP80]], ptr [[TMP105]], align 4
+// CHECK3-NEXT: [[TMP106:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP106]], align 4
+// CHECK3-NEXT: [[TMP107:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 8
+// CHECK3-NEXT: store ptr [[D]], ptr [[TMP107]], align 4
+// CHECK3-NEXT: [[TMP108:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 8
+// CHECK3-NEXT: store ptr [[D]], ptr [[TMP108]], align 4
+// CHECK3-NEXT: [[TMP109:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 8
+// CHECK3-NEXT: store ptr null, ptr [[TMP109]], align 4
+// CHECK3-NEXT: [[TMP110:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 9
+// CHECK3-NEXT: store i32 [[TMP74]], ptr [[TMP110]], align 4
+// CHECK3-NEXT: [[TMP111:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 9
+// CHECK3-NEXT: store i32 [[TMP74]], ptr [[TMP111]], align 4
+// CHECK3-NEXT: [[TMP112:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 9
+// CHECK3-NEXT: store ptr null, ptr [[TMP112]], align 4
+// CHECK3-NEXT: [[TMP113:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP114:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP115:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS17:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP108]], align 4
-// CHECK3-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 1
-// CHECK3-NEXT: store i32 10, ptr [[TMP109]], align 4
-// CHECK3-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP105]], ptr [[TMP110]], align 4
-// CHECK3-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP106]], ptr [[TMP111]], align 4
-// CHECK3-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 4
-// CHECK3-NEXT: store ptr [[TMP107]], ptr [[TMP112]], align 4
-// CHECK3-NEXT: [[TMP113:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP113]], align 4
-// CHECK3-NEXT: [[TMP114:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP114]], align 4
-// CHECK3-NEXT: [[TMP115:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP115]], align 4
-// CHECK3-NEXT: [[TMP116:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP116]], align 8
-// CHECK3-NEXT: [[TMP117:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l140.region_id, ptr [[KERNEL_ARGS17]])
-// CHECK3-NEXT: [[TMP118:%.*]] = icmp ne i32 [[TMP117]], 0
-// CHECK3-NEXT: br i1 [[TMP118]], label [[OMP_OFFLOAD_FAILED18:%.*]], label [[OMP_OFFLOAD_CONT19:%.*]]
+// CHECK3-NEXT: [[TMP116:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP116]], align 4
+// CHECK3-NEXT: [[TMP117:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 1
+// CHECK3-NEXT: store i32 10, ptr [[TMP117]], align 4
+// CHECK3-NEXT: [[TMP118:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP113]], ptr [[TMP118]], align 4
+// CHECK3-NEXT: [[TMP119:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP114]], ptr [[TMP119]], align 4
+// CHECK3-NEXT: [[TMP120:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 4
+// CHECK3-NEXT: store ptr [[TMP115]], ptr [[TMP120]], align 4
+// CHECK3-NEXT: [[TMP121:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP121]], align 4
+// CHECK3-NEXT: [[TMP122:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP122]], align 4
+// CHECK3-NEXT: [[TMP123:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP123]], align 4
+// CHECK3-NEXT: [[TMP124:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP124]], align 8
+// CHECK3-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP125]], align 8
+// CHECK3-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP126]], align 4
+// CHECK3-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP127]], align 4
+// CHECK3-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP128]], align 4
+// CHECK3-NEXT: [[TMP129:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l140.region_id, ptr [[KERNEL_ARGS17]])
+// CHECK3-NEXT: [[TMP130:%.*]] = icmp ne i32 [[TMP129]], 0
+// CHECK3-NEXT: br i1 [[TMP130]], label [[OMP_OFFLOAD_FAILED18:%.*]], label [[OMP_OFFLOAD_CONT19:%.*]]
// CHECK3: omp_offload.failed18:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l140(i32 [[TMP64]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]], i32 [[TMP66]]) #[[ATTR4]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l140(i32 [[TMP72]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]], i32 [[TMP74]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT19]]
// CHECK3: omp_offload.cont19:
// CHECK3-NEXT: br label [[OMP_IF_END21:%.*]]
// CHECK3: omp_if.else20:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l140(i32 [[TMP64]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]], i32 [[TMP66]]) #[[ATTR4]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l140(i32 [[TMP72]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]], i32 [[TMP74]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_IF_END21]]
// CHECK3: omp_if.end21:
-// CHECK3-NEXT: [[TMP119:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: [[TMP120:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK3-NEXT: call void @llvm.stackrestore(ptr [[TMP120]])
-// CHECK3-NEXT: ret i32 [[TMP119]]
+// CHECK3-NEXT: [[TMP131:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: [[TMP132:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK3-NEXT: call void @llvm.stackrestore(ptr [[TMP132]])
+// CHECK3-NEXT: ret i32 [[TMP131]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l96
@@ -2268,7 +2348,7 @@ int bar(int n){
// CHECK3-NEXT: store ptr [[TMP3]], ptr [[DOTTASK_T__ADDR_I]], align 4, !noalias !26
// CHECK3-NEXT: store ptr [[TMP7]], ptr [[__CONTEXT_ADDR_I]], align 4, !noalias !26
// CHECK3-NEXT: [[TMP8:%.*]] = load ptr, ptr [[__CONTEXT_ADDR_I]], align 4, !noalias !26
-// CHECK3-NEXT: store i32 1, ptr [[KERNEL_ARGS_I]], align 4, !noalias !26
+// CHECK3-NEXT: store i32 2, ptr [[KERNEL_ARGS_I]], align 4, !noalias !26
// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP9]], align 4, !noalias !26
// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 2
@@ -2285,9 +2365,17 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP15]], align 4, !noalias !26
// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP16]], align 8, !noalias !26
-// CHECK3-NEXT: [[TMP17:%.*]] = call i32 @__tgt_target_kernel_nowait(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l96.region_id, ptr [[KERNEL_ARGS_I]], i32 0, ptr null, i32 0, ptr null)
-// CHECK3-NEXT: [[TMP18:%.*]] = icmp ne i32 [[TMP17]], 0
-// CHECK3-NEXT: br i1 [[TMP18]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__1_EXIT:%.*]]
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 9
+// CHECK3-NEXT: store i64 1, ptr [[TMP17]], align 8, !noalias !26
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP18]], align 4, !noalias !26
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP19]], align 4, !noalias !26
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP20]], align 4, !noalias !26
+// CHECK3-NEXT: [[TMP21:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l96.region_id, ptr [[KERNEL_ARGS_I]])
+// CHECK3-NEXT: [[TMP22:%.*]] = icmp ne i32 [[TMP21]], 0
+// CHECK3-NEXT: br i1 [[TMP22]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__1_EXIT:%.*]]
// CHECK3: omp_offload.failed.i:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l96() #[[ATTR4]]
// CHECK3-NEXT: br label [[DOTOMP_OUTLINED__1_EXIT]]
@@ -2940,7 +3028,7 @@ int bar(int n){
// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [5 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP29]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP29]], align 4
// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 5, ptr [[TMP30]], align 4
// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2957,9 +3045,17 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP36]], align 4
// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP37]], align 8
-// CHECK3-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK3-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP38]], align 8
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP39]], align 4
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
+// CHECK3-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216(ptr [[THIS1]], i32 [[TMP5]], i32 2, i32 [[TMP1]], ptr [[VLA]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -2969,15 +3065,15 @@ int bar(int n){
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216(ptr [[THIS1]], i32 [[TMP5]], i32 2, i32 [[TMP1]], ptr [[VLA]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
-// CHECK3-NEXT: [[TMP40:%.*]] = mul nsw i32 1, [[TMP1]]
-// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i32 [[TMP40]]
+// CHECK3-NEXT: [[TMP44:%.*]] = mul nsw i32 1, [[TMP1]]
+// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i32 [[TMP44]]
// CHECK3-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i16, ptr [[ARRAYIDX]], i32 1
-// CHECK3-NEXT: [[TMP41:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
-// CHECK3-NEXT: [[CONV:%.*]] = sext i16 [[TMP41]] to i32
-// CHECK3-NEXT: [[TMP42:%.*]] = load i32, ptr [[B]], align 4
-// CHECK3-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP42]]
-// CHECK3-NEXT: [[TMP43:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK3-NEXT: call void @llvm.stackrestore(ptr [[TMP43]])
+// CHECK3-NEXT: [[TMP45:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
+// CHECK3-NEXT: [[CONV:%.*]] = sext i16 [[TMP45]] to i32
+// CHECK3-NEXT: [[TMP46:%.*]] = load i32, ptr [[B]], align 4
+// CHECK3-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP46]]
+// CHECK3-NEXT: [[TMP47:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK3-NEXT: call void @llvm.stackrestore(ptr [[TMP47]])
// CHECK3-NEXT: ret i32 [[ADD3]]
//
//
@@ -3040,7 +3136,7 @@ int bar(int n){
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 4, ptr [[TMP22]], align 4
// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3057,9 +3153,17 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4
// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP29]], align 8
-// CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l195.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK3-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK3-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l195.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK3-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l195(i32 [[TMP1]], i32 [[TMP3]], i32 [[TMP5]], ptr [[B]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -3069,8 +3173,8 @@ int bar(int n){
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l195(i32 [[TMP1]], i32 [[TMP3]], i32 [[TMP5]], ptr [[B]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
-// CHECK3-NEXT: [[TMP32:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: ret i32 [[TMP32]]
+// CHECK3-NEXT: [[TMP36:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: ret i32 [[TMP36]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_Z9ftemplateIiET_i
@@ -3120,7 +3224,7 @@ int bar(int n){
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP16]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP16]], align 4
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 3, ptr [[TMP17]], align 4
// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3137,9 +3241,17 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP23]], align 4
// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP24]], align 8
-// CHECK3-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l178.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK3-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP25]], align 8
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l178.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK3-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l178(i32 [[TMP1]], i32 [[TMP3]], ptr [[B]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -3149,8 +3261,8 @@ int bar(int n){
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l178(i32 [[TMP1]], i32 [[TMP3]], ptr [[B]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
-// CHECK3-NEXT: [[TMP27:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: ret i32 [[TMP27]]
+// CHECK3-NEXT: [[TMP31:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: ret i32 [[TMP31]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216
@@ -3537,7 +3649,7 @@ int bar(int n){
// CHECK5-NEXT: [[TMP30:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP31]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP31]], align 4
// CHECK5-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 3, ptr [[TMP32]], align 4
// CHECK5-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3554,186 +3666,210 @@ int bar(int n){
// CHECK5-NEXT: store ptr null, ptr [[TMP38]], align 8
// CHECK5-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 0, ptr [[TMP39]], align 8
-// CHECK5-NEXT: [[TMP40:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l108.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP41:%.*]] = icmp ne i32 [[TMP40]], 0
-// CHECK5-NEXT: br i1 [[TMP41]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP40]], align 8
+// CHECK5-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP41]], align 4
+// CHECK5-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP42]], align 4
+// CHECK5-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP43]], align 4
+// CHECK5-NEXT: [[TMP44:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l108.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP45:%.*]] = icmp ne i32 [[TMP44]], 0
+// CHECK5-NEXT: br i1 [[TMP45]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l108(i64 [[TMP15]], i64 [[TMP17]], i64 [[TMP19]]) #[[ATTR4]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK5: omp_offload.cont:
-// CHECK5-NEXT: [[TMP42:%.*]] = load i32, ptr [[A]], align 4
-// CHECK5-NEXT: store i32 [[TMP42]], ptr [[A_CASTED3]], align 4
-// CHECK5-NEXT: [[TMP43:%.*]] = load i64, ptr [[A_CASTED3]], align 8
-// CHECK5-NEXT: [[TMP44:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK5-NEXT: store i16 [[TMP44]], ptr [[AA_CASTED4]], align 2
-// CHECK5-NEXT: [[TMP45:%.*]] = load i64, ptr [[AA_CASTED4]], align 8
-// CHECK5-NEXT: [[TMP46:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK5-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP46]], 10
+// CHECK5-NEXT: [[TMP46:%.*]] = load i32, ptr [[A]], align 4
+// CHECK5-NEXT: store i32 [[TMP46]], ptr [[A_CASTED3]], align 4
+// CHECK5-NEXT: [[TMP47:%.*]] = load i64, ptr [[A_CASTED3]], align 8
+// CHECK5-NEXT: [[TMP48:%.*]] = load i16, ptr [[AA]], align 2
+// CHECK5-NEXT: store i16 [[TMP48]], ptr [[AA_CASTED4]], align 2
+// CHECK5-NEXT: [[TMP49:%.*]] = load i64, ptr [[AA_CASTED4]], align 8
+// CHECK5-NEXT: [[TMP50:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK5-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP50]], 10
// CHECK5-NEXT: br i1 [[CMP]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK5: omp_if.then:
-// CHECK5-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK5-NEXT: store i64 [[TMP43]], ptr [[TMP47]], align 8
-// CHECK5-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK5-NEXT: store i64 [[TMP43]], ptr [[TMP48]], align 8
-// CHECK5-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 0
-// CHECK5-NEXT: store ptr null, ptr [[TMP49]], align 8
-// CHECK5-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
-// CHECK5-NEXT: store i64 [[TMP45]], ptr [[TMP50]], align 8
-// CHECK5-NEXT: [[TMP51:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
-// CHECK5-NEXT: store i64 [[TMP45]], ptr [[TMP51]], align 8
-// CHECK5-NEXT: [[TMP52:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 1
-// CHECK5-NEXT: store ptr null, ptr [[TMP52]], align 8
-// CHECK5-NEXT: [[TMP53:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP54:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP51:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK5-NEXT: store i64 [[TMP47]], ptr [[TMP51]], align 8
+// CHECK5-NEXT: [[TMP52:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK5-NEXT: store i64 [[TMP47]], ptr [[TMP52]], align 8
+// CHECK5-NEXT: [[TMP53:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 0
+// CHECK5-NEXT: store ptr null, ptr [[TMP53]], align 8
+// CHECK5-NEXT: [[TMP54:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
+// CHECK5-NEXT: store i64 [[TMP49]], ptr [[TMP54]], align 8
+// CHECK5-NEXT: [[TMP55:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
+// CHECK5-NEXT: store i64 [[TMP49]], ptr [[TMP55]], align 8
+// CHECK5-NEXT: [[TMP56:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 1
+// CHECK5-NEXT: store ptr null, ptr [[TMP56]], align 8
+// CHECK5-NEXT: [[TMP57:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP58:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS8:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP55]], align 4
-// CHECK5-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 1
-// CHECK5-NEXT: store i32 2, ptr [[TMP56]], align 4
-// CHECK5-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 2
-// CHECK5-NEXT: store ptr [[TMP53]], ptr [[TMP57]], align 8
-// CHECK5-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[TMP54]], ptr [[TMP58]], align 8
-// CHECK5-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 4
-// CHECK5-NEXT: store ptr @.offload_sizes.5, ptr [[TMP59]], align 8
-// CHECK5-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 5
-// CHECK5-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP60]], align 8
-// CHECK5-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP61]], align 8
-// CHECK5-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 7
-// CHECK5-NEXT: store ptr null, ptr [[TMP62]], align 8
-// CHECK5-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 8
-// CHECK5-NEXT: store i64 0, ptr [[TMP63]], align 8
-// CHECK5-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l116.region_id, ptr [[KERNEL_ARGS8]])
-// CHECK5-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
-// CHECK5-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED9:%.*]], label [[OMP_OFFLOAD_CONT10:%.*]]
+// CHECK5-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP59]], align 4
+// CHECK5-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 1
+// CHECK5-NEXT: store i32 2, ptr [[TMP60]], align 4
+// CHECK5-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 2
+// CHECK5-NEXT: store ptr [[TMP57]], ptr [[TMP61]], align 8
+// CHECK5-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[TMP58]], ptr [[TMP62]], align 8
+// CHECK5-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 4
+// CHECK5-NEXT: store ptr @.offload_sizes.5, ptr [[TMP63]], align 8
+// CHECK5-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 5
+// CHECK5-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP64]], align 8
+// CHECK5-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP65]], align 8
+// CHECK5-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP66]], align 8
+// CHECK5-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 8
+// CHECK5-NEXT: store i64 0, ptr [[TMP67]], align 8
+// CHECK5-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP68]], align 8
+// CHECK5-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP69]], align 4
+// CHECK5-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP70]], align 4
+// CHECK5-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP71]], align 4
+// CHECK5-NEXT: [[TMP72:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l116.region_id, ptr [[KERNEL_ARGS8]])
+// CHECK5-NEXT: [[TMP73:%.*]] = icmp ne i32 [[TMP72]], 0
+// CHECK5-NEXT: br i1 [[TMP73]], label [[OMP_OFFLOAD_FAILED9:%.*]], label [[OMP_OFFLOAD_CONT10:%.*]]
// CHECK5: omp_offload.failed9:
-// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l116(i64 [[TMP43]], i64 [[TMP45]]) #[[ATTR4]]
+// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l116(i64 [[TMP47]], i64 [[TMP49]]) #[[ATTR4]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT10]]
// CHECK5: omp_offload.cont10:
// CHECK5-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK5: omp_if.else:
-// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l116(i64 [[TMP43]], i64 [[TMP45]]) #[[ATTR4]]
+// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l116(i64 [[TMP47]], i64 [[TMP49]]) #[[ATTR4]]
// CHECK5-NEXT: br label [[OMP_IF_END]]
// CHECK5: omp_if.end:
-// CHECK5-NEXT: [[TMP66:%.*]] = load i32, ptr [[A]], align 4
-// CHECK5-NEXT: store i32 [[TMP66]], ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK5-NEXT: [[TMP67:%.*]] = load i32, ptr [[A]], align 4
-// CHECK5-NEXT: store i32 [[TMP67]], ptr [[A_CASTED11]], align 4
-// CHECK5-NEXT: [[TMP68:%.*]] = load i64, ptr [[A_CASTED11]], align 8
-// CHECK5-NEXT: [[TMP69:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK5-NEXT: store i32 [[TMP69]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK5-NEXT: [[TMP70:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK5-NEXT: [[TMP71:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK5-NEXT: [[CMP12:%.*]] = icmp sgt i32 [[TMP71]], 20
+// CHECK5-NEXT: [[TMP74:%.*]] = load i32, ptr [[A]], align 4
+// CHECK5-NEXT: store i32 [[TMP74]], ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK5-NEXT: [[TMP75:%.*]] = load i32, ptr [[A]], align 4
+// CHECK5-NEXT: store i32 [[TMP75]], ptr [[A_CASTED11]], align 4
+// CHECK5-NEXT: [[TMP76:%.*]] = load i64, ptr [[A_CASTED11]], align 8
+// CHECK5-NEXT: [[TMP77:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK5-NEXT: store i32 [[TMP77]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK5-NEXT: [[TMP78:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK5-NEXT: [[TMP79:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK5-NEXT: [[CMP12:%.*]] = icmp sgt i32 [[TMP79]], 20
// CHECK5-NEXT: br i1 [[CMP12]], label [[OMP_IF_THEN13:%.*]], label [[OMP_IF_ELSE20:%.*]]
// CHECK5: omp_if.then13:
-// CHECK5-NEXT: [[TMP72:%.*]] = mul nuw i64 [[TMP2]], 4
-// CHECK5-NEXT: [[TMP73:%.*]] = mul nuw i64 5, [[TMP5]]
-// CHECK5-NEXT: [[TMP74:%.*]] = mul nuw i64 [[TMP73]], 8
+// CHECK5-NEXT: [[TMP80:%.*]] = mul nuw i64 [[TMP2]], 4
+// CHECK5-NEXT: [[TMP81:%.*]] = mul nuw i64 5, [[TMP5]]
+// CHECK5-NEXT: [[TMP82:%.*]] = mul nuw i64 [[TMP81]], 8
// CHECK5-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES]], ptr align 8 @.offload_sizes.8, i64 80, i1 false)
-// CHECK5-NEXT: [[TMP75:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
-// CHECK5-NEXT: store i64 [[TMP68]], ptr [[TMP75]], align 8
-// CHECK5-NEXT: [[TMP76:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
-// CHECK5-NEXT: store i64 [[TMP68]], ptr [[TMP76]], align 8
-// CHECK5-NEXT: [[TMP77:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 0
-// CHECK5-NEXT: store ptr null, ptr [[TMP77]], align 8
-// CHECK5-NEXT: [[TMP78:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 1
-// CHECK5-NEXT: store ptr [[B]], ptr [[TMP78]], align 8
-// CHECK5-NEXT: [[TMP79:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 1
-// CHECK5-NEXT: store ptr [[B]], ptr [[TMP79]], align 8
-// CHECK5-NEXT: [[TMP80:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 1
-// CHECK5-NEXT: store ptr null, ptr [[TMP80]], align 8
-// CHECK5-NEXT: [[TMP81:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 2
-// CHECK5-NEXT: store i64 [[TMP2]], ptr [[TMP81]], align 8
-// CHECK5-NEXT: [[TMP82:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 2
-// CHECK5-NEXT: store i64 [[TMP2]], ptr [[TMP82]], align 8
-// CHECK5-NEXT: [[TMP83:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 2
-// CHECK5-NEXT: store ptr null, ptr [[TMP83]], align 8
-// CHECK5-NEXT: [[TMP84:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[VLA]], ptr [[TMP84]], align 8
-// CHECK5-NEXT: [[TMP85:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[VLA]], ptr [[TMP85]], align 8
-// CHECK5-NEXT: [[TMP86:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
-// CHECK5-NEXT: store i64 [[TMP72]], ptr [[TMP86]], align 8
-// CHECK5-NEXT: [[TMP87:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 3
-// CHECK5-NEXT: store ptr null, ptr [[TMP87]], align 8
-// CHECK5-NEXT: [[TMP88:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 4
-// CHECK5-NEXT: store ptr [[C]], ptr [[TMP88]], align 8
-// CHECK5-NEXT: [[TMP89:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 4
-// CHECK5-NEXT: store ptr [[C]], ptr [[TMP89]], align 8
-// CHECK5-NEXT: [[TMP90:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 4
-// CHECK5-NEXT: store ptr null, ptr [[TMP90]], align 8
-// CHECK5-NEXT: [[TMP91:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 5
-// CHECK5-NEXT: store i64 5, ptr [[TMP91]], align 8
-// CHECK5-NEXT: [[TMP92:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 5
-// CHECK5-NEXT: store i64 5, ptr [[TMP92]], align 8
-// CHECK5-NEXT: [[TMP93:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 5
-// CHECK5-NEXT: store ptr null, ptr [[TMP93]], align 8
-// CHECK5-NEXT: [[TMP94:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 6
-// CHECK5-NEXT: store i64 [[TMP5]], ptr [[TMP94]], align 8
-// CHECK5-NEXT: [[TMP95:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 6
-// CHECK5-NEXT: store i64 [[TMP5]], ptr [[TMP95]], align 8
-// CHECK5-NEXT: [[TMP96:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP96]], align 8
-// CHECK5-NEXT: [[TMP97:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 7
-// CHECK5-NEXT: store ptr [[VLA1]], ptr [[TMP97]], align 8
-// CHECK5-NEXT: [[TMP98:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 7
-// CHECK5-NEXT: store ptr [[VLA1]], ptr [[TMP98]], align 8
-// CHECK5-NEXT: [[TMP99:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
-// CHECK5-NEXT: store i64 [[TMP74]], ptr [[TMP99]], align 8
-// CHECK5-NEXT: [[TMP100:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 7
-// CHECK5-NEXT: store ptr null, ptr [[TMP100]], align 8
-// CHECK5-NEXT: [[TMP101:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 8
-// CHECK5-NEXT: store ptr [[D]], ptr [[TMP101]], align 8
-// CHECK5-NEXT: [[TMP102:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 8
-// CHECK5-NEXT: store ptr [[D]], ptr [[TMP102]], align 8
-// CHECK5-NEXT: [[TMP103:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 8
-// CHECK5-NEXT: store ptr null, ptr [[TMP103]], align 8
-// CHECK5-NEXT: [[TMP104:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 9
-// CHECK5-NEXT: store i64 [[TMP70]], ptr [[TMP104]], align 8
-// CHECK5-NEXT: [[TMP105:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 9
-// CHECK5-NEXT: store i64 [[TMP70]], ptr [[TMP105]], align 8
-// CHECK5-NEXT: [[TMP106:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 9
-// CHECK5-NEXT: store ptr null, ptr [[TMP106]], align 8
-// CHECK5-NEXT: [[TMP107:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP108:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP109:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP83:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
+// CHECK5-NEXT: store i64 [[TMP76]], ptr [[TMP83]], align 8
+// CHECK5-NEXT: [[TMP84:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
+// CHECK5-NEXT: store i64 [[TMP76]], ptr [[TMP84]], align 8
+// CHECK5-NEXT: [[TMP85:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 0
+// CHECK5-NEXT: store ptr null, ptr [[TMP85]], align 8
+// CHECK5-NEXT: [[TMP86:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 1
+// CHECK5-NEXT: store ptr [[B]], ptr [[TMP86]], align 8
+// CHECK5-NEXT: [[TMP87:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 1
+// CHECK5-NEXT: store ptr [[B]], ptr [[TMP87]], align 8
+// CHECK5-NEXT: [[TMP88:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 1
+// CHECK5-NEXT: store ptr null, ptr [[TMP88]], align 8
+// CHECK5-NEXT: [[TMP89:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 2
+// CHECK5-NEXT: store i64 [[TMP2]], ptr [[TMP89]], align 8
+// CHECK5-NEXT: [[TMP90:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 2
+// CHECK5-NEXT: store i64 [[TMP2]], ptr [[TMP90]], align 8
+// CHECK5-NEXT: [[TMP91:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 2
+// CHECK5-NEXT: store ptr null, ptr [[TMP91]], align 8
+// CHECK5-NEXT: [[TMP92:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[VLA]], ptr [[TMP92]], align 8
+// CHECK5-NEXT: [[TMP93:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[VLA]], ptr [[TMP93]], align 8
+// CHECK5-NEXT: [[TMP94:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
+// CHECK5-NEXT: store i64 [[TMP80]], ptr [[TMP94]], align 8
+// CHECK5-NEXT: [[TMP95:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 3
+// CHECK5-NEXT: store ptr null, ptr [[TMP95]], align 8
+// CHECK5-NEXT: [[TMP96:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 4
+// CHECK5-NEXT: store ptr [[C]], ptr [[TMP96]], align 8
+// CHECK5-NEXT: [[TMP97:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 4
+// CHECK5-NEXT: store ptr [[C]], ptr [[TMP97]], align 8
+// CHECK5-NEXT: [[TMP98:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 4
+// CHECK5-NEXT: store ptr null, ptr [[TMP98]], align 8
+// CHECK5-NEXT: [[TMP99:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 5
+// CHECK5-NEXT: store i64 5, ptr [[TMP99]], align 8
+// CHECK5-NEXT: [[TMP100:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 5
+// CHECK5-NEXT: store i64 5, ptr [[TMP100]], align 8
+// CHECK5-NEXT: [[TMP101:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 5
+// CHECK5-NEXT: store ptr null, ptr [[TMP101]], align 8
+// CHECK5-NEXT: [[TMP102:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 6
+// CHECK5-NEXT: store i64 [[TMP5]], ptr [[TMP102]], align 8
+// CHECK5-NEXT: [[TMP103:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 6
+// CHECK5-NEXT: store i64 [[TMP5]], ptr [[TMP103]], align 8
+// CHECK5-NEXT: [[TMP104:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP104]], align 8
+// CHECK5-NEXT: [[TMP105:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 7
+// CHECK5-NEXT: store ptr [[VLA1]], ptr [[TMP105]], align 8
+// CHECK5-NEXT: [[TMP106:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 7
+// CHECK5-NEXT: store ptr [[VLA1]], ptr [[TMP106]], align 8
+// CHECK5-NEXT: [[TMP107:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
+// CHECK5-NEXT: store i64 [[TMP82]], ptr [[TMP107]], align 8
+// CHECK5-NEXT: [[TMP108:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP108]], align 8
+// CHECK5-NEXT: [[TMP109:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 8
+// CHECK5-NEXT: store ptr [[D]], ptr [[TMP109]], align 8
+// CHECK5-NEXT: [[TMP110:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 8
+// CHECK5-NEXT: store ptr [[D]], ptr [[TMP110]], align 8
+// CHECK5-NEXT: [[TMP111:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 8
+// CHECK5-NEXT: store ptr null, ptr [[TMP111]], align 8
+// CHECK5-NEXT: [[TMP112:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 9
+// CHECK5-NEXT: store i64 [[TMP78]], ptr [[TMP112]], align 8
+// CHECK5-NEXT: [[TMP113:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 9
+// CHECK5-NEXT: store i64 [[TMP78]], ptr [[TMP113]], align 8
+// CHECK5-NEXT: [[TMP114:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 9
+// CHECK5-NEXT: store ptr null, ptr [[TMP114]], align 8
+// CHECK5-NEXT: [[TMP115:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP116:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP117:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS17:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP110]], align 4
-// CHECK5-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 1
-// CHECK5-NEXT: store i32 10, ptr [[TMP111]], align 4
-// CHECK5-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 2
-// CHECK5-NEXT: store ptr [[TMP107]], ptr [[TMP112]], align 8
-// CHECK5-NEXT: [[TMP113:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[TMP108]], ptr [[TMP113]], align 8
-// CHECK5-NEXT: [[TMP114:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 4
-// CHECK5-NEXT: store ptr [[TMP109]], ptr [[TMP114]], align 8
-// CHECK5-NEXT: [[TMP115:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 5
-// CHECK5-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP115]], align 8
-// CHECK5-NEXT: [[TMP116:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP116]], align 8
-// CHECK5-NEXT: [[TMP117:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 7
-// CHECK5-NEXT: store ptr null, ptr [[TMP117]], align 8
-// CHECK5-NEXT: [[TMP118:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 8
-// CHECK5-NEXT: store i64 0, ptr [[TMP118]], align 8
-// CHECK5-NEXT: [[TMP119:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l140.region_id, ptr [[KERNEL_ARGS17]])
-// CHECK5-NEXT: [[TMP120:%.*]] = icmp ne i32 [[TMP119]], 0
-// CHECK5-NEXT: br i1 [[TMP120]], label [[OMP_OFFLOAD_FAILED18:%.*]], label [[OMP_OFFLOAD_CONT19:%.*]]
+// CHECK5-NEXT: [[TMP118:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP118]], align 4
+// CHECK5-NEXT: [[TMP119:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 1
+// CHECK5-NEXT: store i32 10, ptr [[TMP119]], align 4
+// CHECK5-NEXT: [[TMP120:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 2
+// CHECK5-NEXT: store ptr [[TMP115]], ptr [[TMP120]], align 8
+// CHECK5-NEXT: [[TMP121:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[TMP116]], ptr [[TMP121]], align 8
+// CHECK5-NEXT: [[TMP122:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 4
+// CHECK5-NEXT: store ptr [[TMP117]], ptr [[TMP122]], align 8
+// CHECK5-NEXT: [[TMP123:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 5
+// CHECK5-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP123]], align 8
+// CHECK5-NEXT: [[TMP124:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP124]], align 8
+// CHECK5-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP125]], align 8
+// CHECK5-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 8
+// CHECK5-NEXT: store i64 0, ptr [[TMP126]], align 8
+// CHECK5-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP127]], align 8
+// CHECK5-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP128]], align 4
+// CHECK5-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP129]], align 4
+// CHECK5-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP130]], align 4
+// CHECK5-NEXT: [[TMP131:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l140.region_id, ptr [[KERNEL_ARGS17]])
+// CHECK5-NEXT: [[TMP132:%.*]] = icmp ne i32 [[TMP131]], 0
+// CHECK5-NEXT: br i1 [[TMP132]], label [[OMP_OFFLOAD_FAILED18:%.*]], label [[OMP_OFFLOAD_CONT19:%.*]]
// CHECK5: omp_offload.failed18:
-// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l140(i64 [[TMP68]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]], i64 [[TMP70]]) #[[ATTR4]]
+// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l140(i64 [[TMP76]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]], i64 [[TMP78]]) #[[ATTR4]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT19]]
// CHECK5: omp_offload.cont19:
// CHECK5-NEXT: br label [[OMP_IF_END21:%.*]]
// CHECK5: omp_if.else20:
-// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l140(i64 [[TMP68]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]], i64 [[TMP70]]) #[[ATTR4]]
+// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l140(i64 [[TMP76]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]], i64 [[TMP78]]) #[[ATTR4]]
// CHECK5-NEXT: br label [[OMP_IF_END21]]
// CHECK5: omp_if.end21:
-// CHECK5-NEXT: [[TMP121:%.*]] = load i32, ptr [[A]], align 4
-// CHECK5-NEXT: [[TMP122:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK5-NEXT: call void @llvm.stackrestore(ptr [[TMP122]])
-// CHECK5-NEXT: ret i32 [[TMP121]]
+// CHECK5-NEXT: [[TMP133:%.*]] = load i32, ptr [[A]], align 4
+// CHECK5-NEXT: [[TMP134:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK5-NEXT: call void @llvm.stackrestore(ptr [[TMP134]])
+// CHECK5-NEXT: ret i32 [[TMP133]]
//
//
// CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l96
@@ -3841,7 +3977,7 @@ int bar(int n){
// CHECK5-NEXT: store ptr [[TMP3]], ptr [[DOTTASK_T__ADDR_I]], align 8, !noalias !25
// CHECK5-NEXT: store ptr [[TMP7]], ptr [[__CONTEXT_ADDR_I]], align 8, !noalias !25
// CHECK5-NEXT: [[TMP8:%.*]] = load ptr, ptr [[__CONTEXT_ADDR_I]], align 8, !noalias !25
-// CHECK5-NEXT: store i32 1, ptr [[KERNEL_ARGS_I]], align 4, !noalias !25
+// CHECK5-NEXT: store i32 2, ptr [[KERNEL_ARGS_I]], align 4, !noalias !25
// CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 1
// CHECK5-NEXT: store i32 0, ptr [[TMP9]], align 4, !noalias !25
// CHECK5-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 2
@@ -3858,9 +3994,17 @@ int bar(int n){
// CHECK5-NEXT: store ptr null, ptr [[TMP15]], align 8, !noalias !25
// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 8
// CHECK5-NEXT: store i64 0, ptr [[TMP16]], align 8, !noalias !25
-// CHECK5-NEXT: [[TMP17:%.*]] = call i32 @__tgt_target_kernel_nowait(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l96.region_id, ptr [[KERNEL_ARGS_I]], i32 0, ptr null, i32 0, ptr null)
-// CHECK5-NEXT: [[TMP18:%.*]] = icmp ne i32 [[TMP17]], 0
-// CHECK5-NEXT: br i1 [[TMP18]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__1_EXIT:%.*]]
+// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 9
+// CHECK5-NEXT: store i64 1, ptr [[TMP17]], align 8, !noalias !25
+// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP18]], align 4, !noalias !25
+// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP19]], align 4, !noalias !25
+// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP20]], align 4, !noalias !25
+// CHECK5-NEXT: [[TMP21:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l96.region_id, ptr [[KERNEL_ARGS_I]])
+// CHECK5-NEXT: [[TMP22:%.*]] = icmp ne i32 [[TMP21]], 0
+// CHECK5-NEXT: br i1 [[TMP22]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__1_EXIT:%.*]]
// CHECK5: omp_offload.failed.i:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l96() #[[ATTR4]]
// CHECK5-NEXT: br label [[DOTOMP_OUTLINED__1_EXIT]]
@@ -4533,28 +4677,37 @@ int bar(int n){
// CHECK5-NEXT: [[TMP35:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
// CHECK5-NEXT: [[TOBOOL4:%.*]] = trunc i8 [[TMP35]] to i1
// CHECK5-NEXT: [[TMP36:%.*]] = select i1 [[TOBOOL4]], i32 0, i32 1
+// CHECK5-NEXT: [[TMP37:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP36]], 0
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK5-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK5-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK5-NEXT: store i32 6, ptr [[TMP38]], align 4
-// CHECK5-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK5-NEXT: store ptr [[TMP32]], ptr [[TMP39]], align 8
-// CHECK5-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[TMP33]], ptr [[TMP40]], align 8
-// CHECK5-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK5-NEXT: store ptr [[TMP34]], ptr [[TMP41]], align 8
-// CHECK5-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK5-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP42]], align 8
-// CHECK5-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP43]], align 8
-// CHECK5-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK5-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP38]], align 4
+// CHECK5-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK5-NEXT: store i32 6, ptr [[TMP39]], align 4
+// CHECK5-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK5-NEXT: store ptr [[TMP32]], ptr [[TMP40]], align 8
+// CHECK5-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[TMP33]], ptr [[TMP41]], align 8
+// CHECK5-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK5-NEXT: store ptr [[TMP34]], ptr [[TMP42]], align 8
+// CHECK5-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK5-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP43]], align 8
+// CHECK5-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK5-NEXT: store ptr null, ptr [[TMP44]], align 8
-// CHECK5-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK5-NEXT: store i64 0, ptr [[TMP45]], align 8
-// CHECK5-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 [[TMP36]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l214.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK5-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP45]], align 8
+// CHECK5-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK5-NEXT: store i64 0, ptr [[TMP46]], align 8
+// CHECK5-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP47]], align 8
+// CHECK5-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP48]], align 4
+// CHECK5-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [[TMP37]], ptr [[TMP49]], align 4
+// CHECK5-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP50]], align 4
+// CHECK5-NEXT: [[TMP51:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 [[TMP36]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l214.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP52:%.*]] = icmp ne i32 [[TMP51]], 0
+// CHECK5-NEXT: br i1 [[TMP52]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l214(ptr [[THIS1]], i64 [[TMP7]], i64 2, i64 [[TMP2]], ptr [[VLA]], i64 [[TMP9]]) #[[ATTR4]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -4564,15 +4717,15 @@ int bar(int n){
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l214(ptr [[THIS1]], i64 [[TMP7]], i64 2, i64 [[TMP2]], ptr [[VLA]], i64 [[TMP9]]) #[[ATTR4]]
// CHECK5-NEXT: br label [[OMP_IF_END]]
// CHECK5: omp_if.end:
-// CHECK5-NEXT: [[TMP48:%.*]] = mul nsw i64 1, [[TMP2]]
-// CHECK5-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i64 [[TMP48]]
+// CHECK5-NEXT: [[TMP53:%.*]] = mul nsw i64 1, [[TMP2]]
+// CHECK5-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i64 [[TMP53]]
// CHECK5-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds i16, ptr [[ARRAYIDX]], i64 1
-// CHECK5-NEXT: [[TMP49:%.*]] = load i16, ptr [[ARRAYIDX5]], align 2
-// CHECK5-NEXT: [[CONV:%.*]] = sext i16 [[TMP49]] to i32
-// CHECK5-NEXT: [[TMP50:%.*]] = load i32, ptr [[B]], align 4
-// CHECK5-NEXT: [[ADD6:%.*]] = add nsw i32 [[CONV]], [[TMP50]]
-// CHECK5-NEXT: [[TMP51:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK5-NEXT: call void @llvm.stackrestore(ptr [[TMP51]])
+// CHECK5-NEXT: [[TMP54:%.*]] = load i16, ptr [[ARRAYIDX5]], align 2
+// CHECK5-NEXT: [[CONV:%.*]] = sext i16 [[TMP54]] to i32
+// CHECK5-NEXT: [[TMP55:%.*]] = load i32, ptr [[B]], align 4
+// CHECK5-NEXT: [[ADD6:%.*]] = add nsw i32 [[CONV]], [[TMP55]]
+// CHECK5-NEXT: [[TMP56:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK5-NEXT: call void @llvm.stackrestore(ptr [[TMP56]])
// CHECK5-NEXT: ret i32 [[ADD6]]
//
//
@@ -4635,7 +4788,7 @@ int bar(int n){
// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 4, ptr [[TMP22]], align 4
// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -4652,9 +4805,17 @@ int bar(int n){
// CHECK5-NEXT: store ptr null, ptr [[TMP28]], align 8
// CHECK5-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 0, ptr [[TMP29]], align 8
-// CHECK5-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l195.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK5-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK5-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP31]], align 4
+// CHECK5-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK5-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK5-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l195.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK5-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l195(i64 [[TMP1]], i64 [[TMP3]], i64 [[TMP5]], ptr [[B]]) #[[ATTR4]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -4664,8 +4825,8 @@ int bar(int n){
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l195(i64 [[TMP1]], i64 [[TMP3]], i64 [[TMP5]], ptr [[B]]) #[[ATTR4]]
// CHECK5-NEXT: br label [[OMP_IF_END]]
// CHECK5: omp_if.end:
-// CHECK5-NEXT: [[TMP32:%.*]] = load i32, ptr [[A]], align 4
-// CHECK5-NEXT: ret i32 [[TMP32]]
+// CHECK5-NEXT: [[TMP36:%.*]] = load i32, ptr [[A]], align 4
+// CHECK5-NEXT: ret i32 [[TMP36]]
//
//
// CHECK5-LABEL: define {{[^@]+}}@_Z9ftemplateIiET_i
@@ -4715,7 +4876,7 @@ int bar(int n){
// CHECK5-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP16]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP16]], align 4
// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 3, ptr [[TMP17]], align 4
// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -4732,9 +4893,17 @@ int bar(int n){
// CHECK5-NEXT: store ptr null, ptr [[TMP23]], align 8
// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 0, ptr [[TMP24]], align 8
-// CHECK5-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l178.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK5-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP25]], align 8
+// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP27]], align 4
+// CHECK5-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP28]], align 4
+// CHECK5-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l178.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK5-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l178(i64 [[TMP1]], i64 [[TMP3]], ptr [[B]]) #[[ATTR4]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -4744,8 +4913,8 @@ int bar(int n){
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l178(i64 [[TMP1]], i64 [[TMP3]], ptr [[B]]) #[[ATTR4]]
// CHECK5-NEXT: br label [[OMP_IF_END]]
// CHECK5: omp_if.end:
-// CHECK5-NEXT: [[TMP27:%.*]] = load i32, ptr [[A]], align 4
-// CHECK5-NEXT: ret i32 [[TMP27]]
+// CHECK5-NEXT: [[TMP31:%.*]] = load i32, ptr [[A]], align 4
+// CHECK5-NEXT: ret i32 [[TMP31]]
//
//
// CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l214
@@ -5212,7 +5381,7 @@ int bar(int n){
// CHECK7-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK7-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP27]], align 4
+// CHECK7-NEXT: store i32 2, ptr [[TMP27]], align 4
// CHECK7-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK7-NEXT: store i32 3, ptr [[TMP28]], align 4
// CHECK7-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -5229,188 +5398,212 @@ int bar(int n){
// CHECK7-NEXT: store ptr null, ptr [[TMP34]], align 4
// CHECK7-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK7-NEXT: store i64 0, ptr [[TMP35]], align 8
-// CHECK7-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l108.region_id, ptr [[KERNEL_ARGS]])
-// CHECK7-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
-// CHECK7-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK7-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP36]], align 8
+// CHECK7-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP37]], align 4
+// CHECK7-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP38]], align 4
+// CHECK7-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP39]], align 4
+// CHECK7-NEXT: [[TMP40:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l108.region_id, ptr [[KERNEL_ARGS]])
+// CHECK7-NEXT: [[TMP41:%.*]] = icmp ne i32 [[TMP40]], 0
+// CHECK7-NEXT: br i1 [[TMP41]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK7: omp_offload.failed:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l108(i32 [[TMP11]], i32 [[TMP13]], i32 [[TMP15]]) #[[ATTR4]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK7: omp_offload.cont:
-// CHECK7-NEXT: [[TMP38:%.*]] = load i32, ptr [[A]], align 4
-// CHECK7-NEXT: store i32 [[TMP38]], ptr [[A_CASTED3]], align 4
-// CHECK7-NEXT: [[TMP39:%.*]] = load i32, ptr [[A_CASTED3]], align 4
-// CHECK7-NEXT: [[TMP40:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK7-NEXT: store i16 [[TMP40]], ptr [[AA_CASTED4]], align 2
-// CHECK7-NEXT: [[TMP41:%.*]] = load i32, ptr [[AA_CASTED4]], align 4
-// CHECK7-NEXT: [[TMP42:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK7-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP42]], 10
+// CHECK7-NEXT: [[TMP42:%.*]] = load i32, ptr [[A]], align 4
+// CHECK7-NEXT: store i32 [[TMP42]], ptr [[A_CASTED3]], align 4
+// CHECK7-NEXT: [[TMP43:%.*]] = load i32, ptr [[A_CASTED3]], align 4
+// CHECK7-NEXT: [[TMP44:%.*]] = load i16, ptr [[AA]], align 2
+// CHECK7-NEXT: store i16 [[TMP44]], ptr [[AA_CASTED4]], align 2
+// CHECK7-NEXT: [[TMP45:%.*]] = load i32, ptr [[AA_CASTED4]], align 4
+// CHECK7-NEXT: [[TMP46:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK7-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP46]], 10
// CHECK7-NEXT: br i1 [[CMP]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK7: omp_if.then:
-// CHECK7-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK7-NEXT: store i32 [[TMP39]], ptr [[TMP43]], align 4
-// CHECK7-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK7-NEXT: store i32 [[TMP39]], ptr [[TMP44]], align 4
-// CHECK7-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 0
-// CHECK7-NEXT: store ptr null, ptr [[TMP45]], align 4
-// CHECK7-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
-// CHECK7-NEXT: store i32 [[TMP41]], ptr [[TMP46]], align 4
-// CHECK7-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
-// CHECK7-NEXT: store i32 [[TMP41]], ptr [[TMP47]], align 4
-// CHECK7-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 1
-// CHECK7-NEXT: store ptr null, ptr [[TMP48]], align 4
-// CHECK7-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK7-NEXT: store i32 [[TMP43]], ptr [[TMP47]], align 4
+// CHECK7-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK7-NEXT: store i32 [[TMP43]], ptr [[TMP48]], align 4
+// CHECK7-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 0
+// CHECK7-NEXT: store ptr null, ptr [[TMP49]], align 4
+// CHECK7-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
+// CHECK7-NEXT: store i32 [[TMP45]], ptr [[TMP50]], align 4
+// CHECK7-NEXT: [[TMP51:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
+// CHECK7-NEXT: store i32 [[TMP45]], ptr [[TMP51]], align 4
+// CHECK7-NEXT: [[TMP52:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 1
+// CHECK7-NEXT: store ptr null, ptr [[TMP52]], align 4
+// CHECK7-NEXT: [[TMP53:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP54:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS8:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK7-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP51]], align 4
-// CHECK7-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 1
-// CHECK7-NEXT: store i32 2, ptr [[TMP52]], align 4
-// CHECK7-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 2
-// CHECK7-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 4
-// CHECK7-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 3
-// CHECK7-NEXT: store ptr [[TMP50]], ptr [[TMP54]], align 4
-// CHECK7-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 4
-// CHECK7-NEXT: store ptr @.offload_sizes.5, ptr [[TMP55]], align 4
-// CHECK7-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 5
-// CHECK7-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP56]], align 4
-// CHECK7-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 6
-// CHECK7-NEXT: store ptr null, ptr [[TMP57]], align 4
-// CHECK7-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 7
-// CHECK7-NEXT: store ptr null, ptr [[TMP58]], align 4
-// CHECK7-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 8
-// CHECK7-NEXT: store i64 0, ptr [[TMP59]], align 8
-// CHECK7-NEXT: [[TMP60:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l116.region_id, ptr [[KERNEL_ARGS8]])
-// CHECK7-NEXT: [[TMP61:%.*]] = icmp ne i32 [[TMP60]], 0
-// CHECK7-NEXT: br i1 [[TMP61]], label [[OMP_OFFLOAD_FAILED9:%.*]], label [[OMP_OFFLOAD_CONT10:%.*]]
+// CHECK7-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 0
+// CHECK7-NEXT: store i32 2, ptr [[TMP55]], align 4
+// CHECK7-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 1
+// CHECK7-NEXT: store i32 2, ptr [[TMP56]], align 4
+// CHECK7-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 2
+// CHECK7-NEXT: store ptr [[TMP53]], ptr [[TMP57]], align 4
+// CHECK7-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 3
+// CHECK7-NEXT: store ptr [[TMP54]], ptr [[TMP58]], align 4
+// CHECK7-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 4
+// CHECK7-NEXT: store ptr @.offload_sizes.5, ptr [[TMP59]], align 4
+// CHECK7-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 5
+// CHECK7-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP60]], align 4
+// CHECK7-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 6
+// CHECK7-NEXT: store ptr null, ptr [[TMP61]], align 4
+// CHECK7-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 7
+// CHECK7-NEXT: store ptr null, ptr [[TMP62]], align 4
+// CHECK7-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 8
+// CHECK7-NEXT: store i64 0, ptr [[TMP63]], align 8
+// CHECK7-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP64]], align 8
+// CHECK7-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP65]], align 4
+// CHECK7-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP66]], align 4
+// CHECK7-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP67]], align 4
+// CHECK7-NEXT: [[TMP68:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l116.region_id, ptr [[KERNEL_ARGS8]])
+// CHECK7-NEXT: [[TMP69:%.*]] = icmp ne i32 [[TMP68]], 0
+// CHECK7-NEXT: br i1 [[TMP69]], label [[OMP_OFFLOAD_FAILED9:%.*]], label [[OMP_OFFLOAD_CONT10:%.*]]
// CHECK7: omp_offload.failed9:
-// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l116(i32 [[TMP39]], i32 [[TMP41]]) #[[ATTR4]]
+// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l116(i32 [[TMP43]], i32 [[TMP45]]) #[[ATTR4]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT10]]
// CHECK7: omp_offload.cont10:
// CHECK7-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK7: omp_if.else:
-// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l116(i32 [[TMP39]], i32 [[TMP41]]) #[[ATTR4]]
+// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l116(i32 [[TMP43]], i32 [[TMP45]]) #[[ATTR4]]
// CHECK7-NEXT: br label [[OMP_IF_END]]
// CHECK7: omp_if.end:
-// CHECK7-NEXT: [[TMP62:%.*]] = load i32, ptr [[A]], align 4
-// CHECK7-NEXT: store i32 [[TMP62]], ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK7-NEXT: [[TMP63:%.*]] = load i32, ptr [[A]], align 4
-// CHECK7-NEXT: store i32 [[TMP63]], ptr [[A_CASTED11]], align 4
-// CHECK7-NEXT: [[TMP64:%.*]] = load i32, ptr [[A_CASTED11]], align 4
-// CHECK7-NEXT: [[TMP65:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK7-NEXT: store i32 [[TMP65]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK7-NEXT: [[TMP66:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK7-NEXT: [[TMP67:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK7-NEXT: [[CMP12:%.*]] = icmp sgt i32 [[TMP67]], 20
+// CHECK7-NEXT: [[TMP70:%.*]] = load i32, ptr [[A]], align 4
+// CHECK7-NEXT: store i32 [[TMP70]], ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK7-NEXT: [[TMP71:%.*]] = load i32, ptr [[A]], align 4
+// CHECK7-NEXT: store i32 [[TMP71]], ptr [[A_CASTED11]], align 4
+// CHECK7-NEXT: [[TMP72:%.*]] = load i32, ptr [[A_CASTED11]], align 4
+// CHECK7-NEXT: [[TMP73:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK7-NEXT: store i32 [[TMP73]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK7-NEXT: [[TMP74:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK7-NEXT: [[TMP75:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK7-NEXT: [[CMP12:%.*]] = icmp sgt i32 [[TMP75]], 20
// CHECK7-NEXT: br i1 [[CMP12]], label [[OMP_IF_THEN13:%.*]], label [[OMP_IF_ELSE20:%.*]]
// CHECK7: omp_if.then13:
-// CHECK7-NEXT: [[TMP68:%.*]] = mul nuw i32 [[TMP1]], 4
-// CHECK7-NEXT: [[TMP69:%.*]] = sext i32 [[TMP68]] to i64
-// CHECK7-NEXT: [[TMP70:%.*]] = mul nuw i32 5, [[TMP3]]
-// CHECK7-NEXT: [[TMP71:%.*]] = mul nuw i32 [[TMP70]], 8
-// CHECK7-NEXT: [[TMP72:%.*]] = sext i32 [[TMP71]] to i64
+// CHECK7-NEXT: [[TMP76:%.*]] = mul nuw i32 [[TMP1]], 4
+// CHECK7-NEXT: [[TMP77:%.*]] = sext i32 [[TMP76]] to i64
+// CHECK7-NEXT: [[TMP78:%.*]] = mul nuw i32 5, [[TMP3]]
+// CHECK7-NEXT: [[TMP79:%.*]] = mul nuw i32 [[TMP78]], 8
+// CHECK7-NEXT: [[TMP80:%.*]] = sext i32 [[TMP79]] to i64
// CHECK7-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES]], ptr align 4 @.offload_sizes.8, i32 80, i1 false)
-// CHECK7-NEXT: [[TMP73:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
-// CHECK7-NEXT: store i32 [[TMP64]], ptr [[TMP73]], align 4
-// CHECK7-NEXT: [[TMP74:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
-// CHECK7-NEXT: store i32 [[TMP64]], ptr [[TMP74]], align 4
-// CHECK7-NEXT: [[TMP75:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 0
-// CHECK7-NEXT: store ptr null, ptr [[TMP75]], align 4
-// CHECK7-NEXT: [[TMP76:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 1
-// CHECK7-NEXT: store ptr [[B]], ptr [[TMP76]], align 4
-// CHECK7-NEXT: [[TMP77:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 1
-// CHECK7-NEXT: store ptr [[B]], ptr [[TMP77]], align 4
-// CHECK7-NEXT: [[TMP78:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 1
-// CHECK7-NEXT: store ptr null, ptr [[TMP78]], align 4
-// CHECK7-NEXT: [[TMP79:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 2
-// CHECK7-NEXT: store i32 [[TMP1]], ptr [[TMP79]], align 4
-// CHECK7-NEXT: [[TMP80:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 2
-// CHECK7-NEXT: store i32 [[TMP1]], ptr [[TMP80]], align 4
-// CHECK7-NEXT: [[TMP81:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 2
-// CHECK7-NEXT: store ptr null, ptr [[TMP81]], align 4
-// CHECK7-NEXT: [[TMP82:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 3
-// CHECK7-NEXT: store ptr [[VLA]], ptr [[TMP82]], align 4
-// CHECK7-NEXT: [[TMP83:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 3
-// CHECK7-NEXT: store ptr [[VLA]], ptr [[TMP83]], align 4
-// CHECK7-NEXT: [[TMP84:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
-// CHECK7-NEXT: store i64 [[TMP69]], ptr [[TMP84]], align 4
-// CHECK7-NEXT: [[TMP85:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 3
-// CHECK7-NEXT: store ptr null, ptr [[TMP85]], align 4
-// CHECK7-NEXT: [[TMP86:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 4
-// CHECK7-NEXT: store ptr [[C]], ptr [[TMP86]], align 4
-// CHECK7-NEXT: [[TMP87:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 4
-// CHECK7-NEXT: store ptr [[C]], ptr [[TMP87]], align 4
-// CHECK7-NEXT: [[TMP88:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 4
-// CHECK7-NEXT: store ptr null, ptr [[TMP88]], align 4
-// CHECK7-NEXT: [[TMP89:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 5
-// CHECK7-NEXT: store i32 5, ptr [[TMP89]], align 4
-// CHECK7-NEXT: [[TMP90:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 5
-// CHECK7-NEXT: store i32 5, ptr [[TMP90]], align 4
-// CHECK7-NEXT: [[TMP91:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 5
-// CHECK7-NEXT: store ptr null, ptr [[TMP91]], align 4
-// CHECK7-NEXT: [[TMP92:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 6
-// CHECK7-NEXT: store i32 [[TMP3]], ptr [[TMP92]], align 4
-// CHECK7-NEXT: [[TMP93:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 6
-// CHECK7-NEXT: store i32 [[TMP3]], ptr [[TMP93]], align 4
-// CHECK7-NEXT: [[TMP94:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 6
-// CHECK7-NEXT: store ptr null, ptr [[TMP94]], align 4
-// CHECK7-NEXT: [[TMP95:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 7
-// CHECK7-NEXT: store ptr [[VLA1]], ptr [[TMP95]], align 4
-// CHECK7-NEXT: [[TMP96:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 7
-// CHECK7-NEXT: store ptr [[VLA1]], ptr [[TMP96]], align 4
-// CHECK7-NEXT: [[TMP97:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
-// CHECK7-NEXT: store i64 [[TMP72]], ptr [[TMP97]], align 4
-// CHECK7-NEXT: [[TMP98:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 7
-// CHECK7-NEXT: store ptr null, ptr [[TMP98]], align 4
-// CHECK7-NEXT: [[TMP99:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 8
-// CHECK7-NEXT: store ptr [[D]], ptr [[TMP99]], align 4
-// CHECK7-NEXT: [[TMP100:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 8
-// CHECK7-NEXT: store ptr [[D]], ptr [[TMP100]], align 4
-// CHECK7-NEXT: [[TMP101:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 8
-// CHECK7-NEXT: store ptr null, ptr [[TMP101]], align 4
-// CHECK7-NEXT: [[TMP102:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 9
-// CHECK7-NEXT: store i32 [[TMP66]], ptr [[TMP102]], align 4
-// CHECK7-NEXT: [[TMP103:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 9
-// CHECK7-NEXT: store i32 [[TMP66]], ptr [[TMP103]], align 4
-// CHECK7-NEXT: [[TMP104:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 9
-// CHECK7-NEXT: store ptr null, ptr [[TMP104]], align 4
-// CHECK7-NEXT: [[TMP105:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP106:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP107:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP81:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
+// CHECK7-NEXT: store i32 [[TMP72]], ptr [[TMP81]], align 4
+// CHECK7-NEXT: [[TMP82:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
+// CHECK7-NEXT: store i32 [[TMP72]], ptr [[TMP82]], align 4
+// CHECK7-NEXT: [[TMP83:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 0
+// CHECK7-NEXT: store ptr null, ptr [[TMP83]], align 4
+// CHECK7-NEXT: [[TMP84:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 1
+// CHECK7-NEXT: store ptr [[B]], ptr [[TMP84]], align 4
+// CHECK7-NEXT: [[TMP85:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 1
+// CHECK7-NEXT: store ptr [[B]], ptr [[TMP85]], align 4
+// CHECK7-NEXT: [[TMP86:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 1
+// CHECK7-NEXT: store ptr null, ptr [[TMP86]], align 4
+// CHECK7-NEXT: [[TMP87:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 2
+// CHECK7-NEXT: store i32 [[TMP1]], ptr [[TMP87]], align 4
+// CHECK7-NEXT: [[TMP88:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 2
+// CHECK7-NEXT: store i32 [[TMP1]], ptr [[TMP88]], align 4
+// CHECK7-NEXT: [[TMP89:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 2
+// CHECK7-NEXT: store ptr null, ptr [[TMP89]], align 4
+// CHECK7-NEXT: [[TMP90:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 3
+// CHECK7-NEXT: store ptr [[VLA]], ptr [[TMP90]], align 4
+// CHECK7-NEXT: [[TMP91:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 3
+// CHECK7-NEXT: store ptr [[VLA]], ptr [[TMP91]], align 4
+// CHECK7-NEXT: [[TMP92:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
+// CHECK7-NEXT: store i64 [[TMP77]], ptr [[TMP92]], align 4
+// CHECK7-NEXT: [[TMP93:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 3
+// CHECK7-NEXT: store ptr null, ptr [[TMP93]], align 4
+// CHECK7-NEXT: [[TMP94:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 4
+// CHECK7-NEXT: store ptr [[C]], ptr [[TMP94]], align 4
+// CHECK7-NEXT: [[TMP95:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 4
+// CHECK7-NEXT: store ptr [[C]], ptr [[TMP95]], align 4
+// CHECK7-NEXT: [[TMP96:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 4
+// CHECK7-NEXT: store ptr null, ptr [[TMP96]], align 4
+// CHECK7-NEXT: [[TMP97:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 5
+// CHECK7-NEXT: store i32 5, ptr [[TMP97]], align 4
+// CHECK7-NEXT: [[TMP98:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 5
+// CHECK7-NEXT: store i32 5, ptr [[TMP98]], align 4
+// CHECK7-NEXT: [[TMP99:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 5
+// CHECK7-NEXT: store ptr null, ptr [[TMP99]], align 4
+// CHECK7-NEXT: [[TMP100:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 6
+// CHECK7-NEXT: store i32 [[TMP3]], ptr [[TMP100]], align 4
+// CHECK7-NEXT: [[TMP101:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 6
+// CHECK7-NEXT: store i32 [[TMP3]], ptr [[TMP101]], align 4
+// CHECK7-NEXT: [[TMP102:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 6
+// CHECK7-NEXT: store ptr null, ptr [[TMP102]], align 4
+// CHECK7-NEXT: [[TMP103:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 7
+// CHECK7-NEXT: store ptr [[VLA1]], ptr [[TMP103]], align 4
+// CHECK7-NEXT: [[TMP104:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 7
+// CHECK7-NEXT: store ptr [[VLA1]], ptr [[TMP104]], align 4
+// CHECK7-NEXT: [[TMP105:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
+// CHECK7-NEXT: store i64 [[TMP80]], ptr [[TMP105]], align 4
+// CHECK7-NEXT: [[TMP106:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 7
+// CHECK7-NEXT: store ptr null, ptr [[TMP106]], align 4
+// CHECK7-NEXT: [[TMP107:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 8
+// CHECK7-NEXT: store ptr [[D]], ptr [[TMP107]], align 4
+// CHECK7-NEXT: [[TMP108:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 8
+// CHECK7-NEXT: store ptr [[D]], ptr [[TMP108]], align 4
+// CHECK7-NEXT: [[TMP109:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 8
+// CHECK7-NEXT: store ptr null, ptr [[TMP109]], align 4
+// CHECK7-NEXT: [[TMP110:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 9
+// CHECK7-NEXT: store i32 [[TMP74]], ptr [[TMP110]], align 4
+// CHECK7-NEXT: [[TMP111:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 9
+// CHECK7-NEXT: store i32 [[TMP74]], ptr [[TMP111]], align 4
+// CHECK7-NEXT: [[TMP112:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i32 0, i32 9
+// CHECK7-NEXT: store ptr null, ptr [[TMP112]], align 4
+// CHECK7-NEXT: [[TMP113:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP114:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP115:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS17:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK7-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP108]], align 4
-// CHECK7-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 1
-// CHECK7-NEXT: store i32 10, ptr [[TMP109]], align 4
-// CHECK7-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 2
-// CHECK7-NEXT: store ptr [[TMP105]], ptr [[TMP110]], align 4
-// CHECK7-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 3
-// CHECK7-NEXT: store ptr [[TMP106]], ptr [[TMP111]], align 4
-// CHECK7-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 4
-// CHECK7-NEXT: store ptr [[TMP107]], ptr [[TMP112]], align 4
-// CHECK7-NEXT: [[TMP113:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 5
-// CHECK7-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP113]], align 4
-// CHECK7-NEXT: [[TMP114:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 6
-// CHECK7-NEXT: store ptr null, ptr [[TMP114]], align 4
-// CHECK7-NEXT: [[TMP115:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 7
-// CHECK7-NEXT: store ptr null, ptr [[TMP115]], align 4
-// CHECK7-NEXT: [[TMP116:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 8
-// CHECK7-NEXT: store i64 0, ptr [[TMP116]], align 8
-// CHECK7-NEXT: [[TMP117:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l140.region_id, ptr [[KERNEL_ARGS17]])
-// CHECK7-NEXT: [[TMP118:%.*]] = icmp ne i32 [[TMP117]], 0
-// CHECK7-NEXT: br i1 [[TMP118]], label [[OMP_OFFLOAD_FAILED18:%.*]], label [[OMP_OFFLOAD_CONT19:%.*]]
+// CHECK7-NEXT: [[TMP116:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 0
+// CHECK7-NEXT: store i32 2, ptr [[TMP116]], align 4
+// CHECK7-NEXT: [[TMP117:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 1
+// CHECK7-NEXT: store i32 10, ptr [[TMP117]], align 4
+// CHECK7-NEXT: [[TMP118:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 2
+// CHECK7-NEXT: store ptr [[TMP113]], ptr [[TMP118]], align 4
+// CHECK7-NEXT: [[TMP119:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 3
+// CHECK7-NEXT: store ptr [[TMP114]], ptr [[TMP119]], align 4
+// CHECK7-NEXT: [[TMP120:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 4
+// CHECK7-NEXT: store ptr [[TMP115]], ptr [[TMP120]], align 4
+// CHECK7-NEXT: [[TMP121:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 5
+// CHECK7-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP121]], align 4
+// CHECK7-NEXT: [[TMP122:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 6
+// CHECK7-NEXT: store ptr null, ptr [[TMP122]], align 4
+// CHECK7-NEXT: [[TMP123:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 7
+// CHECK7-NEXT: store ptr null, ptr [[TMP123]], align 4
+// CHECK7-NEXT: [[TMP124:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 8
+// CHECK7-NEXT: store i64 0, ptr [[TMP124]], align 8
+// CHECK7-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP125]], align 8
+// CHECK7-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP126]], align 4
+// CHECK7-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP127]], align 4
+// CHECK7-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP128]], align 4
+// CHECK7-NEXT: [[TMP129:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l140.region_id, ptr [[KERNEL_ARGS17]])
+// CHECK7-NEXT: [[TMP130:%.*]] = icmp ne i32 [[TMP129]], 0
+// CHECK7-NEXT: br i1 [[TMP130]], label [[OMP_OFFLOAD_FAILED18:%.*]], label [[OMP_OFFLOAD_CONT19:%.*]]
// CHECK7: omp_offload.failed18:
-// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l140(i32 [[TMP64]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]], i32 [[TMP66]]) #[[ATTR4]]
+// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l140(i32 [[TMP72]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]], i32 [[TMP74]]) #[[ATTR4]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT19]]
// CHECK7: omp_offload.cont19:
// CHECK7-NEXT: br label [[OMP_IF_END21:%.*]]
// CHECK7: omp_if.else20:
-// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l140(i32 [[TMP64]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]], i32 [[TMP66]]) #[[ATTR4]]
+// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l140(i32 [[TMP72]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]], i32 [[TMP74]]) #[[ATTR4]]
// CHECK7-NEXT: br label [[OMP_IF_END21]]
// CHECK7: omp_if.end21:
-// CHECK7-NEXT: [[TMP119:%.*]] = load i32, ptr [[A]], align 4
-// CHECK7-NEXT: [[TMP120:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK7-NEXT: call void @llvm.stackrestore(ptr [[TMP120]])
-// CHECK7-NEXT: ret i32 [[TMP119]]
+// CHECK7-NEXT: [[TMP131:%.*]] = load i32, ptr [[A]], align 4
+// CHECK7-NEXT: [[TMP132:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK7-NEXT: call void @llvm.stackrestore(ptr [[TMP132]])
+// CHECK7-NEXT: ret i32 [[TMP131]]
//
//
// CHECK7-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l96
@@ -5518,7 +5711,7 @@ int bar(int n){
// CHECK7-NEXT: store ptr [[TMP3]], ptr [[DOTTASK_T__ADDR_I]], align 4, !noalias !26
// CHECK7-NEXT: store ptr [[TMP7]], ptr [[__CONTEXT_ADDR_I]], align 4, !noalias !26
// CHECK7-NEXT: [[TMP8:%.*]] = load ptr, ptr [[__CONTEXT_ADDR_I]], align 4, !noalias !26
-// CHECK7-NEXT: store i32 1, ptr [[KERNEL_ARGS_I]], align 4, !noalias !26
+// CHECK7-NEXT: store i32 2, ptr [[KERNEL_ARGS_I]], align 4, !noalias !26
// CHECK7-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 1
// CHECK7-NEXT: store i32 0, ptr [[TMP9]], align 4, !noalias !26
// CHECK7-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 2
@@ -5535,9 +5728,17 @@ int bar(int n){
// CHECK7-NEXT: store ptr null, ptr [[TMP15]], align 4, !noalias !26
// CHECK7-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 8
// CHECK7-NEXT: store i64 0, ptr [[TMP16]], align 8, !noalias !26
-// CHECK7-NEXT: [[TMP17:%.*]] = call i32 @__tgt_target_kernel_nowait(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l96.region_id, ptr [[KERNEL_ARGS_I]], i32 0, ptr null, i32 0, ptr null)
-// CHECK7-NEXT: [[TMP18:%.*]] = icmp ne i32 [[TMP17]], 0
-// CHECK7-NEXT: br i1 [[TMP18]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__1_EXIT:%.*]]
+// CHECK7-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 9
+// CHECK7-NEXT: store i64 1, ptr [[TMP17]], align 8, !noalias !26
+// CHECK7-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP18]], align 4, !noalias !26
+// CHECK7-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP19]], align 4, !noalias !26
+// CHECK7-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP20]], align 4, !noalias !26
+// CHECK7-NEXT: [[TMP21:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l96.region_id, ptr [[KERNEL_ARGS_I]])
+// CHECK7-NEXT: [[TMP22:%.*]] = icmp ne i32 [[TMP21]], 0
+// CHECK7-NEXT: br i1 [[TMP22]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__1_EXIT:%.*]]
// CHECK7: omp_offload.failed.i:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l96() #[[ATTR4]]
// CHECK7-NEXT: br label [[DOTOMP_OUTLINED__1_EXIT]]
@@ -6208,28 +6409,37 @@ int bar(int n){
// CHECK7-NEXT: [[TMP35:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
// CHECK7-NEXT: [[TOBOOL4:%.*]] = trunc i8 [[TMP35]] to i1
// CHECK7-NEXT: [[TMP36:%.*]] = select i1 [[TOBOOL4]], i32 0, i32 1
+// CHECK7-NEXT: [[TMP37:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP36]], 0
// CHECK7-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK7-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK7-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK7-NEXT: store i32 6, ptr [[TMP38]], align 4
-// CHECK7-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK7-NEXT: store ptr [[TMP32]], ptr [[TMP39]], align 4
-// CHECK7-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK7-NEXT: store ptr [[TMP33]], ptr [[TMP40]], align 4
-// CHECK7-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK7-NEXT: store ptr [[TMP34]], ptr [[TMP41]], align 4
-// CHECK7-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK7-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP42]], align 4
-// CHECK7-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK7-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK7-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK7-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK7-NEXT: store i32 2, ptr [[TMP38]], align 4
+// CHECK7-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK7-NEXT: store i32 6, ptr [[TMP39]], align 4
+// CHECK7-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK7-NEXT: store ptr [[TMP32]], ptr [[TMP40]], align 4
+// CHECK7-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK7-NEXT: store ptr [[TMP33]], ptr [[TMP41]], align 4
+// CHECK7-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK7-NEXT: store ptr [[TMP34]], ptr [[TMP42]], align 4
+// CHECK7-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK7-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP43]], align 4
+// CHECK7-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK7-NEXT: store ptr null, ptr [[TMP44]], align 4
-// CHECK7-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK7-NEXT: store i64 0, ptr [[TMP45]], align 8
-// CHECK7-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 [[TMP36]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l214.region_id, ptr [[KERNEL_ARGS]])
-// CHECK7-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK7-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK7-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK7-NEXT: store ptr null, ptr [[TMP45]], align 4
+// CHECK7-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK7-NEXT: store i64 0, ptr [[TMP46]], align 8
+// CHECK7-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP47]], align 8
+// CHECK7-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP48]], align 4
+// CHECK7-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] [[TMP37]], ptr [[TMP49]], align 4
+// CHECK7-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP50]], align 4
+// CHECK7-NEXT: [[TMP51:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 [[TMP36]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l214.region_id, ptr [[KERNEL_ARGS]])
+// CHECK7-NEXT: [[TMP52:%.*]] = icmp ne i32 [[TMP51]], 0
+// CHECK7-NEXT: br i1 [[TMP52]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK7: omp_offload.failed:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l214(ptr [[THIS1]], i32 [[TMP6]], i32 2, i32 [[TMP1]], ptr [[VLA]], i32 [[TMP8]]) #[[ATTR4]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -6239,15 +6449,15 @@ int bar(int n){
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l214(ptr [[THIS1]], i32 [[TMP6]], i32 2, i32 [[TMP1]], ptr [[VLA]], i32 [[TMP8]]) #[[ATTR4]]
// CHECK7-NEXT: br label [[OMP_IF_END]]
// CHECK7: omp_if.end:
-// CHECK7-NEXT: [[TMP48:%.*]] = mul nsw i32 1, [[TMP1]]
-// CHECK7-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i32 [[TMP48]]
+// CHECK7-NEXT: [[TMP53:%.*]] = mul nsw i32 1, [[TMP1]]
+// CHECK7-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i32 [[TMP53]]
// CHECK7-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds i16, ptr [[ARRAYIDX]], i32 1
-// CHECK7-NEXT: [[TMP49:%.*]] = load i16, ptr [[ARRAYIDX5]], align 2
-// CHECK7-NEXT: [[CONV:%.*]] = sext i16 [[TMP49]] to i32
-// CHECK7-NEXT: [[TMP50:%.*]] = load i32, ptr [[B]], align 4
-// CHECK7-NEXT: [[ADD6:%.*]] = add nsw i32 [[CONV]], [[TMP50]]
-// CHECK7-NEXT: [[TMP51:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK7-NEXT: call void @llvm.stackrestore(ptr [[TMP51]])
+// CHECK7-NEXT: [[TMP54:%.*]] = load i16, ptr [[ARRAYIDX5]], align 2
+// CHECK7-NEXT: [[CONV:%.*]] = sext i16 [[TMP54]] to i32
+// CHECK7-NEXT: [[TMP55:%.*]] = load i32, ptr [[B]], align 4
+// CHECK7-NEXT: [[ADD6:%.*]] = add nsw i32 [[CONV]], [[TMP55]]
+// CHECK7-NEXT: [[TMP56:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK7-NEXT: call void @llvm.stackrestore(ptr [[TMP56]])
// CHECK7-NEXT: ret i32 [[ADD6]]
//
//
@@ -6310,7 +6520,7 @@ int bar(int n){
// CHECK7-NEXT: [[TMP20:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK7-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK7-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK7-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK7-NEXT: store i32 4, ptr [[TMP22]], align 4
// CHECK7-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -6327,9 +6537,17 @@ int bar(int n){
// CHECK7-NEXT: store ptr null, ptr [[TMP28]], align 4
// CHECK7-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK7-NEXT: store i64 0, ptr [[TMP29]], align 8
-// CHECK7-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l195.region_id, ptr [[KERNEL_ARGS]])
-// CHECK7-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK7-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK7-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK7-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP31]], align 4
+// CHECK7-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK7-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK7-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l195.region_id, ptr [[KERNEL_ARGS]])
+// CHECK7-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK7-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK7: omp_offload.failed:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l195(i32 [[TMP1]], i32 [[TMP3]], i32 [[TMP5]], ptr [[B]]) #[[ATTR4]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -6339,8 +6557,8 @@ int bar(int n){
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l195(i32 [[TMP1]], i32 [[TMP3]], i32 [[TMP5]], ptr [[B]]) #[[ATTR4]]
// CHECK7-NEXT: br label [[OMP_IF_END]]
// CHECK7: omp_if.end:
-// CHECK7-NEXT: [[TMP32:%.*]] = load i32, ptr [[A]], align 4
-// CHECK7-NEXT: ret i32 [[TMP32]]
+// CHECK7-NEXT: [[TMP36:%.*]] = load i32, ptr [[A]], align 4
+// CHECK7-NEXT: ret i32 [[TMP36]]
//
//
// CHECK7-LABEL: define {{[^@]+}}@_Z9ftemplateIiET_i
@@ -6390,7 +6608,7 @@ int bar(int n){
// CHECK7-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK7-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP16]], align 4
+// CHECK7-NEXT: store i32 2, ptr [[TMP16]], align 4
// CHECK7-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK7-NEXT: store i32 3, ptr [[TMP17]], align 4
// CHECK7-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -6407,9 +6625,17 @@ int bar(int n){
// CHECK7-NEXT: store ptr null, ptr [[TMP23]], align 4
// CHECK7-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK7-NEXT: store i64 0, ptr [[TMP24]], align 8
-// CHECK7-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l178.region_id, ptr [[KERNEL_ARGS]])
-// CHECK7-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK7-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK7-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP25]], align 8
+// CHECK7-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK7-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP27]], align 4
+// CHECK7-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP28]], align 4
+// CHECK7-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l178.region_id, ptr [[KERNEL_ARGS]])
+// CHECK7-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK7-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK7: omp_offload.failed:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l178(i32 [[TMP1]], i32 [[TMP3]], ptr [[B]]) #[[ATTR4]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -6419,8 +6645,8 @@ int bar(int n){
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l178(i32 [[TMP1]], i32 [[TMP3]], ptr [[B]]) #[[ATTR4]]
// CHECK7-NEXT: br label [[OMP_IF_END]]
// CHECK7: omp_if.end:
-// CHECK7-NEXT: [[TMP27:%.*]] = load i32, ptr [[A]], align 4
-// CHECK7-NEXT: ret i32 [[TMP27]]
+// CHECK7-NEXT: [[TMP31:%.*]] = load i32, ptr [[A]], align 4
+// CHECK7-NEXT: ret i32 [[TMP31]]
//
//
// CHECK7-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l214
diff --git a/clang/test/OpenMP/target_parallel_for_simd_depend_codegen.cpp b/clang/test/OpenMP/target_parallel_for_simd_depend_codegen.cpp
index 4a13c711948fb..779e657f33621 100644
--- a/clang/test/OpenMP/target_parallel_for_simd_depend_codegen.cpp
+++ b/clang/test/OpenMP/target_parallel_for_simd_depend_codegen.cpp
@@ -200,7 +200,7 @@ int foo(int n) {
// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2
// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]],
// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64
-// CHECK: [[RET:%.+]] = call i32 @__tgt_target_kernel_nowait(%struct.ident_t* @{{.+}}, i64 [[DEVICE]], i32 1, i32 0, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* %{{.+}})
+// CHECK: [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 [[DEVICE]], i32 1, i32 0, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* %{{.+}})
// CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]]
diff --git a/clang/test/OpenMP/target_parallel_if_codegen.cpp b/clang/test/OpenMP/target_parallel_if_codegen.cpp
index 41b829db57c66..8796862623c4e 100644
--- a/clang/test/OpenMP/target_parallel_if_codegen.cpp
+++ b/clang/test/OpenMP/target_parallel_if_codegen.cpp
@@ -266,97 +266,115 @@ int bar(int n){
// CHECK1-NEXT: [[TMP16:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
// CHECK1-NEXT: [[TOBOOL3:%.*]] = trunc i8 [[TMP16]] to i1
// CHECK1-NEXT: [[TMP17:%.*]] = select i1 [[TOBOOL3]], i32 0, i32 1
+// CHECK1-NEXT: [[TMP18:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP17]], 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP18]], align 4
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK1-NEXT: store i32 3, ptr [[TMP19]], align 4
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP14]], ptr [[TMP20]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP15]], ptr [[TMP21]], align 8
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP22]], align 8
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP19]], align 4
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK1-NEXT: store i32 3, ptr [[TMP20]], align 4
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP14]], ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP15]], ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK1-NEXT: store ptr null, ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 1, i32 [[TMP17]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK1-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP26]], align 8
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP29]], align 4
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP18]], ptr [[TMP30]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP31]], align 4
+// CHECK1-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 1, i32 [[TMP17]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
+// CHECK1-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121(ptr [[THIS1]], i64 [[TMP2]], i64 [[TMP4]]) #[[ATTR3:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
-// CHECK1-NEXT: [[TMP29:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: [[CMP5:%.*]] = icmp sgt i32 [[TMP29]], 5
+// CHECK1-NEXT: [[TMP34:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: [[CMP5:%.*]] = icmp sgt i32 [[TMP34]], 5
// CHECK1-NEXT: [[FROMBOOL6:%.*]] = zext i1 [[CMP5]] to i8
// CHECK1-NEXT: store i8 [[FROMBOOL6]], ptr [[DOTCAPTURE_EXPR_4]], align 1
-// CHECK1-NEXT: [[TMP30:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_4]], align 1
-// CHECK1-NEXT: [[TOBOOL7:%.*]] = trunc i8 [[TMP30]] to i1
+// CHECK1-NEXT: [[TMP35:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_4]], align 1
+// CHECK1-NEXT: [[TOBOOL7:%.*]] = trunc i8 [[TMP35]] to i1
// CHECK1-NEXT: [[FROMBOOL9:%.*]] = zext i1 [[TOBOOL7]] to i8
// CHECK1-NEXT: store i8 [[FROMBOOL9]], ptr [[DOTCAPTURE_EXPR__CASTED8]], align 1
-// CHECK1-NEXT: [[TMP31:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED8]], align 8
-// CHECK1-NEXT: [[TMP32:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: [[CMP10:%.*]] = icmp sgt i32 [[TMP32]], 4
+// CHECK1-NEXT: [[TMP36:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED8]], align 8
+// CHECK1-NEXT: [[TMP37:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: [[CMP10:%.*]] = icmp sgt i32 [[TMP37]], 4
// CHECK1-NEXT: br i1 [[CMP10]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK1: omp_if.then:
// CHECK1-NEXT: [[A11:%.*]] = getelementptr inbounds [[STRUCT_S1]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS12]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS13]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A11]], ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS14]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP35]], align 8
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS12]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP31]], ptr [[TMP36]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS13]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP31]], ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS14]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP38]], align 8
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS12]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS13]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP41:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_4]], align 1
-// CHECK1-NEXT: [[TOBOOL15:%.*]] = trunc i8 [[TMP41]] to i1
-// CHECK1-NEXT: [[TMP42:%.*]] = select i1 [[TOBOOL15]], i32 0, i32 1
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS12]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP38]], align 8
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS13]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A11]], ptr [[TMP39]], align 8
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS14]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS12]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP36]], ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS13]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP36]], ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS14]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS12]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS13]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP46:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_4]], align 1
+// CHECK1-NEXT: [[TOBOOL15:%.*]] = trunc i8 [[TMP46]] to i1
+// CHECK1-NEXT: [[TMP47:%.*]] = select i1 [[TOBOOL15]], i32 0, i32 1
+// CHECK1-NEXT: [[TMP48:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP47]], 0
// CHECK1-NEXT: [[KERNEL_ARGS16:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP43]], align 4
-// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 1
-// CHECK1-NEXT: store i32 2, ptr [[TMP44]], align 4
-// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP39]], ptr [[TMP45]], align 8
-// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP40]], ptr [[TMP46]], align 8
-// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.2, ptr [[TMP47]], align 8
-// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP48]], align 8
-// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP49]], align 8
-// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP50]], align 8
-// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP51]], align 8
-// CHECK1-NEXT: [[TMP52:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 [[TMP42]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126.region_id, ptr [[KERNEL_ARGS16]])
-// CHECK1-NEXT: [[TMP53:%.*]] = icmp ne i32 [[TMP52]], 0
-// CHECK1-NEXT: br i1 [[TMP53]], label [[OMP_OFFLOAD_FAILED17:%.*]], label [[OMP_OFFLOAD_CONT18:%.*]]
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP49]], align 4
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 1
+// CHECK1-NEXT: store i32 2, ptr [[TMP50]], align 4
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP44]], ptr [[TMP51]], align 8
+// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP45]], ptr [[TMP52]], align 8
+// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.2, ptr [[TMP53]], align 8
+// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP54]], align 8
+// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP55]], align 8
+// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP56]], align 8
+// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP57]], align 8
+// CHECK1-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP58]], align 8
+// CHECK1-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP59]], align 4
+// CHECK1-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP48]], ptr [[TMP60]], align 4
+// CHECK1-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP61]], align 4
+// CHECK1-NEXT: [[TMP62:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 [[TMP47]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126.region_id, ptr [[KERNEL_ARGS16]])
+// CHECK1-NEXT: [[TMP63:%.*]] = icmp ne i32 [[TMP62]], 0
+// CHECK1-NEXT: br i1 [[TMP63]], label [[OMP_OFFLOAD_FAILED17:%.*]], label [[OMP_OFFLOAD_CONT18:%.*]]
// CHECK1: omp_offload.failed17:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126(ptr [[THIS1]], i64 [[TMP31]]) #[[ATTR3]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126(ptr [[THIS1]], i64 [[TMP36]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT18]]
// CHECK1: omp_offload.cont18:
// CHECK1-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK1: omp_if.else:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126(ptr [[THIS1]], i64 [[TMP31]]) #[[ATTR3]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126(ptr [[THIS1]], i64 [[TMP36]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
// CHECK1-NEXT: [[A19:%.*]] = getelementptr inbounds [[STRUCT_S1]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP54:%.*]] = load double, ptr [[A19]], align 8
-// CHECK1-NEXT: [[CONV:%.*]] = fptosi double [[TMP54]] to i32
+// CHECK1-NEXT: [[TMP64:%.*]] = load double, ptr [[A19]], align 8
+// CHECK1-NEXT: [[CONV:%.*]] = fptosi double [[TMP64]] to i32
// CHECK1-NEXT: ret i32 [[CONV]]
//
//
@@ -394,28 +412,37 @@ int bar(int n){
// CHECK1-NEXT: [[TMP9:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
// CHECK1-NEXT: [[TOBOOL3:%.*]] = trunc i8 [[TMP9]] to i1
// CHECK1-NEXT: [[TMP10:%.*]] = select i1 [[TOBOOL3]], i32 0, i32 1
+// CHECK1-NEXT: [[TMP11:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP10]], 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP12]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP7]], ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP8]], ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.5, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP13]], align 4
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP7]], ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP8]], ptr [[TMP15]], align 8
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.5, ptr [[TMP16]], align 8
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP17]], align 8
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 [[TMP10]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l104.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP22]], align 4
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP11]], ptr [[TMP23]], align 4
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP24]], align 4
+// CHECK1-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 [[TMP10]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l104.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
+// CHECK1-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l104(i64 [[TMP2]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -425,33 +452,41 @@ int bar(int n){
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l104(i64 [[TMP2]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
-// CHECK1-NEXT: [[TMP22:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP22]], 2
+// CHECK1-NEXT: [[TMP27:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP27]], 2
// CHECK1-NEXT: [[CMP4:%.*]] = icmp sgt i32 [[SUB]], 2
// CHECK1-NEXT: br i1 [[CMP4]], label [[OMP_IF_THEN5:%.*]], label [[OMP_IF_ELSE9:%.*]]
// CHECK1: omp_if.then5:
// CHECK1-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP23]], align 4
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP24]], align 4
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP28]], align 4
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP29]], align 4
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP30]], align 8
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP31]], align 8
-// CHECK1-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108.region_id, ptr [[KERNEL_ARGS6]])
-// CHECK1-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK1-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK1-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP35]], align 8
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP36]], align 8
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP37]], align 8
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP38]], align 4
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP39]], align 4
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP40]], align 4
+// CHECK1-NEXT: [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108.region_id, ptr [[KERNEL_ARGS6]])
+// CHECK1-NEXT: [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
+// CHECK1-NEXT: br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK1: omp_offload.failed7:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108() #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT8]]
@@ -461,8 +496,8 @@ int bar(int n){
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108() #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_IF_END10]]
// CHECK1: omp_if.end10:
-// CHECK1-NEXT: [[TMP34:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP34]], 1
+// CHECK1-NEXT: [[TMP43:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP43]], 1
// CHECK1-NEXT: ret i32 [[ADD]]
//
//
@@ -496,7 +531,7 @@ int bar(int n){
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -513,62 +548,78 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l87.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK1-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP18]], align 4
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l87.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l87(i64 [[TMP1]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: store i16 1, ptr [[B]], align 2
-// CHECK1-NEXT: [[TMP18:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: store i32 [[TMP18]], ptr [[A_CASTED1]], align 4
-// CHECK1-NEXT: [[TMP19:%.*]] = load i64, ptr [[A_CASTED1]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = load i16, ptr [[B]], align 2
-// CHECK1-NEXT: store i16 [[TMP20]], ptr [[B_CASTED]], align 2
-// CHECK1-NEXT: [[TMP21:%.*]] = load i64, ptr [[B_CASTED]], align 8
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP19]], ptr [[TMP22]], align 8
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP19]], ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS4]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP21]], ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP21]], ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS4]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP22:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: store i32 [[TMP22]], ptr [[A_CASTED1]], align 4
+// CHECK1-NEXT: [[TMP23:%.*]] = load i64, ptr [[A_CASTED1]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = load i16, ptr [[B]], align 2
+// CHECK1-NEXT: store i16 [[TMP24]], ptr [[B_CASTED]], align 2
+// CHECK1-NEXT: [[TMP25:%.*]] = load i64, ptr [[B_CASTED]], align 8
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP23]], ptr [[TMP26]], align 8
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP23]], ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS4]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP25]], ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP25]], ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS4]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP30]], align 4
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK1-NEXT: store i32 2, ptr [[TMP31]], align 4
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP28]], ptr [[TMP32]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP29]], ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.12, ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP35]], align 8
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP36]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP38]], align 8
-// CHECK1-NEXT: [[TMP39:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK1-NEXT: [[TMP40:%.*]] = icmp ne i32 [[TMP39]], 0
-// CHECK1-NEXT: br i1 [[TMP40]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP34]], align 4
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK1-NEXT: store i32 2, ptr [[TMP35]], align 4
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP32]], ptr [[TMP36]], align 8
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP33]], ptr [[TMP37]], align 8
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.12, ptr [[TMP38]], align 8
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP39]], align 8
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP43]], align 8
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP44]], align 4
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP45]], align 4
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP46]], align 4
+// CHECK1-NEXT: [[TMP47:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK1-NEXT: [[TMP48:%.*]] = icmp ne i32 [[TMP47]], 0
+// CHECK1-NEXT: br i1 [[TMP48]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK1: omp_offload.failed6:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93(i64 [[TMP19]], i64 [[TMP21]]) #[[ATTR3]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93(i64 [[TMP23]], i64 [[TMP25]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK1: omp_offload.cont7:
-// CHECK1-NEXT: [[TMP41:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: ret i32 [[TMP41]]
+// CHECK1-NEXT: [[TMP49:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: ret i32 [[TMP49]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121
@@ -884,97 +935,115 @@ int bar(int n){
// CHECK3-NEXT: [[TMP16:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
// CHECK3-NEXT: [[TOBOOL3:%.*]] = trunc i8 [[TMP16]] to i1
// CHECK3-NEXT: [[TMP17:%.*]] = select i1 [[TOBOOL3]], i32 0, i32 1
+// CHECK3-NEXT: [[TMP18:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP17]], 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP18]], align 4
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK3-NEXT: store i32 3, ptr [[TMP19]], align 4
-// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP14]], ptr [[TMP20]], align 4
-// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP15]], ptr [[TMP21]], align 4
-// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes, ptr [[TMP22]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP23]], align 4
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP24]], align 4
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP19]], align 4
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK3-NEXT: store i32 3, ptr [[TMP20]], align 4
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP14]], ptr [[TMP21]], align 4
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP15]], ptr [[TMP22]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes, ptr [[TMP23]], align 4
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP24]], align 4
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK3-NEXT: store ptr null, ptr [[TMP25]], align 4
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP26]], align 8
-// CHECK3-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 1, i32 [[TMP17]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK3-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP28]], align 8
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [[TMP18]], ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 1, i32 [[TMP17]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
+// CHECK3-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121(ptr [[THIS1]], i32 [[TMP2]], i32 [[TMP4]]) #[[ATTR3:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
-// CHECK3-NEXT: [[TMP29:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: [[CMP5:%.*]] = icmp sgt i32 [[TMP29]], 5
+// CHECK3-NEXT: [[TMP34:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: [[CMP5:%.*]] = icmp sgt i32 [[TMP34]], 5
// CHECK3-NEXT: [[FROMBOOL6:%.*]] = zext i1 [[CMP5]] to i8
// CHECK3-NEXT: store i8 [[FROMBOOL6]], ptr [[DOTCAPTURE_EXPR_4]], align 1
-// CHECK3-NEXT: [[TMP30:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_4]], align 1
-// CHECK3-NEXT: [[TOBOOL7:%.*]] = trunc i8 [[TMP30]] to i1
+// CHECK3-NEXT: [[TMP35:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_4]], align 1
+// CHECK3-NEXT: [[TOBOOL7:%.*]] = trunc i8 [[TMP35]] to i1
// CHECK3-NEXT: [[FROMBOOL9:%.*]] = zext i1 [[TOBOOL7]] to i8
// CHECK3-NEXT: store i8 [[FROMBOOL9]], ptr [[DOTCAPTURE_EXPR__CASTED8]], align 1
-// CHECK3-NEXT: [[TMP31:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED8]], align 4
-// CHECK3-NEXT: [[TMP32:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: [[CMP10:%.*]] = icmp sgt i32 [[TMP32]], 4
+// CHECK3-NEXT: [[TMP36:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED8]], align 4
+// CHECK3-NEXT: [[TMP37:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: [[CMP10:%.*]] = icmp sgt i32 [[TMP37]], 4
// CHECK3-NEXT: br i1 [[CMP10]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK3: omp_if.then:
// CHECK3-NEXT: [[A11:%.*]] = getelementptr inbounds [[STRUCT_S1]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS12]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP33]], align 4
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS13]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A11]], ptr [[TMP34]], align 4
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS14]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP35]], align 4
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS12]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP31]], ptr [[TMP36]], align 4
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS13]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP31]], ptr [[TMP37]], align 4
-// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS14]], i32 0, i32 1
-// CHECK3-NEXT: store ptr null, ptr [[TMP38]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS12]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS13]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP41:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_4]], align 1
-// CHECK3-NEXT: [[TOBOOL15:%.*]] = trunc i8 [[TMP41]] to i1
-// CHECK3-NEXT: [[TMP42:%.*]] = select i1 [[TOBOOL15]], i32 0, i32 1
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS12]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP38]], align 4
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS13]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A11]], ptr [[TMP39]], align 4
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS14]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS12]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP36]], ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS13]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP36]], ptr [[TMP42]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS14]], i32 0, i32 1
+// CHECK3-NEXT: store ptr null, ptr [[TMP43]], align 4
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS12]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS13]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP46:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_4]], align 1
+// CHECK3-NEXT: [[TOBOOL15:%.*]] = trunc i8 [[TMP46]] to i1
+// CHECK3-NEXT: [[TMP47:%.*]] = select i1 [[TOBOOL15]], i32 0, i32 1
+// CHECK3-NEXT: [[TMP48:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP47]], 0
// CHECK3-NEXT: [[KERNEL_ARGS16:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP43]], align 4
-// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 1
-// CHECK3-NEXT: store i32 2, ptr [[TMP44]], align 4
-// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP39]], ptr [[TMP45]], align 4
-// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP40]], ptr [[TMP46]], align 4
-// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.2, ptr [[TMP47]], align 4
-// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP48]], align 4
-// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP49]], align 4
-// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP50]], align 4
-// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP51]], align 8
-// CHECK3-NEXT: [[TMP52:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 [[TMP42]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126.region_id, ptr [[KERNEL_ARGS16]])
-// CHECK3-NEXT: [[TMP53:%.*]] = icmp ne i32 [[TMP52]], 0
-// CHECK3-NEXT: br i1 [[TMP53]], label [[OMP_OFFLOAD_FAILED17:%.*]], label [[OMP_OFFLOAD_CONT18:%.*]]
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP49]], align 4
+// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 1
+// CHECK3-NEXT: store i32 2, ptr [[TMP50]], align 4
+// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP44]], ptr [[TMP51]], align 4
+// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP45]], ptr [[TMP52]], align 4
+// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.2, ptr [[TMP53]], align 4
+// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP54]], align 4
+// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP55]], align 4
+// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP56]], align 4
+// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP57]], align 8
+// CHECK3-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP58]], align 8
+// CHECK3-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP59]], align 4
+// CHECK3-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [[TMP48]], ptr [[TMP60]], align 4
+// CHECK3-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP61]], align 4
+// CHECK3-NEXT: [[TMP62:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 [[TMP47]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126.region_id, ptr [[KERNEL_ARGS16]])
+// CHECK3-NEXT: [[TMP63:%.*]] = icmp ne i32 [[TMP62]], 0
+// CHECK3-NEXT: br i1 [[TMP63]], label [[OMP_OFFLOAD_FAILED17:%.*]], label [[OMP_OFFLOAD_CONT18:%.*]]
// CHECK3: omp_offload.failed17:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126(ptr [[THIS1]], i32 [[TMP31]]) #[[ATTR3]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126(ptr [[THIS1]], i32 [[TMP36]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT18]]
// CHECK3: omp_offload.cont18:
// CHECK3-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK3: omp_if.else:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126(ptr [[THIS1]], i32 [[TMP31]]) #[[ATTR3]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126(ptr [[THIS1]], i32 [[TMP36]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
// CHECK3-NEXT: [[A19:%.*]] = getelementptr inbounds [[STRUCT_S1]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP54:%.*]] = load double, ptr [[A19]], align 4
-// CHECK3-NEXT: [[CONV:%.*]] = fptosi double [[TMP54]] to i32
+// CHECK3-NEXT: [[TMP64:%.*]] = load double, ptr [[A19]], align 4
+// CHECK3-NEXT: [[CONV:%.*]] = fptosi double [[TMP64]] to i32
// CHECK3-NEXT: ret i32 [[CONV]]
//
//
@@ -1012,28 +1081,37 @@ int bar(int n){
// CHECK3-NEXT: [[TMP9:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
// CHECK3-NEXT: [[TOBOOL3:%.*]] = trunc i8 [[TMP9]] to i1
// CHECK3-NEXT: [[TMP10:%.*]] = select i1 [[TOBOOL3]], i32 0, i32 1
+// CHECK3-NEXT: [[TMP11:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP10]], 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP12]], align 4
-// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP7]], ptr [[TMP13]], align 4
-// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP8]], ptr [[TMP14]], align 4
-// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.5, ptr [[TMP15]], align 4
-// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP16]], align 4
-// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP17]], align 4
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP13]], align 4
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP7]], ptr [[TMP14]], align 4
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP8]], ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.5, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK3-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP19]], align 8
-// CHECK3-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 [[TMP10]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l104.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK3-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP19]], align 4
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP20]], align 8
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP21]], align 8
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP22]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [[TMP11]], ptr [[TMP23]], align 4
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP24]], align 4
+// CHECK3-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 [[TMP10]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l104.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
+// CHECK3-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l104(i32 [[TMP2]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1043,33 +1121,41 @@ int bar(int n){
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l104(i32 [[TMP2]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
-// CHECK3-NEXT: [[TMP22:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP22]], 2
+// CHECK3-NEXT: [[TMP27:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP27]], 2
// CHECK3-NEXT: [[CMP4:%.*]] = icmp sgt i32 [[SUB]], 2
// CHECK3-NEXT: br i1 [[CMP4]], label [[OMP_IF_THEN5:%.*]], label [[OMP_IF_ELSE9:%.*]]
// CHECK3: omp_if.then5:
// CHECK3-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP23]], align 4
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK3-NEXT: store i32 0, ptr [[TMP24]], align 4
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK3-NEXT: store ptr null, ptr [[TMP25]], align 4
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK3-NEXT: store ptr null, ptr [[TMP26]], align 4
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK3-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP29]], align 4
-// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK3-NEXT: store i32 0, ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
// CHECK3-NEXT: store ptr null, ptr [[TMP30]], align 4
-// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP31]], align 8
-// CHECK3-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108.region_id, ptr [[KERNEL_ARGS6]])
-// CHECK3-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK3-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK3-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK3-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK3-NEXT: store ptr null, ptr [[TMP33]], align 4
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP34]], align 4
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP35]], align 4
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP36]], align 8
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP37]], align 8
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP38]], align 4
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP39]], align 4
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108.region_id, ptr [[KERNEL_ARGS6]])
+// CHECK3-NEXT: [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
+// CHECK3-NEXT: br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK3: omp_offload.failed7:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108() #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT8]]
@@ -1079,8 +1165,8 @@ int bar(int n){
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108() #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_IF_END10]]
// CHECK3: omp_if.end10:
-// CHECK3-NEXT: [[TMP34:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP34]], 1
+// CHECK3-NEXT: [[TMP43:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP43]], 1
// CHECK3-NEXT: ret i32 [[ADD]]
//
//
@@ -1114,7 +1200,7 @@ int bar(int n){
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1131,62 +1217,78 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP14]], align 4
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP15]], align 8
-// CHECK3-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l87.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK3-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP18]], align 4
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK3-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l87.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK3-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l87(i32 [[TMP1]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: store i16 1, ptr [[B]], align 2
-// CHECK3-NEXT: [[TMP18:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: store i32 [[TMP18]], ptr [[A_CASTED1]], align 4
-// CHECK3-NEXT: [[TMP19:%.*]] = load i32, ptr [[A_CASTED1]], align 4
-// CHECK3-NEXT: [[TMP20:%.*]] = load i16, ptr [[B]], align 2
-// CHECK3-NEXT: store i16 [[TMP20]], ptr [[B_CASTED]], align 2
-// CHECK3-NEXT: [[TMP21:%.*]] = load i32, ptr [[B_CASTED]], align 4
-// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP19]], ptr [[TMP22]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP19]], ptr [[TMP23]], align 4
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS4]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP24]], align 4
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP21]], ptr [[TMP25]], align 4
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP21]], ptr [[TMP26]], align 4
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS4]], i32 0, i32 1
-// CHECK3-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP22:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: store i32 [[TMP22]], ptr [[A_CASTED1]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = load i32, ptr [[A_CASTED1]], align 4
+// CHECK3-NEXT: [[TMP24:%.*]] = load i16, ptr [[B]], align 2
+// CHECK3-NEXT: store i16 [[TMP24]], ptr [[B_CASTED]], align 2
+// CHECK3-NEXT: [[TMP25:%.*]] = load i32, ptr [[B_CASTED]], align 4
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP23]], ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP23]], ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS4]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP25]], ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP25]], ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS4]], i32 0, i32 1
+// CHECK3-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP30]], align 4
-// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK3-NEXT: store i32 2, ptr [[TMP31]], align 4
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP28]], ptr [[TMP32]], align 4
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP29]], ptr [[TMP33]], align 4
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.12, ptr [[TMP34]], align 4
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP35]], align 4
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP36]], align 4
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP37]], align 4
-// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP38]], align 8
-// CHECK3-NEXT: [[TMP39:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK3-NEXT: [[TMP40:%.*]] = icmp ne i32 [[TMP39]], 0
-// CHECK3-NEXT: br i1 [[TMP40]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP34]], align 4
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK3-NEXT: store i32 2, ptr [[TMP35]], align 4
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP32]], ptr [[TMP36]], align 4
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP33]], ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.12, ptr [[TMP38]], align 4
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP39]], align 4
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP42]], align 8
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP43]], align 8
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP44]], align 4
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP45]], align 4
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP46]], align 4
+// CHECK3-NEXT: [[TMP47:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK3-NEXT: [[TMP48:%.*]] = icmp ne i32 [[TMP47]], 0
+// CHECK3-NEXT: br i1 [[TMP48]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK3: omp_offload.failed6:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93(i32 [[TMP19]], i32 [[TMP21]]) #[[ATTR3]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93(i32 [[TMP23]], i32 [[TMP25]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK3: omp_offload.cont7:
-// CHECK3-NEXT: [[TMP41:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: ret i32 [[TMP41]]
+// CHECK3-NEXT: [[TMP49:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: ret i32 [[TMP49]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121
diff --git a/clang/test/OpenMP/target_parallel_num_threads_codegen.cpp b/clang/test/OpenMP/target_parallel_num_threads_codegen.cpp
index 453624125cc0b..f8be5d02d39c8 100644
--- a/clang/test/OpenMP/target_parallel_num_threads_codegen.cpp
+++ b/clang/test/OpenMP/target_parallel_num_threads_codegen.cpp
@@ -267,70 +267,87 @@ int bar(int n){
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP17]], 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP18]], align 4
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK1-NEXT: store i32 3, ptr [[TMP19]], align 4
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP15]], ptr [[TMP20]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP16]], ptr [[TMP21]], align 8
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP22]], align 8
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP19]], align 4
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK1-NEXT: store i32 3, ptr [[TMP20]], align 4
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP15]], ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP16]], ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK1-NEXT: store ptr null, ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 1, i32 [[TMP17]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK1-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP26]], align 8
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP29]], align 4
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP18]], ptr [[TMP30]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP31]], align 4
+// CHECK1-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 1, i32 [[TMP17]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
+// CHECK1-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121(ptr [[THIS1]], i64 [[TMP3]], i64 [[TMP5]]) #[[ATTR3:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_S1]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP30]], align 8
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP31]], align 8
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP35]], align 8
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP36]], align 8
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP34]], align 4
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP35]], align 4
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP32]], ptr [[TMP36]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP33]], ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.2, ptr [[TMP38]], align 8
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP41]], align 8
-// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP42]], align 8
-// CHECK1-NEXT: [[TMP43:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 1024, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126.region_id, ptr [[KERNEL_ARGS6]])
-// CHECK1-NEXT: [[TMP44:%.*]] = icmp ne i32 [[TMP43]], 0
-// CHECK1-NEXT: br i1 [[TMP44]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP39]], align 4
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP40]], align 4
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP37]], ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP38]], ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.2, ptr [[TMP43]], align 8
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP44]], align 8
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP45]], align 8
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP46]], align 8
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP47]], align 8
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP48]], align 8
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP49]], align 4
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1024, i32 0, i32 0], ptr [[TMP50]], align 4
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP51]], align 4
+// CHECK1-NEXT: [[TMP52:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 1024, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126.region_id, ptr [[KERNEL_ARGS6]])
+// CHECK1-NEXT: [[TMP53:%.*]] = icmp ne i32 [[TMP52]], 0
+// CHECK1-NEXT: br i1 [[TMP53]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK1: omp_offload.failed7:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126(ptr [[THIS1]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK1: omp_offload.cont8:
// CHECK1-NEXT: [[A9:%.*]] = getelementptr inbounds [[STRUCT_S1]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP45:%.*]] = load double, ptr [[A9]], align 8
-// CHECK1-NEXT: [[CONV:%.*]] = fptosi double [[TMP45]] to i32
+// CHECK1-NEXT: [[TMP54:%.*]] = load double, ptr [[A9]], align 8
+// CHECK1-NEXT: [[CONV:%.*]] = fptosi double [[TMP54]] to i32
// CHECK1-NEXT: ret i32 [[CONV]]
//
//
@@ -363,75 +380,93 @@ int bar(int n){
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK1-NEXT: [[TMP9:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP8]], 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP9]], align 4
-// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP10]], align 4
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP6]], ptr [[TMP11]], align 8
-// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP7]], ptr [[TMP12]], align 8
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.5, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP6]], ptr [[TMP12]], align 8
+// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP7]], ptr [[TMP13]], align 8
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.5, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP15]], align 8
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 [[TMP8]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l104.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
-// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP18]], align 8
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP20]], align 4
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP9]], ptr [[TMP21]], align 4
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP22]], align 4
+// CHECK1-NEXT: [[TMP23:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 [[TMP8]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l104.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP24:%.*]] = icmp ne i32 [[TMP23]], 0
+// CHECK1-NEXT: br i1 [[TMP24]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l104(i64 [[TMP2]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
-// CHECK1-NEXT: [[TMP20:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 32, [[TMP20]]
+// CHECK1-NEXT: [[TMP25:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 32, [[TMP25]]
// CHECK1-NEXT: store i32 [[ADD]], ptr [[DOTCAPTURE_EXPR_1]], align 4
-// CHECK1-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
-// CHECK1-NEXT: store i32 [[TMP21]], ptr [[DOTCAPTURE_EXPR__CASTED2]], align 4
-// CHECK1-NEXT: [[TMP22:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED2]], align 8
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP22]], ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP22]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP28:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
+// CHECK1-NEXT: store i32 [[TMP26]], ptr [[DOTCAPTURE_EXPR__CASTED2]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED2]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP27]], ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP27]], ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP33:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
+// CHECK1-NEXT: [[TMP34:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP33]], 0
// CHECK1-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP29]], align 4
-// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP30]], align 4
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP26]], ptr [[TMP31]], align 8
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP27]], ptr [[TMP32]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.8, ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP35]], align 8
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP36]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 [[TMP28]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108.region_id, ptr [[KERNEL_ARGS6]])
-// CHECK1-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK1-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP35]], align 4
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP36]], align 4
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP31]], ptr [[TMP37]], align 8
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP32]], ptr [[TMP38]], align 8
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.8, ptr [[TMP39]], align 8
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP43]], align 8
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP44]], align 8
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP45]], align 4
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP34]], ptr [[TMP46]], align 4
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP47]], align 4
+// CHECK1-NEXT: [[TMP48:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 [[TMP33]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108.region_id, ptr [[KERNEL_ARGS6]])
+// CHECK1-NEXT: [[TMP49:%.*]] = icmp ne i32 [[TMP48]], 0
+// CHECK1-NEXT: br i1 [[TMP49]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK1: omp_offload.failed7:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108(i64 [[TMP22]]) #[[ATTR3]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108(i64 [[TMP27]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK1: omp_offload.cont8:
-// CHECK1-NEXT: [[TMP40:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: [[ADD9:%.*]] = add nsw i32 [[TMP40]], 1
+// CHECK1-NEXT: [[TMP50:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: [[ADD9:%.*]] = add nsw i32 [[TMP50]], 1
// CHECK1-NEXT: ret i32 [[ADD9]]
//
//
@@ -452,7 +487,7 @@ int bar(int n){
// CHECK1-NEXT: store i32 0, ptr [[A]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -469,75 +504,92 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 20, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l88.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 20, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 20, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l88.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l88() #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: store i16 1, ptr [[B]], align 2
-// CHECK1-NEXT: [[TMP11:%.*]] = load i16, ptr [[B]], align 2
-// CHECK1-NEXT: store i16 [[TMP11]], ptr [[DOTCAPTURE_EXPR_]], align 2
-// CHECK1-NEXT: [[TMP12:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: store i32 [[TMP12]], ptr [[A_CASTED]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = load i64, ptr [[A_CASTED]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = load i16, ptr [[B]], align 2
-// CHECK1-NEXT: store i16 [[TMP14]], ptr [[B_CASTED]], align 2
-// CHECK1-NEXT: [[TMP15:%.*]] = load i64, ptr [[B_CASTED]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = load i16, ptr [[DOTCAPTURE_EXPR_]], align 2
-// CHECK1-NEXT: store i16 [[TMP16]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 2
-// CHECK1-NEXT: [[TMP17:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP13]], ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP13]], ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP15]], ptr [[TMP21]], align 8
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP15]], ptr [[TMP22]], align 8
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
-// CHECK1-NEXT: store i64 [[TMP17]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
-// CHECK1-NEXT: store i64 [[TMP17]], ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP29:%.*]] = load i16, ptr [[DOTCAPTURE_EXPR_]], align 2
-// CHECK1-NEXT: [[TMP30:%.*]] = zext i16 [[TMP29]] to i32
+// CHECK1-NEXT: [[TMP15:%.*]] = load i16, ptr [[B]], align 2
+// CHECK1-NEXT: store i16 [[TMP15]], ptr [[DOTCAPTURE_EXPR_]], align 2
+// CHECK1-NEXT: [[TMP16:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: store i32 [[TMP16]], ptr [[A_CASTED]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = load i64, ptr [[A_CASTED]], align 8
+// CHECK1-NEXT: [[TMP18:%.*]] = load i16, ptr [[B]], align 2
+// CHECK1-NEXT: store i16 [[TMP18]], ptr [[B_CASTED]], align 2
+// CHECK1-NEXT: [[TMP19:%.*]] = load i64, ptr [[B_CASTED]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = load i16, ptr [[DOTCAPTURE_EXPR_]], align 2
+// CHECK1-NEXT: store i16 [[TMP20]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 2
+// CHECK1-NEXT: [[TMP21:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP17]], ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP17]], ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP19]], ptr [[TMP25]], align 8
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP19]], ptr [[TMP26]], align 8
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
+// CHECK1-NEXT: store i64 [[TMP21]], ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
+// CHECK1-NEXT: store i64 [[TMP21]], ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 2
+// CHECK1-NEXT: store ptr null, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP33:%.*]] = load i16, ptr [[DOTCAPTURE_EXPR_]], align 2
+// CHECK1-NEXT: [[TMP34:%.*]] = zext i16 [[TMP33]] to i32
+// CHECK1-NEXT: [[TMP35:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP34]], 0
// CHECK1-NEXT: [[KERNEL_ARGS1:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP31]], align 4
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 1
-// CHECK1-NEXT: store i32 3, ptr [[TMP32]], align 4
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP27]], ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP28]], ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.12, ptr [[TMP35]], align 8
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP36]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP38]], align 8
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 [[TMP30]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93.region_id, ptr [[KERNEL_ARGS1]])
-// CHECK1-NEXT: [[TMP41:%.*]] = icmp ne i32 [[TMP40]], 0
-// CHECK1-NEXT: br i1 [[TMP41]], label [[OMP_OFFLOAD_FAILED2:%.*]], label [[OMP_OFFLOAD_CONT3:%.*]]
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP36]], align 4
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 1
+// CHECK1-NEXT: store i32 3, ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP31]], ptr [[TMP38]], align 8
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP32]], ptr [[TMP39]], align 8
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.12, ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP44]], align 8
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP45]], align 8
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP46]], align 4
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP35]], ptr [[TMP47]], align 4
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP48]], align 4
+// CHECK1-NEXT: [[TMP49:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 [[TMP34]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93.region_id, ptr [[KERNEL_ARGS1]])
+// CHECK1-NEXT: [[TMP50:%.*]] = icmp ne i32 [[TMP49]], 0
+// CHECK1-NEXT: br i1 [[TMP50]], label [[OMP_OFFLOAD_FAILED2:%.*]], label [[OMP_OFFLOAD_CONT3:%.*]]
// CHECK1: omp_offload.failed2:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93(i64 [[TMP13]], i64 [[TMP15]], i64 [[TMP17]]) #[[ATTR3]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93(i64 [[TMP17]], i64 [[TMP19]], i64 [[TMP21]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT3]]
// CHECK1: omp_offload.cont3:
-// CHECK1-NEXT: [[TMP42:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: ret i32 [[TMP42]]
+// CHECK1-NEXT: [[TMP51:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: ret i32 [[TMP51]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121
@@ -800,70 +852,87 @@ int bar(int n){
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP17]], 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP18]], align 4
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK3-NEXT: store i32 3, ptr [[TMP19]], align 4
-// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP15]], ptr [[TMP20]], align 4
-// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP16]], ptr [[TMP21]], align 4
-// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes, ptr [[TMP22]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP23]], align 4
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP24]], align 4
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP19]], align 4
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK3-NEXT: store i32 3, ptr [[TMP20]], align 4
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP15]], ptr [[TMP21]], align 4
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP16]], ptr [[TMP22]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes, ptr [[TMP23]], align 4
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP24]], align 4
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK3-NEXT: store ptr null, ptr [[TMP25]], align 4
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP26]], align 8
-// CHECK3-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 1, i32 [[TMP17]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK3-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP28]], align 8
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [[TMP18]], ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 1, i32 [[TMP17]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
+// CHECK3-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121(ptr [[THIS1]], i32 [[TMP3]], i32 [[TMP5]]) #[[ATTR3:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_S1]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP29]], align 4
-// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A2]], ptr [[TMP30]], align 4
-// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP31]], align 4
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP34]], align 4
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A2]], ptr [[TMP35]], align 4
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP36]], align 4
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP34]], align 4
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP35]], align 4
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP32]], ptr [[TMP36]], align 4
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP33]], ptr [[TMP37]], align 4
-// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.2, ptr [[TMP38]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP39]], align 4
-// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP40]], align 4
-// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP41]], align 4
-// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP42]], align 8
-// CHECK3-NEXT: [[TMP43:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 1024, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126.region_id, ptr [[KERNEL_ARGS6]])
-// CHECK3-NEXT: [[TMP44:%.*]] = icmp ne i32 [[TMP43]], 0
-// CHECK3-NEXT: br i1 [[TMP44]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP39]], align 4
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP37]], ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP38]], ptr [[TMP42]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.2, ptr [[TMP43]], align 4
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP44]], align 4
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP45]], align 4
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP46]], align 4
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP47]], align 8
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP48]], align 8
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP49]], align 4
+// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1024, i32 0, i32 0], ptr [[TMP50]], align 4
+// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP51]], align 4
+// CHECK3-NEXT: [[TMP52:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 1024, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126.region_id, ptr [[KERNEL_ARGS6]])
+// CHECK3-NEXT: [[TMP53:%.*]] = icmp ne i32 [[TMP52]], 0
+// CHECK3-NEXT: br i1 [[TMP53]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK3: omp_offload.failed7:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126(ptr [[THIS1]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK3: omp_offload.cont8:
// CHECK3-NEXT: [[A9:%.*]] = getelementptr inbounds [[STRUCT_S1]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP45:%.*]] = load double, ptr [[A9]], align 4
-// CHECK3-NEXT: [[CONV:%.*]] = fptosi double [[TMP45]] to i32
+// CHECK3-NEXT: [[TMP54:%.*]] = load double, ptr [[A9]], align 4
+// CHECK3-NEXT: [[CONV:%.*]] = fptosi double [[TMP54]] to i32
// CHECK3-NEXT: ret i32 [[CONV]]
//
//
@@ -896,75 +965,93 @@ int bar(int n){
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK3-NEXT: [[TMP9:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP8]], 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP9]], align 4
-// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP10]], align 4
-// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP6]], ptr [[TMP11]], align 4
-// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP7]], ptr [[TMP12]], align 4
-// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.5, ptr [[TMP13]], align 4
-// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP14]], align 4
-// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP15]], align 4
-// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP6]], ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP7]], ptr [[TMP13]], align 4
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.5, ptr [[TMP14]], align 4
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK3-NEXT: store ptr null, ptr [[TMP16]], align 4
-// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP17]], align 8
-// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 [[TMP8]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l104.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
-// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP18]], align 8
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP19]], align 8
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP20]], align 4
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [[TMP9]], ptr [[TMP21]], align 4
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP22]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 [[TMP8]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l104.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP24:%.*]] = icmp ne i32 [[TMP23]], 0
+// CHECK3-NEXT: br i1 [[TMP24]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l104(i32 [[TMP2]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
-// CHECK3-NEXT: [[TMP20:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 32, [[TMP20]]
+// CHECK3-NEXT: [[TMP25:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 32, [[TMP25]]
// CHECK3-NEXT: store i32 [[ADD]], ptr [[DOTCAPTURE_EXPR_1]], align 4
-// CHECK3-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
-// CHECK3-NEXT: store i32 [[TMP21]], ptr [[DOTCAPTURE_EXPR__CASTED2]], align 4
-// CHECK3-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED2]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP22]], ptr [[TMP23]], align 4
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP22]], ptr [[TMP24]], align 4
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP25]], align 4
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP28:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
+// CHECK3-NEXT: [[TMP26:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
+// CHECK3-NEXT: store i32 [[TMP26]], ptr [[DOTCAPTURE_EXPR__CASTED2]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED2]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP27]], ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP27]], ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP33:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
+// CHECK3-NEXT: [[TMP34:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP33]], 0
// CHECK3-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP29]], align 4
-// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP30]], align 4
-// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP26]], ptr [[TMP31]], align 4
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP27]], ptr [[TMP32]], align 4
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.8, ptr [[TMP33]], align 4
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP34]], align 4
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP35]], align 4
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP36]], align 4
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP37]], align 8
-// CHECK3-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 [[TMP28]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108.region_id, ptr [[KERNEL_ARGS6]])
-// CHECK3-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK3-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP35]], align 4
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP36]], align 4
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP31]], ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP32]], ptr [[TMP38]], align 4
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.8, ptr [[TMP39]], align 4
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP42]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP43]], align 8
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP44]], align 8
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP45]], align 4
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [[TMP34]], ptr [[TMP46]], align 4
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP47]], align 4
+// CHECK3-NEXT: [[TMP48:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 [[TMP33]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108.region_id, ptr [[KERNEL_ARGS6]])
+// CHECK3-NEXT: [[TMP49:%.*]] = icmp ne i32 [[TMP48]], 0
+// CHECK3-NEXT: br i1 [[TMP49]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK3: omp_offload.failed7:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108(i32 [[TMP22]]) #[[ATTR3]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108(i32 [[TMP27]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK3: omp_offload.cont8:
-// CHECK3-NEXT: [[TMP40:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: [[ADD9:%.*]] = add nsw i32 [[TMP40]], 1
+// CHECK3-NEXT: [[TMP50:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: [[ADD9:%.*]] = add nsw i32 [[TMP50]], 1
// CHECK3-NEXT: ret i32 [[ADD9]]
//
//
@@ -985,7 +1072,7 @@ int bar(int n){
// CHECK3-NEXT: store i32 0, ptr [[A]], align 4
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1002,75 +1089,92 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP8]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 20, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l88.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK3-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 20, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 20, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l88.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK3-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l88() #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: store i16 1, ptr [[B]], align 2
-// CHECK3-NEXT: [[TMP11:%.*]] = load i16, ptr [[B]], align 2
-// CHECK3-NEXT: store i16 [[TMP11]], ptr [[DOTCAPTURE_EXPR_]], align 2
-// CHECK3-NEXT: [[TMP12:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: store i32 [[TMP12]], ptr [[A_CASTED]], align 4
-// CHECK3-NEXT: [[TMP13:%.*]] = load i32, ptr [[A_CASTED]], align 4
-// CHECK3-NEXT: [[TMP14:%.*]] = load i16, ptr [[B]], align 2
-// CHECK3-NEXT: store i16 [[TMP14]], ptr [[B_CASTED]], align 2
-// CHECK3-NEXT: [[TMP15:%.*]] = load i32, ptr [[B_CASTED]], align 4
-// CHECK3-NEXT: [[TMP16:%.*]] = load i16, ptr [[DOTCAPTURE_EXPR_]], align 2
-// CHECK3-NEXT: store i16 [[TMP16]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 2
-// CHECK3-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP13]], ptr [[TMP18]], align 4
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP13]], ptr [[TMP19]], align 4
-// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP20]], align 4
-// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP15]], ptr [[TMP21]], align 4
-// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP15]], ptr [[TMP22]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 1
-// CHECK3-NEXT: store ptr null, ptr [[TMP23]], align 4
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
-// CHECK3-NEXT: store i32 [[TMP17]], ptr [[TMP24]], align 4
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
-// CHECK3-NEXT: store i32 [[TMP17]], ptr [[TMP25]], align 4
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 2
-// CHECK3-NEXT: store ptr null, ptr [[TMP26]], align 4
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP29:%.*]] = load i16, ptr [[DOTCAPTURE_EXPR_]], align 2
-// CHECK3-NEXT: [[TMP30:%.*]] = zext i16 [[TMP29]] to i32
+// CHECK3-NEXT: [[TMP15:%.*]] = load i16, ptr [[B]], align 2
+// CHECK3-NEXT: store i16 [[TMP15]], ptr [[DOTCAPTURE_EXPR_]], align 2
+// CHECK3-NEXT: [[TMP16:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: store i32 [[TMP16]], ptr [[A_CASTED]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = load i32, ptr [[A_CASTED]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = load i16, ptr [[B]], align 2
+// CHECK3-NEXT: store i16 [[TMP18]], ptr [[B_CASTED]], align 2
+// CHECK3-NEXT: [[TMP19:%.*]] = load i32, ptr [[B_CASTED]], align 4
+// CHECK3-NEXT: [[TMP20:%.*]] = load i16, ptr [[DOTCAPTURE_EXPR_]], align 2
+// CHECK3-NEXT: store i16 [[TMP20]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 2
+// CHECK3-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP17]], ptr [[TMP22]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP17]], ptr [[TMP23]], align 4
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP24]], align 4
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP19]], ptr [[TMP25]], align 4
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP19]], ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 1
+// CHECK3-NEXT: store ptr null, ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
+// CHECK3-NEXT: store i32 [[TMP21]], ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
+// CHECK3-NEXT: store i32 [[TMP21]], ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 2
+// CHECK3-NEXT: store ptr null, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP33:%.*]] = load i16, ptr [[DOTCAPTURE_EXPR_]], align 2
+// CHECK3-NEXT: [[TMP34:%.*]] = zext i16 [[TMP33]] to i32
+// CHECK3-NEXT: [[TMP35:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP34]], 0
// CHECK3-NEXT: [[KERNEL_ARGS1:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP31]], align 4
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 1
-// CHECK3-NEXT: store i32 3, ptr [[TMP32]], align 4
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP27]], ptr [[TMP33]], align 4
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP28]], ptr [[TMP34]], align 4
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.12, ptr [[TMP35]], align 4
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP36]], align 4
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP37]], align 4
-// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP38]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP39]], align 8
-// CHECK3-NEXT: [[TMP40:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 [[TMP30]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93.region_id, ptr [[KERNEL_ARGS1]])
-// CHECK3-NEXT: [[TMP41:%.*]] = icmp ne i32 [[TMP40]], 0
-// CHECK3-NEXT: br i1 [[TMP41]], label [[OMP_OFFLOAD_FAILED2:%.*]], label [[OMP_OFFLOAD_CONT3:%.*]]
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP36]], align 4
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 1
+// CHECK3-NEXT: store i32 3, ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP31]], ptr [[TMP38]], align 4
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP32]], ptr [[TMP39]], align 4
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.12, ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP42]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP43]], align 4
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP44]], align 8
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP45]], align 8
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP46]], align 4
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [[TMP35]], ptr [[TMP47]], align 4
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP48]], align 4
+// CHECK3-NEXT: [[TMP49:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 [[TMP34]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93.region_id, ptr [[KERNEL_ARGS1]])
+// CHECK3-NEXT: [[TMP50:%.*]] = icmp ne i32 [[TMP49]], 0
+// CHECK3-NEXT: br i1 [[TMP50]], label [[OMP_OFFLOAD_FAILED2:%.*]], label [[OMP_OFFLOAD_CONT3:%.*]]
// CHECK3: omp_offload.failed2:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93(i32 [[TMP13]], i32 [[TMP15]], i32 [[TMP17]]) #[[ATTR3]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93(i32 [[TMP17]], i32 [[TMP19]], i32 [[TMP21]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT3]]
// CHECK3: omp_offload.cont3:
-// CHECK3-NEXT: [[TMP42:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: ret i32 [[TMP42]]
+// CHECK3-NEXT: [[TMP51:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: ret i32 [[TMP51]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121
diff --git a/clang/test/OpenMP/target_simd_codegen.cpp b/clang/test/OpenMP/target_simd_codegen.cpp
index 8931397e13381..87debe53dbede 100644
--- a/clang/test/OpenMP/target_simd_codegen.cpp
+++ b/clang/test/OpenMP/target_simd_codegen.cpp
@@ -360,7 +360,7 @@ int foo(int n) {
// CHECK-NEXT: }
// CHECK: define internal {{.*}}i32 [[OMP_TASK_ENTRY]](i32 {{.*}}%0, [[KMP_TASK_T_WITH_PRIVATES]]* noalias noundef %1)
-// CHECK: [[RET:%.+]] = call i32 @__tgt_target_kernel_nowait(%struct.ident_t* @{{.+}}, i64 [[DEVICE:.+]], i32 1, i32 1, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]], i32 0, i8* null, i32 0, i8* null)
+// CHECK: [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 [[DEVICE:.+]], i32 1, i32 1, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]])
// CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]]
// CHECK: [[FAIL]]
diff --git a/clang/test/OpenMP/target_simd_depend_codegen.cpp b/clang/test/OpenMP/target_simd_depend_codegen.cpp
index fa7d87c647d5b..5d7c5fa7b323c 100644
--- a/clang/test/OpenMP/target_simd_depend_codegen.cpp
+++ b/clang/test/OpenMP/target_simd_depend_codegen.cpp
@@ -204,7 +204,7 @@ int foo(int n) {
// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2
// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]],
// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64
-// CHECK: [[RET:%.+]] = call i32 @__tgt_target_kernel_nowait(%struct.ident_t* @{{.+}}, i64 [[DEVICE]], i32 1, i32 1, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* %{{.+}})
+// CHECK: [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 [[DEVICE]], i32 1, i32 1, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* %{{.+}})
// CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]]
diff --git a/clang/test/OpenMP/target_teams_codegen.cpp b/clang/test/OpenMP/target_teams_codegen.cpp
index cd85248ef121d..bad0795e5c965 100644
--- a/clang/test/OpenMP/target_teams_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_codegen.cpp
@@ -437,7 +437,7 @@ int bar(int n){
// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP52]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP52]], align 4
// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP53]], align 4
// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -454,250 +454,290 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP59]], align 8
// CHECK1-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP60]], align 8
-// CHECK1-NEXT: [[TMP61:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l111.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP62:%.*]] = icmp ne i32 [[TMP61]], 0
-// CHECK1-NEXT: br i1 [[TMP62]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP61]], align 8
+// CHECK1-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP62]], align 4
+// CHECK1-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP63]], align 4
+// CHECK1-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP64]], align 4
+// CHECK1-NEXT: [[TMP65:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l111.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP66:%.*]] = icmp ne i32 [[TMP65]], 0
+// CHECK1-NEXT: br i1 [[TMP66]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l111(i64 [[TMP46]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
-// CHECK1-NEXT: [[TMP63:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: store i32 [[TMP63]], ptr [[A_CASTED8]], align 4
-// CHECK1-NEXT: [[TMP64:%.*]] = load i64, ptr [[A_CASTED8]], align 8
-// CHECK1-NEXT: [[TMP65:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK1-NEXT: store i16 [[TMP65]], ptr [[AA_CASTED9]], align 2
-// CHECK1-NEXT: [[TMP66:%.*]] = load i64, ptr [[AA_CASTED9]], align 8
-// CHECK1-NEXT: [[TMP67:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP67]], 10
+// CHECK1-NEXT: [[TMP67:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: store i32 [[TMP67]], ptr [[A_CASTED8]], align 4
+// CHECK1-NEXT: [[TMP68:%.*]] = load i64, ptr [[A_CASTED8]], align 8
+// CHECK1-NEXT: [[TMP69:%.*]] = load i16, ptr [[AA]], align 2
+// CHECK1-NEXT: store i16 [[TMP69]], ptr [[AA_CASTED9]], align 2
+// CHECK1-NEXT: [[TMP70:%.*]] = load i64, ptr [[AA_CASTED9]], align 8
+// CHECK1-NEXT: [[TMP71:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP71]], 10
// CHECK1-NEXT: br i1 [[CMP]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK1: omp_if.then:
-// CHECK1-NEXT: [[TMP68:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP64]], ptr [[TMP68]], align 8
-// CHECK1-NEXT: [[TMP69:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP64]], ptr [[TMP69]], align 8
-// CHECK1-NEXT: [[TMP70:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP70]], align 8
-// CHECK1-NEXT: [[TMP71:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP66]], ptr [[TMP71]], align 8
-// CHECK1-NEXT: [[TMP72:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP66]], ptr [[TMP72]], align 8
-// CHECK1-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP73]], align 8
-// CHECK1-NEXT: [[TMP74:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP75:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP72:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP68]], ptr [[TMP72]], align 8
+// CHECK1-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP68]], ptr [[TMP73]], align 8
+// CHECK1-NEXT: [[TMP74:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP74]], align 8
+// CHECK1-NEXT: [[TMP75:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP70]], ptr [[TMP75]], align 8
+// CHECK1-NEXT: [[TMP76:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP70]], ptr [[TMP76]], align 8
+// CHECK1-NEXT: [[TMP77:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP77]], align 8
+// CHECK1-NEXT: [[TMP78:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP79:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS13:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP76]], align 4
-// CHECK1-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 1
-// CHECK1-NEXT: store i32 2, ptr [[TMP77]], align 4
-// CHECK1-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP74]], ptr [[TMP78]], align 8
-// CHECK1-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP75]], ptr [[TMP79]], align 8
-// CHECK1-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.7, ptr [[TMP80]], align 8
-// CHECK1-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP81]], align 8
-// CHECK1-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP82]], align 8
-// CHECK1-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP83]], align 8
-// CHECK1-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP84]], align 8
-// CHECK1-NEXT: [[TMP85:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118.region_id, ptr [[KERNEL_ARGS13]])
-// CHECK1-NEXT: [[TMP86:%.*]] = icmp ne i32 [[TMP85]], 0
-// CHECK1-NEXT: br i1 [[TMP86]], label [[OMP_OFFLOAD_FAILED14:%.*]], label [[OMP_OFFLOAD_CONT15:%.*]]
+// CHECK1-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP80]], align 4
+// CHECK1-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 1
+// CHECK1-NEXT: store i32 2, ptr [[TMP81]], align 4
+// CHECK1-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP78]], ptr [[TMP82]], align 8
+// CHECK1-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP79]], ptr [[TMP83]], align 8
+// CHECK1-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.7, ptr [[TMP84]], align 8
+// CHECK1-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP85]], align 8
+// CHECK1-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP86]], align 8
+// CHECK1-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP87]], align 8
+// CHECK1-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP88]], align 8
+// CHECK1-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP89]], align 8
+// CHECK1-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP90]], align 4
+// CHECK1-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP91]], align 4
+// CHECK1-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP92]], align 4
+// CHECK1-NEXT: [[TMP93:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118.region_id, ptr [[KERNEL_ARGS13]])
+// CHECK1-NEXT: [[TMP94:%.*]] = icmp ne i32 [[TMP93]], 0
+// CHECK1-NEXT: br i1 [[TMP94]], label [[OMP_OFFLOAD_FAILED14:%.*]], label [[OMP_OFFLOAD_CONT15:%.*]]
// CHECK1: omp_offload.failed14:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118(i64 [[TMP64]], i64 [[TMP66]]) #[[ATTR3]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118(i64 [[TMP68]], i64 [[TMP70]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT15]]
// CHECK1: omp_offload.cont15:
// CHECK1-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK1: omp_if.else:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118(i64 [[TMP64]], i64 [[TMP66]]) #[[ATTR3]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118(i64 [[TMP68]], i64 [[TMP70]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
-// CHECK1-NEXT: [[TMP87:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: store i32 [[TMP87]], ptr [[A_CASTED16]], align 4
-// CHECK1-NEXT: [[TMP88:%.*]] = load i64, ptr [[A_CASTED16]], align 8
-// CHECK1-NEXT: [[TMP89:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: [[CMP17:%.*]] = icmp sgt i32 [[TMP89]], 20
+// CHECK1-NEXT: [[TMP95:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: store i32 [[TMP95]], ptr [[A_CASTED16]], align 4
+// CHECK1-NEXT: [[TMP96:%.*]] = load i64, ptr [[A_CASTED16]], align 8
+// CHECK1-NEXT: [[TMP97:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: [[CMP17:%.*]] = icmp sgt i32 [[TMP97]], 20
// CHECK1-NEXT: br i1 [[CMP17]], label [[OMP_IF_THEN18:%.*]], label [[OMP_IF_ELSE25:%.*]]
// CHECK1: omp_if.then18:
-// CHECK1-NEXT: [[TMP90:%.*]] = mul nuw i64 [[TMP2]], 4
-// CHECK1-NEXT: [[TMP91:%.*]] = mul nuw i64 5, [[TMP5]]
-// CHECK1-NEXT: [[TMP92:%.*]] = mul nuw i64 [[TMP91]], 8
+// CHECK1-NEXT: [[TMP98:%.*]] = mul nuw i64 [[TMP2]], 4
+// CHECK1-NEXT: [[TMP99:%.*]] = mul nuw i64 5, [[TMP5]]
+// CHECK1-NEXT: [[TMP100:%.*]] = mul nuw i64 [[TMP99]], 8
// CHECK1-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES]], ptr align 8 @.offload_sizes.10, i64 72, i1 false)
-// CHECK1-NEXT: [[TMP93:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP88]], ptr [[TMP93]], align 8
-// CHECK1-NEXT: [[TMP94:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP88]], ptr [[TMP94]], align 8
-// CHECK1-NEXT: [[TMP95:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP95]], align 8
-// CHECK1-NEXT: [[TMP96:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
-// CHECK1-NEXT: store ptr [[B]], ptr [[TMP96]], align 8
-// CHECK1-NEXT: [[TMP97:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
-// CHECK1-NEXT: store ptr [[B]], ptr [[TMP97]], align 8
-// CHECK1-NEXT: [[TMP98:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP98]], align 8
-// CHECK1-NEXT: [[TMP99:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
-// CHECK1-NEXT: store i64 [[TMP2]], ptr [[TMP99]], align 8
-// CHECK1-NEXT: [[TMP100:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
-// CHECK1-NEXT: store i64 [[TMP2]], ptr [[TMP100]], align 8
-// CHECK1-NEXT: [[TMP101:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP101]], align 8
-// CHECK1-NEXT: [[TMP102:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[VLA]], ptr [[TMP102]], align 8
-// CHECK1-NEXT: [[TMP103:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[VLA]], ptr [[TMP103]], align 8
-// CHECK1-NEXT: [[TMP104:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
-// CHECK1-NEXT: store i64 [[TMP90]], ptr [[TMP104]], align 8
-// CHECK1-NEXT: [[TMP105:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP105]], align 8
-// CHECK1-NEXT: [[TMP106:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 4
-// CHECK1-NEXT: store ptr [[C]], ptr [[TMP106]], align 8
-// CHECK1-NEXT: [[TMP107:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 4
-// CHECK1-NEXT: store ptr [[C]], ptr [[TMP107]], align 8
-// CHECK1-NEXT: [[TMP108:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP108]], align 8
-// CHECK1-NEXT: [[TMP109:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 5
-// CHECK1-NEXT: store i64 5, ptr [[TMP109]], align 8
-// CHECK1-NEXT: [[TMP110:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 5
-// CHECK1-NEXT: store i64 5, ptr [[TMP110]], align 8
-// CHECK1-NEXT: [[TMP111:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP111]], align 8
-// CHECK1-NEXT: [[TMP112:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 6
-// CHECK1-NEXT: store i64 [[TMP5]], ptr [[TMP112]], align 8
-// CHECK1-NEXT: [[TMP113:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 6
-// CHECK1-NEXT: store i64 [[TMP5]], ptr [[TMP113]], align 8
-// CHECK1-NEXT: [[TMP114:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP114]], align 8
-// CHECK1-NEXT: [[TMP115:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 7
-// CHECK1-NEXT: store ptr [[VLA1]], ptr [[TMP115]], align 8
-// CHECK1-NEXT: [[TMP116:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 7
-// CHECK1-NEXT: store ptr [[VLA1]], ptr [[TMP116]], align 8
-// CHECK1-NEXT: [[TMP117:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
-// CHECK1-NEXT: store i64 [[TMP92]], ptr [[TMP117]], align 8
-// CHECK1-NEXT: [[TMP118:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP118]], align 8
-// CHECK1-NEXT: [[TMP119:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 8
-// CHECK1-NEXT: store ptr [[D]], ptr [[TMP119]], align 8
-// CHECK1-NEXT: [[TMP120:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 8
-// CHECK1-NEXT: store ptr [[D]], ptr [[TMP120]], align 8
-// CHECK1-NEXT: [[TMP121:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 8
-// CHECK1-NEXT: store ptr null, ptr [[TMP121]], align 8
-// CHECK1-NEXT: [[TMP122:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP123:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP124:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP101:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP96]], ptr [[TMP101]], align 8
+// CHECK1-NEXT: [[TMP102:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP96]], ptr [[TMP102]], align 8
+// CHECK1-NEXT: [[TMP103:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP103]], align 8
+// CHECK1-NEXT: [[TMP104:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
+// CHECK1-NEXT: store ptr [[B]], ptr [[TMP104]], align 8
+// CHECK1-NEXT: [[TMP105:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
+// CHECK1-NEXT: store ptr [[B]], ptr [[TMP105]], align 8
+// CHECK1-NEXT: [[TMP106:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP106]], align 8
+// CHECK1-NEXT: [[TMP107:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
+// CHECK1-NEXT: store i64 [[TMP2]], ptr [[TMP107]], align 8
+// CHECK1-NEXT: [[TMP108:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
+// CHECK1-NEXT: store i64 [[TMP2]], ptr [[TMP108]], align 8
+// CHECK1-NEXT: [[TMP109:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 2
+// CHECK1-NEXT: store ptr null, ptr [[TMP109]], align 8
+// CHECK1-NEXT: [[TMP110:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[VLA]], ptr [[TMP110]], align 8
+// CHECK1-NEXT: [[TMP111:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[VLA]], ptr [[TMP111]], align 8
+// CHECK1-NEXT: [[TMP112:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
+// CHECK1-NEXT: store i64 [[TMP98]], ptr [[TMP112]], align 8
+// CHECK1-NEXT: [[TMP113:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 3
+// CHECK1-NEXT: store ptr null, ptr [[TMP113]], align 8
+// CHECK1-NEXT: [[TMP114:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 4
+// CHECK1-NEXT: store ptr [[C]], ptr [[TMP114]], align 8
+// CHECK1-NEXT: [[TMP115:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 4
+// CHECK1-NEXT: store ptr [[C]], ptr [[TMP115]], align 8
+// CHECK1-NEXT: [[TMP116:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP116]], align 8
+// CHECK1-NEXT: [[TMP117:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 5
+// CHECK1-NEXT: store i64 5, ptr [[TMP117]], align 8
+// CHECK1-NEXT: [[TMP118:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 5
+// CHECK1-NEXT: store i64 5, ptr [[TMP118]], align 8
+// CHECK1-NEXT: [[TMP119:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP119]], align 8
+// CHECK1-NEXT: [[TMP120:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 6
+// CHECK1-NEXT: store i64 [[TMP5]], ptr [[TMP120]], align 8
+// CHECK1-NEXT: [[TMP121:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 6
+// CHECK1-NEXT: store i64 [[TMP5]], ptr [[TMP121]], align 8
+// CHECK1-NEXT: [[TMP122:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP122]], align 8
+// CHECK1-NEXT: [[TMP123:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 7
+// CHECK1-NEXT: store ptr [[VLA1]], ptr [[TMP123]], align 8
+// CHECK1-NEXT: [[TMP124:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 7
+// CHECK1-NEXT: store ptr [[VLA1]], ptr [[TMP124]], align 8
+// CHECK1-NEXT: [[TMP125:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
+// CHECK1-NEXT: store i64 [[TMP100]], ptr [[TMP125]], align 8
+// CHECK1-NEXT: [[TMP126:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP126]], align 8
+// CHECK1-NEXT: [[TMP127:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 8
+// CHECK1-NEXT: store ptr [[D]], ptr [[TMP127]], align 8
+// CHECK1-NEXT: [[TMP128:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 8
+// CHECK1-NEXT: store ptr [[D]], ptr [[TMP128]], align 8
+// CHECK1-NEXT: [[TMP129:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 8
+// CHECK1-NEXT: store ptr null, ptr [[TMP129]], align 8
+// CHECK1-NEXT: [[TMP130:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP131:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP132:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS22:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP125]], align 4
-// CHECK1-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 1
-// CHECK1-NEXT: store i32 9, ptr [[TMP126]], align 4
-// CHECK1-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP122]], ptr [[TMP127]], align 8
-// CHECK1-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP123]], ptr [[TMP128]], align 8
-// CHECK1-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 4
-// CHECK1-NEXT: store ptr [[TMP124]], ptr [[TMP129]], align 8
-// CHECK1-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.11, ptr [[TMP130]], align 8
-// CHECK1-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP131]], align 8
-// CHECK1-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP132]], align 8
-// CHECK1-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP133]], align 8
-// CHECK1-NEXT: [[TMP134:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142.region_id, ptr [[KERNEL_ARGS22]])
-// CHECK1-NEXT: [[TMP135:%.*]] = icmp ne i32 [[TMP134]], 0
-// CHECK1-NEXT: br i1 [[TMP135]], label [[OMP_OFFLOAD_FAILED23:%.*]], label [[OMP_OFFLOAD_CONT24:%.*]]
+// CHECK1-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP133]], align 4
+// CHECK1-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 1
+// CHECK1-NEXT: store i32 9, ptr [[TMP134]], align 4
+// CHECK1-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP130]], ptr [[TMP135]], align 8
+// CHECK1-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP131]], ptr [[TMP136]], align 8
+// CHECK1-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 4
+// CHECK1-NEXT: store ptr [[TMP132]], ptr [[TMP137]], align 8
+// CHECK1-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.11, ptr [[TMP138]], align 8
+// CHECK1-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP139]], align 8
+// CHECK1-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP140]], align 8
+// CHECK1-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP141]], align 8
+// CHECK1-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP142]], align 8
+// CHECK1-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP143]], align 4
+// CHECK1-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP144]], align 4
+// CHECK1-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP145]], align 4
+// CHECK1-NEXT: [[TMP146:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142.region_id, ptr [[KERNEL_ARGS22]])
+// CHECK1-NEXT: [[TMP147:%.*]] = icmp ne i32 [[TMP146]], 0
+// CHECK1-NEXT: br i1 [[TMP147]], label [[OMP_OFFLOAD_FAILED23:%.*]], label [[OMP_OFFLOAD_CONT24:%.*]]
// CHECK1: omp_offload.failed23:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142(i64 [[TMP88]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]]) #[[ATTR3]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142(i64 [[TMP96]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT24]]
// CHECK1: omp_offload.cont24:
// CHECK1-NEXT: br label [[OMP_IF_END26:%.*]]
// CHECK1: omp_if.else25:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142(i64 [[TMP88]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]]) #[[ATTR3]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142(i64 [[TMP96]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_IF_END26]]
// CHECK1: omp_if.end26:
// CHECK1-NEXT: store i32 0, ptr [[NN]], align 4
-// CHECK1-NEXT: [[TMP136:%.*]] = load i32, ptr [[NN]], align 4
-// CHECK1-NEXT: store i32 [[TMP136]], ptr [[NN_CASTED]], align 4
-// CHECK1-NEXT: [[TMP137:%.*]] = load i64, ptr [[NN_CASTED]], align 8
-// CHECK1-NEXT: [[TMP138:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP137]], ptr [[TMP138]], align 8
-// CHECK1-NEXT: [[TMP139:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP137]], ptr [[TMP139]], align 8
-// CHECK1-NEXT: [[TMP140:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP140]], align 8
-// CHECK1-NEXT: [[TMP141:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP142:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP148:%.*]] = load i32, ptr [[NN]], align 4
+// CHECK1-NEXT: store i32 [[TMP148]], ptr [[NN_CASTED]], align 4
+// CHECK1-NEXT: [[TMP149:%.*]] = load i64, ptr [[NN_CASTED]], align 8
+// CHECK1-NEXT: [[TMP150:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP149]], ptr [[TMP150]], align 8
+// CHECK1-NEXT: [[TMP151:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP149]], ptr [[TMP151]], align 8
+// CHECK1-NEXT: [[TMP152:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP152]], align 8
+// CHECK1-NEXT: [[TMP153:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP154:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS30:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP143]], align 4
-// CHECK1-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP144]], align 4
-// CHECK1-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP141]], ptr [[TMP145]], align 8
-// CHECK1-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP142]], ptr [[TMP146]], align 8
-// CHECK1-NEXT: [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.14, ptr [[TMP147]], align 8
-// CHECK1-NEXT: [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.15, ptr [[TMP148]], align 8
-// CHECK1-NEXT: [[TMP149:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP149]], align 8
-// CHECK1-NEXT: [[TMP150:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP150]], align 8
-// CHECK1-NEXT: [[TMP151:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP151]], align 8
-// CHECK1-NEXT: [[TMP152:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l154.region_id, ptr [[KERNEL_ARGS30]])
-// CHECK1-NEXT: [[TMP153:%.*]] = icmp ne i32 [[TMP152]], 0
-// CHECK1-NEXT: br i1 [[TMP153]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
+// CHECK1-NEXT: [[TMP155:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP155]], align 4
+// CHECK1-NEXT: [[TMP156:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP156]], align 4
+// CHECK1-NEXT: [[TMP157:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP153]], ptr [[TMP157]], align 8
+// CHECK1-NEXT: [[TMP158:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP154]], ptr [[TMP158]], align 8
+// CHECK1-NEXT: [[TMP159:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.14, ptr [[TMP159]], align 8
+// CHECK1-NEXT: [[TMP160:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.15, ptr [[TMP160]], align 8
+// CHECK1-NEXT: [[TMP161:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP161]], align 8
+// CHECK1-NEXT: [[TMP162:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP162]], align 8
+// CHECK1-NEXT: [[TMP163:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP163]], align 8
+// CHECK1-NEXT: [[TMP164:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP164]], align 8
+// CHECK1-NEXT: [[TMP165:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP165]], align 4
+// CHECK1-NEXT: [[TMP166:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP166]], align 4
+// CHECK1-NEXT: [[TMP167:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP167]], align 4
+// CHECK1-NEXT: [[TMP168:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l154.region_id, ptr [[KERNEL_ARGS30]])
+// CHECK1-NEXT: [[TMP169:%.*]] = icmp ne i32 [[TMP168]], 0
+// CHECK1-NEXT: br i1 [[TMP169]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
// CHECK1: omp_offload.failed31:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l154(i64 [[TMP137]]) #[[ATTR3]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l154(i64 [[TMP149]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT32]]
// CHECK1: omp_offload.cont32:
-// CHECK1-NEXT: [[TMP154:%.*]] = load i32, ptr [[NN]], align 4
-// CHECK1-NEXT: store i32 [[TMP154]], ptr [[NN_CASTED33]], align 4
-// CHECK1-NEXT: [[TMP155:%.*]] = load i64, ptr [[NN_CASTED33]], align 8
-// CHECK1-NEXT: [[TMP156:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP155]], ptr [[TMP156]], align 8
-// CHECK1-NEXT: [[TMP157:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP155]], ptr [[TMP157]], align 8
-// CHECK1-NEXT: [[TMP158:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP158]], align 8
-// CHECK1-NEXT: [[TMP159:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP160:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP170:%.*]] = load i32, ptr [[NN]], align 4
+// CHECK1-NEXT: store i32 [[TMP170]], ptr [[NN_CASTED33]], align 4
+// CHECK1-NEXT: [[TMP171:%.*]] = load i64, ptr [[NN_CASTED33]], align 8
+// CHECK1-NEXT: [[TMP172:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP171]], ptr [[TMP172]], align 8
+// CHECK1-NEXT: [[TMP173:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP171]], ptr [[TMP173]], align 8
+// CHECK1-NEXT: [[TMP174:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP174]], align 8
+// CHECK1-NEXT: [[TMP175:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP176:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS37:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP161:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP161]], align 4
-// CHECK1-NEXT: [[TMP162:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP162]], align 4
-// CHECK1-NEXT: [[TMP163:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP159]], ptr [[TMP163]], align 8
-// CHECK1-NEXT: [[TMP164:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP160]], ptr [[TMP164]], align 8
-// CHECK1-NEXT: [[TMP165:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.18, ptr [[TMP165]], align 8
-// CHECK1-NEXT: [[TMP166:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.19, ptr [[TMP166]], align 8
-// CHECK1-NEXT: [[TMP167:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP167]], align 8
-// CHECK1-NEXT: [[TMP168:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP168]], align 8
-// CHECK1-NEXT: [[TMP169:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP169]], align 8
-// CHECK1-NEXT: [[TMP170:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l157.region_id, ptr [[KERNEL_ARGS37]])
-// CHECK1-NEXT: [[TMP171:%.*]] = icmp ne i32 [[TMP170]], 0
-// CHECK1-NEXT: br i1 [[TMP171]], label [[OMP_OFFLOAD_FAILED38:%.*]], label [[OMP_OFFLOAD_CONT39:%.*]]
+// CHECK1-NEXT: [[TMP177:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP177]], align 4
+// CHECK1-NEXT: [[TMP178:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP178]], align 4
+// CHECK1-NEXT: [[TMP179:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP175]], ptr [[TMP179]], align 8
+// CHECK1-NEXT: [[TMP180:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP176]], ptr [[TMP180]], align 8
+// CHECK1-NEXT: [[TMP181:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.18, ptr [[TMP181]], align 8
+// CHECK1-NEXT: [[TMP182:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.19, ptr [[TMP182]], align 8
+// CHECK1-NEXT: [[TMP183:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP183]], align 8
+// CHECK1-NEXT: [[TMP184:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP184]], align 8
+// CHECK1-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP185]], align 8
+// CHECK1-NEXT: [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP186]], align 8
+// CHECK1-NEXT: [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP187]], align 4
+// CHECK1-NEXT: [[TMP188:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP188]], align 4
+// CHECK1-NEXT: [[TMP189:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP189]], align 4
+// CHECK1-NEXT: [[TMP190:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l157.region_id, ptr [[KERNEL_ARGS37]])
+// CHECK1-NEXT: [[TMP191:%.*]] = icmp ne i32 [[TMP190]], 0
+// CHECK1-NEXT: br i1 [[TMP191]], label [[OMP_OFFLOAD_FAILED38:%.*]], label [[OMP_OFFLOAD_CONT39:%.*]]
// CHECK1: omp_offload.failed38:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l157(i64 [[TMP155]]) #[[ATTR3]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l157(i64 [[TMP171]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT39]]
// CHECK1: omp_offload.cont39:
-// CHECK1-NEXT: [[TMP172:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: [[TMP173:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK1-NEXT: call void @llvm.stackrestore(ptr [[TMP173]])
-// CHECK1-NEXT: ret i32 [[TMP172]]
+// CHECK1-NEXT: [[TMP192:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: [[TMP193:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK1-NEXT: call void @llvm.stackrestore(ptr [[TMP193]])
+// CHECK1-NEXT: ret i32 [[TMP192]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l101
@@ -812,37 +852,47 @@ int bar(int n){
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[TMP9]], i32 0, i32 2
// CHECK1-NEXT: [[TMP18:%.*]] = load i32, ptr [[TMP16]], align 4
// CHECK1-NEXT: [[TMP19:%.*]] = load i32, ptr [[TMP17]], align 4
-// CHECK1-NEXT: store i32 1, ptr [[KERNEL_ARGS_I]], align 4, !noalias !24
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 1
-// CHECK1-NEXT: store i32 3, ptr [[TMP20]], align 4, !noalias !24
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP13]], ptr [[TMP21]], align 8, !noalias !24
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP14]], ptr [[TMP22]], align 8, !noalias !24
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 4
-// CHECK1-NEXT: store ptr [[TMP15]], ptr [[TMP23]], align 8, !noalias !24
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP24]], align 8, !noalias !24
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP25]], align 8, !noalias !24
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP26]], align 8, !noalias !24
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP27]], align 8, !noalias !24
-// CHECK1-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel_nowait(ptr @[[GLOB1]], i64 -1, i32 [[TMP18]], i32 [[TMP19]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l101.region_id, ptr [[KERNEL_ARGS_I]], i32 0, ptr null, i32 0, ptr null)
-// CHECK1-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
-// CHECK1-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__1_EXIT:%.*]]
+// CHECK1-NEXT: [[TMP20:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP18]], 0
+// CHECK1-NEXT: [[TMP21:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP19]], 0
+// CHECK1-NEXT: store i32 2, ptr [[KERNEL_ARGS_I]], align 4, !noalias !24
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 1
+// CHECK1-NEXT: store i32 3, ptr [[TMP22]], align 4, !noalias !24
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP13]], ptr [[TMP23]], align 8, !noalias !24
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP14]], ptr [[TMP24]], align 8, !noalias !24
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 4
+// CHECK1-NEXT: store ptr [[TMP15]], ptr [[TMP25]], align 8, !noalias !24
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP26]], align 8, !noalias !24
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8, !noalias !24
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8, !noalias !24
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP29]], align 8, !noalias !24
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 9
+// CHECK1-NEXT: store i64 1, ptr [[TMP30]], align 8, !noalias !24
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [[TMP20]], ptr [[TMP31]], align 4, !noalias !24
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP21]], ptr [[TMP32]], align 4, !noalias !24
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP33]], align 4, !noalias !24
+// CHECK1-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP18]], i32 [[TMP19]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l101.region_id, ptr [[KERNEL_ARGS_I]])
+// CHECK1-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK1-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__1_EXIT:%.*]]
// CHECK1: omp_offload.failed.i:
-// CHECK1-NEXT: [[TMP30:%.*]] = load i16, ptr [[TMP12]], align 2
-// CHECK1-NEXT: store i16 [[TMP30]], ptr [[AA_CASTED_I]], align 2, !noalias !24
-// CHECK1-NEXT: [[TMP31:%.*]] = load i64, ptr [[AA_CASTED_I]], align 8, !noalias !24
-// CHECK1-NEXT: [[TMP32:%.*]] = load i32, ptr [[TMP16]], align 4
-// CHECK1-NEXT: store i32 [[TMP32]], ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 4, !noalias !24
-// CHECK1-NEXT: [[TMP33:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 8, !noalias !24
-// CHECK1-NEXT: [[TMP34:%.*]] = load i32, ptr [[TMP17]], align 4
-// CHECK1-NEXT: store i32 [[TMP34]], ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 4, !noalias !24
-// CHECK1-NEXT: [[TMP35:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 8, !noalias !24
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l101(i64 [[TMP31]], i64 [[TMP33]], i64 [[TMP35]]) #[[ATTR3]]
+// CHECK1-NEXT: [[TMP36:%.*]] = load i16, ptr [[TMP12]], align 2
+// CHECK1-NEXT: store i16 [[TMP36]], ptr [[AA_CASTED_I]], align 2, !noalias !24
+// CHECK1-NEXT: [[TMP37:%.*]] = load i64, ptr [[AA_CASTED_I]], align 8, !noalias !24
+// CHECK1-NEXT: [[TMP38:%.*]] = load i32, ptr [[TMP16]], align 4
+// CHECK1-NEXT: store i32 [[TMP38]], ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 4, !noalias !24
+// CHECK1-NEXT: [[TMP39:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 8, !noalias !24
+// CHECK1-NEXT: [[TMP40:%.*]] = load i32, ptr [[TMP17]], align 4
+// CHECK1-NEXT: store i32 [[TMP40]], ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 4, !noalias !24
+// CHECK1-NEXT: [[TMP41:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 8, !noalias !24
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l101(i64 [[TMP37]], i64 [[TMP39]], i64 [[TMP41]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[DOTOMP_OUTLINED__1_EXIT]]
// CHECK1: .omp_outlined..1.exit:
// CHECK1-NEXT: ret i32 0
@@ -1159,7 +1209,7 @@ int bar(int n){
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1176,9 +1226,17 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z6bazzzziPi_l182.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK1-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z6bazzzziPi_l182.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z6bazzzziPi_l182(i64 [[TMP1]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1316,7 +1374,7 @@ int bar(int n){
// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [5 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP29]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP29]], align 4
// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 5, ptr [[TMP30]], align 4
// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1333,9 +1391,17 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP36]], align 8
// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l227.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK1-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP38]], align 8
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP39]], align 4
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP40]], align 4
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP41]], align 4
+// CHECK1-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l227.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
+// CHECK1-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l227(ptr [[THIS1]], i64 [[TMP6]], i64 2, i64 [[TMP2]], ptr [[VLA]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1345,15 +1411,15 @@ int bar(int n){
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l227(ptr [[THIS1]], i64 [[TMP6]], i64 2, i64 [[TMP2]], ptr [[VLA]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
-// CHECK1-NEXT: [[TMP40:%.*]] = mul nsw i64 1, [[TMP2]]
-// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i64 [[TMP40]]
+// CHECK1-NEXT: [[TMP44:%.*]] = mul nsw i64 1, [[TMP2]]
+// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i64 [[TMP44]]
// CHECK1-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i16, ptr [[ARRAYIDX]], i64 1
-// CHECK1-NEXT: [[TMP41:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
-// CHECK1-NEXT: [[CONV:%.*]] = sext i16 [[TMP41]] to i32
-// CHECK1-NEXT: [[TMP42:%.*]] = load i32, ptr [[B]], align 4
-// CHECK1-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP42]]
-// CHECK1-NEXT: [[TMP43:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK1-NEXT: call void @llvm.stackrestore(ptr [[TMP43]])
+// CHECK1-NEXT: [[TMP45:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
+// CHECK1-NEXT: [[CONV:%.*]] = sext i16 [[TMP45]] to i32
+// CHECK1-NEXT: [[TMP46:%.*]] = load i32, ptr [[B]], align 4
+// CHECK1-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP46]]
+// CHECK1-NEXT: [[TMP47:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK1-NEXT: call void @llvm.stackrestore(ptr [[TMP47]])
// CHECK1-NEXT: ret i32 [[ADD3]]
//
//
@@ -1416,7 +1482,7 @@ int bar(int n){
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 4, ptr [[TMP22]], align 4
// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1433,9 +1499,17 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l209.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK1-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l209.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK1-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l209(i64 [[TMP1]], i64 [[TMP3]], i64 [[TMP5]], ptr [[B]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1445,8 +1519,8 @@ int bar(int n){
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l209(i64 [[TMP1]], i64 [[TMP3]], i64 [[TMP5]], ptr [[B]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
-// CHECK1-NEXT: [[TMP32:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: ret i32 [[TMP32]]
+// CHECK1-NEXT: [[TMP36:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: ret i32 [[TMP36]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_Z9ftemplateIiET_i
@@ -1496,7 +1570,7 @@ int bar(int n){
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP16]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP16]], align 4
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 3, ptr [[TMP17]], align 4
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1513,9 +1587,17 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP23]], align 8
// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l192.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK1-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP25]], align 8
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP28]], align 4
+// CHECK1-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l192.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK1-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l192(i64 [[TMP1]], i64 [[TMP3]], ptr [[B]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1525,8 +1607,8 @@ int bar(int n){
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l192(i64 [[TMP1]], i64 [[TMP3]], ptr [[B]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
-// CHECK1-NEXT: [[TMP27:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: ret i32 [[TMP27]]
+// CHECK1-NEXT: [[TMP31:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: ret i32 [[TMP31]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l227
@@ -1847,7 +1929,7 @@ int bar(int n){
// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP50]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP50]], align 4
// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP51]], align 4
// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1864,252 +1946,292 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP57]], align 4
// CHECK3-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP58]], align 8
-// CHECK3-NEXT: [[TMP59:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l111.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP60:%.*]] = icmp ne i32 [[TMP59]], 0
-// CHECK3-NEXT: br i1 [[TMP60]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP59]], align 8
+// CHECK3-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP60]], align 4
+// CHECK3-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4
+// CHECK3-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP62]], align 4
+// CHECK3-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l111.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
+// CHECK3-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l111(i32 [[TMP44]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
-// CHECK3-NEXT: [[TMP61:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: store i32 [[TMP61]], ptr [[A_CASTED8]], align 4
-// CHECK3-NEXT: [[TMP62:%.*]] = load i32, ptr [[A_CASTED8]], align 4
-// CHECK3-NEXT: [[TMP63:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK3-NEXT: store i16 [[TMP63]], ptr [[AA_CASTED9]], align 2
-// CHECK3-NEXT: [[TMP64:%.*]] = load i32, ptr [[AA_CASTED9]], align 4
-// CHECK3-NEXT: [[TMP65:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP65]], 10
+// CHECK3-NEXT: [[TMP65:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: store i32 [[TMP65]], ptr [[A_CASTED8]], align 4
+// CHECK3-NEXT: [[TMP66:%.*]] = load i32, ptr [[A_CASTED8]], align 4
+// CHECK3-NEXT: [[TMP67:%.*]] = load i16, ptr [[AA]], align 2
+// CHECK3-NEXT: store i16 [[TMP67]], ptr [[AA_CASTED9]], align 2
+// CHECK3-NEXT: [[TMP68:%.*]] = load i32, ptr [[AA_CASTED9]], align 4
+// CHECK3-NEXT: [[TMP69:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP69]], 10
// CHECK3-NEXT: br i1 [[CMP]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK3: omp_if.then:
-// CHECK3-NEXT: [[TMP66:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP62]], ptr [[TMP66]], align 4
-// CHECK3-NEXT: [[TMP67:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP62]], ptr [[TMP67]], align 4
-// CHECK3-NEXT: [[TMP68:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP68]], align 4
-// CHECK3-NEXT: [[TMP69:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP64]], ptr [[TMP69]], align 4
-// CHECK3-NEXT: [[TMP70:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP64]], ptr [[TMP70]], align 4
-// CHECK3-NEXT: [[TMP71:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i32 0, i32 1
-// CHECK3-NEXT: store ptr null, ptr [[TMP71]], align 4
-// CHECK3-NEXT: [[TMP72:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP70:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP66]], ptr [[TMP70]], align 4
+// CHECK3-NEXT: [[TMP71:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP66]], ptr [[TMP71]], align 4
+// CHECK3-NEXT: [[TMP72:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP72]], align 4
+// CHECK3-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP68]], ptr [[TMP73]], align 4
+// CHECK3-NEXT: [[TMP74:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP68]], ptr [[TMP74]], align 4
+// CHECK3-NEXT: [[TMP75:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i32 0, i32 1
+// CHECK3-NEXT: store ptr null, ptr [[TMP75]], align 4
+// CHECK3-NEXT: [[TMP76:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP77:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS13:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP74]], align 4
-// CHECK3-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 1
-// CHECK3-NEXT: store i32 2, ptr [[TMP75]], align 4
-// CHECK3-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP72]], ptr [[TMP76]], align 4
-// CHECK3-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP73]], ptr [[TMP77]], align 4
-// CHECK3-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.7, ptr [[TMP78]], align 4
-// CHECK3-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP79]], align 4
-// CHECK3-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP80]], align 4
-// CHECK3-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP81]], align 4
-// CHECK3-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP82]], align 8
-// CHECK3-NEXT: [[TMP83:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118.region_id, ptr [[KERNEL_ARGS13]])
-// CHECK3-NEXT: [[TMP84:%.*]] = icmp ne i32 [[TMP83]], 0
-// CHECK3-NEXT: br i1 [[TMP84]], label [[OMP_OFFLOAD_FAILED14:%.*]], label [[OMP_OFFLOAD_CONT15:%.*]]
+// CHECK3-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP78]], align 4
+// CHECK3-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 1
+// CHECK3-NEXT: store i32 2, ptr [[TMP79]], align 4
+// CHECK3-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP76]], ptr [[TMP80]], align 4
+// CHECK3-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP77]], ptr [[TMP81]], align 4
+// CHECK3-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.7, ptr [[TMP82]], align 4
+// CHECK3-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP83]], align 4
+// CHECK3-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP84]], align 4
+// CHECK3-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP85]], align 4
+// CHECK3-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP86]], align 8
+// CHECK3-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP87]], align 8
+// CHECK3-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP88]], align 4
+// CHECK3-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP89]], align 4
+// CHECK3-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS13]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP90]], align 4
+// CHECK3-NEXT: [[TMP91:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118.region_id, ptr [[KERNEL_ARGS13]])
+// CHECK3-NEXT: [[TMP92:%.*]] = icmp ne i32 [[TMP91]], 0
+// CHECK3-NEXT: br i1 [[TMP92]], label [[OMP_OFFLOAD_FAILED14:%.*]], label [[OMP_OFFLOAD_CONT15:%.*]]
// CHECK3: omp_offload.failed14:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118(i32 [[TMP62]], i32 [[TMP64]]) #[[ATTR3]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118(i32 [[TMP66]], i32 [[TMP68]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT15]]
// CHECK3: omp_offload.cont15:
// CHECK3-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK3: omp_if.else:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118(i32 [[TMP62]], i32 [[TMP64]]) #[[ATTR3]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118(i32 [[TMP66]], i32 [[TMP68]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
-// CHECK3-NEXT: [[TMP85:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: store i32 [[TMP85]], ptr [[A_CASTED16]], align 4
-// CHECK3-NEXT: [[TMP86:%.*]] = load i32, ptr [[A_CASTED16]], align 4
-// CHECK3-NEXT: [[TMP87:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: [[CMP17:%.*]] = icmp sgt i32 [[TMP87]], 20
+// CHECK3-NEXT: [[TMP93:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: store i32 [[TMP93]], ptr [[A_CASTED16]], align 4
+// CHECK3-NEXT: [[TMP94:%.*]] = load i32, ptr [[A_CASTED16]], align 4
+// CHECK3-NEXT: [[TMP95:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: [[CMP17:%.*]] = icmp sgt i32 [[TMP95]], 20
// CHECK3-NEXT: br i1 [[CMP17]], label [[OMP_IF_THEN18:%.*]], label [[OMP_IF_ELSE25:%.*]]
// CHECK3: omp_if.then18:
-// CHECK3-NEXT: [[TMP88:%.*]] = mul nuw i32 [[TMP1]], 4
-// CHECK3-NEXT: [[TMP89:%.*]] = sext i32 [[TMP88]] to i64
-// CHECK3-NEXT: [[TMP90:%.*]] = mul nuw i32 5, [[TMP3]]
-// CHECK3-NEXT: [[TMP91:%.*]] = mul nuw i32 [[TMP90]], 8
-// CHECK3-NEXT: [[TMP92:%.*]] = sext i32 [[TMP91]] to i64
+// CHECK3-NEXT: [[TMP96:%.*]] = mul nuw i32 [[TMP1]], 4
+// CHECK3-NEXT: [[TMP97:%.*]] = sext i32 [[TMP96]] to i64
+// CHECK3-NEXT: [[TMP98:%.*]] = mul nuw i32 5, [[TMP3]]
+// CHECK3-NEXT: [[TMP99:%.*]] = mul nuw i32 [[TMP98]], 8
+// CHECK3-NEXT: [[TMP100:%.*]] = sext i32 [[TMP99]] to i64
// CHECK3-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES]], ptr align 4 @.offload_sizes.10, i32 72, i1 false)
-// CHECK3-NEXT: [[TMP93:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP86]], ptr [[TMP93]], align 4
-// CHECK3-NEXT: [[TMP94:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP86]], ptr [[TMP94]], align 4
-// CHECK3-NEXT: [[TMP95:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP95]], align 4
-// CHECK3-NEXT: [[TMP96:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
-// CHECK3-NEXT: store ptr [[B]], ptr [[TMP96]], align 4
-// CHECK3-NEXT: [[TMP97:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
-// CHECK3-NEXT: store ptr [[B]], ptr [[TMP97]], align 4
-// CHECK3-NEXT: [[TMP98:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 1
-// CHECK3-NEXT: store ptr null, ptr [[TMP98]], align 4
-// CHECK3-NEXT: [[TMP99:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
-// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP99]], align 4
-// CHECK3-NEXT: [[TMP100:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
-// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP100]], align 4
-// CHECK3-NEXT: [[TMP101:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 2
-// CHECK3-NEXT: store ptr null, ptr [[TMP101]], align 4
-// CHECK3-NEXT: [[TMP102:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[VLA]], ptr [[TMP102]], align 4
-// CHECK3-NEXT: [[TMP103:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[VLA]], ptr [[TMP103]], align 4
-// CHECK3-NEXT: [[TMP104:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
-// CHECK3-NEXT: store i64 [[TMP89]], ptr [[TMP104]], align 4
-// CHECK3-NEXT: [[TMP105:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 3
-// CHECK3-NEXT: store ptr null, ptr [[TMP105]], align 4
-// CHECK3-NEXT: [[TMP106:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 4
-// CHECK3-NEXT: store ptr [[C]], ptr [[TMP106]], align 4
-// CHECK3-NEXT: [[TMP107:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 4
-// CHECK3-NEXT: store ptr [[C]], ptr [[TMP107]], align 4
-// CHECK3-NEXT: [[TMP108:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 4
-// CHECK3-NEXT: store ptr null, ptr [[TMP108]], align 4
-// CHECK3-NEXT: [[TMP109:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 5
-// CHECK3-NEXT: store i32 5, ptr [[TMP109]], align 4
-// CHECK3-NEXT: [[TMP110:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 5
-// CHECK3-NEXT: store i32 5, ptr [[TMP110]], align 4
-// CHECK3-NEXT: [[TMP111:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 5
-// CHECK3-NEXT: store ptr null, ptr [[TMP111]], align 4
-// CHECK3-NEXT: [[TMP112:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 6
-// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP112]], align 4
-// CHECK3-NEXT: [[TMP113:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 6
-// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP113]], align 4
-// CHECK3-NEXT: [[TMP114:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP114]], align 4
-// CHECK3-NEXT: [[TMP115:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 7
-// CHECK3-NEXT: store ptr [[VLA1]], ptr [[TMP115]], align 4
-// CHECK3-NEXT: [[TMP116:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 7
-// CHECK3-NEXT: store ptr [[VLA1]], ptr [[TMP116]], align 4
-// CHECK3-NEXT: [[TMP117:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
-// CHECK3-NEXT: store i64 [[TMP92]], ptr [[TMP117]], align 4
-// CHECK3-NEXT: [[TMP118:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP118]], align 4
-// CHECK3-NEXT: [[TMP119:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 8
-// CHECK3-NEXT: store ptr [[D]], ptr [[TMP119]], align 4
-// CHECK3-NEXT: [[TMP120:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 8
-// CHECK3-NEXT: store ptr [[D]], ptr [[TMP120]], align 4
-// CHECK3-NEXT: [[TMP121:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 8
-// CHECK3-NEXT: store ptr null, ptr [[TMP121]], align 4
-// CHECK3-NEXT: [[TMP122:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP123:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP124:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP101:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP94]], ptr [[TMP101]], align 4
+// CHECK3-NEXT: [[TMP102:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP94]], ptr [[TMP102]], align 4
+// CHECK3-NEXT: [[TMP103:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP103]], align 4
+// CHECK3-NEXT: [[TMP104:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
+// CHECK3-NEXT: store ptr [[B]], ptr [[TMP104]], align 4
+// CHECK3-NEXT: [[TMP105:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
+// CHECK3-NEXT: store ptr [[B]], ptr [[TMP105]], align 4
+// CHECK3-NEXT: [[TMP106:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 1
+// CHECK3-NEXT: store ptr null, ptr [[TMP106]], align 4
+// CHECK3-NEXT: [[TMP107:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
+// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP107]], align 4
+// CHECK3-NEXT: [[TMP108:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
+// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP108]], align 4
+// CHECK3-NEXT: [[TMP109:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 2
+// CHECK3-NEXT: store ptr null, ptr [[TMP109]], align 4
+// CHECK3-NEXT: [[TMP110:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[VLA]], ptr [[TMP110]], align 4
+// CHECK3-NEXT: [[TMP111:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[VLA]], ptr [[TMP111]], align 4
+// CHECK3-NEXT: [[TMP112:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
+// CHECK3-NEXT: store i64 [[TMP97]], ptr [[TMP112]], align 4
+// CHECK3-NEXT: [[TMP113:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 3
+// CHECK3-NEXT: store ptr null, ptr [[TMP113]], align 4
+// CHECK3-NEXT: [[TMP114:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 4
+// CHECK3-NEXT: store ptr [[C]], ptr [[TMP114]], align 4
+// CHECK3-NEXT: [[TMP115:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 4
+// CHECK3-NEXT: store ptr [[C]], ptr [[TMP115]], align 4
+// CHECK3-NEXT: [[TMP116:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 4
+// CHECK3-NEXT: store ptr null, ptr [[TMP116]], align 4
+// CHECK3-NEXT: [[TMP117:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 5
+// CHECK3-NEXT: store i32 5, ptr [[TMP117]], align 4
+// CHECK3-NEXT: [[TMP118:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 5
+// CHECK3-NEXT: store i32 5, ptr [[TMP118]], align 4
+// CHECK3-NEXT: [[TMP119:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 5
+// CHECK3-NEXT: store ptr null, ptr [[TMP119]], align 4
+// CHECK3-NEXT: [[TMP120:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 6
+// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP120]], align 4
+// CHECK3-NEXT: [[TMP121:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 6
+// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP121]], align 4
+// CHECK3-NEXT: [[TMP122:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP122]], align 4
+// CHECK3-NEXT: [[TMP123:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 7
+// CHECK3-NEXT: store ptr [[VLA1]], ptr [[TMP123]], align 4
+// CHECK3-NEXT: [[TMP124:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 7
+// CHECK3-NEXT: store ptr [[VLA1]], ptr [[TMP124]], align 4
+// CHECK3-NEXT: [[TMP125:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
+// CHECK3-NEXT: store i64 [[TMP100]], ptr [[TMP125]], align 4
+// CHECK3-NEXT: [[TMP126:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP126]], align 4
+// CHECK3-NEXT: [[TMP127:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 8
+// CHECK3-NEXT: store ptr [[D]], ptr [[TMP127]], align 4
+// CHECK3-NEXT: [[TMP128:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 8
+// CHECK3-NEXT: store ptr [[D]], ptr [[TMP128]], align 4
+// CHECK3-NEXT: [[TMP129:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 8
+// CHECK3-NEXT: store ptr null, ptr [[TMP129]], align 4
+// CHECK3-NEXT: [[TMP130:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP131:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP132:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS22:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP125]], align 4
-// CHECK3-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 1
-// CHECK3-NEXT: store i32 9, ptr [[TMP126]], align 4
-// CHECK3-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP122]], ptr [[TMP127]], align 4
-// CHECK3-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP123]], ptr [[TMP128]], align 4
-// CHECK3-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 4
-// CHECK3-NEXT: store ptr [[TMP124]], ptr [[TMP129]], align 4
-// CHECK3-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.11, ptr [[TMP130]], align 4
-// CHECK3-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP131]], align 4
-// CHECK3-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP132]], align 4
-// CHECK3-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP133]], align 8
-// CHECK3-NEXT: [[TMP134:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142.region_id, ptr [[KERNEL_ARGS22]])
-// CHECK3-NEXT: [[TMP135:%.*]] = icmp ne i32 [[TMP134]], 0
-// CHECK3-NEXT: br i1 [[TMP135]], label [[OMP_OFFLOAD_FAILED23:%.*]], label [[OMP_OFFLOAD_CONT24:%.*]]
+// CHECK3-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP133]], align 4
+// CHECK3-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 1
+// CHECK3-NEXT: store i32 9, ptr [[TMP134]], align 4
+// CHECK3-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP130]], ptr [[TMP135]], align 4
+// CHECK3-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP131]], ptr [[TMP136]], align 4
+// CHECK3-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 4
+// CHECK3-NEXT: store ptr [[TMP132]], ptr [[TMP137]], align 4
+// CHECK3-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.11, ptr [[TMP138]], align 4
+// CHECK3-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP139]], align 4
+// CHECK3-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP140]], align 4
+// CHECK3-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP141]], align 8
+// CHECK3-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP142]], align 8
+// CHECK3-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP143]], align 4
+// CHECK3-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP144]], align 4
+// CHECK3-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS22]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP145]], align 4
+// CHECK3-NEXT: [[TMP146:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142.region_id, ptr [[KERNEL_ARGS22]])
+// CHECK3-NEXT: [[TMP147:%.*]] = icmp ne i32 [[TMP146]], 0
+// CHECK3-NEXT: br i1 [[TMP147]], label [[OMP_OFFLOAD_FAILED23:%.*]], label [[OMP_OFFLOAD_CONT24:%.*]]
// CHECK3: omp_offload.failed23:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142(i32 [[TMP86]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]]) #[[ATTR3]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142(i32 [[TMP94]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT24]]
// CHECK3: omp_offload.cont24:
// CHECK3-NEXT: br label [[OMP_IF_END26:%.*]]
// CHECK3: omp_if.else25:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142(i32 [[TMP86]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]]) #[[ATTR3]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142(i32 [[TMP94]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_IF_END26]]
// CHECK3: omp_if.end26:
// CHECK3-NEXT: store i32 0, ptr [[NN]], align 4
-// CHECK3-NEXT: [[TMP136:%.*]] = load i32, ptr [[NN]], align 4
-// CHECK3-NEXT: store i32 [[TMP136]], ptr [[NN_CASTED]], align 4
-// CHECK3-NEXT: [[TMP137:%.*]] = load i32, ptr [[NN_CASTED]], align 4
-// CHECK3-NEXT: [[TMP138:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP137]], ptr [[TMP138]], align 4
-// CHECK3-NEXT: [[TMP139:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP137]], ptr [[TMP139]], align 4
-// CHECK3-NEXT: [[TMP140:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP140]], align 4
-// CHECK3-NEXT: [[TMP141:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP142:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP148:%.*]] = load i32, ptr [[NN]], align 4
+// CHECK3-NEXT: store i32 [[TMP148]], ptr [[NN_CASTED]], align 4
+// CHECK3-NEXT: [[TMP149:%.*]] = load i32, ptr [[NN_CASTED]], align 4
+// CHECK3-NEXT: [[TMP150:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP149]], ptr [[TMP150]], align 4
+// CHECK3-NEXT: [[TMP151:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP149]], ptr [[TMP151]], align 4
+// CHECK3-NEXT: [[TMP152:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP152]], align 4
+// CHECK3-NEXT: [[TMP153:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP154:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS30:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP143]], align 4
-// CHECK3-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP144]], align 4
-// CHECK3-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP141]], ptr [[TMP145]], align 4
-// CHECK3-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP142]], ptr [[TMP146]], align 4
-// CHECK3-NEXT: [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.14, ptr [[TMP147]], align 4
-// CHECK3-NEXT: [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.15, ptr [[TMP148]], align 4
-// CHECK3-NEXT: [[TMP149:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP149]], align 4
-// CHECK3-NEXT: [[TMP150:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP150]], align 4
-// CHECK3-NEXT: [[TMP151:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP151]], align 8
-// CHECK3-NEXT: [[TMP152:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l154.region_id, ptr [[KERNEL_ARGS30]])
-// CHECK3-NEXT: [[TMP153:%.*]] = icmp ne i32 [[TMP152]], 0
-// CHECK3-NEXT: br i1 [[TMP153]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
+// CHECK3-NEXT: [[TMP155:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP155]], align 4
+// CHECK3-NEXT: [[TMP156:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP156]], align 4
+// CHECK3-NEXT: [[TMP157:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP153]], ptr [[TMP157]], align 4
+// CHECK3-NEXT: [[TMP158:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP154]], ptr [[TMP158]], align 4
+// CHECK3-NEXT: [[TMP159:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.14, ptr [[TMP159]], align 4
+// CHECK3-NEXT: [[TMP160:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.15, ptr [[TMP160]], align 4
+// CHECK3-NEXT: [[TMP161:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP161]], align 4
+// CHECK3-NEXT: [[TMP162:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP162]], align 4
+// CHECK3-NEXT: [[TMP163:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP163]], align 8
+// CHECK3-NEXT: [[TMP164:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP164]], align 8
+// CHECK3-NEXT: [[TMP165:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP165]], align 4
+// CHECK3-NEXT: [[TMP166:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP166]], align 4
+// CHECK3-NEXT: [[TMP167:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP167]], align 4
+// CHECK3-NEXT: [[TMP168:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l154.region_id, ptr [[KERNEL_ARGS30]])
+// CHECK3-NEXT: [[TMP169:%.*]] = icmp ne i32 [[TMP168]], 0
+// CHECK3-NEXT: br i1 [[TMP169]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
// CHECK3: omp_offload.failed31:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l154(i32 [[TMP137]]) #[[ATTR3]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l154(i32 [[TMP149]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT32]]
// CHECK3: omp_offload.cont32:
-// CHECK3-NEXT: [[TMP154:%.*]] = load i32, ptr [[NN]], align 4
-// CHECK3-NEXT: store i32 [[TMP154]], ptr [[NN_CASTED33]], align 4
-// CHECK3-NEXT: [[TMP155:%.*]] = load i32, ptr [[NN_CASTED33]], align 4
-// CHECK3-NEXT: [[TMP156:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP155]], ptr [[TMP156]], align 4
-// CHECK3-NEXT: [[TMP157:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP155]], ptr [[TMP157]], align 4
-// CHECK3-NEXT: [[TMP158:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP158]], align 4
-// CHECK3-NEXT: [[TMP159:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP160:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP170:%.*]] = load i32, ptr [[NN]], align 4
+// CHECK3-NEXT: store i32 [[TMP170]], ptr [[NN_CASTED33]], align 4
+// CHECK3-NEXT: [[TMP171:%.*]] = load i32, ptr [[NN_CASTED33]], align 4
+// CHECK3-NEXT: [[TMP172:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP171]], ptr [[TMP172]], align 4
+// CHECK3-NEXT: [[TMP173:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP171]], ptr [[TMP173]], align 4
+// CHECK3-NEXT: [[TMP174:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP174]], align 4
+// CHECK3-NEXT: [[TMP175:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP176:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS37:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP161:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP161]], align 4
-// CHECK3-NEXT: [[TMP162:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP162]], align 4
-// CHECK3-NEXT: [[TMP163:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP159]], ptr [[TMP163]], align 4
-// CHECK3-NEXT: [[TMP164:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP160]], ptr [[TMP164]], align 4
-// CHECK3-NEXT: [[TMP165:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.18, ptr [[TMP165]], align 4
-// CHECK3-NEXT: [[TMP166:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.19, ptr [[TMP166]], align 4
-// CHECK3-NEXT: [[TMP167:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP167]], align 4
-// CHECK3-NEXT: [[TMP168:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP168]], align 4
-// CHECK3-NEXT: [[TMP169:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP169]], align 8
-// CHECK3-NEXT: [[TMP170:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l157.region_id, ptr [[KERNEL_ARGS37]])
-// CHECK3-NEXT: [[TMP171:%.*]] = icmp ne i32 [[TMP170]], 0
-// CHECK3-NEXT: br i1 [[TMP171]], label [[OMP_OFFLOAD_FAILED38:%.*]], label [[OMP_OFFLOAD_CONT39:%.*]]
+// CHECK3-NEXT: [[TMP177:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP177]], align 4
+// CHECK3-NEXT: [[TMP178:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP178]], align 4
+// CHECK3-NEXT: [[TMP179:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP175]], ptr [[TMP179]], align 4
+// CHECK3-NEXT: [[TMP180:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP176]], ptr [[TMP180]], align 4
+// CHECK3-NEXT: [[TMP181:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.18, ptr [[TMP181]], align 4
+// CHECK3-NEXT: [[TMP182:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.19, ptr [[TMP182]], align 4
+// CHECK3-NEXT: [[TMP183:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP183]], align 4
+// CHECK3-NEXT: [[TMP184:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP184]], align 4
+// CHECK3-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP185]], align 8
+// CHECK3-NEXT: [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP186]], align 8
+// CHECK3-NEXT: [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP187]], align 4
+// CHECK3-NEXT: [[TMP188:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP188]], align 4
+// CHECK3-NEXT: [[TMP189:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS37]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP189]], align 4
+// CHECK3-NEXT: [[TMP190:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l157.region_id, ptr [[KERNEL_ARGS37]])
+// CHECK3-NEXT: [[TMP191:%.*]] = icmp ne i32 [[TMP190]], 0
+// CHECK3-NEXT: br i1 [[TMP191]], label [[OMP_OFFLOAD_FAILED38:%.*]], label [[OMP_OFFLOAD_CONT39:%.*]]
// CHECK3: omp_offload.failed38:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l157(i32 [[TMP155]]) #[[ATTR3]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l157(i32 [[TMP171]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT39]]
// CHECK3: omp_offload.cont39:
-// CHECK3-NEXT: [[TMP172:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: [[TMP173:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK3-NEXT: call void @llvm.stackrestore(ptr [[TMP173]])
-// CHECK3-NEXT: ret i32 [[TMP172]]
+// CHECK3-NEXT: [[TMP192:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: [[TMP193:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK3-NEXT: call void @llvm.stackrestore(ptr [[TMP193]])
+// CHECK3-NEXT: ret i32 [[TMP192]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l101
@@ -2224,37 +2346,47 @@ int bar(int n){
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[TMP9]], i32 0, i32 2
// CHECK3-NEXT: [[TMP18:%.*]] = load i32, ptr [[TMP16]], align 4
// CHECK3-NEXT: [[TMP19:%.*]] = load i32, ptr [[TMP17]], align 4
-// CHECK3-NEXT: store i32 1, ptr [[KERNEL_ARGS_I]], align 4, !noalias !25
-// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 1
-// CHECK3-NEXT: store i32 3, ptr [[TMP20]], align 4, !noalias !25
-// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP13]], ptr [[TMP21]], align 4, !noalias !25
-// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP14]], ptr [[TMP22]], align 4, !noalias !25
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 4
-// CHECK3-NEXT: store ptr [[TMP15]], ptr [[TMP23]], align 4, !noalias !25
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP24]], align 4, !noalias !25
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP25]], align 4, !noalias !25
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP26]], align 4, !noalias !25
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP27]], align 8, !noalias !25
-// CHECK3-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel_nowait(ptr @[[GLOB1]], i64 -1, i32 [[TMP18]], i32 [[TMP19]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l101.region_id, ptr [[KERNEL_ARGS_I]], i32 0, ptr null, i32 0, ptr null)
-// CHECK3-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
-// CHECK3-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__1_EXIT:%.*]]
+// CHECK3-NEXT: [[TMP20:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP18]], 0
+// CHECK3-NEXT: [[TMP21:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP19]], 0
+// CHECK3-NEXT: store i32 2, ptr [[KERNEL_ARGS_I]], align 4, !noalias !25
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 1
+// CHECK3-NEXT: store i32 3, ptr [[TMP22]], align 4, !noalias !25
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP13]], ptr [[TMP23]], align 4, !noalias !25
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP14]], ptr [[TMP24]], align 4, !noalias !25
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 4
+// CHECK3-NEXT: store ptr [[TMP15]], ptr [[TMP25]], align 4, !noalias !25
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP26]], align 4, !noalias !25
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP27]], align 4, !noalias !25
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4, !noalias !25
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP29]], align 8, !noalias !25
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 9
+// CHECK3-NEXT: store i64 1, ptr [[TMP30]], align 8, !noalias !25
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [[TMP20]], ptr [[TMP31]], align 4, !noalias !25
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [[TMP21]], ptr [[TMP32]], align 4, !noalias !25
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP33]], align 4, !noalias !25
+// CHECK3-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP18]], i32 [[TMP19]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l101.region_id, ptr [[KERNEL_ARGS_I]])
+// CHECK3-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK3-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__1_EXIT:%.*]]
// CHECK3: omp_offload.failed.i:
-// CHECK3-NEXT: [[TMP30:%.*]] = load i16, ptr [[TMP12]], align 2
-// CHECK3-NEXT: store i16 [[TMP30]], ptr [[AA_CASTED_I]], align 2, !noalias !25
-// CHECK3-NEXT: [[TMP31:%.*]] = load i32, ptr [[AA_CASTED_I]], align 4, !noalias !25
-// CHECK3-NEXT: [[TMP32:%.*]] = load i32, ptr [[TMP16]], align 4
-// CHECK3-NEXT: store i32 [[TMP32]], ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 4, !noalias !25
-// CHECK3-NEXT: [[TMP33:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 4, !noalias !25
-// CHECK3-NEXT: [[TMP34:%.*]] = load i32, ptr [[TMP17]], align 4
-// CHECK3-NEXT: store i32 [[TMP34]], ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 4, !noalias !25
-// CHECK3-NEXT: [[TMP35:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 4, !noalias !25
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l101(i32 [[TMP31]], i32 [[TMP33]], i32 [[TMP35]]) #[[ATTR3]]
+// CHECK3-NEXT: [[TMP36:%.*]] = load i16, ptr [[TMP12]], align 2
+// CHECK3-NEXT: store i16 [[TMP36]], ptr [[AA_CASTED_I]], align 2, !noalias !25
+// CHECK3-NEXT: [[TMP37:%.*]] = load i32, ptr [[AA_CASTED_I]], align 4, !noalias !25
+// CHECK3-NEXT: [[TMP38:%.*]] = load i32, ptr [[TMP16]], align 4
+// CHECK3-NEXT: store i32 [[TMP38]], ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 4, !noalias !25
+// CHECK3-NEXT: [[TMP39:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 4, !noalias !25
+// CHECK3-NEXT: [[TMP40:%.*]] = load i32, ptr [[TMP17]], align 4
+// CHECK3-NEXT: store i32 [[TMP40]], ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 4, !noalias !25
+// CHECK3-NEXT: [[TMP41:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 4, !noalias !25
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l101(i32 [[TMP37]], i32 [[TMP39]], i32 [[TMP41]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[DOTOMP_OUTLINED__1_EXIT]]
// CHECK3: .omp_outlined..1.exit:
// CHECK3-NEXT: ret i32 0
@@ -2570,7 +2702,7 @@ int bar(int n){
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2587,9 +2719,17 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP13]], align 4
// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
-// CHECK3-NEXT: [[TMP15:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z6bazzzziPi_l182.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP16:%.*]] = icmp ne i32 [[TMP15]], 0
-// CHECK3-NEXT: br i1 [[TMP16]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP15]], align 8
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP18]], align 4
+// CHECK3-NEXT: [[TMP19:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z6bazzzziPi_l182.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP20:%.*]] = icmp ne i32 [[TMP19]], 0
+// CHECK3-NEXT: br i1 [[TMP20]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z6bazzzziPi_l182(i32 [[TMP0]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -2727,7 +2867,7 @@ int bar(int n){
// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [5 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP29]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP29]], align 4
// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 5, ptr [[TMP30]], align 4
// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2744,9 +2884,17 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP36]], align 4
// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP37]], align 8
-// CHECK3-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l227.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK3-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP38]], align 8
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP39]], align 4
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l227.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
+// CHECK3-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l227(ptr [[THIS1]], i32 [[TMP5]], i32 2, i32 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -2756,15 +2904,15 @@ int bar(int n){
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l227(ptr [[THIS1]], i32 [[TMP5]], i32 2, i32 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
-// CHECK3-NEXT: [[TMP40:%.*]] = mul nsw i32 1, [[TMP1]]
-// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i32 [[TMP40]]
+// CHECK3-NEXT: [[TMP44:%.*]] = mul nsw i32 1, [[TMP1]]
+// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i32 [[TMP44]]
// CHECK3-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i16, ptr [[ARRAYIDX]], i32 1
-// CHECK3-NEXT: [[TMP41:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
-// CHECK3-NEXT: [[CONV:%.*]] = sext i16 [[TMP41]] to i32
-// CHECK3-NEXT: [[TMP42:%.*]] = load i32, ptr [[B]], align 4
-// CHECK3-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP42]]
-// CHECK3-NEXT: [[TMP43:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK3-NEXT: call void @llvm.stackrestore(ptr [[TMP43]])
+// CHECK3-NEXT: [[TMP45:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
+// CHECK3-NEXT: [[CONV:%.*]] = sext i16 [[TMP45]] to i32
+// CHECK3-NEXT: [[TMP46:%.*]] = load i32, ptr [[B]], align 4
+// CHECK3-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP46]]
+// CHECK3-NEXT: [[TMP47:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK3-NEXT: call void @llvm.stackrestore(ptr [[TMP47]])
// CHECK3-NEXT: ret i32 [[ADD3]]
//
//
@@ -2827,7 +2975,7 @@ int bar(int n){
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 4, ptr [[TMP22]], align 4
// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2844,9 +2992,17 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4
// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP29]], align 8
-// CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l209.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK3-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK3-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l209.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK3-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l209(i32 [[TMP1]], i32 [[TMP3]], i32 [[TMP5]], ptr [[B]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -2856,8 +3012,8 @@ int bar(int n){
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l209(i32 [[TMP1]], i32 [[TMP3]], i32 [[TMP5]], ptr [[B]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
-// CHECK3-NEXT: [[TMP32:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: ret i32 [[TMP32]]
+// CHECK3-NEXT: [[TMP36:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: ret i32 [[TMP36]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_Z9ftemplateIiET_i
@@ -2907,7 +3063,7 @@ int bar(int n){
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP16]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP16]], align 4
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 3, ptr [[TMP17]], align 4
// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2924,9 +3080,17 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP23]], align 4
// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP24]], align 8
-// CHECK3-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l192.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK3-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP25]], align 8
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l192.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK3-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l192(i32 [[TMP1]], i32 [[TMP3]], ptr [[B]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -2936,8 +3100,8 @@ int bar(int n){
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l192(i32 [[TMP1]], i32 [[TMP3]], ptr [[B]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
-// CHECK3-NEXT: [[TMP27:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: ret i32 [[TMP27]]
+// CHECK3-NEXT: [[TMP31:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: ret i32 [[TMP31]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l227
diff --git a/clang/test/OpenMP/target_teams_depend_codegen.cpp b/clang/test/OpenMP/target_teams_depend_codegen.cpp
index cdd220a1023af..ec0d01653bbbc 100644
--- a/clang/test/OpenMP/target_teams_depend_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_depend_codegen.cpp
@@ -204,7 +204,7 @@ int foo(int n) {
// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2
// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]],
// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64
-// CHECK: [[RET:%.+]] = call i32 @__tgt_target_kernel_nowait(%struct.ident_t* @{{.+}}, i64 [[DEVICE]], i32 0, i32 0, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* %{{.+}})
+// CHECK: [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 [[DEVICE]], i32 0, i32 0, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* %{{.+}})
// CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]]
diff --git a/clang/test/OpenMP/target_teams_distribute_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_codegen.cpp
index f8a9086923cad..a37039588afbd 100644
--- a/clang/test/OpenMP/target_teams_distribute_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_codegen.cpp
@@ -425,7 +425,7 @@ int bar(int n){
// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP52]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP52]], align 4
// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP53]], align 4
// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -442,186 +442,210 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP59]], align 8
// CHECK1-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 10, ptr [[TMP60]], align 8
-// CHECK1-NEXT: [[TMP61:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l113.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP62:%.*]] = icmp ne i32 [[TMP61]], 0
-// CHECK1-NEXT: br i1 [[TMP62]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP61]], align 8
+// CHECK1-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP62]], align 4
+// CHECK1-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP63]], align 4
+// CHECK1-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP64]], align 4
+// CHECK1-NEXT: [[TMP65:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l113.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP66:%.*]] = icmp ne i32 [[TMP65]], 0
+// CHECK1-NEXT: br i1 [[TMP66]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l113(i64 [[TMP46]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
-// CHECK1-NEXT: [[TMP63:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: store i32 [[TMP63]], ptr [[A_CASTED8]], align 4
-// CHECK1-NEXT: [[TMP64:%.*]] = load i64, ptr [[A_CASTED8]], align 8
-// CHECK1-NEXT: [[TMP65:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK1-NEXT: store i16 [[TMP65]], ptr [[AA_CASTED9]], align 2
-// CHECK1-NEXT: [[TMP66:%.*]] = load i64, ptr [[AA_CASTED9]], align 8
-// CHECK1-NEXT: [[TMP67:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP67]], 10
+// CHECK1-NEXT: [[TMP67:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: store i32 [[TMP67]], ptr [[A_CASTED8]], align 4
+// CHECK1-NEXT: [[TMP68:%.*]] = load i64, ptr [[A_CASTED8]], align 8
+// CHECK1-NEXT: [[TMP69:%.*]] = load i16, ptr [[AA]], align 2
+// CHECK1-NEXT: store i16 [[TMP69]], ptr [[AA_CASTED9]], align 2
+// CHECK1-NEXT: [[TMP70:%.*]] = load i64, ptr [[AA_CASTED9]], align 8
+// CHECK1-NEXT: [[TMP71:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP71]], 10
// CHECK1-NEXT: br i1 [[CMP]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK1: omp_if.then:
-// CHECK1-NEXT: [[TMP68:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP64]], ptr [[TMP68]], align 8
-// CHECK1-NEXT: [[TMP69:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP64]], ptr [[TMP69]], align 8
-// CHECK1-NEXT: [[TMP70:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP70]], align 8
-// CHECK1-NEXT: [[TMP71:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP66]], ptr [[TMP71]], align 8
-// CHECK1-NEXT: [[TMP72:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP66]], ptr [[TMP72]], align 8
-// CHECK1-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP73]], align 8
-// CHECK1-NEXT: [[TMP74:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP75:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP72:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP68]], ptr [[TMP72]], align 8
+// CHECK1-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP68]], ptr [[TMP73]], align 8
+// CHECK1-NEXT: [[TMP74:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP74]], align 8
+// CHECK1-NEXT: [[TMP75:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP70]], ptr [[TMP75]], align 8
+// CHECK1-NEXT: [[TMP76:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP70]], ptr [[TMP76]], align 8
+// CHECK1-NEXT: [[TMP77:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP77]], align 8
+// CHECK1-NEXT: [[TMP78:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP79:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS14:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP76]], align 4
-// CHECK1-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
-// CHECK1-NEXT: store i32 2, ptr [[TMP77]], align 4
-// CHECK1-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP74]], ptr [[TMP78]], align 8
-// CHECK1-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP75]], ptr [[TMP79]], align 8
-// CHECK1-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.7, ptr [[TMP80]], align 8
-// CHECK1-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP81]], align 8
-// CHECK1-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP82]], align 8
-// CHECK1-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP83]], align 8
-// CHECK1-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
-// CHECK1-NEXT: store i64 10, ptr [[TMP84]], align 8
-// CHECK1-NEXT: [[TMP85:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l120.region_id, ptr [[KERNEL_ARGS14]])
-// CHECK1-NEXT: [[TMP86:%.*]] = icmp ne i32 [[TMP85]], 0
-// CHECK1-NEXT: br i1 [[TMP86]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
+// CHECK1-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP80]], align 4
+// CHECK1-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
+// CHECK1-NEXT: store i32 2, ptr [[TMP81]], align 4
+// CHECK1-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP78]], ptr [[TMP82]], align 8
+// CHECK1-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP79]], ptr [[TMP83]], align 8
+// CHECK1-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.7, ptr [[TMP84]], align 8
+// CHECK1-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP85]], align 8
+// CHECK1-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP86]], align 8
+// CHECK1-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP87]], align 8
+// CHECK1-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
+// CHECK1-NEXT: store i64 10, ptr [[TMP88]], align 8
+// CHECK1-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP89]], align 8
+// CHECK1-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP90]], align 4
+// CHECK1-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP91]], align 4
+// CHECK1-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP92]], align 4
+// CHECK1-NEXT: [[TMP93:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l120.region_id, ptr [[KERNEL_ARGS14]])
+// CHECK1-NEXT: [[TMP94:%.*]] = icmp ne i32 [[TMP93]], 0
+// CHECK1-NEXT: br i1 [[TMP94]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
// CHECK1: omp_offload.failed15:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l120(i64 [[TMP64]], i64 [[TMP66]]) #[[ATTR3]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l120(i64 [[TMP68]], i64 [[TMP70]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT16]]
// CHECK1: omp_offload.cont16:
// CHECK1-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK1: omp_if.else:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l120(i64 [[TMP64]], i64 [[TMP66]]) #[[ATTR3]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l120(i64 [[TMP68]], i64 [[TMP70]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
-// CHECK1-NEXT: [[TMP87:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: store i32 [[TMP87]], ptr [[DOTCAPTURE_EXPR_17]], align 4
-// CHECK1-NEXT: [[TMP88:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: store i32 [[TMP88]], ptr [[A_CASTED18]], align 4
-// CHECK1-NEXT: [[TMP89:%.*]] = load i64, ptr [[A_CASTED18]], align 8
-// CHECK1-NEXT: [[TMP90:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_17]], align 4
-// CHECK1-NEXT: store i32 [[TMP90]], ptr [[DOTCAPTURE_EXPR__CASTED19]], align 4
-// CHECK1-NEXT: [[TMP91:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED19]], align 8
-// CHECK1-NEXT: [[TMP92:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: [[CMP20:%.*]] = icmp sgt i32 [[TMP92]], 20
+// CHECK1-NEXT: [[TMP95:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: store i32 [[TMP95]], ptr [[DOTCAPTURE_EXPR_17]], align 4
+// CHECK1-NEXT: [[TMP96:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: store i32 [[TMP96]], ptr [[A_CASTED18]], align 4
+// CHECK1-NEXT: [[TMP97:%.*]] = load i64, ptr [[A_CASTED18]], align 8
+// CHECK1-NEXT: [[TMP98:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_17]], align 4
+// CHECK1-NEXT: store i32 [[TMP98]], ptr [[DOTCAPTURE_EXPR__CASTED19]], align 4
+// CHECK1-NEXT: [[TMP99:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED19]], align 8
+// CHECK1-NEXT: [[TMP100:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: [[CMP20:%.*]] = icmp sgt i32 [[TMP100]], 20
// CHECK1-NEXT: br i1 [[CMP20]], label [[OMP_IF_THEN21:%.*]], label [[OMP_IF_ELSE29:%.*]]
// CHECK1: omp_if.then21:
-// CHECK1-NEXT: [[TMP93:%.*]] = mul nuw i64 [[TMP2]], 4
-// CHECK1-NEXT: [[TMP94:%.*]] = mul nuw i64 5, [[TMP5]]
-// CHECK1-NEXT: [[TMP95:%.*]] = mul nuw i64 [[TMP94]], 8
+// CHECK1-NEXT: [[TMP101:%.*]] = mul nuw i64 [[TMP2]], 4
+// CHECK1-NEXT: [[TMP102:%.*]] = mul nuw i64 5, [[TMP5]]
+// CHECK1-NEXT: [[TMP103:%.*]] = mul nuw i64 [[TMP102]], 8
// CHECK1-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES]], ptr align 8 @.offload_sizes.10, i64 80, i1 false)
-// CHECK1-NEXT: [[TMP96:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP89]], ptr [[TMP96]], align 8
-// CHECK1-NEXT: [[TMP97:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP89]], ptr [[TMP97]], align 8
-// CHECK1-NEXT: [[TMP98:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP98]], align 8
-// CHECK1-NEXT: [[TMP99:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 1
-// CHECK1-NEXT: store ptr [[B]], ptr [[TMP99]], align 8
-// CHECK1-NEXT: [[TMP100:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 1
-// CHECK1-NEXT: store ptr [[B]], ptr [[TMP100]], align 8
-// CHECK1-NEXT: [[TMP101:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP101]], align 8
-// CHECK1-NEXT: [[TMP102:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 2
-// CHECK1-NEXT: store i64 [[TMP2]], ptr [[TMP102]], align 8
-// CHECK1-NEXT: [[TMP103:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 2
-// CHECK1-NEXT: store i64 [[TMP2]], ptr [[TMP103]], align 8
-// CHECK1-NEXT: [[TMP104:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i64 0, i64 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP104]], align 8
-// CHECK1-NEXT: [[TMP105:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[VLA]], ptr [[TMP105]], align 8
-// CHECK1-NEXT: [[TMP106:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[VLA]], ptr [[TMP106]], align 8
-// CHECK1-NEXT: [[TMP107:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
-// CHECK1-NEXT: store i64 [[TMP93]], ptr [[TMP107]], align 8
-// CHECK1-NEXT: [[TMP108:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i64 0, i64 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP108]], align 8
-// CHECK1-NEXT: [[TMP109:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 4
-// CHECK1-NEXT: store ptr [[C]], ptr [[TMP109]], align 8
-// CHECK1-NEXT: [[TMP110:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 4
-// CHECK1-NEXT: store ptr [[C]], ptr [[TMP110]], align 8
-// CHECK1-NEXT: [[TMP111:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i64 0, i64 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP111]], align 8
-// CHECK1-NEXT: [[TMP112:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 5
-// CHECK1-NEXT: store i64 5, ptr [[TMP112]], align 8
-// CHECK1-NEXT: [[TMP113:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 5
-// CHECK1-NEXT: store i64 5, ptr [[TMP113]], align 8
-// CHECK1-NEXT: [[TMP114:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i64 0, i64 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP114]], align 8
-// CHECK1-NEXT: [[TMP115:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 6
-// CHECK1-NEXT: store i64 [[TMP5]], ptr [[TMP115]], align 8
-// CHECK1-NEXT: [[TMP116:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 6
-// CHECK1-NEXT: store i64 [[TMP5]], ptr [[TMP116]], align 8
-// CHECK1-NEXT: [[TMP117:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i64 0, i64 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP117]], align 8
-// CHECK1-NEXT: [[TMP118:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 7
-// CHECK1-NEXT: store ptr [[VLA1]], ptr [[TMP118]], align 8
-// CHECK1-NEXT: [[TMP119:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 7
-// CHECK1-NEXT: store ptr [[VLA1]], ptr [[TMP119]], align 8
-// CHECK1-NEXT: [[TMP120:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
-// CHECK1-NEXT: store i64 [[TMP95]], ptr [[TMP120]], align 8
-// CHECK1-NEXT: [[TMP121:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i64 0, i64 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP121]], align 8
-// CHECK1-NEXT: [[TMP122:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 8
-// CHECK1-NEXT: store ptr [[D]], ptr [[TMP122]], align 8
-// CHECK1-NEXT: [[TMP123:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 8
-// CHECK1-NEXT: store ptr [[D]], ptr [[TMP123]], align 8
-// CHECK1-NEXT: [[TMP124:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i64 0, i64 8
-// CHECK1-NEXT: store ptr null, ptr [[TMP124]], align 8
-// CHECK1-NEXT: [[TMP125:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 9
-// CHECK1-NEXT: store i64 [[TMP91]], ptr [[TMP125]], align 8
-// CHECK1-NEXT: [[TMP126:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 9
-// CHECK1-NEXT: store i64 [[TMP91]], ptr [[TMP126]], align 8
-// CHECK1-NEXT: [[TMP127:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i64 0, i64 9
-// CHECK1-NEXT: store ptr null, ptr [[TMP127]], align 8
-// CHECK1-NEXT: [[TMP128:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP129:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP130:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP104:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP97]], ptr [[TMP104]], align 8
+// CHECK1-NEXT: [[TMP105:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP97]], ptr [[TMP105]], align 8
+// CHECK1-NEXT: [[TMP106:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP106]], align 8
+// CHECK1-NEXT: [[TMP107:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 1
+// CHECK1-NEXT: store ptr [[B]], ptr [[TMP107]], align 8
+// CHECK1-NEXT: [[TMP108:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 1
+// CHECK1-NEXT: store ptr [[B]], ptr [[TMP108]], align 8
+// CHECK1-NEXT: [[TMP109:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP109]], align 8
+// CHECK1-NEXT: [[TMP110:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 2
+// CHECK1-NEXT: store i64 [[TMP2]], ptr [[TMP110]], align 8
+// CHECK1-NEXT: [[TMP111:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 2
+// CHECK1-NEXT: store i64 [[TMP2]], ptr [[TMP111]], align 8
+// CHECK1-NEXT: [[TMP112:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i64 0, i64 2
+// CHECK1-NEXT: store ptr null, ptr [[TMP112]], align 8
+// CHECK1-NEXT: [[TMP113:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[VLA]], ptr [[TMP113]], align 8
+// CHECK1-NEXT: [[TMP114:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[VLA]], ptr [[TMP114]], align 8
+// CHECK1-NEXT: [[TMP115:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
+// CHECK1-NEXT: store i64 [[TMP101]], ptr [[TMP115]], align 8
+// CHECK1-NEXT: [[TMP116:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i64 0, i64 3
+// CHECK1-NEXT: store ptr null, ptr [[TMP116]], align 8
+// CHECK1-NEXT: [[TMP117:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 4
+// CHECK1-NEXT: store ptr [[C]], ptr [[TMP117]], align 8
+// CHECK1-NEXT: [[TMP118:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 4
+// CHECK1-NEXT: store ptr [[C]], ptr [[TMP118]], align 8
+// CHECK1-NEXT: [[TMP119:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i64 0, i64 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP119]], align 8
+// CHECK1-NEXT: [[TMP120:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 5
+// CHECK1-NEXT: store i64 5, ptr [[TMP120]], align 8
+// CHECK1-NEXT: [[TMP121:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 5
+// CHECK1-NEXT: store i64 5, ptr [[TMP121]], align 8
+// CHECK1-NEXT: [[TMP122:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i64 0, i64 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP122]], align 8
+// CHECK1-NEXT: [[TMP123:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 6
+// CHECK1-NEXT: store i64 [[TMP5]], ptr [[TMP123]], align 8
+// CHECK1-NEXT: [[TMP124:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 6
+// CHECK1-NEXT: store i64 [[TMP5]], ptr [[TMP124]], align 8
+// CHECK1-NEXT: [[TMP125:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i64 0, i64 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP125]], align 8
+// CHECK1-NEXT: [[TMP126:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 7
+// CHECK1-NEXT: store ptr [[VLA1]], ptr [[TMP126]], align 8
+// CHECK1-NEXT: [[TMP127:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 7
+// CHECK1-NEXT: store ptr [[VLA1]], ptr [[TMP127]], align 8
+// CHECK1-NEXT: [[TMP128:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
+// CHECK1-NEXT: store i64 [[TMP103]], ptr [[TMP128]], align 8
+// CHECK1-NEXT: [[TMP129:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i64 0, i64 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP129]], align 8
+// CHECK1-NEXT: [[TMP130:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 8
+// CHECK1-NEXT: store ptr [[D]], ptr [[TMP130]], align 8
+// CHECK1-NEXT: [[TMP131:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 8
+// CHECK1-NEXT: store ptr [[D]], ptr [[TMP131]], align 8
+// CHECK1-NEXT: [[TMP132:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i64 0, i64 8
+// CHECK1-NEXT: store ptr null, ptr [[TMP132]], align 8
+// CHECK1-NEXT: [[TMP133:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 9
+// CHECK1-NEXT: store i64 [[TMP99]], ptr [[TMP133]], align 8
+// CHECK1-NEXT: [[TMP134:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 9
+// CHECK1-NEXT: store i64 [[TMP99]], ptr [[TMP134]], align 8
+// CHECK1-NEXT: [[TMP135:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i64 0, i64 9
+// CHECK1-NEXT: store ptr null, ptr [[TMP135]], align 8
+// CHECK1-NEXT: [[TMP136:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP137:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP138:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS26:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP131]], align 4
-// CHECK1-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 1
-// CHECK1-NEXT: store i32 10, ptr [[TMP132]], align 4
-// CHECK1-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP128]], ptr [[TMP133]], align 8
-// CHECK1-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP129]], ptr [[TMP134]], align 8
-// CHECK1-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 4
-// CHECK1-NEXT: store ptr [[TMP130]], ptr [[TMP135]], align 8
-// CHECK1-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.11, ptr [[TMP136]], align 8
-// CHECK1-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP137]], align 8
-// CHECK1-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP138]], align 8
-// CHECK1-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 8
-// CHECK1-NEXT: store i64 10, ptr [[TMP139]], align 8
-// CHECK1-NEXT: [[TMP140:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l145.region_id, ptr [[KERNEL_ARGS26]])
-// CHECK1-NEXT: [[TMP141:%.*]] = icmp ne i32 [[TMP140]], 0
-// CHECK1-NEXT: br i1 [[TMP141]], label [[OMP_OFFLOAD_FAILED27:%.*]], label [[OMP_OFFLOAD_CONT28:%.*]]
+// CHECK1-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP139]], align 4
+// CHECK1-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 1
+// CHECK1-NEXT: store i32 10, ptr [[TMP140]], align 4
+// CHECK1-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP136]], ptr [[TMP141]], align 8
+// CHECK1-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP137]], ptr [[TMP142]], align 8
+// CHECK1-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 4
+// CHECK1-NEXT: store ptr [[TMP138]], ptr [[TMP143]], align 8
+// CHECK1-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.11, ptr [[TMP144]], align 8
+// CHECK1-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP145]], align 8
+// CHECK1-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP146]], align 8
+// CHECK1-NEXT: [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 8
+// CHECK1-NEXT: store i64 10, ptr [[TMP147]], align 8
+// CHECK1-NEXT: [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP148]], align 8
+// CHECK1-NEXT: [[TMP149:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP149]], align 4
+// CHECK1-NEXT: [[TMP150:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP150]], align 4
+// CHECK1-NEXT: [[TMP151:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP151]], align 4
+// CHECK1-NEXT: [[TMP152:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l145.region_id, ptr [[KERNEL_ARGS26]])
+// CHECK1-NEXT: [[TMP153:%.*]] = icmp ne i32 [[TMP152]], 0
+// CHECK1-NEXT: br i1 [[TMP153]], label [[OMP_OFFLOAD_FAILED27:%.*]], label [[OMP_OFFLOAD_CONT28:%.*]]
// CHECK1: omp_offload.failed27:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l145(i64 [[TMP89]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]], i64 [[TMP91]]) #[[ATTR3]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l145(i64 [[TMP97]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]], i64 [[TMP99]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT28]]
// CHECK1: omp_offload.cont28:
// CHECK1-NEXT: br label [[OMP_IF_END30:%.*]]
// CHECK1: omp_if.else29:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l145(i64 [[TMP89]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]], i64 [[TMP91]]) #[[ATTR3]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l145(i64 [[TMP97]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]], i64 [[TMP99]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_IF_END30]]
// CHECK1: omp_if.end30:
-// CHECK1-NEXT: [[TMP142:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: [[TMP143:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK1-NEXT: call void @llvm.stackrestore(ptr [[TMP143]])
-// CHECK1-NEXT: ret i32 [[TMP142]]
+// CHECK1-NEXT: [[TMP154:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: [[TMP155:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK1-NEXT: call void @llvm.stackrestore(ptr [[TMP155]])
+// CHECK1-NEXT: ret i32 [[TMP154]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l103
@@ -786,37 +810,47 @@ int bar(int n){
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[TMP9]], i32 0, i32 2
// CHECK1-NEXT: [[TMP18:%.*]] = load i32, ptr [[TMP16]], align 4
// CHECK1-NEXT: [[TMP19:%.*]] = load i32, ptr [[TMP17]], align 4
-// CHECK1-NEXT: store i32 1, ptr [[KERNEL_ARGS_I]], align 4, !noalias !21
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 1
-// CHECK1-NEXT: store i32 3, ptr [[TMP20]], align 4, !noalias !21
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP13]], ptr [[TMP21]], align 8, !noalias !21
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP14]], ptr [[TMP22]], align 8, !noalias !21
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 4
-// CHECK1-NEXT: store ptr [[TMP15]], ptr [[TMP23]], align 8, !noalias !21
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP24]], align 8, !noalias !21
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP25]], align 8, !noalias !21
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP26]], align 8, !noalias !21
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 8
-// CHECK1-NEXT: store i64 10, ptr [[TMP27]], align 8, !noalias !21
-// CHECK1-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel_nowait(ptr @[[GLOB2]], i64 -1, i32 [[TMP18]], i32 [[TMP19]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l103.region_id, ptr [[KERNEL_ARGS_I]], i32 0, ptr null, i32 0, ptr null)
-// CHECK1-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
-// CHECK1-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__1_EXIT:%.*]]
+// CHECK1-NEXT: [[TMP20:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP18]], 0
+// CHECK1-NEXT: [[TMP21:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP19]], 0
+// CHECK1-NEXT: store i32 2, ptr [[KERNEL_ARGS_I]], align 4, !noalias !21
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 1
+// CHECK1-NEXT: store i32 3, ptr [[TMP22]], align 4, !noalias !21
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP13]], ptr [[TMP23]], align 8, !noalias !21
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP14]], ptr [[TMP24]], align 8, !noalias !21
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 4
+// CHECK1-NEXT: store ptr [[TMP15]], ptr [[TMP25]], align 8, !noalias !21
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP26]], align 8, !noalias !21
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8, !noalias !21
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8, !noalias !21
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 8
+// CHECK1-NEXT: store i64 10, ptr [[TMP29]], align 8, !noalias !21
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 9
+// CHECK1-NEXT: store i64 1, ptr [[TMP30]], align 8, !noalias !21
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [[TMP20]], ptr [[TMP31]], align 4, !noalias !21
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP21]], ptr [[TMP32]], align 4, !noalias !21
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP33]], align 4, !noalias !21
+// CHECK1-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 [[TMP18]], i32 [[TMP19]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l103.region_id, ptr [[KERNEL_ARGS_I]])
+// CHECK1-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK1-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__1_EXIT:%.*]]
// CHECK1: omp_offload.failed.i:
-// CHECK1-NEXT: [[TMP30:%.*]] = load i16, ptr [[TMP12]], align 2
-// CHECK1-NEXT: store i16 [[TMP30]], ptr [[AA_CASTED_I]], align 2, !noalias !21
-// CHECK1-NEXT: [[TMP31:%.*]] = load i64, ptr [[AA_CASTED_I]], align 8, !noalias !21
-// CHECK1-NEXT: [[TMP32:%.*]] = load i32, ptr [[TMP16]], align 4
-// CHECK1-NEXT: store i32 [[TMP32]], ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 4, !noalias !21
-// CHECK1-NEXT: [[TMP33:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 8, !noalias !21
-// CHECK1-NEXT: [[TMP34:%.*]] = load i32, ptr [[TMP17]], align 4
-// CHECK1-NEXT: store i32 [[TMP34]], ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 4, !noalias !21
-// CHECK1-NEXT: [[TMP35:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 8, !noalias !21
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l103(i64 [[TMP31]], i64 [[TMP33]], i64 [[TMP35]]) #[[ATTR3]]
+// CHECK1-NEXT: [[TMP36:%.*]] = load i16, ptr [[TMP12]], align 2
+// CHECK1-NEXT: store i16 [[TMP36]], ptr [[AA_CASTED_I]], align 2, !noalias !21
+// CHECK1-NEXT: [[TMP37:%.*]] = load i64, ptr [[AA_CASTED_I]], align 8, !noalias !21
+// CHECK1-NEXT: [[TMP38:%.*]] = load i32, ptr [[TMP16]], align 4
+// CHECK1-NEXT: store i32 [[TMP38]], ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 4, !noalias !21
+// CHECK1-NEXT: [[TMP39:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 8, !noalias !21
+// CHECK1-NEXT: [[TMP40:%.*]] = load i32, ptr [[TMP17]], align 4
+// CHECK1-NEXT: store i32 [[TMP40]], ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 4, !noalias !21
+// CHECK1-NEXT: [[TMP41:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 8, !noalias !21
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l103(i64 [[TMP37]], i64 [[TMP39]], i64 [[TMP41]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[DOTOMP_OUTLINED__1_EXIT]]
// CHECK1: .omp_outlined..1.exit:
// CHECK1-NEXT: ret i32 0
@@ -1363,7 +1397,7 @@ int bar(int n){
// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [5 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP29]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP29]], align 4
// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 5, ptr [[TMP30]], align 4
// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1380,9 +1414,17 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP36]], align 8
// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 10, ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l218.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK1-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP38]], align 8
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP39]], align 4
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP40]], align 4
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP41]], align 4
+// CHECK1-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l218.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
+// CHECK1-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l218(ptr [[THIS1]], i64 [[TMP6]], i64 2, i64 [[TMP2]], ptr [[VLA]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1392,15 +1434,15 @@ int bar(int n){
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l218(ptr [[THIS1]], i64 [[TMP6]], i64 2, i64 [[TMP2]], ptr [[VLA]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
-// CHECK1-NEXT: [[TMP40:%.*]] = mul nsw i64 1, [[TMP2]]
-// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i64 [[TMP40]]
+// CHECK1-NEXT: [[TMP44:%.*]] = mul nsw i64 1, [[TMP2]]
+// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i64 [[TMP44]]
// CHECK1-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i16, ptr [[ARRAYIDX]], i64 1
-// CHECK1-NEXT: [[TMP41:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
-// CHECK1-NEXT: [[CONV:%.*]] = sext i16 [[TMP41]] to i32
-// CHECK1-NEXT: [[TMP42:%.*]] = load i32, ptr [[B]], align 4
-// CHECK1-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP42]]
-// CHECK1-NEXT: [[TMP43:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK1-NEXT: call void @llvm.stackrestore(ptr [[TMP43]])
+// CHECK1-NEXT: [[TMP45:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
+// CHECK1-NEXT: [[CONV:%.*]] = sext i16 [[TMP45]] to i32
+// CHECK1-NEXT: [[TMP46:%.*]] = load i32, ptr [[B]], align 4
+// CHECK1-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP46]]
+// CHECK1-NEXT: [[TMP47:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK1-NEXT: call void @llvm.stackrestore(ptr [[TMP47]])
// CHECK1-NEXT: ret i32 [[ADD3]]
//
//
@@ -1492,7 +1534,7 @@ int bar(int n){
// CHECK1-NEXT: [[TMP31:%.*]] = zext i32 [[ADD5]] to i64
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP32]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP32]], align 4
// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 5, ptr [[TMP33]], align 4
// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1509,9 +1551,17 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP39]], align 8
// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 [[TMP31]], ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l200.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
-// CHECK1-NEXT: br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP42]], align 4
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP43]], align 4
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP44]], align 4
+// CHECK1-NEXT: [[TMP45:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l200.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP46:%.*]] = icmp ne i32 [[TMP45]], 0
+// CHECK1-NEXT: br i1 [[TMP46]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l200(i64 [[TMP1]], i64 [[TMP3]], i64 [[TMP5]], i64 [[TMP7]], ptr [[B]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1521,8 +1571,8 @@ int bar(int n){
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l200(i64 [[TMP1]], i64 [[TMP3]], i64 [[TMP5]], i64 [[TMP7]], ptr [[B]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
-// CHECK1-NEXT: [[TMP43:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: ret i32 [[TMP43]]
+// CHECK1-NEXT: [[TMP47:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: ret i32 [[TMP47]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_Z9ftemplateIiET_i
@@ -1573,7 +1623,7 @@ int bar(int n){
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP16]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP16]], align 4
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 3, ptr [[TMP17]], align 4
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1590,9 +1640,17 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP23]], align 8
// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 10, ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l183.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK1-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP25]], align 8
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP28]], align 4
+// CHECK1-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l183.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK1-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l183(i64 [[TMP1]], i64 [[TMP3]], ptr [[B]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1602,8 +1660,8 @@ int bar(int n){
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l183(i64 [[TMP1]], i64 [[TMP3]], ptr [[B]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
-// CHECK1-NEXT: [[TMP27:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: ret i32 [[TMP27]]
+// CHECK1-NEXT: [[TMP31:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: ret i32 [[TMP31]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l218
@@ -2110,7 +2168,7 @@ int bar(int n){
// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP50]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP50]], align 4
// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP51]], align 4
// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2127,188 +2185,212 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP57]], align 4
// CHECK3-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 10, ptr [[TMP58]], align 8
-// CHECK3-NEXT: [[TMP59:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l113.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP60:%.*]] = icmp ne i32 [[TMP59]], 0
-// CHECK3-NEXT: br i1 [[TMP60]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP59]], align 8
+// CHECK3-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP60]], align 4
+// CHECK3-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4
+// CHECK3-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP62]], align 4
+// CHECK3-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l113.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
+// CHECK3-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l113(i32 [[TMP44]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
-// CHECK3-NEXT: [[TMP61:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: store i32 [[TMP61]], ptr [[A_CASTED8]], align 4
-// CHECK3-NEXT: [[TMP62:%.*]] = load i32, ptr [[A_CASTED8]], align 4
-// CHECK3-NEXT: [[TMP63:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK3-NEXT: store i16 [[TMP63]], ptr [[AA_CASTED9]], align 2
-// CHECK3-NEXT: [[TMP64:%.*]] = load i32, ptr [[AA_CASTED9]], align 4
-// CHECK3-NEXT: [[TMP65:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP65]], 10
+// CHECK3-NEXT: [[TMP65:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: store i32 [[TMP65]], ptr [[A_CASTED8]], align 4
+// CHECK3-NEXT: [[TMP66:%.*]] = load i32, ptr [[A_CASTED8]], align 4
+// CHECK3-NEXT: [[TMP67:%.*]] = load i16, ptr [[AA]], align 2
+// CHECK3-NEXT: store i16 [[TMP67]], ptr [[AA_CASTED9]], align 2
+// CHECK3-NEXT: [[TMP68:%.*]] = load i32, ptr [[AA_CASTED9]], align 4
+// CHECK3-NEXT: [[TMP69:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP69]], 10
// CHECK3-NEXT: br i1 [[CMP]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK3: omp_if.then:
-// CHECK3-NEXT: [[TMP66:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP62]], ptr [[TMP66]], align 4
-// CHECK3-NEXT: [[TMP67:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP62]], ptr [[TMP67]], align 4
-// CHECK3-NEXT: [[TMP68:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP68]], align 4
-// CHECK3-NEXT: [[TMP69:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP64]], ptr [[TMP69]], align 4
-// CHECK3-NEXT: [[TMP70:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP64]], ptr [[TMP70]], align 4
-// CHECK3-NEXT: [[TMP71:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i32 0, i32 1
-// CHECK3-NEXT: store ptr null, ptr [[TMP71]], align 4
-// CHECK3-NEXT: [[TMP72:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP70:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP66]], ptr [[TMP70]], align 4
+// CHECK3-NEXT: [[TMP71:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP66]], ptr [[TMP71]], align 4
+// CHECK3-NEXT: [[TMP72:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP72]], align 4
+// CHECK3-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP68]], ptr [[TMP73]], align 4
+// CHECK3-NEXT: [[TMP74:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP68]], ptr [[TMP74]], align 4
+// CHECK3-NEXT: [[TMP75:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i32 0, i32 1
+// CHECK3-NEXT: store ptr null, ptr [[TMP75]], align 4
+// CHECK3-NEXT: [[TMP76:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP77:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS14:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP74]], align 4
-// CHECK3-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
-// CHECK3-NEXT: store i32 2, ptr [[TMP75]], align 4
-// CHECK3-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP72]], ptr [[TMP76]], align 4
-// CHECK3-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP73]], ptr [[TMP77]], align 4
-// CHECK3-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.7, ptr [[TMP78]], align 4
-// CHECK3-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP79]], align 4
-// CHECK3-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP80]], align 4
-// CHECK3-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP81]], align 4
-// CHECK3-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
-// CHECK3-NEXT: store i64 10, ptr [[TMP82]], align 8
-// CHECK3-NEXT: [[TMP83:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l120.region_id, ptr [[KERNEL_ARGS14]])
-// CHECK3-NEXT: [[TMP84:%.*]] = icmp ne i32 [[TMP83]], 0
-// CHECK3-NEXT: br i1 [[TMP84]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
+// CHECK3-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP78]], align 4
+// CHECK3-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
+// CHECK3-NEXT: store i32 2, ptr [[TMP79]], align 4
+// CHECK3-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP76]], ptr [[TMP80]], align 4
+// CHECK3-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP77]], ptr [[TMP81]], align 4
+// CHECK3-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.7, ptr [[TMP82]], align 4
+// CHECK3-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP83]], align 4
+// CHECK3-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP84]], align 4
+// CHECK3-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP85]], align 4
+// CHECK3-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
+// CHECK3-NEXT: store i64 10, ptr [[TMP86]], align 8
+// CHECK3-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP87]], align 8
+// CHECK3-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP88]], align 4
+// CHECK3-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP89]], align 4
+// CHECK3-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP90]], align 4
+// CHECK3-NEXT: [[TMP91:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l120.region_id, ptr [[KERNEL_ARGS14]])
+// CHECK3-NEXT: [[TMP92:%.*]] = icmp ne i32 [[TMP91]], 0
+// CHECK3-NEXT: br i1 [[TMP92]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
// CHECK3: omp_offload.failed15:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l120(i32 [[TMP62]], i32 [[TMP64]]) #[[ATTR3]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l120(i32 [[TMP66]], i32 [[TMP68]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT16]]
// CHECK3: omp_offload.cont16:
// CHECK3-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK3: omp_if.else:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l120(i32 [[TMP62]], i32 [[TMP64]]) #[[ATTR3]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l120(i32 [[TMP66]], i32 [[TMP68]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
-// CHECK3-NEXT: [[TMP85:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: store i32 [[TMP85]], ptr [[DOTCAPTURE_EXPR_17]], align 4
-// CHECK3-NEXT: [[TMP86:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: store i32 [[TMP86]], ptr [[A_CASTED18]], align 4
-// CHECK3-NEXT: [[TMP87:%.*]] = load i32, ptr [[A_CASTED18]], align 4
-// CHECK3-NEXT: [[TMP88:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_17]], align 4
-// CHECK3-NEXT: store i32 [[TMP88]], ptr [[DOTCAPTURE_EXPR__CASTED19]], align 4
-// CHECK3-NEXT: [[TMP89:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED19]], align 4
-// CHECK3-NEXT: [[TMP90:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: [[CMP20:%.*]] = icmp sgt i32 [[TMP90]], 20
+// CHECK3-NEXT: [[TMP93:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: store i32 [[TMP93]], ptr [[DOTCAPTURE_EXPR_17]], align 4
+// CHECK3-NEXT: [[TMP94:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: store i32 [[TMP94]], ptr [[A_CASTED18]], align 4
+// CHECK3-NEXT: [[TMP95:%.*]] = load i32, ptr [[A_CASTED18]], align 4
+// CHECK3-NEXT: [[TMP96:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_17]], align 4
+// CHECK3-NEXT: store i32 [[TMP96]], ptr [[DOTCAPTURE_EXPR__CASTED19]], align 4
+// CHECK3-NEXT: [[TMP97:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED19]], align 4
+// CHECK3-NEXT: [[TMP98:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: [[CMP20:%.*]] = icmp sgt i32 [[TMP98]], 20
// CHECK3-NEXT: br i1 [[CMP20]], label [[OMP_IF_THEN21:%.*]], label [[OMP_IF_ELSE29:%.*]]
// CHECK3: omp_if.then21:
-// CHECK3-NEXT: [[TMP91:%.*]] = mul nuw i32 [[TMP1]], 4
-// CHECK3-NEXT: [[TMP92:%.*]] = sext i32 [[TMP91]] to i64
-// CHECK3-NEXT: [[TMP93:%.*]] = mul nuw i32 5, [[TMP3]]
-// CHECK3-NEXT: [[TMP94:%.*]] = mul nuw i32 [[TMP93]], 8
-// CHECK3-NEXT: [[TMP95:%.*]] = sext i32 [[TMP94]] to i64
+// CHECK3-NEXT: [[TMP99:%.*]] = mul nuw i32 [[TMP1]], 4
+// CHECK3-NEXT: [[TMP100:%.*]] = sext i32 [[TMP99]] to i64
+// CHECK3-NEXT: [[TMP101:%.*]] = mul nuw i32 5, [[TMP3]]
+// CHECK3-NEXT: [[TMP102:%.*]] = mul nuw i32 [[TMP101]], 8
+// CHECK3-NEXT: [[TMP103:%.*]] = sext i32 [[TMP102]] to i64
// CHECK3-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES]], ptr align 4 @.offload_sizes.10, i32 80, i1 false)
-// CHECK3-NEXT: [[TMP96:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP87]], ptr [[TMP96]], align 4
-// CHECK3-NEXT: [[TMP97:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP87]], ptr [[TMP97]], align 4
-// CHECK3-NEXT: [[TMP98:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP98]], align 4
-// CHECK3-NEXT: [[TMP99:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 1
-// CHECK3-NEXT: store ptr [[B]], ptr [[TMP99]], align 4
-// CHECK3-NEXT: [[TMP100:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 1
-// CHECK3-NEXT: store ptr [[B]], ptr [[TMP100]], align 4
-// CHECK3-NEXT: [[TMP101:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i32 0, i32 1
-// CHECK3-NEXT: store ptr null, ptr [[TMP101]], align 4
-// CHECK3-NEXT: [[TMP102:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 2
-// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP102]], align 4
-// CHECK3-NEXT: [[TMP103:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 2
-// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP103]], align 4
-// CHECK3-NEXT: [[TMP104:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i32 0, i32 2
-// CHECK3-NEXT: store ptr null, ptr [[TMP104]], align 4
-// CHECK3-NEXT: [[TMP105:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[VLA]], ptr [[TMP105]], align 4
-// CHECK3-NEXT: [[TMP106:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[VLA]], ptr [[TMP106]], align 4
-// CHECK3-NEXT: [[TMP107:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
-// CHECK3-NEXT: store i64 [[TMP92]], ptr [[TMP107]], align 4
-// CHECK3-NEXT: [[TMP108:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i32 0, i32 3
-// CHECK3-NEXT: store ptr null, ptr [[TMP108]], align 4
-// CHECK3-NEXT: [[TMP109:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 4
-// CHECK3-NEXT: store ptr [[C]], ptr [[TMP109]], align 4
-// CHECK3-NEXT: [[TMP110:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 4
-// CHECK3-NEXT: store ptr [[C]], ptr [[TMP110]], align 4
-// CHECK3-NEXT: [[TMP111:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i32 0, i32 4
-// CHECK3-NEXT: store ptr null, ptr [[TMP111]], align 4
-// CHECK3-NEXT: [[TMP112:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 5
-// CHECK3-NEXT: store i32 5, ptr [[TMP112]], align 4
-// CHECK3-NEXT: [[TMP113:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 5
-// CHECK3-NEXT: store i32 5, ptr [[TMP113]], align 4
-// CHECK3-NEXT: [[TMP114:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i32 0, i32 5
-// CHECK3-NEXT: store ptr null, ptr [[TMP114]], align 4
-// CHECK3-NEXT: [[TMP115:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 6
-// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP115]], align 4
-// CHECK3-NEXT: [[TMP116:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 6
-// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP116]], align 4
-// CHECK3-NEXT: [[TMP117:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP117]], align 4
-// CHECK3-NEXT: [[TMP118:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 7
-// CHECK3-NEXT: store ptr [[VLA1]], ptr [[TMP118]], align 4
-// CHECK3-NEXT: [[TMP119:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 7
-// CHECK3-NEXT: store ptr [[VLA1]], ptr [[TMP119]], align 4
-// CHECK3-NEXT: [[TMP120:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
-// CHECK3-NEXT: store i64 [[TMP95]], ptr [[TMP120]], align 4
-// CHECK3-NEXT: [[TMP121:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP121]], align 4
-// CHECK3-NEXT: [[TMP122:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 8
-// CHECK3-NEXT: store ptr [[D]], ptr [[TMP122]], align 4
-// CHECK3-NEXT: [[TMP123:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 8
-// CHECK3-NEXT: store ptr [[D]], ptr [[TMP123]], align 4
-// CHECK3-NEXT: [[TMP124:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i32 0, i32 8
-// CHECK3-NEXT: store ptr null, ptr [[TMP124]], align 4
-// CHECK3-NEXT: [[TMP125:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 9
-// CHECK3-NEXT: store i32 [[TMP89]], ptr [[TMP125]], align 4
-// CHECK3-NEXT: [[TMP126:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 9
-// CHECK3-NEXT: store i32 [[TMP89]], ptr [[TMP126]], align 4
-// CHECK3-NEXT: [[TMP127:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i32 0, i32 9
-// CHECK3-NEXT: store ptr null, ptr [[TMP127]], align 4
-// CHECK3-NEXT: [[TMP128:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP129:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP130:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP104:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP95]], ptr [[TMP104]], align 4
+// CHECK3-NEXT: [[TMP105:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP95]], ptr [[TMP105]], align 4
+// CHECK3-NEXT: [[TMP106:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP106]], align 4
+// CHECK3-NEXT: [[TMP107:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 1
+// CHECK3-NEXT: store ptr [[B]], ptr [[TMP107]], align 4
+// CHECK3-NEXT: [[TMP108:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 1
+// CHECK3-NEXT: store ptr [[B]], ptr [[TMP108]], align 4
+// CHECK3-NEXT: [[TMP109:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i32 0, i32 1
+// CHECK3-NEXT: store ptr null, ptr [[TMP109]], align 4
+// CHECK3-NEXT: [[TMP110:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 2
+// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP110]], align 4
+// CHECK3-NEXT: [[TMP111:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 2
+// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP111]], align 4
+// CHECK3-NEXT: [[TMP112:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i32 0, i32 2
+// CHECK3-NEXT: store ptr null, ptr [[TMP112]], align 4
+// CHECK3-NEXT: [[TMP113:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[VLA]], ptr [[TMP113]], align 4
+// CHECK3-NEXT: [[TMP114:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[VLA]], ptr [[TMP114]], align 4
+// CHECK3-NEXT: [[TMP115:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
+// CHECK3-NEXT: store i64 [[TMP100]], ptr [[TMP115]], align 4
+// CHECK3-NEXT: [[TMP116:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i32 0, i32 3
+// CHECK3-NEXT: store ptr null, ptr [[TMP116]], align 4
+// CHECK3-NEXT: [[TMP117:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 4
+// CHECK3-NEXT: store ptr [[C]], ptr [[TMP117]], align 4
+// CHECK3-NEXT: [[TMP118:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 4
+// CHECK3-NEXT: store ptr [[C]], ptr [[TMP118]], align 4
+// CHECK3-NEXT: [[TMP119:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i32 0, i32 4
+// CHECK3-NEXT: store ptr null, ptr [[TMP119]], align 4
+// CHECK3-NEXT: [[TMP120:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 5
+// CHECK3-NEXT: store i32 5, ptr [[TMP120]], align 4
+// CHECK3-NEXT: [[TMP121:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 5
+// CHECK3-NEXT: store i32 5, ptr [[TMP121]], align 4
+// CHECK3-NEXT: [[TMP122:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i32 0, i32 5
+// CHECK3-NEXT: store ptr null, ptr [[TMP122]], align 4
+// CHECK3-NEXT: [[TMP123:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 6
+// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP123]], align 4
+// CHECK3-NEXT: [[TMP124:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 6
+// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP124]], align 4
+// CHECK3-NEXT: [[TMP125:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP125]], align 4
+// CHECK3-NEXT: [[TMP126:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 7
+// CHECK3-NEXT: store ptr [[VLA1]], ptr [[TMP126]], align 4
+// CHECK3-NEXT: [[TMP127:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 7
+// CHECK3-NEXT: store ptr [[VLA1]], ptr [[TMP127]], align 4
+// CHECK3-NEXT: [[TMP128:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
+// CHECK3-NEXT: store i64 [[TMP103]], ptr [[TMP128]], align 4
+// CHECK3-NEXT: [[TMP129:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP129]], align 4
+// CHECK3-NEXT: [[TMP130:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 8
+// CHECK3-NEXT: store ptr [[D]], ptr [[TMP130]], align 4
+// CHECK3-NEXT: [[TMP131:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 8
+// CHECK3-NEXT: store ptr [[D]], ptr [[TMP131]], align 4
+// CHECK3-NEXT: [[TMP132:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i32 0, i32 8
+// CHECK3-NEXT: store ptr null, ptr [[TMP132]], align 4
+// CHECK3-NEXT: [[TMP133:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 9
+// CHECK3-NEXT: store i32 [[TMP97]], ptr [[TMP133]], align 4
+// CHECK3-NEXT: [[TMP134:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 9
+// CHECK3-NEXT: store i32 [[TMP97]], ptr [[TMP134]], align 4
+// CHECK3-NEXT: [[TMP135:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_MAPPERS24]], i32 0, i32 9
+// CHECK3-NEXT: store ptr null, ptr [[TMP135]], align 4
+// CHECK3-NEXT: [[TMP136:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_BASEPTRS22]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP137:%.*]] = getelementptr inbounds [10 x ptr], ptr [[DOTOFFLOAD_PTRS23]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP138:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS26:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP131]], align 4
-// CHECK3-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 1
-// CHECK3-NEXT: store i32 10, ptr [[TMP132]], align 4
-// CHECK3-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP128]], ptr [[TMP133]], align 4
-// CHECK3-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP129]], ptr [[TMP134]], align 4
-// CHECK3-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 4
-// CHECK3-NEXT: store ptr [[TMP130]], ptr [[TMP135]], align 4
-// CHECK3-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.11, ptr [[TMP136]], align 4
-// CHECK3-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP137]], align 4
-// CHECK3-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP138]], align 4
-// CHECK3-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 8
-// CHECK3-NEXT: store i64 10, ptr [[TMP139]], align 8
-// CHECK3-NEXT: [[TMP140:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l145.region_id, ptr [[KERNEL_ARGS26]])
-// CHECK3-NEXT: [[TMP141:%.*]] = icmp ne i32 [[TMP140]], 0
-// CHECK3-NEXT: br i1 [[TMP141]], label [[OMP_OFFLOAD_FAILED27:%.*]], label [[OMP_OFFLOAD_CONT28:%.*]]
+// CHECK3-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP139]], align 4
+// CHECK3-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 1
+// CHECK3-NEXT: store i32 10, ptr [[TMP140]], align 4
+// CHECK3-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP136]], ptr [[TMP141]], align 4
+// CHECK3-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP137]], ptr [[TMP142]], align 4
+// CHECK3-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 4
+// CHECK3-NEXT: store ptr [[TMP138]], ptr [[TMP143]], align 4
+// CHECK3-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.11, ptr [[TMP144]], align 4
+// CHECK3-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP145]], align 4
+// CHECK3-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP146]], align 4
+// CHECK3-NEXT: [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 8
+// CHECK3-NEXT: store i64 10, ptr [[TMP147]], align 8
+// CHECK3-NEXT: [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP148]], align 8
+// CHECK3-NEXT: [[TMP149:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP149]], align 4
+// CHECK3-NEXT: [[TMP150:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP150]], align 4
+// CHECK3-NEXT: [[TMP151:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS26]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP151]], align 4
+// CHECK3-NEXT: [[TMP152:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l145.region_id, ptr [[KERNEL_ARGS26]])
+// CHECK3-NEXT: [[TMP153:%.*]] = icmp ne i32 [[TMP152]], 0
+// CHECK3-NEXT: br i1 [[TMP153]], label [[OMP_OFFLOAD_FAILED27:%.*]], label [[OMP_OFFLOAD_CONT28:%.*]]
// CHECK3: omp_offload.failed27:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l145(i32 [[TMP87]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]], i32 [[TMP89]]) #[[ATTR3]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l145(i32 [[TMP95]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]], i32 [[TMP97]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT28]]
// CHECK3: omp_offload.cont28:
// CHECK3-NEXT: br label [[OMP_IF_END30:%.*]]
// CHECK3: omp_if.else29:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l145(i32 [[TMP87]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]], i32 [[TMP89]]) #[[ATTR3]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l145(i32 [[TMP95]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]], i32 [[TMP97]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_IF_END30]]
// CHECK3: omp_if.end30:
-// CHECK3-NEXT: [[TMP142:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: [[TMP143:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK3-NEXT: call void @llvm.stackrestore(ptr [[TMP143]])
-// CHECK3-NEXT: ret i32 [[TMP142]]
+// CHECK3-NEXT: [[TMP154:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: [[TMP155:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK3-NEXT: call void @llvm.stackrestore(ptr [[TMP155]])
+// CHECK3-NEXT: ret i32 [[TMP154]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l103
@@ -2473,37 +2555,47 @@ int bar(int n){
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[TMP9]], i32 0, i32 2
// CHECK3-NEXT: [[TMP18:%.*]] = load i32, ptr [[TMP16]], align 4
// CHECK3-NEXT: [[TMP19:%.*]] = load i32, ptr [[TMP17]], align 4
-// CHECK3-NEXT: store i32 1, ptr [[KERNEL_ARGS_I]], align 4, !noalias !22
-// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 1
-// CHECK3-NEXT: store i32 3, ptr [[TMP20]], align 4, !noalias !22
-// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP13]], ptr [[TMP21]], align 4, !noalias !22
-// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP14]], ptr [[TMP22]], align 4, !noalias !22
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 4
-// CHECK3-NEXT: store ptr [[TMP15]], ptr [[TMP23]], align 4, !noalias !22
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP24]], align 4, !noalias !22
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP25]], align 4, !noalias !22
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP26]], align 4, !noalias !22
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 8
-// CHECK3-NEXT: store i64 10, ptr [[TMP27]], align 8, !noalias !22
-// CHECK3-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel_nowait(ptr @[[GLOB2]], i64 -1, i32 [[TMP18]], i32 [[TMP19]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l103.region_id, ptr [[KERNEL_ARGS_I]], i32 0, ptr null, i32 0, ptr null)
-// CHECK3-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
-// CHECK3-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__1_EXIT:%.*]]
+// CHECK3-NEXT: [[TMP20:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP18]], 0
+// CHECK3-NEXT: [[TMP21:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP19]], 0
+// CHECK3-NEXT: store i32 2, ptr [[KERNEL_ARGS_I]], align 4, !noalias !22
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 1
+// CHECK3-NEXT: store i32 3, ptr [[TMP22]], align 4, !noalias !22
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP13]], ptr [[TMP23]], align 4, !noalias !22
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP14]], ptr [[TMP24]], align 4, !noalias !22
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 4
+// CHECK3-NEXT: store ptr [[TMP15]], ptr [[TMP25]], align 4, !noalias !22
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP26]], align 4, !noalias !22
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP27]], align 4, !noalias !22
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4, !noalias !22
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 8
+// CHECK3-NEXT: store i64 10, ptr [[TMP29]], align 8, !noalias !22
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 9
+// CHECK3-NEXT: store i64 1, ptr [[TMP30]], align 8, !noalias !22
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [[TMP20]], ptr [[TMP31]], align 4, !noalias !22
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [[TMP21]], ptr [[TMP32]], align 4, !noalias !22
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP33]], align 4, !noalias !22
+// CHECK3-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 [[TMP18]], i32 [[TMP19]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l103.region_id, ptr [[KERNEL_ARGS_I]])
+// CHECK3-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK3-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__1_EXIT:%.*]]
// CHECK3: omp_offload.failed.i:
-// CHECK3-NEXT: [[TMP30:%.*]] = load i16, ptr [[TMP12]], align 2
-// CHECK3-NEXT: store i16 [[TMP30]], ptr [[AA_CASTED_I]], align 2, !noalias !22
-// CHECK3-NEXT: [[TMP31:%.*]] = load i32, ptr [[AA_CASTED_I]], align 4, !noalias !22
-// CHECK3-NEXT: [[TMP32:%.*]] = load i32, ptr [[TMP16]], align 4
-// CHECK3-NEXT: store i32 [[TMP32]], ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 4, !noalias !22
-// CHECK3-NEXT: [[TMP33:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 4, !noalias !22
-// CHECK3-NEXT: [[TMP34:%.*]] = load i32, ptr [[TMP17]], align 4
-// CHECK3-NEXT: store i32 [[TMP34]], ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 4, !noalias !22
-// CHECK3-NEXT: [[TMP35:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 4, !noalias !22
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l103(i32 [[TMP31]], i32 [[TMP33]], i32 [[TMP35]]) #[[ATTR3]]
+// CHECK3-NEXT: [[TMP36:%.*]] = load i16, ptr [[TMP12]], align 2
+// CHECK3-NEXT: store i16 [[TMP36]], ptr [[AA_CASTED_I]], align 2, !noalias !22
+// CHECK3-NEXT: [[TMP37:%.*]] = load i32, ptr [[AA_CASTED_I]], align 4, !noalias !22
+// CHECK3-NEXT: [[TMP38:%.*]] = load i32, ptr [[TMP16]], align 4
+// CHECK3-NEXT: store i32 [[TMP38]], ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 4, !noalias !22
+// CHECK3-NEXT: [[TMP39:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 4, !noalias !22
+// CHECK3-NEXT: [[TMP40:%.*]] = load i32, ptr [[TMP17]], align 4
+// CHECK3-NEXT: store i32 [[TMP40]], ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 4, !noalias !22
+// CHECK3-NEXT: [[TMP41:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 4, !noalias !22
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l103(i32 [[TMP37]], i32 [[TMP39]], i32 [[TMP41]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[DOTOMP_OUTLINED__1_EXIT]]
// CHECK3: .omp_outlined..1.exit:
// CHECK3-NEXT: ret i32 0
@@ -3050,7 +3142,7 @@ int bar(int n){
// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [5 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP29]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP29]], align 4
// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 5, ptr [[TMP30]], align 4
// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3067,9 +3159,17 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP36]], align 4
// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 10, ptr [[TMP37]], align 8
-// CHECK3-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l218.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK3-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP38]], align 8
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP39]], align 4
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l218.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
+// CHECK3-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l218(ptr [[THIS1]], i32 [[TMP5]], i32 2, i32 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -3079,15 +3179,15 @@ int bar(int n){
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l218(ptr [[THIS1]], i32 [[TMP5]], i32 2, i32 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
-// CHECK3-NEXT: [[TMP40:%.*]] = mul nsw i32 1, [[TMP1]]
-// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i32 [[TMP40]]
+// CHECK3-NEXT: [[TMP44:%.*]] = mul nsw i32 1, [[TMP1]]
+// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i32 [[TMP44]]
// CHECK3-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i16, ptr [[ARRAYIDX]], i32 1
-// CHECK3-NEXT: [[TMP41:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
-// CHECK3-NEXT: [[CONV:%.*]] = sext i16 [[TMP41]] to i32
-// CHECK3-NEXT: [[TMP42:%.*]] = load i32, ptr [[B]], align 4
-// CHECK3-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP42]]
-// CHECK3-NEXT: [[TMP43:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK3-NEXT: call void @llvm.stackrestore(ptr [[TMP43]])
+// CHECK3-NEXT: [[TMP45:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
+// CHECK3-NEXT: [[CONV:%.*]] = sext i16 [[TMP45]] to i32
+// CHECK3-NEXT: [[TMP46:%.*]] = load i32, ptr [[B]], align 4
+// CHECK3-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP46]]
+// CHECK3-NEXT: [[TMP47:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK3-NEXT: call void @llvm.stackrestore(ptr [[TMP47]])
// CHECK3-NEXT: ret i32 [[ADD3]]
//
//
@@ -3179,7 +3279,7 @@ int bar(int n){
// CHECK3-NEXT: [[TMP31:%.*]] = zext i32 [[ADD5]] to i64
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP32]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP32]], align 4
// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 5, ptr [[TMP33]], align 4
// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3196,9 +3296,17 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP39]], align 4
// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 [[TMP31]], ptr [[TMP40]], align 8
-// CHECK3-NEXT: [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l200.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
-// CHECK3-NEXT: br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP41]], align 8
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP42]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP43]], align 4
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP44]], align 4
+// CHECK3-NEXT: [[TMP45:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l200.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP46:%.*]] = icmp ne i32 [[TMP45]], 0
+// CHECK3-NEXT: br i1 [[TMP46]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l200(i32 [[TMP1]], i32 [[TMP3]], i32 [[TMP5]], i32 [[TMP7]], ptr [[B]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -3208,8 +3316,8 @@ int bar(int n){
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l200(i32 [[TMP1]], i32 [[TMP3]], i32 [[TMP5]], i32 [[TMP7]], ptr [[B]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
-// CHECK3-NEXT: [[TMP43:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: ret i32 [[TMP43]]
+// CHECK3-NEXT: [[TMP47:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: ret i32 [[TMP47]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_Z9ftemplateIiET_i
@@ -3260,7 +3368,7 @@ int bar(int n){
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP16]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP16]], align 4
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 3, ptr [[TMP17]], align 4
// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3277,9 +3385,17 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP23]], align 4
// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 10, ptr [[TMP24]], align 8
-// CHECK3-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l183.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK3-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP25]], align 8
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l183.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK3-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l183(i32 [[TMP1]], i32 [[TMP3]], ptr [[B]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -3289,8 +3405,8 @@ int bar(int n){
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l183(i32 [[TMP1]], i32 [[TMP3]], ptr [[B]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
-// CHECK3-NEXT: [[TMP27:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: ret i32 [[TMP27]]
+// CHECK3-NEXT: [[TMP31:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: ret i32 [[TMP31]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l218
diff --git a/clang/test/OpenMP/target_teams_distribute_collapse_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_collapse_codegen.cpp
index b39440c39598a..67b2acf3fa485 100644
--- a/clang/test/OpenMP/target_teams_distribute_collapse_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_collapse_codegen.cpp
@@ -124,7 +124,7 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -141,9 +141,17 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 56088, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -151,8 +159,8 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[A3:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x [456 x i32]], ptr [[A3]], i64 0, i64 0
// CHECK1-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds [456 x i32], ptr [[ARRAYIDX]], i64 0, i64 0
-// CHECK1-NEXT: [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
-// CHECK1-NEXT: ret i32 [[TMP16]]
+// CHECK1-NEXT: [[TMP20:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
+// CHECK1-NEXT: ret i32 [[TMP20]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -284,7 +292,7 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -301,9 +309,17 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 56088, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -311,8 +327,8 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[A3:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x [456 x i32]], ptr [[A3]], i32 0, i32 0
// CHECK3-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds [456 x i32], ptr [[ARRAYIDX]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
-// CHECK3-NEXT: ret i32 [[TMP16]]
+// CHECK3-NEXT: [[TMP20:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
+// CHECK3-NEXT: ret i32 [[TMP20]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -512,7 +528,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[ADD:%.*]] = add nsw i64 [[TMP35]], 1
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP36]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP36]], align 4
// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 5, ptr [[TMP37]], align 4
// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -529,20 +545,28 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP43]], align 8
// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 [[ADD]], ptr [[TMP44]], align 8
-// CHECK9-NEXT: [[TMP45:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l80.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP46:%.*]] = icmp ne i32 [[TMP45]], 0
-// CHECK9-NEXT: br i1 [[TMP46]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP46]], align 4
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP47]], align 4
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP48]], align 4
+// CHECK9-NEXT: [[TMP49:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l80.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP50:%.*]] = icmp ne i32 [[TMP49]], 0
+// CHECK9-NEXT: br i1 [[TMP50]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l80(i64 [[TMP7]], i64 [[TMP9]], i64 [[TMP1]], i64 [[TMP3]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP47:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef signext [[TMP47]])
+// CHECK9-NEXT: [[TMP51:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef signext [[TMP51]])
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK9-NEXT: [[TMP48:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP48]])
-// CHECK9-NEXT: [[TMP49:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP49]]
+// CHECK9-NEXT: [[TMP52:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP52]])
+// CHECK9-NEXT: [[TMP53:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP53]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l80
@@ -740,7 +764,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK9-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK9-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -757,9 +781,17 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 20, ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l67.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK9-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK9-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l67.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK9-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l67(ptr [[A]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -964,7 +996,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[ADD:%.*]] = add nsw i64 [[TMP34]], 1
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP35]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP35]], align 4
// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 5, ptr [[TMP36]], align 4
// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -981,20 +1013,28 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP42]], align 4
// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 [[ADD]], ptr [[TMP43]], align 8
-// CHECK11-NEXT: [[TMP44:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l80.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP45:%.*]] = icmp ne i32 [[TMP44]], 0
-// CHECK11-NEXT: br i1 [[TMP45]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP44]], align 8
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP47]], align 4
+// CHECK11-NEXT: [[TMP48:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l80.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP49:%.*]] = icmp ne i32 [[TMP48]], 0
+// CHECK11-NEXT: br i1 [[TMP49]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l80(i32 [[TMP5]], i32 [[TMP7]], i32 [[TMP0]], i32 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP46:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef [[TMP46]])
+// CHECK11-NEXT: [[TMP50:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef [[TMP50]])
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK11-NEXT: [[TMP47:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP47]])
-// CHECK11-NEXT: [[TMP48:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP48]]
+// CHECK11-NEXT: [[TMP51:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP51]])
+// CHECK11-NEXT: [[TMP52:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP52]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l80
@@ -1190,7 +1230,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK11-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK11-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1207,9 +1247,17 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK11-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 20, ptr [[TMP13]], align 8
-// CHECK11-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l67.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK11-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK11-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l67.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK11-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l67(ptr [[A]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/target_teams_distribute_depend_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_depend_codegen.cpp
index 56d28b2f4ce4e..6b66fab3ec427 100644
--- a/clang/test/OpenMP/target_teams_distribute_depend_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_depend_codegen.cpp
@@ -200,7 +200,7 @@ int foo(int n) {
// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2
// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]],
// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64
-// CHECK: [[RET:%.+]] = call i32 @__tgt_target_kernel_nowait(%struct.ident_t* @{{.+}}, i64 [[DEVICE]], i32 0, i32 0, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* %{{.+}})
+// CHECK: [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 [[DEVICE]], i32 0, i32 0, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* %{{.+}})
// CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]]
diff --git a/clang/test/OpenMP/target_teams_distribute_dist_schedule_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_dist_schedule_codegen.cpp
index 0f67103597c19..4c4dab839a365 100644
--- a/clang/test/OpenMP/target_teams_distribute_dist_schedule_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_dist_schedule_codegen.cpp
@@ -159,7 +159,7 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -176,87 +176,111 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.2, ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l32.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.2, ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l32.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK1-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK1-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK1: omp_offload.failed8:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l32(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK1: omp_offload.cont9:
// CHECK1-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A10]], ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A10]], ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.5, ptr [[TMP41]], align 8
-// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP42]], align 8
-// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP43]], align 8
-// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP44]], align 8
-// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK1-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK1-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK1-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 8
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 8
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.5, ptr [[TMP49]], align 8
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP50]], align 8
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP51]], align 8
+// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP52]], align 8
+// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK1-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK1-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK1-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK1: omp_offload.failed16:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK1: omp_offload.cont17:
// CHECK1-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A18]], i64 0, i64 0
-// CHECK1-NEXT: [[TMP48:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK1-NEXT: ret i32 [[TMP48]]
+// CHECK1-NEXT: [[TMP60:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK1-NEXT: ret i32 [[TMP60]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -554,7 +578,7 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -571,87 +595,111 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 4
-// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A2]], ptr [[TMP17]], align 4
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 4
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A2]], ptr [[TMP21]], align 4
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.2, ptr [[TMP25]], align 4
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP26]], align 4
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK3-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l32.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK3-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK3-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.2, ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK3-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l32.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK3-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK3-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK3: omp_offload.failed8:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l32(ptr [[THIS1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK3: omp_offload.cont9:
// CHECK3-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 4
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A10]], ptr [[TMP33]], align 4
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP34]], align 4
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A10]], ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP42]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 4
-// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 4
-// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.5, ptr [[TMP41]], align 4
-// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP42]], align 4
-// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP44]], align 4
-// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK3-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK3-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK3-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK3-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 4
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 4
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.5, ptr [[TMP49]], align 4
+// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP50]], align 4
+// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP52]], align 4
+// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK3-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK3-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK3-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK3-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK3: omp_offload.failed16:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36(ptr [[THIS1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK3: omp_offload.cont17:
// CHECK3-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A18]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP48:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK3-NEXT: ret i32 [[TMP48]]
+// CHECK3-NEXT: [[TMP60:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK3-NEXT: ret i32 [[TMP60]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -994,7 +1042,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1011,159 +1059,183 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK9-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK9-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK9-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94(i64 [[TMP4]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK9-NEXT: [[TMP35:%.*]] = load i64, ptr [[N_CASTED3]], align 8
-// CHECK9-NEXT: [[TMP36:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK9-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK9-NEXT: [[TMP39:%.*]] = load i64, ptr [[N_CASTED3]], align 8
+// CHECK9-NEXT: [[TMP40:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK9-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES7]], ptr align 8 @.offload_sizes.2, i64 24, i1 false)
-// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP35]], ptr [[TMP37]], align 8
-// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP35]], ptr [[TMP38]], align 8
-// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP40]], align 8
-// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP41]], align 8
-// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP42]], align 8
-// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP43]], align 8
-// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 8
-// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK9-NEXT: store i64 [[TMP36]], ptr [[TMP45]], align 8
-// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP39]], ptr [[TMP41]], align 8
+// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP39]], ptr [[TMP42]], align 8
+// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP44]], align 8
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
// CHECK9-NEXT: store ptr null, ptr [[TMP46]], align 8
-// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP50:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP50]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK9-NEXT: [[TMP51:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK9-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP51]], 0
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP47]], align 8
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 8
+// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK9-NEXT: store i64 [[TMP40]], ptr [[TMP49]], align 8
+// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP50]], align 8
+// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP54:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP54]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK9-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK9-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP55]], 0
// CHECK9-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK9-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK9-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK9-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK9-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP52]], 1
-// CHECK9-NEXT: [[TMP53:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK9-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK9-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP56]], 1
+// CHECK9-NEXT: [[TMP57:%.*]] = zext i32 [[ADD14]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK9-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK9-NEXT: store i32 3, ptr [[TMP55]], align 4
-// CHECK9-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP47]], ptr [[TMP56]], align 8
-// CHECK9-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 8
-// CHECK9-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 8
-// CHECK9-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP59]], align 8
-// CHECK9-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP60]], align 8
-// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP61]], align 8
-// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP53]], ptr [[TMP62]], align 8
-// CHECK9-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l98.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK9-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
-// CHECK9-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK9-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP58]], align 4
+// CHECK9-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK9-NEXT: store i32 3, ptr [[TMP59]], align 4
+// CHECK9-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP51]], ptr [[TMP60]], align 8
+// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 8
+// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 8
+// CHECK9-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP63]], align 8
+// CHECK9-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP64]], align 8
+// CHECK9-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP65]], align 8
+// CHECK9-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP57]], ptr [[TMP66]], align 8
+// CHECK9-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP67]], align 8
+// CHECK9-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP68]], align 4
+// CHECK9-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK9-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP70]], align 4
+// CHECK9-NEXT: [[TMP71:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l98.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK9-NEXT: [[TMP72:%.*]] = icmp ne i32 [[TMP71]], 0
+// CHECK9-NEXT: br i1 [[TMP72]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK9: omp_offload.failed16:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l98(i64 [[TMP35]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l98(i64 [[TMP39]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK9: omp_offload.cont17:
-// CHECK9-NEXT: [[TMP65:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP65]], ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK9-NEXT: [[TMP66:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP66]], ptr [[N_CASTED19]], align 4
-// CHECK9-NEXT: [[TMP67:%.*]] = load i64, ptr [[N_CASTED19]], align 8
-// CHECK9-NEXT: [[TMP68:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK9-NEXT: store i32 [[TMP68]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK9-NEXT: [[TMP69:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK9-NEXT: [[TMP70:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK9-NEXT: [[TMP73:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP73]], ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK9-NEXT: [[TMP74:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP74]], ptr [[N_CASTED19]], align 4
+// CHECK9-NEXT: [[TMP75:%.*]] = load i64, ptr [[N_CASTED19]], align 8
+// CHECK9-NEXT: [[TMP76:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK9-NEXT: store i32 [[TMP76]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK9-NEXT: [[TMP77:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK9-NEXT: [[TMP78:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK9-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES23]], ptr align 8 @.offload_sizes.5, i64 32, i1 false)
-// CHECK9-NEXT: [[TMP71:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP67]], ptr [[TMP71]], align 8
-// CHECK9-NEXT: [[TMP72:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP67]], ptr [[TMP72]], align 8
-// CHECK9-NEXT: [[TMP73:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP73]], align 8
-// CHECK9-NEXT: [[TMP74:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP74]], align 8
-// CHECK9-NEXT: [[TMP75:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP75]], align 8
-// CHECK9-NEXT: [[TMP76:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP76]], align 8
-// CHECK9-NEXT: [[TMP77:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP77]], align 8
-// CHECK9-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP78]], align 8
-// CHECK9-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
-// CHECK9-NEXT: store i64 [[TMP70]], ptr [[TMP79]], align 8
-// CHECK9-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP80]], align 8
-// CHECK9-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
-// CHECK9-NEXT: store i64 [[TMP69]], ptr [[TMP81]], align 8
-// CHECK9-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
-// CHECK9-NEXT: store i64 [[TMP69]], ptr [[TMP82]], align 8
-// CHECK9-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP83]], align 8
-// CHECK9-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP87:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP87]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK9-NEXT: [[TMP88:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK9-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP88]], 0
+// CHECK9-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP75]], ptr [[TMP79]], align 8
+// CHECK9-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP75]], ptr [[TMP80]], align 8
+// CHECK9-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP81]], align 8
+// CHECK9-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP82]], align 8
+// CHECK9-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP83]], align 8
+// CHECK9-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP84]], align 8
+// CHECK9-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP85]], align 8
+// CHECK9-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP86]], align 8
+// CHECK9-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
+// CHECK9-NEXT: store i64 [[TMP78]], ptr [[TMP87]], align 8
+// CHECK9-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP88]], align 8
+// CHECK9-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
+// CHECK9-NEXT: store i64 [[TMP77]], ptr [[TMP89]], align 8
+// CHECK9-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
+// CHECK9-NEXT: store i64 [[TMP77]], ptr [[TMP90]], align 8
+// CHECK9-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP91]], align 8
+// CHECK9-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP94:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP95:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP95]], ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK9-NEXT: [[TMP96:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK9-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP96]], 0
// CHECK9-NEXT: [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1
// CHECK9-NEXT: [[SUB29:%.*]] = sub nsw i32 [[DIV28]], 1
// CHECK9-NEXT: store i32 [[SUB29]], ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK9-NEXT: [[TMP89:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK9-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP89]], 1
-// CHECK9-NEXT: [[TMP90:%.*]] = zext i32 [[ADD30]] to i64
+// CHECK9-NEXT: [[TMP97:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
+// CHECK9-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP97]], 1
+// CHECK9-NEXT: [[TMP98:%.*]] = zext i32 [[ADD30]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP91]], align 4
-// CHECK9-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK9-NEXT: store i32 4, ptr [[TMP92]], align 4
-// CHECK9-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP84]], ptr [[TMP93]], align 8
-// CHECK9-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP85]], ptr [[TMP94]], align 8
-// CHECK9-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP86]], ptr [[TMP95]], align 8
-// CHECK9-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP96]], align 8
-// CHECK9-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP97]], align 8
-// CHECK9-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP98]], align 8
-// CHECK9-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP90]], ptr [[TMP99]], align 8
-// CHECK9-NEXT: [[TMP100:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK9-NEXT: [[TMP101:%.*]] = icmp ne i32 [[TMP100]], 0
-// CHECK9-NEXT: br i1 [[TMP101]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK9-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP99]], align 4
+// CHECK9-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK9-NEXT: store i32 4, ptr [[TMP100]], align 4
+// CHECK9-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP92]], ptr [[TMP101]], align 8
+// CHECK9-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP93]], ptr [[TMP102]], align 8
+// CHECK9-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP94]], ptr [[TMP103]], align 8
+// CHECK9-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP104]], align 8
+// CHECK9-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP105]], align 8
+// CHECK9-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP106]], align 8
+// CHECK9-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP98]], ptr [[TMP107]], align 8
+// CHECK9-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP108]], align 8
+// CHECK9-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP109]], align 4
+// CHECK9-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP110]], align 4
+// CHECK9-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP111]], align 4
+// CHECK9-NEXT: [[TMP112:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK9-NEXT: [[TMP113:%.*]] = icmp ne i32 [[TMP112]], 0
+// CHECK9-NEXT: br i1 [[TMP113]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK9: omp_offload.failed32:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102(i64 [[TMP67]], i64 [[TMP1]], ptr [[VLA]], i64 [[TMP69]]) #[[ATTR3]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102(i64 [[TMP75]], i64 [[TMP1]], ptr [[VLA]], i64 [[TMP77]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK9: omp_offload.cont33:
-// CHECK9-NEXT: [[TMP102:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP102]])
+// CHECK9-NEXT: [[TMP114:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP114]])
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK9-NEXT: [[TMP103:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP103]])
-// CHECK9-NEXT: [[TMP104:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP104]]
+// CHECK9-NEXT: [[TMP115:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP115]])
+// CHECK9-NEXT: [[TMP116:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP116]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94
@@ -1559,7 +1631,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK9-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK9-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1576,77 +1648,101 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l76.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK9-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK9-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l76.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK9-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l76(ptr [[A]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP16]], align 8
-// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP17]], align 8
-// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP20]], align 8
+// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP21]], align 8
+// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK9-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.11, ptr [[TMP25]], align 8
-// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP26]], align 8
-// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK9-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK9-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l80.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK9-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK9-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK9-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.11, ptr [[TMP29]], align 8
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP30]], align 8
+// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK9-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK9-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l80.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK9-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK9-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK9: omp_offload.failed6:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l80(ptr [[A]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK9: omp_offload.cont7:
-// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP32]], align 8
-// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP33]], align 8
-// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP34]], align 8
-// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP40]], align 8
+// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP41]], align 8
+// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP42]], align 8
+// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK9-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
-// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
-// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.14, ptr [[TMP41]], align 8
-// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.15, ptr [[TMP42]], align 8
-// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP43]], align 8
-// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP44]], align 8
-// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK9-NEXT: store i64 10, ptr [[TMP45]], align 8
-// CHECK9-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK9-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK9-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK9-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 8
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 8
+// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.14, ptr [[TMP49]], align 8
+// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.15, ptr [[TMP50]], align 8
+// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP51]], align 8
+// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP52]], align 8
+// CHECK9-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK9-NEXT: store i64 10, ptr [[TMP53]], align 8
+// CHECK9-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK9-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK9-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK9-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK9-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK9-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK9-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK9: omp_offload.failed13:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84(ptr [[A]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT14]]
@@ -1994,7 +2090,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2011,161 +2107,185 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK11-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK11-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK11-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94(i32 [[TMP3]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK11-NEXT: [[TMP35:%.*]] = load i32, ptr [[N_CASTED3]], align 4
-// CHECK11-NEXT: [[TMP36:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK11-NEXT: [[TMP37:%.*]] = sext i32 [[TMP36]] to i64
+// CHECK11-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK11-NEXT: [[TMP39:%.*]] = load i32, ptr [[N_CASTED3]], align 4
+// CHECK11-NEXT: [[TMP40:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK11-NEXT: [[TMP41:%.*]] = sext i32 [[TMP40]] to i64
// CHECK11-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES7]], ptr align 4 @.offload_sizes.2, i32 24, i1 false)
-// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP35]], ptr [[TMP38]], align 4
-// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP35]], ptr [[TMP39]], align 4
-// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP40]], align 4
-// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP41]], align 4
-// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP42]], align 4
-// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 4
-// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP45]], align 4
-// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK11-NEXT: store i64 [[TMP37]], ptr [[TMP46]], align 4
-// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP39]], ptr [[TMP42]], align 4
+// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP39]], ptr [[TMP43]], align 4
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP44]], align 4
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
// CHECK11-NEXT: store ptr null, ptr [[TMP47]], align 4
-// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP51:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP51]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK11-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK11-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP52]], 0
+// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 4
+// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP49]], align 4
+// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK11-NEXT: store i64 [[TMP41]], ptr [[TMP50]], align 4
+// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP54:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP55:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP55]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK11-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK11-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP56]], 0
// CHECK11-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK11-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK11-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK11-NEXT: [[TMP53:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK11-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP53]], 1
-// CHECK11-NEXT: [[TMP54:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK11-NEXT: [[TMP57:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK11-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP57]], 1
+// CHECK11-NEXT: [[TMP58:%.*]] = zext i32 [[ADD14]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP55]], align 4
-// CHECK11-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK11-NEXT: store i32 3, ptr [[TMP56]], align 4
-// CHECK11-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 4
-// CHECK11-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 4
-// CHECK11-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP50]], ptr [[TMP59]], align 4
-// CHECK11-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP60]], align 4
-// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP61]], align 4
-// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP62]], align 4
-// CHECK11-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP54]], ptr [[TMP63]], align 8
-// CHECK11-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l98.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK11-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
-// CHECK11-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK11-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP59]], align 4
+// CHECK11-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK11-NEXT: store i32 3, ptr [[TMP60]], align 4
+// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 4
+// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 4
+// CHECK11-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP54]], ptr [[TMP63]], align 4
+// CHECK11-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP64]], align 4
+// CHECK11-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP65]], align 4
+// CHECK11-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP66]], align 4
+// CHECK11-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP58]], ptr [[TMP67]], align 8
+// CHECK11-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP68]], align 8
+// CHECK11-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK11-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP70]], align 4
+// CHECK11-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP71]], align 4
+// CHECK11-NEXT: [[TMP72:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l98.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK11-NEXT: [[TMP73:%.*]] = icmp ne i32 [[TMP72]], 0
+// CHECK11-NEXT: br i1 [[TMP73]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK11: omp_offload.failed16:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l98(i32 [[TMP35]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l98(i32 [[TMP39]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK11: omp_offload.cont17:
-// CHECK11-NEXT: [[TMP66:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP66]], ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK11-NEXT: [[TMP67:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP67]], ptr [[N_CASTED19]], align 4
-// CHECK11-NEXT: [[TMP68:%.*]] = load i32, ptr [[N_CASTED19]], align 4
-// CHECK11-NEXT: [[TMP69:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK11-NEXT: store i32 [[TMP69]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK11-NEXT: [[TMP70:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK11-NEXT: [[TMP71:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK11-NEXT: [[TMP72:%.*]] = sext i32 [[TMP71]] to i64
+// CHECK11-NEXT: [[TMP74:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP74]], ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK11-NEXT: [[TMP75:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP75]], ptr [[N_CASTED19]], align 4
+// CHECK11-NEXT: [[TMP76:%.*]] = load i32, ptr [[N_CASTED19]], align 4
+// CHECK11-NEXT: [[TMP77:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK11-NEXT: store i32 [[TMP77]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK11-NEXT: [[TMP78:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK11-NEXT: [[TMP79:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK11-NEXT: [[TMP80:%.*]] = sext i32 [[TMP79]] to i64
// CHECK11-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES23]], ptr align 4 @.offload_sizes.5, i32 32, i1 false)
-// CHECK11-NEXT: [[TMP73:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP68]], ptr [[TMP73]], align 4
-// CHECK11-NEXT: [[TMP74:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP68]], ptr [[TMP74]], align 4
-// CHECK11-NEXT: [[TMP75:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP75]], align 4
-// CHECK11-NEXT: [[TMP76:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP76]], align 4
-// CHECK11-NEXT: [[TMP77:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP77]], align 4
-// CHECK11-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP78]], align 4
-// CHECK11-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP79]], align 4
-// CHECK11-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP80]], align 4
-// CHECK11-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
-// CHECK11-NEXT: store i64 [[TMP72]], ptr [[TMP81]], align 4
-// CHECK11-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP82]], align 4
-// CHECK11-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
-// CHECK11-NEXT: store i32 [[TMP70]], ptr [[TMP83]], align 4
-// CHECK11-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
-// CHECK11-NEXT: store i32 [[TMP70]], ptr [[TMP84]], align 4
-// CHECK11-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP85]], align 4
-// CHECK11-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP89:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP89]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK11-NEXT: [[TMP90:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK11-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP90]], 0
+// CHECK11-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP76]], ptr [[TMP81]], align 4
+// CHECK11-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP76]], ptr [[TMP82]], align 4
+// CHECK11-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP83]], align 4
+// CHECK11-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP84]], align 4
+// CHECK11-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP85]], align 4
+// CHECK11-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP86]], align 4
+// CHECK11-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP87]], align 4
+// CHECK11-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP88]], align 4
+// CHECK11-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
+// CHECK11-NEXT: store i64 [[TMP80]], ptr [[TMP89]], align 4
+// CHECK11-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP90]], align 4
+// CHECK11-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
+// CHECK11-NEXT: store i32 [[TMP78]], ptr [[TMP91]], align 4
+// CHECK11-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
+// CHECK11-NEXT: store i32 [[TMP78]], ptr [[TMP92]], align 4
+// CHECK11-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP93]], align 4
+// CHECK11-NEXT: [[TMP94:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP95:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP96:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP97:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP97]], ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK11-NEXT: [[TMP98:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK11-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP98]], 0
// CHECK11-NEXT: [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1
// CHECK11-NEXT: [[SUB29:%.*]] = sub nsw i32 [[DIV28]], 1
// CHECK11-NEXT: store i32 [[SUB29]], ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK11-NEXT: [[TMP91:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK11-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP91]], 1
-// CHECK11-NEXT: [[TMP92:%.*]] = zext i32 [[ADD30]] to i64
+// CHECK11-NEXT: [[TMP99:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
+// CHECK11-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP99]], 1
+// CHECK11-NEXT: [[TMP100:%.*]] = zext i32 [[ADD30]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP93]], align 4
-// CHECK11-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK11-NEXT: store i32 4, ptr [[TMP94]], align 4
-// CHECK11-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP86]], ptr [[TMP95]], align 4
-// CHECK11-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP87]], ptr [[TMP96]], align 4
-// CHECK11-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP88]], ptr [[TMP97]], align 4
-// CHECK11-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP98]], align 4
-// CHECK11-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP99]], align 4
-// CHECK11-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP100]], align 4
-// CHECK11-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP92]], ptr [[TMP101]], align 8
-// CHECK11-NEXT: [[TMP102:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK11-NEXT: [[TMP103:%.*]] = icmp ne i32 [[TMP102]], 0
-// CHECK11-NEXT: br i1 [[TMP103]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK11-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP101]], align 4
+// CHECK11-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK11-NEXT: store i32 4, ptr [[TMP102]], align 4
+// CHECK11-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP94]], ptr [[TMP103]], align 4
+// CHECK11-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP95]], ptr [[TMP104]], align 4
+// CHECK11-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP96]], ptr [[TMP105]], align 4
+// CHECK11-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP106]], align 4
+// CHECK11-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP107]], align 4
+// CHECK11-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP108]], align 4
+// CHECK11-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP100]], ptr [[TMP109]], align 8
+// CHECK11-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP110]], align 8
+// CHECK11-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP111]], align 4
+// CHECK11-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP112]], align 4
+// CHECK11-NEXT: [[TMP113:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP113]], align 4
+// CHECK11-NEXT: [[TMP114:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK11-NEXT: [[TMP115:%.*]] = icmp ne i32 [[TMP114]], 0
+// CHECK11-NEXT: br i1 [[TMP115]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK11: omp_offload.failed32:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102(i32 [[TMP68]], i32 [[TMP0]], ptr [[VLA]], i32 [[TMP70]]) #[[ATTR3]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102(i32 [[TMP76]], i32 [[TMP0]], ptr [[VLA]], i32 [[TMP78]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK11: omp_offload.cont33:
-// CHECK11-NEXT: [[TMP104:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP104]])
+// CHECK11-NEXT: [[TMP116:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP116]])
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK11-NEXT: [[TMP105:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP105]])
-// CHECK11-NEXT: [[TMP106:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP106]]
+// CHECK11-NEXT: [[TMP117:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP117]])
+// CHECK11-NEXT: [[TMP118:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP118]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94
@@ -2558,7 +2678,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK11-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK11-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2575,77 +2695,101 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK11-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK11-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l76.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK11-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK11-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l76.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK11-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l76(ptr [[A]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP16]], align 4
-// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP17]], align 4
-// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP20]], align 4
+// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP21]], align 4
+// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK11-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.11, ptr [[TMP25]], align 4
-// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP26]], align 4
-// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK11-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK11-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l80.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK11-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK11-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK11-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.11, ptr [[TMP29]], align 4
+// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP30]], align 4
+// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK11-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK11-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l80.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK11-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK11-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK11: omp_offload.failed6:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l80(ptr [[A]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK11: omp_offload.cont7:
-// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP32]], align 4
-// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP33]], align 4
-// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP34]], align 4
-// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP40]], align 4
+// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP41]], align 4
+// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP42]], align 4
+// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK11-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 4
-// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 4
-// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.14, ptr [[TMP41]], align 4
-// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.15, ptr [[TMP42]], align 4
-// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP44]], align 4
-// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK11-NEXT: store i64 10, ptr [[TMP45]], align 8
-// CHECK11-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK11-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK11-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK11-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 4
+// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 4
+// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.14, ptr [[TMP49]], align 4
+// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.15, ptr [[TMP50]], align 4
+// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP52]], align 4
+// CHECK11-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK11-NEXT: store i64 10, ptr [[TMP53]], align 8
+// CHECK11-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK11-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK11-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK11-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK11-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK11-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK11-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK11: omp_offload.failed13:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84(ptr [[A]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT14]]
diff --git a/clang/test/OpenMP/target_teams_distribute_firstprivate_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_firstprivate_codegen.cpp
index db0aff10e21c3..93a869acd1369 100644
--- a/clang/test/OpenMP/target_teams_distribute_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_firstprivate_codegen.cpp
@@ -303,7 +303,7 @@ int main() {
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 5, ptr [[TMP22]], align 4
// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -320,9 +320,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK1-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK1-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92(ptr @vec, i64 [[TMP1]], ptr @s_arr, ptr @var, i64 [[TMP3]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -574,7 +582,7 @@ int main() {
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -591,27 +599,35 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP25]], align 8
// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK1-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP29]], align 4
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK1-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56(ptr [[VEC]], i64 [[TMP2]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK1-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK1: arraydestroy.body:
-// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP29]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP33]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK1-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK1-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK1: arraydestroy.done2:
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK1-NEXT: [[TMP30:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK1-NEXT: ret i32 [[TMP30]]
+// CHECK1-NEXT: [[TMP34:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK1-NEXT: ret i32 [[TMP34]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN2StC2Ev
@@ -1088,7 +1104,7 @@ int main() {
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 5, ptr [[TMP22]], align 4
// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1105,9 +1121,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4
// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP29]], align 8
-// CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK3-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK3-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK3-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92(ptr @vec, i32 [[TMP1]], ptr @s_arr, ptr @var, i32 [[TMP3]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1357,7 +1381,7 @@ int main() {
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1374,27 +1398,35 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP25]], align 4
// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP26]], align 8
-// CHECK3-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK3-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK3-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56(ptr [[VEC]], i32 [[TMP2]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK3-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK3-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK3: arraydestroy.body:
-// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP29]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP33]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK3-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK3-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK3: arraydestroy.done2:
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK3-NEXT: [[TMP30:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK3-NEXT: ret i32 [[TMP30]]
+// CHECK3-NEXT: [[TMP34:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK3-NEXT: ret i32 [[TMP34]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN2StC2Ev
diff --git a/clang/test/OpenMP/target_teams_distribute_lastprivate_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_lastprivate_codegen.cpp
index d5d493a12bfa7..03bf0ec637c62 100644
--- a/clang/test/OpenMP/target_teams_distribute_lastprivate_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_lastprivate_codegen.cpp
@@ -536,7 +536,7 @@ int main() {
// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 5, ptr [[TMP24]], align 4
// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -553,9 +553,17 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP31]], align 8
-// CHECK9-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK9-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK9-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94(ptr [[VEC]], i64 [[TMP2]], ptr [[S_ARR]], ptr [[TMP3]], i64 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -563,18 +571,18 @@ int main() {
// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiET_v()
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP34]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP38]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP35:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP35]]
+// CHECK9-NEXT: [[TMP39:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP39]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -843,7 +851,7 @@ int main() {
// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -860,27 +868,35 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP25]], align 8
// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP26]], align 8
-// CHECK9-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK9-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP29]], align 4
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK9-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK9-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49(ptr [[VEC]], i64 [[TMP2]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP29]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP33]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP30:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP30]]
+// CHECK9-NEXT: [[TMP34:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP34]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
@@ -1228,7 +1244,7 @@ int main() {
// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 5, ptr [[TMP24]], align 4
// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1245,9 +1261,17 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP31]], align 8
-// CHECK11-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK11-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK11-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94(ptr [[VEC]], i32 [[TMP2]], ptr [[S_ARR]], ptr [[TMP3]], i32 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1255,18 +1279,18 @@ int main() {
// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiET_v()
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP34]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP38]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP35:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP35]]
+// CHECK11-NEXT: [[TMP39:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP39]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -1533,7 +1557,7 @@ int main() {
// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1550,27 +1574,35 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP25]], align 4
// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP26]], align 8
-// CHECK11-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK11-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP29]], align 4
+// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK11-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK11-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49(ptr [[VEC]], i32 [[TMP2]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP29]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP33]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP30:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP30]]
+// CHECK11-NEXT: [[TMP34:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP34]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
diff --git a/clang/test/OpenMP/target_teams_distribute_parallel_for_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_parallel_for_codegen.cpp
index caffcb48f13a2..1a09e0e93be6e 100644
--- a/clang/test/OpenMP/target_teams_distribute_parallel_for_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_parallel_for_codegen.cpp
@@ -648,95 +648,113 @@ int target_teams_fun(int *g){
// CHECK2-NEXT: [[TMP27:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_4]], align 4
// CHECK2-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP27]], 1
// CHECK2-NEXT: [[TMP28:%.*]] = zext i32 [[ADD]] to i64
+// CHECK2-NEXT: [[TMP29:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP23]], 0
+// CHECK2-NEXT: [[TMP30:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP24]], 0
// CHECK2-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK2-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK2-NEXT: store i32 1, ptr [[TMP29]], align 4
-// CHECK2-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK2-NEXT: store i32 4, ptr [[TMP30]], align 4
-// CHECK2-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK2-NEXT: store ptr [[TMP21]], ptr [[TMP31]], align 8
-// CHECK2-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK2-NEXT: store ptr [[TMP22]], ptr [[TMP32]], align 8
-// CHECK2-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK2-NEXT: store ptr @.offload_sizes, ptr [[TMP33]], align 8
-// CHECK2-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK2-NEXT: store ptr @.offload_maptypes, ptr [[TMP34]], align 8
-// CHECK2-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK2-NEXT: store ptr null, ptr [[TMP35]], align 8
-// CHECK2-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
-// CHECK2-NEXT: store ptr null, ptr [[TMP36]], align 8
-// CHECK2-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK2-NEXT: store i64 [[TMP28]], ptr [[TMP37]], align 8
-// CHECK2-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 [[TMP23]], i32 [[TMP24]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l51.region_id, ptr [[KERNEL_ARGS]])
-// CHECK2-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK2-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK2-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK2-NEXT: store i32 2, ptr [[TMP31]], align 4
+// CHECK2-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK2-NEXT: store i32 4, ptr [[TMP32]], align 4
+// CHECK2-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK2-NEXT: store ptr [[TMP21]], ptr [[TMP33]], align 8
+// CHECK2-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK2-NEXT: store ptr [[TMP22]], ptr [[TMP34]], align 8
+// CHECK2-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK2-NEXT: store ptr @.offload_sizes, ptr [[TMP35]], align 8
+// CHECK2-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK2-NEXT: store ptr @.offload_maptypes, ptr [[TMP36]], align 8
+// CHECK2-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
+// CHECK2-NEXT: store ptr null, ptr [[TMP37]], align 8
+// CHECK2-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK2-NEXT: store ptr null, ptr [[TMP38]], align 8
+// CHECK2-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK2-NEXT: store i64 [[TMP28]], ptr [[TMP39]], align 8
+// CHECK2-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK2-NEXT: store i64 0, ptr [[TMP40]], align 8
+// CHECK2-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK2-NEXT: store [3 x i32] [[TMP29]], ptr [[TMP41]], align 4
+// CHECK2-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK2-NEXT: store [3 x i32] [[TMP30]], ptr [[TMP42]], align 4
+// CHECK2-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK2-NEXT: store i32 0, ptr [[TMP43]], align 4
+// CHECK2-NEXT: [[TMP44:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 [[TMP23]], i32 [[TMP24]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l51.region_id, ptr [[KERNEL_ARGS]])
+// CHECK2-NEXT: [[TMP45:%.*]] = icmp ne i32 [[TMP44]], 0
+// CHECK2-NEXT: br i1 [[TMP45]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK2: omp_offload.failed:
// CHECK2-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l51(i64 [[TMP4]], ptr [[A]], i64 [[TMP6]], i64 [[TMP8]]) #[[ATTR2:[0-9]+]]
// CHECK2-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK2: omp_offload.cont:
-// CHECK2-NEXT: [[TMP40:%.*]] = load i32, ptr [[N]], align 4
-// CHECK2-NEXT: store i32 [[TMP40]], ptr [[N_CASTED7]], align 4
-// CHECK2-NEXT: [[TMP41:%.*]] = load i64, ptr [[N_CASTED7]], align 8
-// CHECK2-NEXT: [[TMP42:%.*]] = load ptr, ptr [[G_ADDR]], align 8
-// CHECK2-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK2-NEXT: store i64 [[TMP41]], ptr [[TMP43]], align 8
-// CHECK2-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK2-NEXT: store i64 [[TMP41]], ptr [[TMP44]], align 8
-// CHECK2-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
-// CHECK2-NEXT: store ptr null, ptr [[TMP45]], align 8
-// CHECK2-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
-// CHECK2-NEXT: store ptr [[A]], ptr [[TMP46]], align 8
-// CHECK2-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
-// CHECK2-NEXT: store ptr [[A]], ptr [[TMP47]], align 8
-// CHECK2-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
-// CHECK2-NEXT: store ptr null, ptr [[TMP48]], align 8
-// CHECK2-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 2
-// CHECK2-NEXT: store ptr [[TMP42]], ptr [[TMP49]], align 8
-// CHECK2-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 2
-// CHECK2-NEXT: store ptr [[TMP42]], ptr [[TMP50]], align 8
-// CHECK2-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 2
+// CHECK2-NEXT: [[TMP46:%.*]] = load i32, ptr [[N]], align 4
+// CHECK2-NEXT: store i32 [[TMP46]], ptr [[N_CASTED7]], align 4
+// CHECK2-NEXT: [[TMP47:%.*]] = load i64, ptr [[N_CASTED7]], align 8
+// CHECK2-NEXT: [[TMP48:%.*]] = load ptr, ptr [[G_ADDR]], align 8
+// CHECK2-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK2-NEXT: store i64 [[TMP47]], ptr [[TMP49]], align 8
+// CHECK2-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK2-NEXT: store i64 [[TMP47]], ptr [[TMP50]], align 8
+// CHECK2-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
// CHECK2-NEXT: store ptr null, ptr [[TMP51]], align 8
-// CHECK2-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK2-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK2-NEXT: [[TMP54:%.*]] = load i32, ptr [[N]], align 4
-// CHECK2-NEXT: store i32 [[TMP54]], ptr [[DOTCAPTURE_EXPR_12]], align 4
-// CHECK2-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_12]], align 4
-// CHECK2-NEXT: [[SUB14:%.*]] = sub nsw i32 [[TMP55]], 0
+// CHECK2-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
+// CHECK2-NEXT: store ptr [[A]], ptr [[TMP52]], align 8
+// CHECK2-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
+// CHECK2-NEXT: store ptr [[A]], ptr [[TMP53]], align 8
+// CHECK2-NEXT: [[TMP54:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
+// CHECK2-NEXT: store ptr null, ptr [[TMP54]], align 8
+// CHECK2-NEXT: [[TMP55:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 2
+// CHECK2-NEXT: store ptr [[TMP48]], ptr [[TMP55]], align 8
+// CHECK2-NEXT: [[TMP56:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 2
+// CHECK2-NEXT: store ptr [[TMP48]], ptr [[TMP56]], align 8
+// CHECK2-NEXT: [[TMP57:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 2
+// CHECK2-NEXT: store ptr null, ptr [[TMP57]], align 8
+// CHECK2-NEXT: [[TMP58:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK2-NEXT: [[TMP59:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK2-NEXT: [[TMP60:%.*]] = load i32, ptr [[N]], align 4
+// CHECK2-NEXT: store i32 [[TMP60]], ptr [[DOTCAPTURE_EXPR_12]], align 4
+// CHECK2-NEXT: [[TMP61:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_12]], align 4
+// CHECK2-NEXT: [[SUB14:%.*]] = sub nsw i32 [[TMP61]], 0
// CHECK2-NEXT: [[DIV15:%.*]] = sdiv i32 [[SUB14]], 1
// CHECK2-NEXT: [[SUB16:%.*]] = sub nsw i32 [[DIV15]], 1
// CHECK2-NEXT: store i32 [[SUB16]], ptr [[DOTCAPTURE_EXPR_13]], align 4
-// CHECK2-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_13]], align 4
-// CHECK2-NEXT: [[ADD17:%.*]] = add nsw i32 [[TMP56]], 1
-// CHECK2-NEXT: [[TMP57:%.*]] = zext i32 [[ADD17]] to i64
+// CHECK2-NEXT: [[TMP62:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_13]], align 4
+// CHECK2-NEXT: [[ADD17:%.*]] = add nsw i32 [[TMP62]], 1
+// CHECK2-NEXT: [[TMP63:%.*]] = zext i32 [[ADD17]] to i64
// CHECK2-NEXT: [[KERNEL_ARGS18:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK2-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 0
-// CHECK2-NEXT: store i32 1, ptr [[TMP58]], align 4
-// CHECK2-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 1
-// CHECK2-NEXT: store i32 3, ptr [[TMP59]], align 4
-// CHECK2-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 2
-// CHECK2-NEXT: store ptr [[TMP52]], ptr [[TMP60]], align 8
-// CHECK2-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 3
-// CHECK2-NEXT: store ptr [[TMP53]], ptr [[TMP61]], align 8
-// CHECK2-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 4
-// CHECK2-NEXT: store ptr @.offload_sizes.4, ptr [[TMP62]], align 8
-// CHECK2-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 5
-// CHECK2-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP63]], align 8
-// CHECK2-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 6
-// CHECK2-NEXT: store ptr null, ptr [[TMP64]], align 8
-// CHECK2-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 7
-// CHECK2-NEXT: store ptr null, ptr [[TMP65]], align 8
-// CHECK2-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 8
-// CHECK2-NEXT: store i64 [[TMP57]], ptr [[TMP66]], align 8
-// CHECK2-NEXT: [[TMP67:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l58.region_id, ptr [[KERNEL_ARGS18]])
-// CHECK2-NEXT: [[TMP68:%.*]] = icmp ne i32 [[TMP67]], 0
-// CHECK2-NEXT: br i1 [[TMP68]], label [[OMP_OFFLOAD_FAILED19:%.*]], label [[OMP_OFFLOAD_CONT20:%.*]]
+// CHECK2-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 0
+// CHECK2-NEXT: store i32 2, ptr [[TMP64]], align 4
+// CHECK2-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 1
+// CHECK2-NEXT: store i32 3, ptr [[TMP65]], align 4
+// CHECK2-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 2
+// CHECK2-NEXT: store ptr [[TMP58]], ptr [[TMP66]], align 8
+// CHECK2-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 3
+// CHECK2-NEXT: store ptr [[TMP59]], ptr [[TMP67]], align 8
+// CHECK2-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 4
+// CHECK2-NEXT: store ptr @.offload_sizes.4, ptr [[TMP68]], align 8
+// CHECK2-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 5
+// CHECK2-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP69]], align 8
+// CHECK2-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 6
+// CHECK2-NEXT: store ptr null, ptr [[TMP70]], align 8
+// CHECK2-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 7
+// CHECK2-NEXT: store ptr null, ptr [[TMP71]], align 8
+// CHECK2-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 8
+// CHECK2-NEXT: store i64 [[TMP63]], ptr [[TMP72]], align 8
+// CHECK2-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 9
+// CHECK2-NEXT: store i64 0, ptr [[TMP73]], align 8
+// CHECK2-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 10
+// CHECK2-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP74]], align 4
+// CHECK2-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 11
+// CHECK2-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP75]], align 4
+// CHECK2-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 12
+// CHECK2-NEXT: store i32 0, ptr [[TMP76]], align 4
+// CHECK2-NEXT: [[TMP77:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l58.region_id, ptr [[KERNEL_ARGS18]])
+// CHECK2-NEXT: [[TMP78:%.*]] = icmp ne i32 [[TMP77]], 0
+// CHECK2-NEXT: br i1 [[TMP78]], label [[OMP_OFFLOAD_FAILED19:%.*]], label [[OMP_OFFLOAD_CONT20:%.*]]
// CHECK2: omp_offload.failed19:
-// CHECK2-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l58(i64 [[TMP41]], ptr [[A]], ptr [[TMP42]]) #[[ATTR2]]
+// CHECK2-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l58(i64 [[TMP47]], ptr [[A]], ptr [[TMP48]]) #[[ATTR2]]
// CHECK2-NEXT: br label [[OMP_OFFLOAD_CONT20]]
// CHECK2: omp_offload.cont20:
// CHECK2-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i32], ptr [[A]], i64 0, i64 0
-// CHECK2-NEXT: [[TMP69:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK2-NEXT: ret i32 [[TMP69]]
+// CHECK2-NEXT: [[TMP79:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK2-NEXT: ret i32 [[TMP79]]
//
//
// CHECK2-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l51
@@ -1279,95 +1297,113 @@ int target_teams_fun(int *g){
// CHECK4-NEXT: [[TMP27:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_4]], align 4
// CHECK4-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP27]], 1
// CHECK4-NEXT: [[TMP28:%.*]] = zext i32 [[ADD]] to i64
+// CHECK4-NEXT: [[TMP29:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP23]], 0
+// CHECK4-NEXT: [[TMP30:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP24]], 0
// CHECK4-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK4-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK4-NEXT: store i32 1, ptr [[TMP29]], align 4
-// CHECK4-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK4-NEXT: store i32 4, ptr [[TMP30]], align 4
-// CHECK4-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK4-NEXT: store ptr [[TMP21]], ptr [[TMP31]], align 4
-// CHECK4-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK4-NEXT: store ptr [[TMP22]], ptr [[TMP32]], align 4
-// CHECK4-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK4-NEXT: store ptr @.offload_sizes, ptr [[TMP33]], align 4
-// CHECK4-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK4-NEXT: store ptr @.offload_maptypes, ptr [[TMP34]], align 4
-// CHECK4-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK4-NEXT: store ptr null, ptr [[TMP35]], align 4
-// CHECK4-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
-// CHECK4-NEXT: store ptr null, ptr [[TMP36]], align 4
-// CHECK4-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK4-NEXT: store i64 [[TMP28]], ptr [[TMP37]], align 8
-// CHECK4-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 [[TMP23]], i32 [[TMP24]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l51.region_id, ptr [[KERNEL_ARGS]])
-// CHECK4-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK4-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK4-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK4-NEXT: store i32 2, ptr [[TMP31]], align 4
+// CHECK4-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK4-NEXT: store i32 4, ptr [[TMP32]], align 4
+// CHECK4-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK4-NEXT: store ptr [[TMP21]], ptr [[TMP33]], align 4
+// CHECK4-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK4-NEXT: store ptr [[TMP22]], ptr [[TMP34]], align 4
+// CHECK4-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK4-NEXT: store ptr @.offload_sizes, ptr [[TMP35]], align 4
+// CHECK4-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK4-NEXT: store ptr @.offload_maptypes, ptr [[TMP36]], align 4
+// CHECK4-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
+// CHECK4-NEXT: store ptr null, ptr [[TMP37]], align 4
+// CHECK4-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK4-NEXT: store ptr null, ptr [[TMP38]], align 4
+// CHECK4-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK4-NEXT: store i64 [[TMP28]], ptr [[TMP39]], align 8
+// CHECK4-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK4-NEXT: store i64 0, ptr [[TMP40]], align 8
+// CHECK4-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK4-NEXT: store [3 x i32] [[TMP29]], ptr [[TMP41]], align 4
+// CHECK4-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK4-NEXT: store [3 x i32] [[TMP30]], ptr [[TMP42]], align 4
+// CHECK4-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK4-NEXT: store i32 0, ptr [[TMP43]], align 4
+// CHECK4-NEXT: [[TMP44:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 [[TMP23]], i32 [[TMP24]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l51.region_id, ptr [[KERNEL_ARGS]])
+// CHECK4-NEXT: [[TMP45:%.*]] = icmp ne i32 [[TMP44]], 0
+// CHECK4-NEXT: br i1 [[TMP45]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK4: omp_offload.failed:
// CHECK4-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l51(i32 [[TMP4]], ptr [[A]], i32 [[TMP6]], i32 [[TMP8]]) #[[ATTR2:[0-9]+]]
// CHECK4-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK4: omp_offload.cont:
-// CHECK4-NEXT: [[TMP40:%.*]] = load i32, ptr [[N]], align 4
-// CHECK4-NEXT: store i32 [[TMP40]], ptr [[N_CASTED7]], align 4
-// CHECK4-NEXT: [[TMP41:%.*]] = load i32, ptr [[N_CASTED7]], align 4
-// CHECK4-NEXT: [[TMP42:%.*]] = load ptr, ptr [[G_ADDR]], align 4
-// CHECK4-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK4-NEXT: store i32 [[TMP41]], ptr [[TMP43]], align 4
-// CHECK4-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK4-NEXT: store i32 [[TMP41]], ptr [[TMP44]], align 4
-// CHECK4-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
-// CHECK4-NEXT: store ptr null, ptr [[TMP45]], align 4
-// CHECK4-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
-// CHECK4-NEXT: store ptr [[A]], ptr [[TMP46]], align 4
-// CHECK4-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
-// CHECK4-NEXT: store ptr [[A]], ptr [[TMP47]], align 4
-// CHECK4-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1
-// CHECK4-NEXT: store ptr null, ptr [[TMP48]], align 4
-// CHECK4-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 2
-// CHECK4-NEXT: store ptr [[TMP42]], ptr [[TMP49]], align 4
-// CHECK4-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 2
-// CHECK4-NEXT: store ptr [[TMP42]], ptr [[TMP50]], align 4
-// CHECK4-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 2
+// CHECK4-NEXT: [[TMP46:%.*]] = load i32, ptr [[N]], align 4
+// CHECK4-NEXT: store i32 [[TMP46]], ptr [[N_CASTED7]], align 4
+// CHECK4-NEXT: [[TMP47:%.*]] = load i32, ptr [[N_CASTED7]], align 4
+// CHECK4-NEXT: [[TMP48:%.*]] = load ptr, ptr [[G_ADDR]], align 4
+// CHECK4-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK4-NEXT: store i32 [[TMP47]], ptr [[TMP49]], align 4
+// CHECK4-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK4-NEXT: store i32 [[TMP47]], ptr [[TMP50]], align 4
+// CHECK4-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
// CHECK4-NEXT: store ptr null, ptr [[TMP51]], align 4
-// CHECK4-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK4-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK4-NEXT: [[TMP54:%.*]] = load i32, ptr [[N]], align 4
-// CHECK4-NEXT: store i32 [[TMP54]], ptr [[DOTCAPTURE_EXPR_12]], align 4
-// CHECK4-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_12]], align 4
-// CHECK4-NEXT: [[SUB14:%.*]] = sub nsw i32 [[TMP55]], 0
+// CHECK4-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
+// CHECK4-NEXT: store ptr [[A]], ptr [[TMP52]], align 4
+// CHECK4-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
+// CHECK4-NEXT: store ptr [[A]], ptr [[TMP53]], align 4
+// CHECK4-NEXT: [[TMP54:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1
+// CHECK4-NEXT: store ptr null, ptr [[TMP54]], align 4
+// CHECK4-NEXT: [[TMP55:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 2
+// CHECK4-NEXT: store ptr [[TMP48]], ptr [[TMP55]], align 4
+// CHECK4-NEXT: [[TMP56:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 2
+// CHECK4-NEXT: store ptr [[TMP48]], ptr [[TMP56]], align 4
+// CHECK4-NEXT: [[TMP57:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 2
+// CHECK4-NEXT: store ptr null, ptr [[TMP57]], align 4
+// CHECK4-NEXT: [[TMP58:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK4-NEXT: [[TMP59:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK4-NEXT: [[TMP60:%.*]] = load i32, ptr [[N]], align 4
+// CHECK4-NEXT: store i32 [[TMP60]], ptr [[DOTCAPTURE_EXPR_12]], align 4
+// CHECK4-NEXT: [[TMP61:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_12]], align 4
+// CHECK4-NEXT: [[SUB14:%.*]] = sub nsw i32 [[TMP61]], 0
// CHECK4-NEXT: [[DIV15:%.*]] = sdiv i32 [[SUB14]], 1
// CHECK4-NEXT: [[SUB16:%.*]] = sub nsw i32 [[DIV15]], 1
// CHECK4-NEXT: store i32 [[SUB16]], ptr [[DOTCAPTURE_EXPR_13]], align 4
-// CHECK4-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_13]], align 4
-// CHECK4-NEXT: [[ADD17:%.*]] = add nsw i32 [[TMP56]], 1
-// CHECK4-NEXT: [[TMP57:%.*]] = zext i32 [[ADD17]] to i64
+// CHECK4-NEXT: [[TMP62:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_13]], align 4
+// CHECK4-NEXT: [[ADD17:%.*]] = add nsw i32 [[TMP62]], 1
+// CHECK4-NEXT: [[TMP63:%.*]] = zext i32 [[ADD17]] to i64
// CHECK4-NEXT: [[KERNEL_ARGS18:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK4-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 0
-// CHECK4-NEXT: store i32 1, ptr [[TMP58]], align 4
-// CHECK4-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 1
-// CHECK4-NEXT: store i32 3, ptr [[TMP59]], align 4
-// CHECK4-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 2
-// CHECK4-NEXT: store ptr [[TMP52]], ptr [[TMP60]], align 4
-// CHECK4-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 3
-// CHECK4-NEXT: store ptr [[TMP53]], ptr [[TMP61]], align 4
-// CHECK4-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 4
-// CHECK4-NEXT: store ptr @.offload_sizes.4, ptr [[TMP62]], align 4
-// CHECK4-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 5
-// CHECK4-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP63]], align 4
-// CHECK4-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 6
-// CHECK4-NEXT: store ptr null, ptr [[TMP64]], align 4
-// CHECK4-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 7
-// CHECK4-NEXT: store ptr null, ptr [[TMP65]], align 4
-// CHECK4-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 8
-// CHECK4-NEXT: store i64 [[TMP57]], ptr [[TMP66]], align 8
-// CHECK4-NEXT: [[TMP67:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l58.region_id, ptr [[KERNEL_ARGS18]])
-// CHECK4-NEXT: [[TMP68:%.*]] = icmp ne i32 [[TMP67]], 0
-// CHECK4-NEXT: br i1 [[TMP68]], label [[OMP_OFFLOAD_FAILED19:%.*]], label [[OMP_OFFLOAD_CONT20:%.*]]
+// CHECK4-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 0
+// CHECK4-NEXT: store i32 2, ptr [[TMP64]], align 4
+// CHECK4-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 1
+// CHECK4-NEXT: store i32 3, ptr [[TMP65]], align 4
+// CHECK4-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 2
+// CHECK4-NEXT: store ptr [[TMP58]], ptr [[TMP66]], align 4
+// CHECK4-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 3
+// CHECK4-NEXT: store ptr [[TMP59]], ptr [[TMP67]], align 4
+// CHECK4-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 4
+// CHECK4-NEXT: store ptr @.offload_sizes.4, ptr [[TMP68]], align 4
+// CHECK4-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 5
+// CHECK4-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP69]], align 4
+// CHECK4-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 6
+// CHECK4-NEXT: store ptr null, ptr [[TMP70]], align 4
+// CHECK4-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 7
+// CHECK4-NEXT: store ptr null, ptr [[TMP71]], align 4
+// CHECK4-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 8
+// CHECK4-NEXT: store i64 [[TMP63]], ptr [[TMP72]], align 8
+// CHECK4-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 9
+// CHECK4-NEXT: store i64 0, ptr [[TMP73]], align 8
+// CHECK4-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 10
+// CHECK4-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP74]], align 4
+// CHECK4-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 11
+// CHECK4-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP75]], align 4
+// CHECK4-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 12
+// CHECK4-NEXT: store i32 0, ptr [[TMP76]], align 4
+// CHECK4-NEXT: [[TMP77:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l58.region_id, ptr [[KERNEL_ARGS18]])
+// CHECK4-NEXT: [[TMP78:%.*]] = icmp ne i32 [[TMP77]], 0
+// CHECK4-NEXT: br i1 [[TMP78]], label [[OMP_OFFLOAD_FAILED19:%.*]], label [[OMP_OFFLOAD_CONT20:%.*]]
// CHECK4: omp_offload.failed19:
-// CHECK4-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l58(i32 [[TMP41]], ptr [[A]], ptr [[TMP42]]) #[[ATTR2]]
+// CHECK4-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l58(i32 [[TMP47]], ptr [[A]], ptr [[TMP48]]) #[[ATTR2]]
// CHECK4-NEXT: br label [[OMP_OFFLOAD_CONT20]]
// CHECK4: omp_offload.cont20:
// CHECK4-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i32], ptr [[A]], i32 0, i32 0
-// CHECK4-NEXT: [[TMP69:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK4-NEXT: ret i32 [[TMP69]]
+// CHECK4-NEXT: [[TMP79:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK4-NEXT: ret i32 [[TMP79]]
//
//
// CHECK4-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l51
diff --git a/clang/test/OpenMP/target_teams_distribute_parallel_for_collapse_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_parallel_for_collapse_codegen.cpp
index 3cf52915884a6..a6b2cfba59584 100644
--- a/clang/test/OpenMP/target_teams_distribute_parallel_for_collapse_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_parallel_for_collapse_codegen.cpp
@@ -129,7 +129,7 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -146,9 +146,17 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 56088, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -156,8 +164,8 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[A3:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x [456 x i32]], ptr [[A3]], i64 0, i64 0
// CHECK1-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds [456 x i32], ptr [[ARRAYIDX]], i64 0, i64 0
-// CHECK1-NEXT: [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
-// CHECK1-NEXT: ret i32 [[TMP16]]
+// CHECK1-NEXT: [[TMP20:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
+// CHECK1-NEXT: ret i32 [[TMP20]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -364,7 +372,7 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -381,9 +389,17 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 56088, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -391,8 +407,8 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[A3:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x [456 x i32]], ptr [[A3]], i32 0, i32 0
// CHECK3-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds [456 x i32], ptr [[ARRAYIDX]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
-// CHECK3-NEXT: ret i32 [[TMP16]]
+// CHECK3-NEXT: [[TMP20:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
+// CHECK3-NEXT: ret i32 [[TMP20]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -663,7 +679,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[ADD:%.*]] = add nsw i64 [[TMP35]], 1
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP36]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP36]], align 4
// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 5, ptr [[TMP37]], align 4
// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -680,20 +696,28 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP43]], align 8
// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 [[ADD]], ptr [[TMP44]], align 8
-// CHECK9-NEXT: [[TMP45:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP46:%.*]] = icmp ne i32 [[TMP45]], 0
-// CHECK9-NEXT: br i1 [[TMP46]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP46]], align 4
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP47]], align 4
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP48]], align 4
+// CHECK9-NEXT: [[TMP49:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP50:%.*]] = icmp ne i32 [[TMP49]], 0
+// CHECK9-NEXT: br i1 [[TMP50]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81(i64 [[TMP7]], i64 [[TMP9]], i64 [[TMP1]], i64 [[TMP3]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP47:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef signext [[TMP47]])
+// CHECK9-NEXT: [[TMP51:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef signext [[TMP51]])
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK9-NEXT: [[TMP48:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP48]])
-// CHECK9-NEXT: [[TMP49:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP49]]
+// CHECK9-NEXT: [[TMP52:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP52]])
+// CHECK9-NEXT: [[TMP53:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP53]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81
@@ -1017,7 +1041,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK9-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK9-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1034,9 +1058,17 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 20, ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l68.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK9-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK9-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l68.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK9-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l68(ptr [[A]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1316,7 +1348,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[ADD:%.*]] = add nsw i64 [[TMP34]], 1
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP35]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP35]], align 4
// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 5, ptr [[TMP36]], align 4
// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1333,20 +1365,28 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP42]], align 4
// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 [[ADD]], ptr [[TMP43]], align 8
-// CHECK11-NEXT: [[TMP44:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP45:%.*]] = icmp ne i32 [[TMP44]], 0
-// CHECK11-NEXT: br i1 [[TMP45]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP44]], align 8
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP47]], align 4
+// CHECK11-NEXT: [[TMP48:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP49:%.*]] = icmp ne i32 [[TMP48]], 0
+// CHECK11-NEXT: br i1 [[TMP49]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81(i32 [[TMP5]], i32 [[TMP7]], i32 [[TMP0]], i32 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP46:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef [[TMP46]])
+// CHECK11-NEXT: [[TMP50:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef [[TMP50]])
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK11-NEXT: [[TMP47:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP47]])
-// CHECK11-NEXT: [[TMP48:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP48]]
+// CHECK11-NEXT: [[TMP51:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP51]])
+// CHECK11-NEXT: [[TMP52:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP52]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81
@@ -1672,7 +1712,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK11-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK11-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1689,9 +1729,17 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK11-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 20, ptr [[TMP13]], align 8
-// CHECK11-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l68.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK11-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK11-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l68.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK11-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l68(ptr [[A]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/target_teams_distribute_parallel_for_depend_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_parallel_for_depend_codegen.cpp
index 5cb55b6204795..e09bd99f2f5d7 100644
--- a/clang/test/OpenMP/target_teams_distribute_parallel_for_depend_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_parallel_for_depend_codegen.cpp
@@ -200,7 +200,7 @@ int foo(int n) {
// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2
// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]],
// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64
-// CHECK: [[RET:%.+]] = call i32 @__tgt_target_kernel_nowait(%struct.ident_t* @{{.+}}, i64 [[DEVICE]], i32 0, i32 0, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* %{{.+}})
+// CHECK: [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 [[DEVICE]], i32 0, i32 0, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* %{{.+}})
// CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]]
diff --git a/clang/test/OpenMP/target_teams_distribute_parallel_for_dist_schedule_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_parallel_for_dist_schedule_codegen.cpp
index f180df59bf69f..a35a2fd44c554 100644
--- a/clang/test/OpenMP/target_teams_distribute_parallel_for_dist_schedule_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_parallel_for_dist_schedule_codegen.cpp
@@ -171,7 +171,7 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -188,87 +188,111 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.4, ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l32.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.4, ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l32.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK1-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK1-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK1: omp_offload.failed8:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l32(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK1: omp_offload.cont9:
// CHECK1-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A10]], ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A10]], ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.8, ptr [[TMP41]], align 8
-// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP42]], align 8
-// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP43]], align 8
-// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP44]], align 8
-// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK1-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK1-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK1-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 8
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 8
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.8, ptr [[TMP49]], align 8
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP50]], align 8
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP51]], align 8
+// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP52]], align 8
+// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK1-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK1-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK1-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK1: omp_offload.failed16:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK1: omp_offload.cont17:
// CHECK1-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A18]], i64 0, i64 0
-// CHECK1-NEXT: [[TMP48:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK1-NEXT: ret i32 [[TMP48]]
+// CHECK1-NEXT: [[TMP60:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK1-NEXT: ret i32 [[TMP60]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -788,7 +812,7 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -805,87 +829,111 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 4
-// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A2]], ptr [[TMP17]], align 4
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 4
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A2]], ptr [[TMP21]], align 4
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.4, ptr [[TMP25]], align 4
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP26]], align 4
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK3-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l32.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK3-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK3-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.4, ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK3-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l32.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK3-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK3-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK3: omp_offload.failed8:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l32(ptr [[THIS1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK3: omp_offload.cont9:
// CHECK3-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 4
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A10]], ptr [[TMP33]], align 4
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP34]], align 4
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A10]], ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP42]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 4
-// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 4
-// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.8, ptr [[TMP41]], align 4
-// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP42]], align 4
-// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP44]], align 4
-// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK3-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK3-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK3-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK3-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 4
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 4
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.8, ptr [[TMP49]], align 4
+// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP50]], align 4
+// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP52]], align 4
+// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK3-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK3-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK3-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK3-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK3: omp_offload.failed16:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36(ptr [[THIS1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK3: omp_offload.cont17:
// CHECK3-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A18]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP48:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK3-NEXT: ret i32 [[TMP48]]
+// CHECK3-NEXT: [[TMP60:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK3-NEXT: ret i32 [[TMP60]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -1440,7 +1488,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1457,159 +1505,183 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK9-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK9-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK9-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100(i64 [[TMP4]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK9-NEXT: [[TMP35:%.*]] = load i64, ptr [[N_CASTED3]], align 8
-// CHECK9-NEXT: [[TMP36:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK9-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK9-NEXT: [[TMP39:%.*]] = load i64, ptr [[N_CASTED3]], align 8
+// CHECK9-NEXT: [[TMP40:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK9-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES7]], ptr align 8 @.offload_sizes.4, i64 24, i1 false)
-// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP35]], ptr [[TMP37]], align 8
-// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP35]], ptr [[TMP38]], align 8
-// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP40]], align 8
-// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP41]], align 8
-// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP42]], align 8
-// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP43]], align 8
-// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 8
-// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK9-NEXT: store i64 [[TMP36]], ptr [[TMP45]], align 8
-// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP39]], ptr [[TMP41]], align 8
+// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP39]], ptr [[TMP42]], align 8
+// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP44]], align 8
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
// CHECK9-NEXT: store ptr null, ptr [[TMP46]], align 8
-// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP50:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP50]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK9-NEXT: [[TMP51:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK9-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP51]], 0
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP47]], align 8
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 8
+// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK9-NEXT: store i64 [[TMP40]], ptr [[TMP49]], align 8
+// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP50]], align 8
+// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP54:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP54]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK9-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK9-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP55]], 0
// CHECK9-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK9-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK9-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK9-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK9-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP52]], 1
-// CHECK9-NEXT: [[TMP53:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK9-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK9-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP56]], 1
+// CHECK9-NEXT: [[TMP57:%.*]] = zext i32 [[ADD14]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK9-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK9-NEXT: store i32 3, ptr [[TMP55]], align 4
-// CHECK9-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP47]], ptr [[TMP56]], align 8
-// CHECK9-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 8
-// CHECK9-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 8
-// CHECK9-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP59]], align 8
-// CHECK9-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP60]], align 8
-// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP61]], align 8
-// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP53]], ptr [[TMP62]], align 8
-// CHECK9-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l104.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK9-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
-// CHECK9-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK9-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP58]], align 4
+// CHECK9-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK9-NEXT: store i32 3, ptr [[TMP59]], align 4
+// CHECK9-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP51]], ptr [[TMP60]], align 8
+// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 8
+// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 8
+// CHECK9-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP63]], align 8
+// CHECK9-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP64]], align 8
+// CHECK9-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP65]], align 8
+// CHECK9-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP57]], ptr [[TMP66]], align 8
+// CHECK9-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP67]], align 8
+// CHECK9-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP68]], align 4
+// CHECK9-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK9-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP70]], align 4
+// CHECK9-NEXT: [[TMP71:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l104.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK9-NEXT: [[TMP72:%.*]] = icmp ne i32 [[TMP71]], 0
+// CHECK9-NEXT: br i1 [[TMP72]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK9: omp_offload.failed16:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l104(i64 [[TMP35]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l104(i64 [[TMP39]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK9: omp_offload.cont17:
-// CHECK9-NEXT: [[TMP65:%.*]] = load i32, ptr [[M]], align 4
-// CHECK9-NEXT: store i32 [[TMP65]], ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK9-NEXT: [[TMP66:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP66]], ptr [[N_CASTED19]], align 4
-// CHECK9-NEXT: [[TMP67:%.*]] = load i64, ptr [[N_CASTED19]], align 8
-// CHECK9-NEXT: [[TMP68:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK9-NEXT: store i32 [[TMP68]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK9-NEXT: [[TMP69:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK9-NEXT: [[TMP70:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK9-NEXT: [[TMP73:%.*]] = load i32, ptr [[M]], align 4
+// CHECK9-NEXT: store i32 [[TMP73]], ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK9-NEXT: [[TMP74:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP74]], ptr [[N_CASTED19]], align 4
+// CHECK9-NEXT: [[TMP75:%.*]] = load i64, ptr [[N_CASTED19]], align 8
+// CHECK9-NEXT: [[TMP76:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK9-NEXT: store i32 [[TMP76]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK9-NEXT: [[TMP77:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK9-NEXT: [[TMP78:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK9-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES23]], ptr align 8 @.offload_sizes.8, i64 32, i1 false)
-// CHECK9-NEXT: [[TMP71:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP67]], ptr [[TMP71]], align 8
-// CHECK9-NEXT: [[TMP72:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP67]], ptr [[TMP72]], align 8
-// CHECK9-NEXT: [[TMP73:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP73]], align 8
-// CHECK9-NEXT: [[TMP74:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP74]], align 8
-// CHECK9-NEXT: [[TMP75:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP75]], align 8
-// CHECK9-NEXT: [[TMP76:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP76]], align 8
-// CHECK9-NEXT: [[TMP77:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP77]], align 8
-// CHECK9-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP78]], align 8
-// CHECK9-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
-// CHECK9-NEXT: store i64 [[TMP70]], ptr [[TMP79]], align 8
-// CHECK9-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP80]], align 8
-// CHECK9-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
-// CHECK9-NEXT: store i64 [[TMP69]], ptr [[TMP81]], align 8
-// CHECK9-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
-// CHECK9-NEXT: store i64 [[TMP69]], ptr [[TMP82]], align 8
-// CHECK9-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP83]], align 8
-// CHECK9-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP87:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP87]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK9-NEXT: [[TMP88:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK9-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP88]], 0
+// CHECK9-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP75]], ptr [[TMP79]], align 8
+// CHECK9-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP75]], ptr [[TMP80]], align 8
+// CHECK9-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP81]], align 8
+// CHECK9-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP82]], align 8
+// CHECK9-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP83]], align 8
+// CHECK9-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP84]], align 8
+// CHECK9-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP85]], align 8
+// CHECK9-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP86]], align 8
+// CHECK9-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
+// CHECK9-NEXT: store i64 [[TMP78]], ptr [[TMP87]], align 8
+// CHECK9-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP88]], align 8
+// CHECK9-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
+// CHECK9-NEXT: store i64 [[TMP77]], ptr [[TMP89]], align 8
+// CHECK9-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
+// CHECK9-NEXT: store i64 [[TMP77]], ptr [[TMP90]], align 8
+// CHECK9-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP91]], align 8
+// CHECK9-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP94:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP95:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP95]], ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK9-NEXT: [[TMP96:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK9-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP96]], 0
// CHECK9-NEXT: [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1
// CHECK9-NEXT: [[SUB29:%.*]] = sub nsw i32 [[DIV28]], 1
// CHECK9-NEXT: store i32 [[SUB29]], ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK9-NEXT: [[TMP89:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK9-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP89]], 1
-// CHECK9-NEXT: [[TMP90:%.*]] = zext i32 [[ADD30]] to i64
+// CHECK9-NEXT: [[TMP97:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
+// CHECK9-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP97]], 1
+// CHECK9-NEXT: [[TMP98:%.*]] = zext i32 [[ADD30]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP91]], align 4
-// CHECK9-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK9-NEXT: store i32 4, ptr [[TMP92]], align 4
-// CHECK9-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP84]], ptr [[TMP93]], align 8
-// CHECK9-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP85]], ptr [[TMP94]], align 8
-// CHECK9-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP86]], ptr [[TMP95]], align 8
-// CHECK9-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP96]], align 8
-// CHECK9-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP97]], align 8
-// CHECK9-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP98]], align 8
-// CHECK9-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP90]], ptr [[TMP99]], align 8
-// CHECK9-NEXT: [[TMP100:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l108.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK9-NEXT: [[TMP101:%.*]] = icmp ne i32 [[TMP100]], 0
-// CHECK9-NEXT: br i1 [[TMP101]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK9-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP99]], align 4
+// CHECK9-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK9-NEXT: store i32 4, ptr [[TMP100]], align 4
+// CHECK9-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP92]], ptr [[TMP101]], align 8
+// CHECK9-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP93]], ptr [[TMP102]], align 8
+// CHECK9-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP94]], ptr [[TMP103]], align 8
+// CHECK9-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP104]], align 8
+// CHECK9-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP105]], align 8
+// CHECK9-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP106]], align 8
+// CHECK9-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP98]], ptr [[TMP107]], align 8
+// CHECK9-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP108]], align 8
+// CHECK9-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP109]], align 4
+// CHECK9-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP110]], align 4
+// CHECK9-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP111]], align 4
+// CHECK9-NEXT: [[TMP112:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l108.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK9-NEXT: [[TMP113:%.*]] = icmp ne i32 [[TMP112]], 0
+// CHECK9-NEXT: br i1 [[TMP113]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK9: omp_offload.failed32:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l108(i64 [[TMP67]], i64 [[TMP1]], ptr [[VLA]], i64 [[TMP69]]) #[[ATTR3]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l108(i64 [[TMP75]], i64 [[TMP1]], ptr [[VLA]], i64 [[TMP77]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK9: omp_offload.cont33:
-// CHECK9-NEXT: [[TMP102:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP102]])
+// CHECK9-NEXT: [[TMP114:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP114]])
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK9-NEXT: [[TMP103:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP103]])
-// CHECK9-NEXT: [[TMP104:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP104]]
+// CHECK9-NEXT: [[TMP115:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP115]])
+// CHECK9-NEXT: [[TMP116:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP116]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100
@@ -2334,7 +2406,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK9-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK9-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2351,90 +2423,114 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l81.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK9-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK9-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l81.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK9-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l81(ptr [[A]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP16]], align 8
-// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP17]], align 8
-// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP20]], align 8
+// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP21]], align 8
+// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK9-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.16, ptr [[TMP25]], align 8
-// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP26]], align 8
-// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK9-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK9-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l85.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK9-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK9-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK9-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.16, ptr [[TMP29]], align 8
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP30]], align 8
+// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK9-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK9-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l85.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK9-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK9-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK9: omp_offload.failed6:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l85(ptr [[A]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK9: omp_offload.cont7:
-// CHECK9-NEXT: [[TMP32:%.*]] = load i32, ptr [[M]], align 4
-// CHECK9-NEXT: store i32 [[TMP32]], ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK9-NEXT: [[TMP33:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK9-NEXT: store i32 [[TMP33]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK9-NEXT: [[TMP34:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP35]], align 8
-// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP36]], align 8
-// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP37]], align 8
-// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP34]], ptr [[TMP38]], align 8
-// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP34]], ptr [[TMP39]], align 8
-// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP40]], align 8
-// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP40:%.*]] = load i32, ptr [[M]], align 4
+// CHECK9-NEXT: store i32 [[TMP40]], ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK9-NEXT: [[TMP41:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK9-NEXT: store i32 [[TMP41]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK9-NEXT: [[TMP42:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP43]], align 8
+// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP44]], align 8
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP42]], ptr [[TMP46]], align 8
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP42]], ptr [[TMP47]], align 8
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP48]], align 8
+// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP43]], align 4
-// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK9-NEXT: store i32 2, ptr [[TMP44]], align 4
-// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP41]], ptr [[TMP45]], align 8
-// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP42]], ptr [[TMP46]], align 8
-// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.20, ptr [[TMP47]], align 8
-// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.21, ptr [[TMP48]], align 8
-// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP49]], align 8
-// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP50]], align 8
-// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK9-NEXT: store i64 10, ptr [[TMP51]], align 8
-// CHECK9-NEXT: [[TMP52:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK9-NEXT: [[TMP53:%.*]] = icmp ne i32 [[TMP52]], 0
-// CHECK9-NEXT: br i1 [[TMP53]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP51]], align 4
+// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK9-NEXT: store i32 2, ptr [[TMP52]], align 4
+// CHECK9-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 8
+// CHECK9-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP50]], ptr [[TMP54]], align 8
+// CHECK9-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.20, ptr [[TMP55]], align 8
+// CHECK9-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.21, ptr [[TMP56]], align 8
+// CHECK9-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP57]], align 8
+// CHECK9-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP58]], align 8
+// CHECK9-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK9-NEXT: store i64 10, ptr [[TMP59]], align 8
+// CHECK9-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP60]], align 8
+// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4
+// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP62]], align 4
+// CHECK9-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP63]], align 4
+// CHECK9-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK9-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
+// CHECK9-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK9: omp_offload.failed13:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89(ptr [[A]], i64 [[TMP34]]) #[[ATTR3]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89(ptr [[A]], i64 [[TMP42]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT14]]
// CHECK9: omp_offload.cont14:
// CHECK9-NEXT: ret i32 0
@@ -3019,7 +3115,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3036,161 +3132,185 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK11-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK11-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK11-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100(i32 [[TMP3]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK11-NEXT: [[TMP35:%.*]] = load i32, ptr [[N_CASTED3]], align 4
-// CHECK11-NEXT: [[TMP36:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK11-NEXT: [[TMP37:%.*]] = sext i32 [[TMP36]] to i64
+// CHECK11-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK11-NEXT: [[TMP39:%.*]] = load i32, ptr [[N_CASTED3]], align 4
+// CHECK11-NEXT: [[TMP40:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK11-NEXT: [[TMP41:%.*]] = sext i32 [[TMP40]] to i64
// CHECK11-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES7]], ptr align 4 @.offload_sizes.4, i32 24, i1 false)
-// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP35]], ptr [[TMP38]], align 4
-// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP35]], ptr [[TMP39]], align 4
-// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP40]], align 4
-// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP41]], align 4
-// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP42]], align 4
-// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 4
-// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP45]], align 4
-// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK11-NEXT: store i64 [[TMP37]], ptr [[TMP46]], align 4
-// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP39]], ptr [[TMP42]], align 4
+// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP39]], ptr [[TMP43]], align 4
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP44]], align 4
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
// CHECK11-NEXT: store ptr null, ptr [[TMP47]], align 4
-// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP51:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP51]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK11-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK11-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP52]], 0
+// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 4
+// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP49]], align 4
+// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK11-NEXT: store i64 [[TMP41]], ptr [[TMP50]], align 4
+// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP54:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP55:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP55]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK11-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK11-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP56]], 0
// CHECK11-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK11-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK11-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK11-NEXT: [[TMP53:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK11-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP53]], 1
-// CHECK11-NEXT: [[TMP54:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK11-NEXT: [[TMP57:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK11-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP57]], 1
+// CHECK11-NEXT: [[TMP58:%.*]] = zext i32 [[ADD14]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP55]], align 4
-// CHECK11-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK11-NEXT: store i32 3, ptr [[TMP56]], align 4
-// CHECK11-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 4
-// CHECK11-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 4
-// CHECK11-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP50]], ptr [[TMP59]], align 4
-// CHECK11-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP60]], align 4
-// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP61]], align 4
-// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP62]], align 4
-// CHECK11-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP54]], ptr [[TMP63]], align 8
-// CHECK11-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l104.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK11-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
-// CHECK11-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK11-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP59]], align 4
+// CHECK11-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK11-NEXT: store i32 3, ptr [[TMP60]], align 4
+// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 4
+// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 4
+// CHECK11-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP54]], ptr [[TMP63]], align 4
+// CHECK11-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP64]], align 4
+// CHECK11-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP65]], align 4
+// CHECK11-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP66]], align 4
+// CHECK11-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP58]], ptr [[TMP67]], align 8
+// CHECK11-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP68]], align 8
+// CHECK11-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK11-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP70]], align 4
+// CHECK11-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP71]], align 4
+// CHECK11-NEXT: [[TMP72:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l104.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK11-NEXT: [[TMP73:%.*]] = icmp ne i32 [[TMP72]], 0
+// CHECK11-NEXT: br i1 [[TMP73]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK11: omp_offload.failed16:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l104(i32 [[TMP35]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l104(i32 [[TMP39]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK11: omp_offload.cont17:
-// CHECK11-NEXT: [[TMP66:%.*]] = load i32, ptr [[M]], align 4
-// CHECK11-NEXT: store i32 [[TMP66]], ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK11-NEXT: [[TMP67:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP67]], ptr [[N_CASTED19]], align 4
-// CHECK11-NEXT: [[TMP68:%.*]] = load i32, ptr [[N_CASTED19]], align 4
-// CHECK11-NEXT: [[TMP69:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK11-NEXT: store i32 [[TMP69]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK11-NEXT: [[TMP70:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK11-NEXT: [[TMP71:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK11-NEXT: [[TMP72:%.*]] = sext i32 [[TMP71]] to i64
+// CHECK11-NEXT: [[TMP74:%.*]] = load i32, ptr [[M]], align 4
+// CHECK11-NEXT: store i32 [[TMP74]], ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK11-NEXT: [[TMP75:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP75]], ptr [[N_CASTED19]], align 4
+// CHECK11-NEXT: [[TMP76:%.*]] = load i32, ptr [[N_CASTED19]], align 4
+// CHECK11-NEXT: [[TMP77:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK11-NEXT: store i32 [[TMP77]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK11-NEXT: [[TMP78:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK11-NEXT: [[TMP79:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK11-NEXT: [[TMP80:%.*]] = sext i32 [[TMP79]] to i64
// CHECK11-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES23]], ptr align 4 @.offload_sizes.8, i32 32, i1 false)
-// CHECK11-NEXT: [[TMP73:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP68]], ptr [[TMP73]], align 4
-// CHECK11-NEXT: [[TMP74:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP68]], ptr [[TMP74]], align 4
-// CHECK11-NEXT: [[TMP75:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP75]], align 4
-// CHECK11-NEXT: [[TMP76:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP76]], align 4
-// CHECK11-NEXT: [[TMP77:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP77]], align 4
-// CHECK11-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP78]], align 4
-// CHECK11-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP79]], align 4
-// CHECK11-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP80]], align 4
-// CHECK11-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
-// CHECK11-NEXT: store i64 [[TMP72]], ptr [[TMP81]], align 4
-// CHECK11-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP82]], align 4
-// CHECK11-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
-// CHECK11-NEXT: store i32 [[TMP70]], ptr [[TMP83]], align 4
-// CHECK11-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
-// CHECK11-NEXT: store i32 [[TMP70]], ptr [[TMP84]], align 4
-// CHECK11-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP85]], align 4
-// CHECK11-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP89:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP89]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK11-NEXT: [[TMP90:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK11-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP90]], 0
+// CHECK11-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP76]], ptr [[TMP81]], align 4
+// CHECK11-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP76]], ptr [[TMP82]], align 4
+// CHECK11-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP83]], align 4
+// CHECK11-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP84]], align 4
+// CHECK11-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP85]], align 4
+// CHECK11-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP86]], align 4
+// CHECK11-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP87]], align 4
+// CHECK11-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP88]], align 4
+// CHECK11-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
+// CHECK11-NEXT: store i64 [[TMP80]], ptr [[TMP89]], align 4
+// CHECK11-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP90]], align 4
+// CHECK11-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
+// CHECK11-NEXT: store i32 [[TMP78]], ptr [[TMP91]], align 4
+// CHECK11-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
+// CHECK11-NEXT: store i32 [[TMP78]], ptr [[TMP92]], align 4
+// CHECK11-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP93]], align 4
+// CHECK11-NEXT: [[TMP94:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP95:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP96:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP97:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP97]], ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK11-NEXT: [[TMP98:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK11-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP98]], 0
// CHECK11-NEXT: [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1
// CHECK11-NEXT: [[SUB29:%.*]] = sub nsw i32 [[DIV28]], 1
// CHECK11-NEXT: store i32 [[SUB29]], ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK11-NEXT: [[TMP91:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK11-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP91]], 1
-// CHECK11-NEXT: [[TMP92:%.*]] = zext i32 [[ADD30]] to i64
+// CHECK11-NEXT: [[TMP99:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
+// CHECK11-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP99]], 1
+// CHECK11-NEXT: [[TMP100:%.*]] = zext i32 [[ADD30]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP93]], align 4
-// CHECK11-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK11-NEXT: store i32 4, ptr [[TMP94]], align 4
-// CHECK11-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP86]], ptr [[TMP95]], align 4
-// CHECK11-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP87]], ptr [[TMP96]], align 4
-// CHECK11-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP88]], ptr [[TMP97]], align 4
-// CHECK11-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP98]], align 4
-// CHECK11-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP99]], align 4
-// CHECK11-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP100]], align 4
-// CHECK11-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP92]], ptr [[TMP101]], align 8
-// CHECK11-NEXT: [[TMP102:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l108.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK11-NEXT: [[TMP103:%.*]] = icmp ne i32 [[TMP102]], 0
-// CHECK11-NEXT: br i1 [[TMP103]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK11-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP101]], align 4
+// CHECK11-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK11-NEXT: store i32 4, ptr [[TMP102]], align 4
+// CHECK11-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP94]], ptr [[TMP103]], align 4
+// CHECK11-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP95]], ptr [[TMP104]], align 4
+// CHECK11-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP96]], ptr [[TMP105]], align 4
+// CHECK11-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP106]], align 4
+// CHECK11-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP107]], align 4
+// CHECK11-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP108]], align 4
+// CHECK11-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP100]], ptr [[TMP109]], align 8
+// CHECK11-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP110]], align 8
+// CHECK11-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP111]], align 4
+// CHECK11-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP112]], align 4
+// CHECK11-NEXT: [[TMP113:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP113]], align 4
+// CHECK11-NEXT: [[TMP114:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l108.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK11-NEXT: [[TMP115:%.*]] = icmp ne i32 [[TMP114]], 0
+// CHECK11-NEXT: br i1 [[TMP115]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK11: omp_offload.failed32:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l108(i32 [[TMP68]], i32 [[TMP0]], ptr [[VLA]], i32 [[TMP70]]) #[[ATTR3]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l108(i32 [[TMP76]], i32 [[TMP0]], ptr [[VLA]], i32 [[TMP78]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK11: omp_offload.cont33:
-// CHECK11-NEXT: [[TMP104:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP104]])
+// CHECK11-NEXT: [[TMP116:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP116]])
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK11-NEXT: [[TMP105:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP105]])
-// CHECK11-NEXT: [[TMP106:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP106]]
+// CHECK11-NEXT: [[TMP117:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP117]])
+// CHECK11-NEXT: [[TMP118:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP118]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100
@@ -3900,7 +4020,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK11-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK11-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3917,90 +4037,114 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK11-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK11-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l81.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK11-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK11-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l81.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK11-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l81(ptr [[A]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP16]], align 4
-// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP17]], align 4
-// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP20]], align 4
+// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP21]], align 4
+// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK11-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.16, ptr [[TMP25]], align 4
-// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP26]], align 4
-// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK11-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK11-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l85.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK11-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK11-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK11-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.16, ptr [[TMP29]], align 4
+// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP30]], align 4
+// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK11-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK11-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l85.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK11-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK11-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK11: omp_offload.failed6:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l85(ptr [[A]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK11: omp_offload.cont7:
-// CHECK11-NEXT: [[TMP32:%.*]] = load i32, ptr [[M]], align 4
-// CHECK11-NEXT: store i32 [[TMP32]], ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK11-NEXT: [[TMP33:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK11-NEXT: store i32 [[TMP33]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK11-NEXT: [[TMP34:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP35]], align 4
-// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP36]], align 4
-// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP37]], align 4
-// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP34]], ptr [[TMP38]], align 4
-// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP34]], ptr [[TMP39]], align 4
-// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP40]], align 4
-// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP40:%.*]] = load i32, ptr [[M]], align 4
+// CHECK11-NEXT: store i32 [[TMP40]], ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK11-NEXT: [[TMP41:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK11-NEXT: store i32 [[TMP41]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK11-NEXT: [[TMP42:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP43]], align 4
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP44]], align 4
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP42]], ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP42]], ptr [[TMP47]], align 4
+// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP48]], align 4
+// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP43]], align 4
-// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK11-NEXT: store i32 2, ptr [[TMP44]], align 4
-// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP41]], ptr [[TMP45]], align 4
-// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP42]], ptr [[TMP46]], align 4
-// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.20, ptr [[TMP47]], align 4
-// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.21, ptr [[TMP48]], align 4
-// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP49]], align 4
-// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP50]], align 4
-// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK11-NEXT: store i64 10, ptr [[TMP51]], align 8
-// CHECK11-NEXT: [[TMP52:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK11-NEXT: [[TMP53:%.*]] = icmp ne i32 [[TMP52]], 0
-// CHECK11-NEXT: br i1 [[TMP53]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP51]], align 4
+// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK11-NEXT: store i32 2, ptr [[TMP52]], align 4
+// CHECK11-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 4
+// CHECK11-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP50]], ptr [[TMP54]], align 4
+// CHECK11-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.20, ptr [[TMP55]], align 4
+// CHECK11-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.21, ptr [[TMP56]], align 4
+// CHECK11-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP57]], align 4
+// CHECK11-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP58]], align 4
+// CHECK11-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK11-NEXT: store i64 10, ptr [[TMP59]], align 8
+// CHECK11-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP60]], align 8
+// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4
+// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP62]], align 4
+// CHECK11-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP63]], align 4
+// CHECK11-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK11-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
+// CHECK11-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK11: omp_offload.failed13:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89(ptr [[A]], i32 [[TMP34]]) #[[ATTR3]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89(ptr [[A]], i32 [[TMP42]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT14]]
// CHECK11: omp_offload.cont14:
// CHECK11-NEXT: ret i32 0
diff --git a/clang/test/OpenMP/target_teams_distribute_parallel_for_firstprivate_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_parallel_for_firstprivate_codegen.cpp
index 411b71fbef7ef..a01d42a683620 100644
--- a/clang/test/OpenMP/target_teams_distribute_parallel_for_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_parallel_for_firstprivate_codegen.cpp
@@ -364,7 +364,7 @@ int main() {
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 5, ptr [[TMP22]], align 4
// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -381,9 +381,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l122.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK1-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l122.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK1-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l122(ptr @vec, i64 [[TMP1]], ptr @s_arr, ptr @var, i64 [[TMP3]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -765,7 +773,7 @@ int main() {
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -782,27 +790,35 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP25]], align 8
// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l81.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK1-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP29]], align 4
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l81.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK1-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l81(ptr [[VEC]], i64 [[TMP2]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK1-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK1: arraydestroy.body:
-// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP29]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP33]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK1-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK1-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK1: arraydestroy.done2:
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK1-NEXT: [[TMP30:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK1-NEXT: ret i32 [[TMP30]]
+// CHECK1-NEXT: [[TMP34:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK1-NEXT: ret i32 [[TMP34]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN2StC2Ev
@@ -1409,7 +1425,7 @@ int main() {
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 5, ptr [[TMP22]], align 4
// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1426,9 +1442,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4
// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP29]], align 8
-// CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l122.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK3-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK3-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l122.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK3-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l122(ptr @vec, i32 [[TMP1]], ptr @s_arr, ptr @var, i32 [[TMP3]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1804,7 +1828,7 @@ int main() {
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1821,27 +1845,35 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP25]], align 4
// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP26]], align 8
-// CHECK3-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l81.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK3-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l81.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK3-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l81(ptr [[VEC]], i32 [[TMP2]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK3-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK3-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK3: arraydestroy.body:
-// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP29]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP33]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK3-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK3-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK3: arraydestroy.done2:
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK3-NEXT: [[TMP30:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK3-NEXT: ret i32 [[TMP30]]
+// CHECK3-NEXT: [[TMP34:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK3-NEXT: ret i32 [[TMP34]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN2StC2Ev
diff --git a/clang/test/OpenMP/target_teams_distribute_parallel_for_if_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_parallel_for_if_codegen.cpp
index b7e8f3d9c7de3..0cdb8bc0f5b19 100644
--- a/clang/test/OpenMP/target_teams_distribute_parallel_for_if_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_parallel_for_if_codegen.cpp
@@ -112,7 +112,7 @@ int main() {
// CHECK1-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -129,35 +129,51 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l48.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l48.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l48() #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l51.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l51.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK1-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l51() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -459,7 +475,7 @@ int main() {
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -476,71 +492,88 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l76.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l76.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l76() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l83() #[[ATTR2]]
-// CHECK1-NEXT: [[TMP11:%.*]] = load i32, ptr @Arg, align 4
-// CHECK1-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP11]], 0
+// CHECK1-NEXT: [[TMP15:%.*]] = load i32, ptr @Arg, align 4
+// CHECK1-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP15]], 0
// CHECK1-NEXT: [[FROMBOOL:%.*]] = zext i1 [[TOBOOL]] to i8
// CHECK1-NEXT: store i8 [[FROMBOOL]], ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TMP12:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TOBOOL1:%.*]] = trunc i8 [[TMP12]] to i1
+// CHECK1-NEXT: [[TMP16:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK1-NEXT: [[TOBOOL1:%.*]] = trunc i8 [[TMP16]] to i1
// CHECK1-NEXT: [[FROMBOOL2:%.*]] = zext i1 [[TOBOOL1]] to i8
// CHECK1-NEXT: store i8 [[FROMBOOL2]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 1
-// CHECK1-NEXT: [[TMP13:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TOBOOL3:%.*]] = trunc i8 [[TMP14]] to i1
+// CHECK1-NEXT: [[TMP17:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK1-NEXT: [[TMP18:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK1-NEXT: [[TOBOOL3:%.*]] = trunc i8 [[TMP18]] to i1
// CHECK1-NEXT: br i1 [[TOBOOL3]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK1: omp_if.then:
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP13]], ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP13]], ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP20:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TOBOOL4:%.*]] = trunc i8 [[TMP20]] to i1
-// CHECK1-NEXT: [[TMP21:%.*]] = select i1 [[TOBOOL4]], i32 0, i32 1
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP17]], ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP17]], ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP24:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK1-NEXT: [[TOBOOL4:%.*]] = trunc i8 [[TMP24]] to i1
+// CHECK1-NEXT: [[TMP25:%.*]] = select i1 [[TOBOOL4]], i32 0, i32 1
+// CHECK1-NEXT: [[TMP26:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP25]], 0
// CHECK1-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP23]], align 4
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP18]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP19]], ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP27]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP30]], align 8
-// CHECK1-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP21]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l90.region_id, ptr [[KERNEL_ARGS6]])
-// CHECK1-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
-// CHECK1-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP28]], align 4
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP22]], ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP23]], ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP35]], align 8
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP36]], align 8
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP26]], ptr [[TMP38]], align 4
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP39]], align 4
+// CHECK1-NEXT: [[TMP40:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP25]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l90.region_id, ptr [[KERNEL_ARGS6]])
+// CHECK1-NEXT: [[TMP41:%.*]] = icmp ne i32 [[TMP40]], 0
+// CHECK1-NEXT: br i1 [[TMP41]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK1: omp_offload.failed7:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l90(i64 [[TMP13]]) #[[ATTR2]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l90(i64 [[TMP17]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK1: omp_offload.cont8:
// CHECK1-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK1: omp_if.else:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l90(i64 [[TMP13]]) #[[ATTR2]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l90(i64 [[TMP17]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
-// CHECK1-NEXT: [[TMP33:%.*]] = load i32, ptr @Arg, align 4
-// CHECK1-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP33]])
+// CHECK1-NEXT: [[TMP42:%.*]] = load i32, ptr @Arg, align 4
+// CHECK1-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP42]])
// CHECK1-NEXT: ret i32 [[CALL]]
//
//
@@ -1001,7 +1034,7 @@ int main() {
// CHECK1-NEXT: store i32 [[ARG]], ptr [[ARG_ADDR]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1018,58 +1051,75 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l60.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l60.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l60() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l64() #[[ATTR2]]
-// CHECK1-NEXT: [[TMP11:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
-// CHECK1-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP11]], 0
+// CHECK1-NEXT: [[TMP15:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
+// CHECK1-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP15]], 0
// CHECK1-NEXT: [[FROMBOOL:%.*]] = zext i1 [[TOBOOL]] to i8
// CHECK1-NEXT: store i8 [[FROMBOOL]], ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TMP12:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TOBOOL1:%.*]] = trunc i8 [[TMP12]] to i1
+// CHECK1-NEXT: [[TMP16:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK1-NEXT: [[TOBOOL1:%.*]] = trunc i8 [[TMP16]] to i1
// CHECK1-NEXT: [[FROMBOOL2:%.*]] = zext i1 [[TOBOOL1]] to i8
// CHECK1-NEXT: store i8 [[FROMBOOL2]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 1
-// CHECK1-NEXT: [[TMP13:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP13]], ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP13]], ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP19:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TOBOOL3:%.*]] = trunc i8 [[TMP19]] to i1
-// CHECK1-NEXT: [[TMP20:%.*]] = select i1 [[TOBOOL3]], i32 0, i32 1
+// CHECK1-NEXT: [[TMP17:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP17]], ptr [[TMP18]], align 8
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP17]], ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP23:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK1-NEXT: [[TOBOOL3:%.*]] = trunc i8 [[TMP23]] to i1
+// CHECK1-NEXT: [[TMP24:%.*]] = select i1 [[TOBOOL3]], i32 0, i32 1
+// CHECK1-NEXT: [[TMP25:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP24]], 0
// CHECK1-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP17]], ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP18]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.16, ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP20]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l68.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP21]], ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP22]], ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.16, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP35]], align 8
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP25]], ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP38]], align 4
+// CHECK1-NEXT: [[TMP39:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP24]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l68.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK1-NEXT: [[TMP40:%.*]] = icmp ne i32 [[TMP39]], 0
+// CHECK1-NEXT: br i1 [[TMP40]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK1: omp_offload.failed6:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l68(i64 [[TMP13]]) #[[ATTR2]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l68(i64 [[TMP17]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK1: omp_offload.cont7:
// CHECK1-NEXT: ret i32 0
diff --git a/clang/test/OpenMP/target_teams_distribute_parallel_for_lastprivate_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_parallel_for_lastprivate_codegen.cpp
index 2a8a92a3cb512..cc3d28be30ce7 100644
--- a/clang/test/OpenMP/target_teams_distribute_parallel_for_lastprivate_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_parallel_for_lastprivate_codegen.cpp
@@ -797,7 +797,7 @@ int main() {
// CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 5, ptr [[TMP24]], align 4
// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -814,9 +814,17 @@ int main() {
// CHECK5-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK5-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 2, ptr [[TMP31]], align 8
-// CHECK5-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l106.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK5-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK5-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK5-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK5-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK5-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l106.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK5-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l106(ptr [[VEC]], i64 [[TMP2]], ptr [[S_ARR]], ptr [[TMP3]], i64 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -824,18 +832,18 @@ int main() {
// CHECK5-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiET_v()
// CHECK5-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK5-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK5-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK5-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK5: arraydestroy.body:
-// CHECK5-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP34]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK5-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP38]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK5-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK5-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK5-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK5-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK5: arraydestroy.done2:
// CHECK5-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK5-NEXT: [[TMP35:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK5-NEXT: ret i32 [[TMP35]]
+// CHECK5-NEXT: [[TMP39:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK5-NEXT: ret i32 [[TMP39]]
//
//
// CHECK5-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -1258,7 +1266,7 @@ int main() {
// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1275,27 +1283,35 @@ int main() {
// CHECK5-NEXT: store ptr null, ptr [[TMP25]], align 8
// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 2, ptr [[TMP26]], align 8
-// CHECK5-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l50.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK5-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK5-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK5-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP29]], align 4
+// CHECK5-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK5-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l50.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK5-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l50(ptr [[VEC]], i64 [[TMP2]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK5: omp_offload.cont:
// CHECK5-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK5-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK5-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK5-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK5: arraydestroy.body:
-// CHECK5-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP29]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK5-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP33]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK5-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK5-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK5-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK5-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK5: arraydestroy.done2:
// CHECK5-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK5-NEXT: [[TMP30:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK5-NEXT: ret i32 [[TMP30]]
+// CHECK5-NEXT: [[TMP34:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK5-NEXT: ret i32 [[TMP34]]
//
//
// CHECK5-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
@@ -1788,7 +1804,7 @@ int main() {
// CHECK7-NEXT: [[TMP22:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK7-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK7-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK7-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK7-NEXT: store i32 5, ptr [[TMP24]], align 4
// CHECK7-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1805,9 +1821,17 @@ int main() {
// CHECK7-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK7-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK7-NEXT: store i64 2, ptr [[TMP31]], align 8
-// CHECK7-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l106.region_id, ptr [[KERNEL_ARGS]])
-// CHECK7-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK7-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK7-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK7-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK7-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK7-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK7-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l106.region_id, ptr [[KERNEL_ARGS]])
+// CHECK7-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK7-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK7: omp_offload.failed:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l106(ptr [[VEC]], i32 [[TMP2]], ptr [[S_ARR]], ptr [[TMP3]], i32 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1815,18 +1839,18 @@ int main() {
// CHECK7-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiET_v()
// CHECK7-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK7-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK7-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK7-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK7: arraydestroy.body:
-// CHECK7-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP34]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK7-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP38]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK7-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK7-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK7-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK7-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK7: arraydestroy.done2:
// CHECK7-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK7-NEXT: [[TMP35:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK7-NEXT: ret i32 [[TMP35]]
+// CHECK7-NEXT: [[TMP39:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK7-NEXT: ret i32 [[TMP39]]
//
//
// CHECK7-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -2243,7 +2267,7 @@ int main() {
// CHECK7-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK7-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK7-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK7-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK7-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK7-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2260,27 +2284,35 @@ int main() {
// CHECK7-NEXT: store ptr null, ptr [[TMP25]], align 4
// CHECK7-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK7-NEXT: store i64 2, ptr [[TMP26]], align 8
-// CHECK7-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l50.region_id, ptr [[KERNEL_ARGS]])
-// CHECK7-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK7-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK7-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK7-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK7-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP29]], align 4
+// CHECK7-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK7-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l50.region_id, ptr [[KERNEL_ARGS]])
+// CHECK7-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK7-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK7: omp_offload.failed:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l50(ptr [[VEC]], i32 [[TMP2]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK7: omp_offload.cont:
// CHECK7-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK7-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK7-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK7-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK7: arraydestroy.body:
-// CHECK7-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP29]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK7-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP33]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK7-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK7-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK7-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK7-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK7: arraydestroy.done2:
// CHECK7-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK7-NEXT: [[TMP30:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK7-NEXT: ret i32 [[TMP30]]
+// CHECK7-NEXT: [[TMP34:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK7-NEXT: ret i32 [[TMP34]]
//
//
// CHECK7-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
diff --git a/clang/test/OpenMP/target_teams_distribute_parallel_for_order_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_parallel_for_order_codegen.cpp
index 4c5524a979665..f3478cf5a2e15 100644
--- a/clang/test/OpenMP/target_teams_distribute_parallel_for_order_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_parallel_for_order_codegen.cpp
@@ -27,7 +27,7 @@ void gtid_test() {
// CHECK1-NEXT: [[TMP:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -44,9 +44,17 @@ void gtid_test() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l16.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l16.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l16() #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/target_teams_distribute_parallel_for_private_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_parallel_for_private_codegen.cpp
index fe311a4446f73..375e79d0736a0 100644
--- a/clang/test/OpenMP/target_teams_distribute_parallel_for_private_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_parallel_for_private_codegen.cpp
@@ -301,7 +301,7 @@ int main() {
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -318,9 +318,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l124.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l124.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l124() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -563,7 +571,7 @@ int main() {
// CHECK1-NEXT: store ptr undef, ptr [[_TMP1]], align 8
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -580,27 +588,35 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l80.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l80.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l80() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK1-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK1: arraydestroy.body:
-// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK1-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK1-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK1: arraydestroy.done2:
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK1-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK1-NEXT: ret i32 [[TMP12]]
+// CHECK1-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK1-NEXT: ret i32 [[TMP16]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN1SIiEC1Ev
@@ -1025,7 +1041,7 @@ int main() {
// CHECK3-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1042,9 +1058,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l124.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK3-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l124.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK3-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l124() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1281,7 +1305,7 @@ int main() {
// CHECK3-NEXT: store ptr undef, ptr [[_TMP1]], align 4
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1298,27 +1322,35 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l80.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK3-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l80.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK3-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l80() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK3-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK3-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK3: arraydestroy.body:
-// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK3-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK3-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK3: arraydestroy.done2:
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK3-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK3-NEXT: ret i32 [[TMP12]]
+// CHECK3-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK3-NEXT: ret i32 [[TMP16]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN1SIiEC1Ev
diff --git a/clang/test/OpenMP/target_teams_distribute_parallel_for_proc_bind_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_parallel_for_proc_bind_codegen.cpp
index a86b8aebcc937..f1bc5360d465b 100644
--- a/clang/test/OpenMP/target_teams_distribute_parallel_for_proc_bind_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_parallel_for_proc_bind_codegen.cpp
@@ -57,7 +57,7 @@ int main() {
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -74,35 +74,51 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 1000, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l36.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l36.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l36() #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 1000, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l38.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 1000, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l38.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK1-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l38() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -393,7 +409,7 @@ int main() {
// CHECK1-NEXT: [[TMP:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -410,9 +426,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 1000, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l30.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l30.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l30() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/target_teams_distribute_parallel_for_reduction_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_parallel_for_reduction_codegen.cpp
index 69ec5d9187408..2dc01b18aac20 100644
--- a/clang/test/OpenMP/target_teams_distribute_parallel_for_reduction_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_parallel_for_reduction_codegen.cpp
@@ -109,7 +109,7 @@ int main() {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -126,9 +126,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66(ptr @_ZZ4mainE5sivar) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -392,7 +400,7 @@ int main() {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -409,9 +417,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32(ptr [[T_VAR]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -679,7 +695,7 @@ int main() {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -696,9 +712,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66(ptr @_ZZ4mainE5sivar) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -958,7 +982,7 @@ int main() {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -975,9 +999,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32(ptr [[T_VAR]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/target_teams_distribute_parallel_for_schedule_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_parallel_for_schedule_codegen.cpp
index 7551b751fc6c6..7f663d81ffe59 100644
--- a/clang/test/OpenMP/target_teams_distribute_parallel_for_schedule_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_parallel_for_schedule_codegen.cpp
@@ -239,7 +239,7 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -256,157 +256,197 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.4, ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.4, ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK1-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK1-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK1: omp_offload.failed8:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK1: omp_offload.cont9:
// CHECK1-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A10]], ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A10]], ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.8, ptr [[TMP41]], align 8
-// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP42]], align 8
-// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP43]], align 8
-// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP44]], align 8
-// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK1-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK1-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK1-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 8
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 8
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.8, ptr [[TMP49]], align 8
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP50]], align 8
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP51]], align 8
+// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP52]], align 8
+// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK1-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK1-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK1-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK1: omp_offload.failed16:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK1: omp_offload.cont17:
// CHECK1-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP48]], align 8
-// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A18]], ptr [[TMP49]], align 8
-// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP50]], align 8
-// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP60:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP60]], align 8
+// CHECK1-NEXT: [[TMP61:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A18]], ptr [[TMP61]], align 8
+// CHECK1-NEXT: [[TMP62:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP62]], align 8
+// CHECK1-NEXT: [[TMP63:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS23:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP53]], align 4
-// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP51]], ptr [[TMP55]], align 8
-// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP52]], ptr [[TMP56]], align 8
-// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.12, ptr [[TMP57]], align 8
-// CHECK1-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP58]], align 8
-// CHECK1-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP59]], align 8
-// CHECK1-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP60]], align 8
-// CHECK1-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP61]], align 8
-// CHECK1-NEXT: [[TMP62:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.region_id, ptr [[KERNEL_ARGS23]])
-// CHECK1-NEXT: [[TMP63:%.*]] = icmp ne i32 [[TMP62]], 0
-// CHECK1-NEXT: br i1 [[TMP63]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
+// CHECK1-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP65]], align 4
+// CHECK1-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP66]], align 4
+// CHECK1-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP63]], ptr [[TMP67]], align 8
+// CHECK1-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP64]], ptr [[TMP68]], align 8
+// CHECK1-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.12, ptr [[TMP69]], align 8
+// CHECK1-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP70]], align 8
+// CHECK1-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP71]], align 8
+// CHECK1-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP72]], align 8
+// CHECK1-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP73]], align 8
+// CHECK1-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP74]], align 8
+// CHECK1-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP75]], align 4
+// CHECK1-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP76]], align 4
+// CHECK1-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP77]], align 4
+// CHECK1-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.region_id, ptr [[KERNEL_ARGS23]])
+// CHECK1-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
+// CHECK1-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
// CHECK1: omp_offload.failed24:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT25]]
// CHECK1: omp_offload.cont25:
// CHECK1-NEXT: [[A26:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP64]], align 8
-// CHECK1-NEXT: [[TMP65:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A26]], ptr [[TMP65]], align 8
-// CHECK1-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP66]], align 8
-// CHECK1-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP80:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP80]], align 8
+// CHECK1-NEXT: [[TMP81:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A26]], ptr [[TMP81]], align 8
+// CHECK1-NEXT: [[TMP82:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP82]], align 8
+// CHECK1-NEXT: [[TMP83:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP84:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP69]], align 4
-// CHECK1-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP70]], align 4
-// CHECK1-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP67]], ptr [[TMP71]], align 8
-// CHECK1-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 8
-// CHECK1-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.16, ptr [[TMP73]], align 8
-// CHECK1-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP74]], align 8
-// CHECK1-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP75]], align 8
-// CHECK1-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP76]], align 8
-// CHECK1-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP77]], align 8
-// CHECK1-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK1-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
-// CHECK1-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK1-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP85]], align 4
+// CHECK1-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP86]], align 4
+// CHECK1-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP83]], ptr [[TMP87]], align 8
+// CHECK1-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP84]], ptr [[TMP88]], align 8
+// CHECK1-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.16, ptr [[TMP89]], align 8
+// CHECK1-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP90]], align 8
+// CHECK1-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP91]], align 8
+// CHECK1-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP92]], align 8
+// CHECK1-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP93]], align 8
+// CHECK1-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP94]], align 8
+// CHECK1-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP95]], align 4
+// CHECK1-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP96]], align 4
+// CHECK1-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP97]], align 4
+// CHECK1-NEXT: [[TMP98:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK1-NEXT: [[TMP99:%.*]] = icmp ne i32 [[TMP98]], 0
+// CHECK1-NEXT: br i1 [[TMP99]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK1: omp_offload.failed32:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK1: omp_offload.cont33:
// CHECK1-NEXT: [[A34:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A34]], i64 0, i64 0
-// CHECK1-NEXT: [[TMP80:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK1-NEXT: ret i32 [[TMP80]]
+// CHECK1-NEXT: [[TMP100:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK1-NEXT: ret i32 [[TMP100]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36
@@ -1233,7 +1273,7 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1250,157 +1290,197 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 4
-// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A2]], ptr [[TMP17]], align 4
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 4
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A2]], ptr [[TMP21]], align 4
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.4, ptr [[TMP25]], align 4
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP26]], align 4
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK3-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK3-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK3-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.4, ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK3-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK3-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK3-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK3: omp_offload.failed8:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40(ptr [[THIS1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK3: omp_offload.cont9:
// CHECK3-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 4
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A10]], ptr [[TMP33]], align 4
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP34]], align 4
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A10]], ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP42]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 4
-// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 4
-// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.8, ptr [[TMP41]], align 4
-// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP42]], align 4
-// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP44]], align 4
-// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK3-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK3-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK3-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK3-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 4
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 4
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.8, ptr [[TMP49]], align 4
+// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP50]], align 4
+// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP52]], align 4
+// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK3-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK3-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK3-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK3-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK3: omp_offload.failed16:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44(ptr [[THIS1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK3: omp_offload.cont17:
// CHECK3-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP48]], align 4
-// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A18]], ptr [[TMP49]], align 4
-// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP50]], align 4
-// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP60:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP60]], align 4
+// CHECK3-NEXT: [[TMP61:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A18]], ptr [[TMP61]], align 4
+// CHECK3-NEXT: [[TMP62:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP62]], align 4
+// CHECK3-NEXT: [[TMP63:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS23:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP53]], align 4
-// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP51]], ptr [[TMP55]], align 4
-// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP52]], ptr [[TMP56]], align 4
-// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.12, ptr [[TMP57]], align 4
-// CHECK3-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP58]], align 4
-// CHECK3-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP59]], align 4
-// CHECK3-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP60]], align 4
-// CHECK3-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
-// CHECK3-NEXT: store i64 123, ptr [[TMP61]], align 8
-// CHECK3-NEXT: [[TMP62:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.region_id, ptr [[KERNEL_ARGS23]])
-// CHECK3-NEXT: [[TMP63:%.*]] = icmp ne i32 [[TMP62]], 0
-// CHECK3-NEXT: br i1 [[TMP63]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
+// CHECK3-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP65]], align 4
+// CHECK3-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP66]], align 4
+// CHECK3-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP63]], ptr [[TMP67]], align 4
+// CHECK3-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP64]], ptr [[TMP68]], align 4
+// CHECK3-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.12, ptr [[TMP69]], align 4
+// CHECK3-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP70]], align 4
+// CHECK3-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP71]], align 4
+// CHECK3-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP72]], align 4
+// CHECK3-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
+// CHECK3-NEXT: store i64 123, ptr [[TMP73]], align 8
+// CHECK3-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP74]], align 8
+// CHECK3-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP75]], align 4
+// CHECK3-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP76]], align 4
+// CHECK3-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP77]], align 4
+// CHECK3-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.region_id, ptr [[KERNEL_ARGS23]])
+// CHECK3-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
+// CHECK3-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
// CHECK3: omp_offload.failed24:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49(ptr [[THIS1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT25]]
// CHECK3: omp_offload.cont25:
// CHECK3-NEXT: [[A26:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP64]], align 4
-// CHECK3-NEXT: [[TMP65:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A26]], ptr [[TMP65]], align 4
-// CHECK3-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP66]], align 4
-// CHECK3-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP80:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP80]], align 4
+// CHECK3-NEXT: [[TMP81:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A26]], ptr [[TMP81]], align 4
+// CHECK3-NEXT: [[TMP82:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP82]], align 4
+// CHECK3-NEXT: [[TMP83:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP84:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP69]], align 4
-// CHECK3-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP70]], align 4
-// CHECK3-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP67]], ptr [[TMP71]], align 4
-// CHECK3-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 4
-// CHECK3-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.16, ptr [[TMP73]], align 4
-// CHECK3-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP74]], align 4
-// CHECK3-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP75]], align 4
-// CHECK3-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP76]], align 4
-// CHECK3-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK3-NEXT: store i64 123, ptr [[TMP77]], align 8
-// CHECK3-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK3-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
-// CHECK3-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK3-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP85]], align 4
+// CHECK3-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP86]], align 4
+// CHECK3-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP83]], ptr [[TMP87]], align 4
+// CHECK3-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP84]], ptr [[TMP88]], align 4
+// CHECK3-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.16, ptr [[TMP89]], align 4
+// CHECK3-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP90]], align 4
+// CHECK3-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP91]], align 4
+// CHECK3-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP92]], align 4
+// CHECK3-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK3-NEXT: store i64 123, ptr [[TMP93]], align 8
+// CHECK3-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP94]], align 8
+// CHECK3-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP95]], align 4
+// CHECK3-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP96]], align 4
+// CHECK3-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP97]], align 4
+// CHECK3-NEXT: [[TMP98:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK3-NEXT: [[TMP99:%.*]] = icmp ne i32 [[TMP98]], 0
+// CHECK3-NEXT: br i1 [[TMP99]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK3: omp_offload.failed32:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54(ptr [[THIS1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK3: omp_offload.cont33:
// CHECK3-NEXT: [[A34:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A34]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP80:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK3-NEXT: ret i32 [[TMP80]]
+// CHECK3-NEXT: [[TMP100:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK3-NEXT: ret i32 [[TMP100]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36
@@ -2200,7 +2280,7 @@ int main (int argc, char **argv) {
// CHECK5-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK5-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK5-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2217,157 +2297,197 @@ int main (int argc, char **argv) {
// CHECK5-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK5-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK5-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK5-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK5-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK5-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK5-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK5: omp_offload.cont:
// CHECK5-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 8
-// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[A2]], ptr [[TMP17]], align 8
-// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
-// CHECK5-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 8
+// CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[A2]], ptr [[TMP21]], align 8
+// CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
+// CHECK5-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK5-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK5-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK5-NEXT: store ptr @.offload_sizes.4, ptr [[TMP25]], align 8
-// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK5-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP26]], align 8
-// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK5-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK5-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK5-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK5-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK5-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK5-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK5-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK5-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK5-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK5-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK5-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK5-NEXT: store ptr @.offload_sizes.4, ptr [[TMP29]], align 8
+// CHECK5-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK5-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP30]], align 8
+// CHECK5-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK5-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK5-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK5-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK5-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK5-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK5-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK5-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK5-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK5-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK5-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK5: omp_offload.failed8:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40(ptr [[THIS1]]) #[[ATTR2]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK5: omp_offload.cont9:
// CHECK5-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 8
-// CHECK5-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[A10]], ptr [[TMP33]], align 8
-// CHECK5-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
-// CHECK5-NEXT: store ptr null, ptr [[TMP34]], align 8
-// CHECK5-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 8
+// CHECK5-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[A10]], ptr [[TMP41]], align 8
+// CHECK5-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
+// CHECK5-NEXT: store ptr null, ptr [[TMP42]], align 8
+// CHECK5-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK5-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK5-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK5-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK5-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
-// CHECK5-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
-// CHECK5-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK5-NEXT: store ptr @.offload_sizes.8, ptr [[TMP41]], align 8
-// CHECK5-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK5-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP42]], align 8
-// CHECK5-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP43]], align 8
-// CHECK5-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK5-NEXT: store ptr null, ptr [[TMP44]], align 8
-// CHECK5-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK5-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK5-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK5-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK5-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK5-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK5-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK5-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK5-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK5-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 8
+// CHECK5-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 8
+// CHECK5-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK5-NEXT: store ptr @.offload_sizes.8, ptr [[TMP49]], align 8
+// CHECK5-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK5-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP50]], align 8
+// CHECK5-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP51]], align 8
+// CHECK5-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP52]], align 8
+// CHECK5-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK5-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK5-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK5-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK5-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK5-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK5-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK5-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK5-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK5: omp_offload.failed16:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44(ptr [[THIS1]]) #[[ATTR2]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK5: omp_offload.cont17:
// CHECK5-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP48:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP48]], align 8
-// CHECK5-NEXT: [[TMP49:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[A18]], ptr [[TMP49]], align 8
-// CHECK5-NEXT: [[TMP50:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
-// CHECK5-NEXT: store ptr null, ptr [[TMP50]], align 8
-// CHECK5-NEXT: [[TMP51:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP52:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP60:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP60]], align 8
+// CHECK5-NEXT: [[TMP61:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[A18]], ptr [[TMP61]], align 8
+// CHECK5-NEXT: [[TMP62:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
+// CHECK5-NEXT: store ptr null, ptr [[TMP62]], align 8
+// CHECK5-NEXT: [[TMP63:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS23:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP53]], align 4
-// CHECK5-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
-// CHECK5-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK5-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
-// CHECK5-NEXT: store ptr [[TMP51]], ptr [[TMP55]], align 8
-// CHECK5-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[TMP52]], ptr [[TMP56]], align 8
-// CHECK5-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
-// CHECK5-NEXT: store ptr @.offload_sizes.12, ptr [[TMP57]], align 8
-// CHECK5-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
-// CHECK5-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP58]], align 8
-// CHECK5-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP59]], align 8
-// CHECK5-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
-// CHECK5-NEXT: store ptr null, ptr [[TMP60]], align 8
-// CHECK5-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
-// CHECK5-NEXT: store i64 123, ptr [[TMP61]], align 8
-// CHECK5-NEXT: [[TMP62:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.region_id, ptr [[KERNEL_ARGS23]])
-// CHECK5-NEXT: [[TMP63:%.*]] = icmp ne i32 [[TMP62]], 0
-// CHECK5-NEXT: br i1 [[TMP63]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
+// CHECK5-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP65]], align 4
+// CHECK5-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
+// CHECK5-NEXT: store i32 1, ptr [[TMP66]], align 4
+// CHECK5-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
+// CHECK5-NEXT: store ptr [[TMP63]], ptr [[TMP67]], align 8
+// CHECK5-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[TMP64]], ptr [[TMP68]], align 8
+// CHECK5-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
+// CHECK5-NEXT: store ptr @.offload_sizes.12, ptr [[TMP69]], align 8
+// CHECK5-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
+// CHECK5-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP70]], align 8
+// CHECK5-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP71]], align 8
+// CHECK5-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP72]], align 8
+// CHECK5-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
+// CHECK5-NEXT: store i64 123, ptr [[TMP73]], align 8
+// CHECK5-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP74]], align 8
+// CHECK5-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP75]], align 4
+// CHECK5-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP76]], align 4
+// CHECK5-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP77]], align 4
+// CHECK5-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.region_id, ptr [[KERNEL_ARGS23]])
+// CHECK5-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
+// CHECK5-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
// CHECK5: omp_offload.failed24:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49(ptr [[THIS1]]) #[[ATTR2]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT25]]
// CHECK5: omp_offload.cont25:
// CHECK5-NEXT: [[A26:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP64]], align 8
-// CHECK5-NEXT: [[TMP65:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[A26]], ptr [[TMP65]], align 8
-// CHECK5-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i64 0, i64 0
-// CHECK5-NEXT: store ptr null, ptr [[TMP66]], align 8
-// CHECK5-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP80:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP80]], align 8
+// CHECK5-NEXT: [[TMP81:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[A26]], ptr [[TMP81]], align 8
+// CHECK5-NEXT: [[TMP82:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i64 0, i64 0
+// CHECK5-NEXT: store ptr null, ptr [[TMP82]], align 8
+// CHECK5-NEXT: [[TMP83:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP84:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP69]], align 4
-// CHECK5-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK5-NEXT: store i32 1, ptr [[TMP70]], align 4
-// CHECK5-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK5-NEXT: store ptr [[TMP67]], ptr [[TMP71]], align 8
-// CHECK5-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 8
-// CHECK5-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK5-NEXT: store ptr @.offload_sizes.16, ptr [[TMP73]], align 8
-// CHECK5-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK5-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP74]], align 8
-// CHECK5-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP75]], align 8
-// CHECK5-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK5-NEXT: store ptr null, ptr [[TMP76]], align 8
-// CHECK5-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK5-NEXT: store i64 123, ptr [[TMP77]], align 8
-// CHECK5-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK5-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
-// CHECK5-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK5-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP85]], align 4
+// CHECK5-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK5-NEXT: store i32 1, ptr [[TMP86]], align 4
+// CHECK5-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK5-NEXT: store ptr [[TMP83]], ptr [[TMP87]], align 8
+// CHECK5-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[TMP84]], ptr [[TMP88]], align 8
+// CHECK5-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK5-NEXT: store ptr @.offload_sizes.16, ptr [[TMP89]], align 8
+// CHECK5-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK5-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP90]], align 8
+// CHECK5-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP91]], align 8
+// CHECK5-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP92]], align 8
+// CHECK5-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK5-NEXT: store i64 123, ptr [[TMP93]], align 8
+// CHECK5-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP94]], align 8
+// CHECK5-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP95]], align 4
+// CHECK5-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP96]], align 4
+// CHECK5-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP97]], align 4
+// CHECK5-NEXT: [[TMP98:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK5-NEXT: [[TMP99:%.*]] = icmp ne i32 [[TMP98]], 0
+// CHECK5-NEXT: br i1 [[TMP99]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK5: omp_offload.failed32:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54(ptr [[THIS1]]) #[[ATTR2]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK5: omp_offload.cont33:
// CHECK5-NEXT: [[A34:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK5-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A34]], i64 0, i64 0
-// CHECK5-NEXT: [[TMP80:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK5-NEXT: ret i32 [[TMP80]]
+// CHECK5-NEXT: [[TMP100:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK5-NEXT: ret i32 [[TMP100]]
//
//
// CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36
@@ -3194,7 +3314,7 @@ int main (int argc, char **argv) {
// CHECK7-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK7-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK7-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK7-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK7-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK7-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3211,157 +3331,197 @@ int main (int argc, char **argv) {
// CHECK7-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK7-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK7-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK7-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS]])
-// CHECK7-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK7-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK7-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK7-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK7-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK7-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK7-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS]])
+// CHECK7-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK7-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK7: omp_offload.failed:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK7: omp_offload.cont:
// CHECK7-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 4
-// CHECK7-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK7-NEXT: store ptr [[A2]], ptr [[TMP17]], align 4
-// CHECK7-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
-// CHECK7-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK7-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 4
+// CHECK7-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK7-NEXT: store ptr [[A2]], ptr [[TMP21]], align 4
+// CHECK7-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
+// CHECK7-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK7-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK7-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK7-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK7-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK7-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK7-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK7-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK7-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK7-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK7-NEXT: store ptr @.offload_sizes.4, ptr [[TMP25]], align 4
-// CHECK7-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK7-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP26]], align 4
-// CHECK7-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK7-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK7-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK7-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK7-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK7-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK7-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK7-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK7-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK7-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK7-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK7-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK7-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK7-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK7-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK7-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK7-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK7-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK7-NEXT: store ptr @.offload_sizes.4, ptr [[TMP29]], align 4
+// CHECK7-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK7-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP30]], align 4
+// CHECK7-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK7-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK7-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK7-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK7-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK7-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK7-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK7-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK7-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK7-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK7-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK7-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK7-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK7: omp_offload.failed8:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40(ptr [[THIS1]]) #[[ATTR2]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK7: omp_offload.cont9:
// CHECK7-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 4
-// CHECK7-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK7-NEXT: store ptr [[A10]], ptr [[TMP33]], align 4
-// CHECK7-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
-// CHECK7-NEXT: store ptr null, ptr [[TMP34]], align 4
-// CHECK7-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 4
+// CHECK7-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK7-NEXT: store ptr [[A10]], ptr [[TMP41]], align 4
+// CHECK7-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
+// CHECK7-NEXT: store ptr null, ptr [[TMP42]], align 4
+// CHECK7-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK7-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK7-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK7-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK7-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK7-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 4
-// CHECK7-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK7-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 4
-// CHECK7-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK7-NEXT: store ptr @.offload_sizes.8, ptr [[TMP41]], align 4
-// CHECK7-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK7-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP42]], align 4
-// CHECK7-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK7-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK7-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK7-NEXT: store ptr null, ptr [[TMP44]], align 4
-// CHECK7-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK7-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK7-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK7-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK7-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK7-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK7-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK7-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK7-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK7-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK7-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 4
+// CHECK7-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK7-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 4
+// CHECK7-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK7-NEXT: store ptr @.offload_sizes.8, ptr [[TMP49]], align 4
+// CHECK7-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK7-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP50]], align 4
+// CHECK7-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK7-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK7-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK7-NEXT: store ptr null, ptr [[TMP52]], align 4
+// CHECK7-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK7-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK7-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK7-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK7-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK7-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK7-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK7-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK7-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK7: omp_offload.failed16:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44(ptr [[THIS1]]) #[[ATTR2]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK7: omp_offload.cont17:
// CHECK7-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP48:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP48]], align 4
-// CHECK7-NEXT: [[TMP49:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK7-NEXT: store ptr [[A18]], ptr [[TMP49]], align 4
-// CHECK7-NEXT: [[TMP50:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
-// CHECK7-NEXT: store ptr null, ptr [[TMP50]], align 4
-// CHECK7-NEXT: [[TMP51:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP52:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP60:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP60]], align 4
+// CHECK7-NEXT: [[TMP61:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK7-NEXT: store ptr [[A18]], ptr [[TMP61]], align 4
+// CHECK7-NEXT: [[TMP62:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
+// CHECK7-NEXT: store ptr null, ptr [[TMP62]], align 4
+// CHECK7-NEXT: [[TMP63:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS23:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK7-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP53]], align 4
-// CHECK7-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
-// CHECK7-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK7-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
-// CHECK7-NEXT: store ptr [[TMP51]], ptr [[TMP55]], align 4
-// CHECK7-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
-// CHECK7-NEXT: store ptr [[TMP52]], ptr [[TMP56]], align 4
-// CHECK7-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
-// CHECK7-NEXT: store ptr @.offload_sizes.12, ptr [[TMP57]], align 4
-// CHECK7-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
-// CHECK7-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP58]], align 4
-// CHECK7-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
-// CHECK7-NEXT: store ptr null, ptr [[TMP59]], align 4
-// CHECK7-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
-// CHECK7-NEXT: store ptr null, ptr [[TMP60]], align 4
-// CHECK7-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
-// CHECK7-NEXT: store i64 123, ptr [[TMP61]], align 8
-// CHECK7-NEXT: [[TMP62:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.region_id, ptr [[KERNEL_ARGS23]])
-// CHECK7-NEXT: [[TMP63:%.*]] = icmp ne i32 [[TMP62]], 0
-// CHECK7-NEXT: br i1 [[TMP63]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
+// CHECK7-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
+// CHECK7-NEXT: store i32 2, ptr [[TMP65]], align 4
+// CHECK7-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
+// CHECK7-NEXT: store i32 1, ptr [[TMP66]], align 4
+// CHECK7-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
+// CHECK7-NEXT: store ptr [[TMP63]], ptr [[TMP67]], align 4
+// CHECK7-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
+// CHECK7-NEXT: store ptr [[TMP64]], ptr [[TMP68]], align 4
+// CHECK7-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
+// CHECK7-NEXT: store ptr @.offload_sizes.12, ptr [[TMP69]], align 4
+// CHECK7-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
+// CHECK7-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP70]], align 4
+// CHECK7-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
+// CHECK7-NEXT: store ptr null, ptr [[TMP71]], align 4
+// CHECK7-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
+// CHECK7-NEXT: store ptr null, ptr [[TMP72]], align 4
+// CHECK7-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
+// CHECK7-NEXT: store i64 123, ptr [[TMP73]], align 8
+// CHECK7-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP74]], align 8
+// CHECK7-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP75]], align 4
+// CHECK7-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP76]], align 4
+// CHECK7-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP77]], align 4
+// CHECK7-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.region_id, ptr [[KERNEL_ARGS23]])
+// CHECK7-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
+// CHECK7-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
// CHECK7: omp_offload.failed24:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49(ptr [[THIS1]]) #[[ATTR2]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT25]]
// CHECK7: omp_offload.cont25:
// CHECK7-NEXT: [[A26:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP64]], align 4
-// CHECK7-NEXT: [[TMP65:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
-// CHECK7-NEXT: store ptr [[A26]], ptr [[TMP65]], align 4
-// CHECK7-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i32 0, i32 0
-// CHECK7-NEXT: store ptr null, ptr [[TMP66]], align 4
-// CHECK7-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP80:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP80]], align 4
+// CHECK7-NEXT: [[TMP81:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK7-NEXT: store ptr [[A26]], ptr [[TMP81]], align 4
+// CHECK7-NEXT: [[TMP82:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i32 0, i32 0
+// CHECK7-NEXT: store ptr null, ptr [[TMP82]], align 4
+// CHECK7-NEXT: [[TMP83:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP84:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK7-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP69]], align 4
-// CHECK7-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK7-NEXT: store i32 1, ptr [[TMP70]], align 4
-// CHECK7-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK7-NEXT: store ptr [[TMP67]], ptr [[TMP71]], align 4
-// CHECK7-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK7-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 4
-// CHECK7-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK7-NEXT: store ptr @.offload_sizes.16, ptr [[TMP73]], align 4
-// CHECK7-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK7-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP74]], align 4
-// CHECK7-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK7-NEXT: store ptr null, ptr [[TMP75]], align 4
-// CHECK7-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK7-NEXT: store ptr null, ptr [[TMP76]], align 4
-// CHECK7-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK7-NEXT: store i64 123, ptr [[TMP77]], align 8
-// CHECK7-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK7-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
-// CHECK7-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK7-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK7-NEXT: store i32 2, ptr [[TMP85]], align 4
+// CHECK7-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK7-NEXT: store i32 1, ptr [[TMP86]], align 4
+// CHECK7-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK7-NEXT: store ptr [[TMP83]], ptr [[TMP87]], align 4
+// CHECK7-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK7-NEXT: store ptr [[TMP84]], ptr [[TMP88]], align 4
+// CHECK7-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK7-NEXT: store ptr @.offload_sizes.16, ptr [[TMP89]], align 4
+// CHECK7-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK7-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP90]], align 4
+// CHECK7-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK7-NEXT: store ptr null, ptr [[TMP91]], align 4
+// CHECK7-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK7-NEXT: store ptr null, ptr [[TMP92]], align 4
+// CHECK7-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK7-NEXT: store i64 123, ptr [[TMP93]], align 8
+// CHECK7-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP94]], align 8
+// CHECK7-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP95]], align 4
+// CHECK7-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP96]], align 4
+// CHECK7-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP97]], align 4
+// CHECK7-NEXT: [[TMP98:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK7-NEXT: [[TMP99:%.*]] = icmp ne i32 [[TMP98]], 0
+// CHECK7-NEXT: br i1 [[TMP99]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK7: omp_offload.failed32:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54(ptr [[THIS1]]) #[[ATTR2]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK7: omp_offload.cont33:
// CHECK7-NEXT: [[A34:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK7-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A34]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP80:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK7-NEXT: ret i32 [[TMP80]]
+// CHECK7-NEXT: [[TMP100:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK7-NEXT: ret i32 [[TMP100]]
//
//
// CHECK7-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36
@@ -4221,7 +4381,7 @@ int main (int argc, char **argv) {
// CHECK13-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK13-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK13-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK13-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK13-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK13-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK13-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -4238,298 +4398,338 @@ int main (int argc, char **argv) {
// CHECK13-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK13-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK13-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK13-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.region_id, ptr [[KERNEL_ARGS]])
-// CHECK13-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK13-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK13-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK13-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK13-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK13-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK13-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.region_id, ptr [[KERNEL_ARGS]])
+// CHECK13-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK13-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK13: omp_offload.failed:
// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139(i64 [[TMP4]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK13: omp_offload.cont:
-// CHECK13-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK13-NEXT: [[TMP35:%.*]] = load i64, ptr [[N_CASTED3]], align 8
-// CHECK13-NEXT: [[TMP36:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK13-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK13-NEXT: [[TMP39:%.*]] = load i64, ptr [[N_CASTED3]], align 8
+// CHECK13-NEXT: [[TMP40:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK13-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES7]], ptr align 8 @.offload_sizes.4, i64 24, i1 false)
-// CHECK13-NEXT: [[TMP37:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP35]], ptr [[TMP37]], align 8
-// CHECK13-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP35]], ptr [[TMP38]], align 8
-// CHECK13-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK13-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP40]], align 8
-// CHECK13-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP41]], align 8
-// CHECK13-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
-// CHECK13-NEXT: store ptr null, ptr [[TMP42]], align 8
-// CHECK13-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP43]], align 8
-// CHECK13-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 8
-// CHECK13-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK13-NEXT: store i64 [[TMP36]], ptr [[TMP45]], align 8
-// CHECK13-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK13-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP39]], ptr [[TMP41]], align 8
+// CHECK13-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP39]], ptr [[TMP42]], align 8
+// CHECK13-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK13-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP44]], align 8
+// CHECK13-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP45]], align 8
+// CHECK13-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
// CHECK13-NEXT: store ptr null, ptr [[TMP46]], align 8
-// CHECK13-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP50:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP50]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK13-NEXT: [[TMP51:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK13-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP51]], 0
+// CHECK13-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP47]], align 8
+// CHECK13-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 8
+// CHECK13-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK13-NEXT: store i64 [[TMP40]], ptr [[TMP49]], align 8
+// CHECK13-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK13-NEXT: store ptr null, ptr [[TMP50]], align 8
+// CHECK13-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP54:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP54]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK13-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK13-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP55]], 0
// CHECK13-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK13-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK13-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK13-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK13-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP52]], 1
-// CHECK13-NEXT: [[TMP53:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK13-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK13-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP56]], 1
+// CHECK13-NEXT: [[TMP57:%.*]] = zext i32 [[ADD14]] to i64
// CHECK13-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK13-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK13-NEXT: store i32 3, ptr [[TMP55]], align 4
-// CHECK13-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP47]], ptr [[TMP56]], align 8
-// CHECK13-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 8
-// CHECK13-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK13-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 8
-// CHECK13-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP59]], align 8
-// CHECK13-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP60]], align 8
-// CHECK13-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP61]], align 8
-// CHECK13-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK13-NEXT: store i64 [[TMP53]], ptr [[TMP62]], align 8
-// CHECK13-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK13-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
-// CHECK13-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK13-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP58]], align 4
+// CHECK13-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK13-NEXT: store i32 3, ptr [[TMP59]], align 4
+// CHECK13-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP51]], ptr [[TMP60]], align 8
+// CHECK13-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 8
+// CHECK13-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK13-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 8
+// CHECK13-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP63]], align 8
+// CHECK13-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP64]], align 8
+// CHECK13-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP65]], align 8
+// CHECK13-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK13-NEXT: store i64 [[TMP57]], ptr [[TMP66]], align 8
+// CHECK13-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP67]], align 8
+// CHECK13-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP68]], align 4
+// CHECK13-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK13-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP70]], align 4
+// CHECK13-NEXT: [[TMP71:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK13-NEXT: [[TMP72:%.*]] = icmp ne i32 [[TMP71]], 0
+// CHECK13-NEXT: br i1 [[TMP72]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK13: omp_offload.failed16:
-// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143(i64 [[TMP35]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143(i64 [[TMP39]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK13: omp_offload.cont17:
-// CHECK13-NEXT: [[TMP65:%.*]] = load i32, ptr [[M]], align 4
-// CHECK13-NEXT: store i32 [[TMP65]], ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK13-NEXT: [[TMP66:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP66]], ptr [[N_CASTED19]], align 4
-// CHECK13-NEXT: [[TMP67:%.*]] = load i64, ptr [[N_CASTED19]], align 8
-// CHECK13-NEXT: [[TMP68:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK13-NEXT: store i32 [[TMP68]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK13-NEXT: [[TMP69:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK13-NEXT: [[TMP70:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK13-NEXT: [[TMP73:%.*]] = load i32, ptr [[M]], align 4
+// CHECK13-NEXT: store i32 [[TMP73]], ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK13-NEXT: [[TMP74:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP74]], ptr [[N_CASTED19]], align 4
+// CHECK13-NEXT: [[TMP75:%.*]] = load i64, ptr [[N_CASTED19]], align 8
+// CHECK13-NEXT: [[TMP76:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK13-NEXT: store i32 [[TMP76]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK13-NEXT: [[TMP77:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK13-NEXT: [[TMP78:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK13-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES23]], ptr align 8 @.offload_sizes.8, i64 32, i1 false)
-// CHECK13-NEXT: [[TMP71:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP67]], ptr [[TMP71]], align 8
-// CHECK13-NEXT: [[TMP72:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP67]], ptr [[TMP72]], align 8
-// CHECK13-NEXT: [[TMP73:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP73]], align 8
-// CHECK13-NEXT: [[TMP74:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP74]], align 8
-// CHECK13-NEXT: [[TMP75:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP75]], align 8
-// CHECK13-NEXT: [[TMP76:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 1
-// CHECK13-NEXT: store ptr null, ptr [[TMP76]], align 8
-// CHECK13-NEXT: [[TMP77:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP77]], align 8
-// CHECK13-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP78]], align 8
-// CHECK13-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
-// CHECK13-NEXT: store i64 [[TMP70]], ptr [[TMP79]], align 8
-// CHECK13-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 2
-// CHECK13-NEXT: store ptr null, ptr [[TMP80]], align 8
-// CHECK13-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
-// CHECK13-NEXT: store i64 [[TMP69]], ptr [[TMP81]], align 8
-// CHECK13-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
-// CHECK13-NEXT: store i64 [[TMP69]], ptr [[TMP82]], align 8
-// CHECK13-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 3
-// CHECK13-NEXT: store ptr null, ptr [[TMP83]], align 8
-// CHECK13-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP87:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP87]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK13-NEXT: [[TMP88:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK13-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP88]], 0
+// CHECK13-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP75]], ptr [[TMP79]], align 8
+// CHECK13-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP75]], ptr [[TMP80]], align 8
+// CHECK13-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP81]], align 8
+// CHECK13-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP82]], align 8
+// CHECK13-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP83]], align 8
+// CHECK13-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 1
+// CHECK13-NEXT: store ptr null, ptr [[TMP84]], align 8
+// CHECK13-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP85]], align 8
+// CHECK13-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP86]], align 8
+// CHECK13-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
+// CHECK13-NEXT: store i64 [[TMP78]], ptr [[TMP87]], align 8
+// CHECK13-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 2
+// CHECK13-NEXT: store ptr null, ptr [[TMP88]], align 8
+// CHECK13-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
+// CHECK13-NEXT: store i64 [[TMP77]], ptr [[TMP89]], align 8
+// CHECK13-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
+// CHECK13-NEXT: store i64 [[TMP77]], ptr [[TMP90]], align 8
+// CHECK13-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 3
+// CHECK13-NEXT: store ptr null, ptr [[TMP91]], align 8
+// CHECK13-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP94:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP95:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP95]], ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK13-NEXT: [[TMP96:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK13-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP96]], 0
// CHECK13-NEXT: [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1
// CHECK13-NEXT: [[SUB29:%.*]] = sub nsw i32 [[DIV28]], 1
// CHECK13-NEXT: store i32 [[SUB29]], ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK13-NEXT: [[TMP89:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK13-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP89]], 1
-// CHECK13-NEXT: [[TMP90:%.*]] = zext i32 [[ADD30]] to i64
+// CHECK13-NEXT: [[TMP97:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
+// CHECK13-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP97]], 1
+// CHECK13-NEXT: [[TMP98:%.*]] = zext i32 [[ADD30]] to i64
// CHECK13-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP91]], align 4
-// CHECK13-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK13-NEXT: store i32 4, ptr [[TMP92]], align 4
-// CHECK13-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP84]], ptr [[TMP93]], align 8
-// CHECK13-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP85]], ptr [[TMP94]], align 8
-// CHECK13-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK13-NEXT: store ptr [[TMP86]], ptr [[TMP95]], align 8
-// CHECK13-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP96]], align 8
-// CHECK13-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP97]], align 8
-// CHECK13-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP98]], align 8
-// CHECK13-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK13-NEXT: store i64 [[TMP90]], ptr [[TMP99]], align 8
-// CHECK13-NEXT: [[TMP100:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK13-NEXT: [[TMP101:%.*]] = icmp ne i32 [[TMP100]], 0
-// CHECK13-NEXT: br i1 [[TMP101]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK13-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP99]], align 4
+// CHECK13-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK13-NEXT: store i32 4, ptr [[TMP100]], align 4
+// CHECK13-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP92]], ptr [[TMP101]], align 8
+// CHECK13-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP93]], ptr [[TMP102]], align 8
+// CHECK13-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK13-NEXT: store ptr [[TMP94]], ptr [[TMP103]], align 8
+// CHECK13-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP104]], align 8
+// CHECK13-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP105]], align 8
+// CHECK13-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP106]], align 8
+// CHECK13-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK13-NEXT: store i64 [[TMP98]], ptr [[TMP107]], align 8
+// CHECK13-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP108]], align 8
+// CHECK13-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP109]], align 4
+// CHECK13-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP110]], align 4
+// CHECK13-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP111]], align 4
+// CHECK13-NEXT: [[TMP112:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK13-NEXT: [[TMP113:%.*]] = icmp ne i32 [[TMP112]], 0
+// CHECK13-NEXT: br i1 [[TMP113]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK13: omp_offload.failed32:
-// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147(i64 [[TMP67]], i64 [[TMP1]], ptr [[VLA]], i64 [[TMP69]]) #[[ATTR3]]
+// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147(i64 [[TMP75]], i64 [[TMP1]], ptr [[VLA]], i64 [[TMP77]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK13: omp_offload.cont33:
-// CHECK13-NEXT: [[TMP102:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP102]], ptr [[N_CASTED34]], align 4
-// CHECK13-NEXT: [[TMP103:%.*]] = load i64, ptr [[N_CASTED34]], align 8
-// CHECK13-NEXT: [[TMP104:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK13-NEXT: [[TMP114:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP114]], ptr [[N_CASTED34]], align 4
+// CHECK13-NEXT: [[TMP115:%.*]] = load i64, ptr [[N_CASTED34]], align 8
+// CHECK13-NEXT: [[TMP116:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK13-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES38]], ptr align 8 @.offload_sizes.12, i64 24, i1 false)
-// CHECK13-NEXT: [[TMP105:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP103]], ptr [[TMP105]], align 8
-// CHECK13-NEXT: [[TMP106:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP103]], ptr [[TMP106]], align 8
-// CHECK13-NEXT: [[TMP107:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP107]], align 8
-// CHECK13-NEXT: [[TMP108:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP108]], align 8
-// CHECK13-NEXT: [[TMP109:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP109]], align 8
-// CHECK13-NEXT: [[TMP110:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 1
-// CHECK13-NEXT: store ptr null, ptr [[TMP110]], align 8
-// CHECK13-NEXT: [[TMP111:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP111]], align 8
-// CHECK13-NEXT: [[TMP112:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP112]], align 8
-// CHECK13-NEXT: [[TMP113:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 2
-// CHECK13-NEXT: store i64 [[TMP104]], ptr [[TMP113]], align 8
-// CHECK13-NEXT: [[TMP114:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 2
-// CHECK13-NEXT: store ptr null, ptr [[TMP114]], align 8
-// CHECK13-NEXT: [[TMP115:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP116:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP117:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP118:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP118]], ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK13-NEXT: [[TMP119:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK13-NEXT: [[SUB42:%.*]] = sub nsw i32 [[TMP119]], 0
+// CHECK13-NEXT: [[TMP117:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP115]], ptr [[TMP117]], align 8
+// CHECK13-NEXT: [[TMP118:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP115]], ptr [[TMP118]], align 8
+// CHECK13-NEXT: [[TMP119:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP119]], align 8
+// CHECK13-NEXT: [[TMP120:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP120]], align 8
+// CHECK13-NEXT: [[TMP121:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP121]], align 8
+// CHECK13-NEXT: [[TMP122:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 1
+// CHECK13-NEXT: store ptr null, ptr [[TMP122]], align 8
+// CHECK13-NEXT: [[TMP123:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP123]], align 8
+// CHECK13-NEXT: [[TMP124:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP124]], align 8
+// CHECK13-NEXT: [[TMP125:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 2
+// CHECK13-NEXT: store i64 [[TMP116]], ptr [[TMP125]], align 8
+// CHECK13-NEXT: [[TMP126:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 2
+// CHECK13-NEXT: store ptr null, ptr [[TMP126]], align 8
+// CHECK13-NEXT: [[TMP127:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP128:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP129:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP130:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP130]], ptr [[DOTCAPTURE_EXPR_40]], align 4
+// CHECK13-NEXT: [[TMP131:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
+// CHECK13-NEXT: [[SUB42:%.*]] = sub nsw i32 [[TMP131]], 0
// CHECK13-NEXT: [[DIV43:%.*]] = sdiv i32 [[SUB42]], 1
// CHECK13-NEXT: [[SUB44:%.*]] = sub nsw i32 [[DIV43]], 1
// CHECK13-NEXT: store i32 [[SUB44]], ptr [[DOTCAPTURE_EXPR_41]], align 4
-// CHECK13-NEXT: [[TMP120:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_41]], align 4
-// CHECK13-NEXT: [[ADD45:%.*]] = add nsw i32 [[TMP120]], 1
-// CHECK13-NEXT: [[TMP121:%.*]] = zext i32 [[ADD45]] to i64
+// CHECK13-NEXT: [[TMP132:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_41]], align 4
+// CHECK13-NEXT: [[ADD45:%.*]] = add nsw i32 [[TMP132]], 1
+// CHECK13-NEXT: [[TMP133:%.*]] = zext i32 [[ADD45]] to i64
// CHECK13-NEXT: [[KERNEL_ARGS46:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP122:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP122]], align 4
-// CHECK13-NEXT: [[TMP123:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 1
-// CHECK13-NEXT: store i32 3, ptr [[TMP123]], align 4
-// CHECK13-NEXT: [[TMP124:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP115]], ptr [[TMP124]], align 8
-// CHECK13-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP116]], ptr [[TMP125]], align 8
-// CHECK13-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 4
-// CHECK13-NEXT: store ptr [[TMP117]], ptr [[TMP126]], align 8
-// CHECK13-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP127]], align 8
-// CHECK13-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP128]], align 8
-// CHECK13-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP129]], align 8
-// CHECK13-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 8
-// CHECK13-NEXT: store i64 [[TMP121]], ptr [[TMP130]], align 8
-// CHECK13-NEXT: [[TMP131:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.region_id, ptr [[KERNEL_ARGS46]])
-// CHECK13-NEXT: [[TMP132:%.*]] = icmp ne i32 [[TMP131]], 0
-// CHECK13-NEXT: br i1 [[TMP132]], label [[OMP_OFFLOAD_FAILED47:%.*]], label [[OMP_OFFLOAD_CONT48:%.*]]
+// CHECK13-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP134]], align 4
+// CHECK13-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 1
+// CHECK13-NEXT: store i32 3, ptr [[TMP135]], align 4
+// CHECK13-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP127]], ptr [[TMP136]], align 8
+// CHECK13-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP128]], ptr [[TMP137]], align 8
+// CHECK13-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 4
+// CHECK13-NEXT: store ptr [[TMP129]], ptr [[TMP138]], align 8
+// CHECK13-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP139]], align 8
+// CHECK13-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP140]], align 8
+// CHECK13-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP141]], align 8
+// CHECK13-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 8
+// CHECK13-NEXT: store i64 [[TMP133]], ptr [[TMP142]], align 8
+// CHECK13-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP143]], align 8
+// CHECK13-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP144]], align 4
+// CHECK13-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP145]], align 4
+// CHECK13-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP146]], align 4
+// CHECK13-NEXT: [[TMP147:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.region_id, ptr [[KERNEL_ARGS46]])
+// CHECK13-NEXT: [[TMP148:%.*]] = icmp ne i32 [[TMP147]], 0
+// CHECK13-NEXT: br i1 [[TMP148]], label [[OMP_OFFLOAD_FAILED47:%.*]], label [[OMP_OFFLOAD_CONT48:%.*]]
// CHECK13: omp_offload.failed47:
-// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151(i64 [[TMP103]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151(i64 [[TMP115]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT48]]
// CHECK13: omp_offload.cont48:
-// CHECK13-NEXT: [[TMP133:%.*]] = load i32, ptr [[M]], align 4
-// CHECK13-NEXT: store i32 [[TMP133]], ptr [[DOTCAPTURE_EXPR_49]], align 4
-// CHECK13-NEXT: [[TMP134:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP134]], ptr [[N_CASTED50]], align 4
-// CHECK13-NEXT: [[TMP135:%.*]] = load i64, ptr [[N_CASTED50]], align 8
-// CHECK13-NEXT: [[TMP136:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_49]], align 4
-// CHECK13-NEXT: store i32 [[TMP136]], ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4
-// CHECK13-NEXT: [[TMP137:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED51]], align 8
-// CHECK13-NEXT: [[TMP138:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK13-NEXT: [[TMP149:%.*]] = load i32, ptr [[M]], align 4
+// CHECK13-NEXT: store i32 [[TMP149]], ptr [[DOTCAPTURE_EXPR_49]], align 4
+// CHECK13-NEXT: [[TMP150:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP150]], ptr [[N_CASTED50]], align 4
+// CHECK13-NEXT: [[TMP151:%.*]] = load i64, ptr [[N_CASTED50]], align 8
+// CHECK13-NEXT: [[TMP152:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_49]], align 4
+// CHECK13-NEXT: store i32 [[TMP152]], ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4
+// CHECK13-NEXT: [[TMP153:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED51]], align 8
+// CHECK13-NEXT: [[TMP154:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK13-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES55]], ptr align 8 @.offload_sizes.16, i64 32, i1 false)
-// CHECK13-NEXT: [[TMP139:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP135]], ptr [[TMP139]], align 8
-// CHECK13-NEXT: [[TMP140:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP135]], ptr [[TMP140]], align 8
-// CHECK13-NEXT: [[TMP141:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP141]], align 8
-// CHECK13-NEXT: [[TMP142:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP142]], align 8
-// CHECK13-NEXT: [[TMP143:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP143]], align 8
-// CHECK13-NEXT: [[TMP144:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 1
-// CHECK13-NEXT: store ptr null, ptr [[TMP144]], align 8
-// CHECK13-NEXT: [[TMP145:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP145]], align 8
-// CHECK13-NEXT: [[TMP146:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP146]], align 8
-// CHECK13-NEXT: [[TMP147:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 2
-// CHECK13-NEXT: store i64 [[TMP138]], ptr [[TMP147]], align 8
-// CHECK13-NEXT: [[TMP148:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 2
-// CHECK13-NEXT: store ptr null, ptr [[TMP148]], align 8
-// CHECK13-NEXT: [[TMP149:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 3
-// CHECK13-NEXT: store i64 [[TMP137]], ptr [[TMP149]], align 8
-// CHECK13-NEXT: [[TMP150:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 3
-// CHECK13-NEXT: store i64 [[TMP137]], ptr [[TMP150]], align 8
-// CHECK13-NEXT: [[TMP151:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 3
-// CHECK13-NEXT: store ptr null, ptr [[TMP151]], align 8
-// CHECK13-NEXT: [[TMP152:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP153:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP154:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP155:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP155]], ptr [[DOTCAPTURE_EXPR_57]], align 4
-// CHECK13-NEXT: [[TMP156:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_57]], align 4
-// CHECK13-NEXT: [[SUB59:%.*]] = sub nsw i32 [[TMP156]], 0
+// CHECK13-NEXT: [[TMP155:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP151]], ptr [[TMP155]], align 8
+// CHECK13-NEXT: [[TMP156:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP151]], ptr [[TMP156]], align 8
+// CHECK13-NEXT: [[TMP157:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP157]], align 8
+// CHECK13-NEXT: [[TMP158:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP158]], align 8
+// CHECK13-NEXT: [[TMP159:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP159]], align 8
+// CHECK13-NEXT: [[TMP160:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 1
+// CHECK13-NEXT: store ptr null, ptr [[TMP160]], align 8
+// CHECK13-NEXT: [[TMP161:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP161]], align 8
+// CHECK13-NEXT: [[TMP162:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP162]], align 8
+// CHECK13-NEXT: [[TMP163:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 2
+// CHECK13-NEXT: store i64 [[TMP154]], ptr [[TMP163]], align 8
+// CHECK13-NEXT: [[TMP164:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 2
+// CHECK13-NEXT: store ptr null, ptr [[TMP164]], align 8
+// CHECK13-NEXT: [[TMP165:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 3
+// CHECK13-NEXT: store i64 [[TMP153]], ptr [[TMP165]], align 8
+// CHECK13-NEXT: [[TMP166:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 3
+// CHECK13-NEXT: store i64 [[TMP153]], ptr [[TMP166]], align 8
+// CHECK13-NEXT: [[TMP167:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 3
+// CHECK13-NEXT: store ptr null, ptr [[TMP167]], align 8
+// CHECK13-NEXT: [[TMP168:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP169:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP170:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP171:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP171]], ptr [[DOTCAPTURE_EXPR_57]], align 4
+// CHECK13-NEXT: [[TMP172:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_57]], align 4
+// CHECK13-NEXT: [[SUB59:%.*]] = sub nsw i32 [[TMP172]], 0
// CHECK13-NEXT: [[DIV60:%.*]] = sdiv i32 [[SUB59]], 1
// CHECK13-NEXT: [[SUB61:%.*]] = sub nsw i32 [[DIV60]], 1
// CHECK13-NEXT: store i32 [[SUB61]], ptr [[DOTCAPTURE_EXPR_58]], align 4
-// CHECK13-NEXT: [[TMP157:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_58]], align 4
-// CHECK13-NEXT: [[ADD62:%.*]] = add nsw i32 [[TMP157]], 1
-// CHECK13-NEXT: [[TMP158:%.*]] = zext i32 [[ADD62]] to i64
+// CHECK13-NEXT: [[TMP173:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_58]], align 4
+// CHECK13-NEXT: [[ADD62:%.*]] = add nsw i32 [[TMP173]], 1
+// CHECK13-NEXT: [[TMP174:%.*]] = zext i32 [[ADD62]] to i64
// CHECK13-NEXT: [[KERNEL_ARGS63:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP159:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP159]], align 4
-// CHECK13-NEXT: [[TMP160:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 1
-// CHECK13-NEXT: store i32 4, ptr [[TMP160]], align 4
-// CHECK13-NEXT: [[TMP161:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP152]], ptr [[TMP161]], align 8
-// CHECK13-NEXT: [[TMP162:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP153]], ptr [[TMP162]], align 8
-// CHECK13-NEXT: [[TMP163:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 4
-// CHECK13-NEXT: store ptr [[TMP154]], ptr [[TMP163]], align 8
-// CHECK13-NEXT: [[TMP164:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP164]], align 8
-// CHECK13-NEXT: [[TMP165:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP165]], align 8
-// CHECK13-NEXT: [[TMP166:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP166]], align 8
-// CHECK13-NEXT: [[TMP167:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 8
-// CHECK13-NEXT: store i64 [[TMP158]], ptr [[TMP167]], align 8
-// CHECK13-NEXT: [[TMP168:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.region_id, ptr [[KERNEL_ARGS63]])
-// CHECK13-NEXT: [[TMP169:%.*]] = icmp ne i32 [[TMP168]], 0
-// CHECK13-NEXT: br i1 [[TMP169]], label [[OMP_OFFLOAD_FAILED64:%.*]], label [[OMP_OFFLOAD_CONT65:%.*]]
+// CHECK13-NEXT: [[TMP175:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP175]], align 4
+// CHECK13-NEXT: [[TMP176:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 1
+// CHECK13-NEXT: store i32 4, ptr [[TMP176]], align 4
+// CHECK13-NEXT: [[TMP177:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP168]], ptr [[TMP177]], align 8
+// CHECK13-NEXT: [[TMP178:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP169]], ptr [[TMP178]], align 8
+// CHECK13-NEXT: [[TMP179:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 4
+// CHECK13-NEXT: store ptr [[TMP170]], ptr [[TMP179]], align 8
+// CHECK13-NEXT: [[TMP180:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP180]], align 8
+// CHECK13-NEXT: [[TMP181:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP181]], align 8
+// CHECK13-NEXT: [[TMP182:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP182]], align 8
+// CHECK13-NEXT: [[TMP183:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 8
+// CHECK13-NEXT: store i64 [[TMP174]], ptr [[TMP183]], align 8
+// CHECK13-NEXT: [[TMP184:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP184]], align 8
+// CHECK13-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP185]], align 4
+// CHECK13-NEXT: [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP186]], align 4
+// CHECK13-NEXT: [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP187]], align 4
+// CHECK13-NEXT: [[TMP188:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.region_id, ptr [[KERNEL_ARGS63]])
+// CHECK13-NEXT: [[TMP189:%.*]] = icmp ne i32 [[TMP188]], 0
+// CHECK13-NEXT: br i1 [[TMP189]], label [[OMP_OFFLOAD_FAILED64:%.*]], label [[OMP_OFFLOAD_CONT65:%.*]]
// CHECK13: omp_offload.failed64:
-// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155(i64 [[TMP135]], i64 [[TMP1]], ptr [[VLA]], i64 [[TMP137]]) #[[ATTR3]]
+// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155(i64 [[TMP151]], i64 [[TMP1]], ptr [[VLA]], i64 [[TMP153]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT65]]
// CHECK13: omp_offload.cont65:
-// CHECK13-NEXT: [[TMP170:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK13-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP170]])
+// CHECK13-NEXT: [[TMP190:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK13-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP190]])
// CHECK13-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK13-NEXT: [[TMP171:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK13-NEXT: call void @llvm.stackrestore(ptr [[TMP171]])
-// CHECK13-NEXT: [[TMP172:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK13-NEXT: ret i32 [[TMP172]]
+// CHECK13-NEXT: [[TMP191:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK13-NEXT: call void @llvm.stackrestore(ptr [[TMP191]])
+// CHECK13-NEXT: [[TMP192:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK13-NEXT: ret i32 [[TMP192]]
//
//
// CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139
@@ -5705,7 +5905,7 @@ int main (int argc, char **argv) {
// CHECK13-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK13-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK13-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK13-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK13-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK13-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK13-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -5722,169 +5922,209 @@ int main (int argc, char **argv) {
// CHECK13-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK13-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK13-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK13-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.region_id, ptr [[KERNEL_ARGS]])
-// CHECK13-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK13-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK13-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK13-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK13-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK13-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK13-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.region_id, ptr [[KERNEL_ARGS]])
+// CHECK13-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK13-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK13: omp_offload.failed:
// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112(ptr [[A]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK13: omp_offload.cont:
-// CHECK13-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP16]], align 8
-// CHECK13-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP17]], align 8
-// CHECK13-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK13-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP20]], align 8
+// CHECK13-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP21]], align 8
+// CHECK13-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK13-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK13-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK13-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK13-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK13-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK13-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK13-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK13-NEXT: store ptr @.offload_sizes.24, ptr [[TMP25]], align 8
-// CHECK13-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP26]], align 8
-// CHECK13-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK13-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK13-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK13-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK13-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK13-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK13-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK13-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK13-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK13-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK13-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK13-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK13-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK13-NEXT: store ptr @.offload_sizes.24, ptr [[TMP29]], align 8
+// CHECK13-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP30]], align 8
+// CHECK13-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK13-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK13-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK13-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK13-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK13-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK13-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK13-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK13-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK13-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK13-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK13: omp_offload.failed6:
// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116(ptr [[A]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK13: omp_offload.cont7:
-// CHECK13-NEXT: [[TMP32:%.*]] = load i32, ptr [[M]], align 4
-// CHECK13-NEXT: store i32 [[TMP32]], ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK13-NEXT: [[TMP33:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK13-NEXT: store i32 [[TMP33]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK13-NEXT: [[TMP34:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK13-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP35]], align 8
-// CHECK13-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP36]], align 8
-// CHECK13-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP37]], align 8
-// CHECK13-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP34]], ptr [[TMP38]], align 8
-// CHECK13-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP34]], ptr [[TMP39]], align 8
-// CHECK13-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
-// CHECK13-NEXT: store ptr null, ptr [[TMP40]], align 8
-// CHECK13-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP40:%.*]] = load i32, ptr [[M]], align 4
+// CHECK13-NEXT: store i32 [[TMP40]], ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK13-NEXT: [[TMP41:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK13-NEXT: store i32 [[TMP41]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK13-NEXT: [[TMP42:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK13-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP43]], align 8
+// CHECK13-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP44]], align 8
+// CHECK13-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP45]], align 8
+// CHECK13-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP42]], ptr [[TMP46]], align 8
+// CHECK13-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP42]], ptr [[TMP47]], align 8
+// CHECK13-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
+// CHECK13-NEXT: store ptr null, ptr [[TMP48]], align 8
+// CHECK13-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK13-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP43]], align 4
-// CHECK13-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK13-NEXT: store i32 2, ptr [[TMP44]], align 4
-// CHECK13-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP41]], ptr [[TMP45]], align 8
-// CHECK13-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP42]], ptr [[TMP46]], align 8
-// CHECK13-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK13-NEXT: store ptr @.offload_sizes.28, ptr [[TMP47]], align 8
-// CHECK13-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP48]], align 8
-// CHECK13-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP49]], align 8
-// CHECK13-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP50]], align 8
-// CHECK13-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK13-NEXT: store i64 10, ptr [[TMP51]], align 8
-// CHECK13-NEXT: [[TMP52:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK13-NEXT: [[TMP53:%.*]] = icmp ne i32 [[TMP52]], 0
-// CHECK13-NEXT: br i1 [[TMP53]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK13-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP51]], align 4
+// CHECK13-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK13-NEXT: store i32 2, ptr [[TMP52]], align 4
+// CHECK13-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 8
+// CHECK13-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP50]], ptr [[TMP54]], align 8
+// CHECK13-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK13-NEXT: store ptr @.offload_sizes.28, ptr [[TMP55]], align 8
+// CHECK13-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP56]], align 8
+// CHECK13-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP57]], align 8
+// CHECK13-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP58]], align 8
+// CHECK13-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK13-NEXT: store i64 10, ptr [[TMP59]], align 8
+// CHECK13-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP60]], align 8
+// CHECK13-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4
+// CHECK13-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP62]], align 4
+// CHECK13-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP63]], align 4
+// CHECK13-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK13-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
+// CHECK13-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK13: omp_offload.failed13:
-// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120(ptr [[A]], i64 [[TMP34]]) #[[ATTR3]]
+// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120(ptr [[A]], i64 [[TMP42]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT14]]
// CHECK13: omp_offload.cont14:
-// CHECK13-NEXT: [[TMP54:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP54]], align 8
-// CHECK13-NEXT: [[TMP55:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP55]], align 8
-// CHECK13-NEXT: [[TMP56:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP56]], align 8
-// CHECK13-NEXT: [[TMP57:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP58:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP66]], align 8
+// CHECK13-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP67]], align 8
+// CHECK13-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP68]], align 8
+// CHECK13-NEXT: [[TMP69:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP70:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
// CHECK13-NEXT: [[KERNEL_ARGS19:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP59]], align 4
-// CHECK13-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
-// CHECK13-NEXT: store i32 1, ptr [[TMP60]], align 4
-// CHECK13-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP57]], ptr [[TMP61]], align 8
-// CHECK13-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP58]], ptr [[TMP62]], align 8
-// CHECK13-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
-// CHECK13-NEXT: store ptr @.offload_sizes.32, ptr [[TMP63]], align 8
-// CHECK13-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP64]], align 8
-// CHECK13-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP65]], align 8
-// CHECK13-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP66]], align 8
-// CHECK13-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
-// CHECK13-NEXT: store i64 10, ptr [[TMP67]], align 8
-// CHECK13-NEXT: [[TMP68:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.region_id, ptr [[KERNEL_ARGS19]])
-// CHECK13-NEXT: [[TMP69:%.*]] = icmp ne i32 [[TMP68]], 0
-// CHECK13-NEXT: br i1 [[TMP69]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
+// CHECK13-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP71]], align 4
+// CHECK13-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
+// CHECK13-NEXT: store i32 1, ptr [[TMP72]], align 4
+// CHECK13-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP69]], ptr [[TMP73]], align 8
+// CHECK13-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP70]], ptr [[TMP74]], align 8
+// CHECK13-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
+// CHECK13-NEXT: store ptr @.offload_sizes.32, ptr [[TMP75]], align 8
+// CHECK13-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP76]], align 8
+// CHECK13-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP77]], align 8
+// CHECK13-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP78]], align 8
+// CHECK13-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
+// CHECK13-NEXT: store i64 10, ptr [[TMP79]], align 8
+// CHECK13-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP80]], align 8
+// CHECK13-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP81]], align 4
+// CHECK13-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP82]], align 4
+// CHECK13-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP83]], align 4
+// CHECK13-NEXT: [[TMP84:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.region_id, ptr [[KERNEL_ARGS19]])
+// CHECK13-NEXT: [[TMP85:%.*]] = icmp ne i32 [[TMP84]], 0
+// CHECK13-NEXT: br i1 [[TMP85]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
// CHECK13: omp_offload.failed20:
// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124(ptr [[A]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT21]]
// CHECK13: omp_offload.cont21:
-// CHECK13-NEXT: [[TMP70:%.*]] = load i32, ptr [[M]], align 4
-// CHECK13-NEXT: store i32 [[TMP70]], ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK13-NEXT: [[TMP71:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK13-NEXT: store i32 [[TMP71]], ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4
-// CHECK13-NEXT: [[TMP72:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED23]], align 8
-// CHECK13-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP73]], align 8
-// CHECK13-NEXT: [[TMP74:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP74]], align 8
-// CHECK13-NEXT: [[TMP75:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP75]], align 8
-// CHECK13-NEXT: [[TMP76:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP72]], ptr [[TMP76]], align 8
-// CHECK13-NEXT: [[TMP77:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP72]], ptr [[TMP77]], align 8
-// CHECK13-NEXT: [[TMP78:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 1
-// CHECK13-NEXT: store ptr null, ptr [[TMP78]], align 8
-// CHECK13-NEXT: [[TMP79:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP80:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP86:%.*]] = load i32, ptr [[M]], align 4
+// CHECK13-NEXT: store i32 [[TMP86]], ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK13-NEXT: [[TMP87:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK13-NEXT: store i32 [[TMP87]], ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4
+// CHECK13-NEXT: [[TMP88:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED23]], align 8
+// CHECK13-NEXT: [[TMP89:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP89]], align 8
+// CHECK13-NEXT: [[TMP90:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP90]], align 8
+// CHECK13-NEXT: [[TMP91:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP91]], align 8
+// CHECK13-NEXT: [[TMP92:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP88]], ptr [[TMP92]], align 8
+// CHECK13-NEXT: [[TMP93:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP88]], ptr [[TMP93]], align 8
+// CHECK13-NEXT: [[TMP94:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 1
+// CHECK13-NEXT: store ptr null, ptr [[TMP94]], align 8
+// CHECK13-NEXT: [[TMP95:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP96:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
// CHECK13-NEXT: [[KERNEL_ARGS28:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP81]], align 4
-// CHECK13-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
-// CHECK13-NEXT: store i32 2, ptr [[TMP82]], align 4
-// CHECK13-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP79]], ptr [[TMP83]], align 8
-// CHECK13-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP80]], ptr [[TMP84]], align 8
-// CHECK13-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
-// CHECK13-NEXT: store ptr @.offload_sizes.36, ptr [[TMP85]], align 8
-// CHECK13-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP86]], align 8
-// CHECK13-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP87]], align 8
-// CHECK13-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP88]], align 8
-// CHECK13-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
-// CHECK13-NEXT: store i64 10, ptr [[TMP89]], align 8
-// CHECK13-NEXT: [[TMP90:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.region_id, ptr [[KERNEL_ARGS28]])
-// CHECK13-NEXT: [[TMP91:%.*]] = icmp ne i32 [[TMP90]], 0
-// CHECK13-NEXT: br i1 [[TMP91]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
+// CHECK13-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP97]], align 4
+// CHECK13-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
+// CHECK13-NEXT: store i32 2, ptr [[TMP98]], align 4
+// CHECK13-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP95]], ptr [[TMP99]], align 8
+// CHECK13-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP96]], ptr [[TMP100]], align 8
+// CHECK13-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
+// CHECK13-NEXT: store ptr @.offload_sizes.36, ptr [[TMP101]], align 8
+// CHECK13-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP102]], align 8
+// CHECK13-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP103]], align 8
+// CHECK13-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP104]], align 8
+// CHECK13-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
+// CHECK13-NEXT: store i64 10, ptr [[TMP105]], align 8
+// CHECK13-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP106]], align 8
+// CHECK13-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP107]], align 4
+// CHECK13-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP108]], align 4
+// CHECK13-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP109]], align 4
+// CHECK13-NEXT: [[TMP110:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.region_id, ptr [[KERNEL_ARGS28]])
+// CHECK13-NEXT: [[TMP111:%.*]] = icmp ne i32 [[TMP110]], 0
+// CHECK13-NEXT: br i1 [[TMP111]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
// CHECK13: omp_offload.failed29:
-// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128(ptr [[A]], i64 [[TMP72]]) #[[ATTR3]]
+// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128(ptr [[A]], i64 [[TMP88]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT30]]
// CHECK13: omp_offload.cont30:
// CHECK13-NEXT: ret i32 0
@@ -6799,7 +7039,7 @@ int main (int argc, char **argv) {
// CHECK15-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK15-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK15-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK15-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK15-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK15-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK15-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -6816,302 +7056,342 @@ int main (int argc, char **argv) {
// CHECK15-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK15-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK15-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK15-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.region_id, ptr [[KERNEL_ARGS]])
-// CHECK15-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK15-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK15-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK15-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK15-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK15-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK15-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.region_id, ptr [[KERNEL_ARGS]])
+// CHECK15-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK15-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK15: omp_offload.failed:
// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139(i32 [[TMP3]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK15: omp_offload.cont:
-// CHECK15-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK15-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK15-NEXT: [[TMP35:%.*]] = load i32, ptr [[N_CASTED3]], align 4
-// CHECK15-NEXT: [[TMP36:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK15-NEXT: [[TMP37:%.*]] = sext i32 [[TMP36]] to i64
+// CHECK15-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK15-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK15-NEXT: [[TMP39:%.*]] = load i32, ptr [[N_CASTED3]], align 4
+// CHECK15-NEXT: [[TMP40:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK15-NEXT: [[TMP41:%.*]] = sext i32 [[TMP40]] to i64
// CHECK15-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES7]], ptr align 4 @.offload_sizes.4, i32 24, i1 false)
-// CHECK15-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK15-NEXT: store i32 [[TMP35]], ptr [[TMP38]], align 4
-// CHECK15-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK15-NEXT: store i32 [[TMP35]], ptr [[TMP39]], align 4
-// CHECK15-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
-// CHECK15-NEXT: store ptr null, ptr [[TMP40]], align 4
-// CHECK15-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP41]], align 4
-// CHECK15-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP42]], align 4
-// CHECK15-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
-// CHECK15-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK15-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 4
-// CHECK15-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP45]], align 4
-// CHECK15-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK15-NEXT: store i64 [[TMP37]], ptr [[TMP46]], align 4
-// CHECK15-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK15-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK15-NEXT: store i32 [[TMP39]], ptr [[TMP42]], align 4
+// CHECK15-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK15-NEXT: store i32 [[TMP39]], ptr [[TMP43]], align 4
+// CHECK15-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
+// CHECK15-NEXT: store ptr null, ptr [[TMP44]], align 4
+// CHECK15-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP45]], align 4
+// CHECK15-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP46]], align 4
+// CHECK15-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
// CHECK15-NEXT: store ptr null, ptr [[TMP47]], align 4
-// CHECK15-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP51:%.*]] = load i32, ptr [[N]], align 4
-// CHECK15-NEXT: store i32 [[TMP51]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK15-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK15-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP52]], 0
+// CHECK15-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 4
+// CHECK15-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP49]], align 4
+// CHECK15-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK15-NEXT: store i64 [[TMP41]], ptr [[TMP50]], align 4
+// CHECK15-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK15-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK15-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP54:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP55:%.*]] = load i32, ptr [[N]], align 4
+// CHECK15-NEXT: store i32 [[TMP55]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK15-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK15-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP56]], 0
// CHECK15-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK15-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK15-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK15-NEXT: [[TMP53:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK15-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP53]], 1
-// CHECK15-NEXT: [[TMP54:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK15-NEXT: [[TMP57:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK15-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP57]], 1
+// CHECK15-NEXT: [[TMP58:%.*]] = zext i32 [[ADD14]] to i64
// CHECK15-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK15-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP55]], align 4
-// CHECK15-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK15-NEXT: store i32 3, ptr [[TMP56]], align 4
-// CHECK15-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 4
-// CHECK15-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK15-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 4
-// CHECK15-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK15-NEXT: store ptr [[TMP50]], ptr [[TMP59]], align 4
-// CHECK15-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK15-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP60]], align 4
-// CHECK15-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK15-NEXT: store ptr null, ptr [[TMP61]], align 4
-// CHECK15-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK15-NEXT: store ptr null, ptr [[TMP62]], align 4
-// CHECK15-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK15-NEXT: store i64 [[TMP54]], ptr [[TMP63]], align 8
-// CHECK15-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK15-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
-// CHECK15-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK15-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK15-NEXT: store i32 2, ptr [[TMP59]], align 4
+// CHECK15-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK15-NEXT: store i32 3, ptr [[TMP60]], align 4
+// CHECK15-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 4
+// CHECK15-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK15-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 4
+// CHECK15-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK15-NEXT: store ptr [[TMP54]], ptr [[TMP63]], align 4
+// CHECK15-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK15-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP64]], align 4
+// CHECK15-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK15-NEXT: store ptr null, ptr [[TMP65]], align 4
+// CHECK15-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK15-NEXT: store ptr null, ptr [[TMP66]], align 4
+// CHECK15-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK15-NEXT: store i64 [[TMP58]], ptr [[TMP67]], align 8
+// CHECK15-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP68]], align 8
+// CHECK15-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK15-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP70]], align 4
+// CHECK15-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP71]], align 4
+// CHECK15-NEXT: [[TMP72:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK15-NEXT: [[TMP73:%.*]] = icmp ne i32 [[TMP72]], 0
+// CHECK15-NEXT: br i1 [[TMP73]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK15: omp_offload.failed16:
-// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143(i32 [[TMP35]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143(i32 [[TMP39]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK15: omp_offload.cont17:
-// CHECK15-NEXT: [[TMP66:%.*]] = load i32, ptr [[M]], align 4
-// CHECK15-NEXT: store i32 [[TMP66]], ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK15-NEXT: [[TMP67:%.*]] = load i32, ptr [[N]], align 4
-// CHECK15-NEXT: store i32 [[TMP67]], ptr [[N_CASTED19]], align 4
-// CHECK15-NEXT: [[TMP68:%.*]] = load i32, ptr [[N_CASTED19]], align 4
-// CHECK15-NEXT: [[TMP69:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK15-NEXT: store i32 [[TMP69]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK15-NEXT: [[TMP70:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK15-NEXT: [[TMP71:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK15-NEXT: [[TMP72:%.*]] = sext i32 [[TMP71]] to i64
+// CHECK15-NEXT: [[TMP74:%.*]] = load i32, ptr [[M]], align 4
+// CHECK15-NEXT: store i32 [[TMP74]], ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK15-NEXT: [[TMP75:%.*]] = load i32, ptr [[N]], align 4
+// CHECK15-NEXT: store i32 [[TMP75]], ptr [[N_CASTED19]], align 4
+// CHECK15-NEXT: [[TMP76:%.*]] = load i32, ptr [[N_CASTED19]], align 4
+// CHECK15-NEXT: [[TMP77:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK15-NEXT: store i32 [[TMP77]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK15-NEXT: [[TMP78:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK15-NEXT: [[TMP79:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK15-NEXT: [[TMP80:%.*]] = sext i32 [[TMP79]] to i64
// CHECK15-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES23]], ptr align 4 @.offload_sizes.8, i32 32, i1 false)
-// CHECK15-NEXT: [[TMP73:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK15-NEXT: store i32 [[TMP68]], ptr [[TMP73]], align 4
-// CHECK15-NEXT: [[TMP74:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK15-NEXT: store i32 [[TMP68]], ptr [[TMP74]], align 4
-// CHECK15-NEXT: [[TMP75:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 0
-// CHECK15-NEXT: store ptr null, ptr [[TMP75]], align 4
-// CHECK15-NEXT: [[TMP76:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP76]], align 4
-// CHECK15-NEXT: [[TMP77:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP77]], align 4
-// CHECK15-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 1
-// CHECK15-NEXT: store ptr null, ptr [[TMP78]], align 4
-// CHECK15-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP79]], align 4
-// CHECK15-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP80]], align 4
-// CHECK15-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
-// CHECK15-NEXT: store i64 [[TMP72]], ptr [[TMP81]], align 4
-// CHECK15-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 2
-// CHECK15-NEXT: store ptr null, ptr [[TMP82]], align 4
-// CHECK15-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
-// CHECK15-NEXT: store i32 [[TMP70]], ptr [[TMP83]], align 4
-// CHECK15-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
-// CHECK15-NEXT: store i32 [[TMP70]], ptr [[TMP84]], align 4
-// CHECK15-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 3
-// CHECK15-NEXT: store ptr null, ptr [[TMP85]], align 4
-// CHECK15-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP89:%.*]] = load i32, ptr [[N]], align 4
-// CHECK15-NEXT: store i32 [[TMP89]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK15-NEXT: [[TMP90:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK15-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP90]], 0
+// CHECK15-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK15-NEXT: store i32 [[TMP76]], ptr [[TMP81]], align 4
+// CHECK15-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK15-NEXT: store i32 [[TMP76]], ptr [[TMP82]], align 4
+// CHECK15-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 0
+// CHECK15-NEXT: store ptr null, ptr [[TMP83]], align 4
+// CHECK15-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP84]], align 4
+// CHECK15-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP85]], align 4
+// CHECK15-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 1
+// CHECK15-NEXT: store ptr null, ptr [[TMP86]], align 4
+// CHECK15-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP87]], align 4
+// CHECK15-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP88]], align 4
+// CHECK15-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
+// CHECK15-NEXT: store i64 [[TMP80]], ptr [[TMP89]], align 4
+// CHECK15-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 2
+// CHECK15-NEXT: store ptr null, ptr [[TMP90]], align 4
+// CHECK15-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
+// CHECK15-NEXT: store i32 [[TMP78]], ptr [[TMP91]], align 4
+// CHECK15-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
+// CHECK15-NEXT: store i32 [[TMP78]], ptr [[TMP92]], align 4
+// CHECK15-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 3
+// CHECK15-NEXT: store ptr null, ptr [[TMP93]], align 4
+// CHECK15-NEXT: [[TMP94:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP95:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP96:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP97:%.*]] = load i32, ptr [[N]], align 4
+// CHECK15-NEXT: store i32 [[TMP97]], ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK15-NEXT: [[TMP98:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK15-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP98]], 0
// CHECK15-NEXT: [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1
// CHECK15-NEXT: [[SUB29:%.*]] = sub nsw i32 [[DIV28]], 1
// CHECK15-NEXT: store i32 [[SUB29]], ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK15-NEXT: [[TMP91:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK15-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP91]], 1
-// CHECK15-NEXT: [[TMP92:%.*]] = zext i32 [[ADD30]] to i64
+// CHECK15-NEXT: [[TMP99:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
+// CHECK15-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP99]], 1
+// CHECK15-NEXT: [[TMP100:%.*]] = zext i32 [[ADD30]] to i64
// CHECK15-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK15-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP93]], align 4
-// CHECK15-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK15-NEXT: store i32 4, ptr [[TMP94]], align 4
-// CHECK15-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[TMP86]], ptr [[TMP95]], align 4
-// CHECK15-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK15-NEXT: store ptr [[TMP87]], ptr [[TMP96]], align 4
-// CHECK15-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK15-NEXT: store ptr [[TMP88]], ptr [[TMP97]], align 4
-// CHECK15-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK15-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP98]], align 4
-// CHECK15-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK15-NEXT: store ptr null, ptr [[TMP99]], align 4
-// CHECK15-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK15-NEXT: store ptr null, ptr [[TMP100]], align 4
-// CHECK15-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK15-NEXT: store i64 [[TMP92]], ptr [[TMP101]], align 8
-// CHECK15-NEXT: [[TMP102:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK15-NEXT: [[TMP103:%.*]] = icmp ne i32 [[TMP102]], 0
-// CHECK15-NEXT: br i1 [[TMP103]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK15-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK15-NEXT: store i32 2, ptr [[TMP101]], align 4
+// CHECK15-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK15-NEXT: store i32 4, ptr [[TMP102]], align 4
+// CHECK15-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[TMP94]], ptr [[TMP103]], align 4
+// CHECK15-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK15-NEXT: store ptr [[TMP95]], ptr [[TMP104]], align 4
+// CHECK15-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK15-NEXT: store ptr [[TMP96]], ptr [[TMP105]], align 4
+// CHECK15-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK15-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP106]], align 4
+// CHECK15-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK15-NEXT: store ptr null, ptr [[TMP107]], align 4
+// CHECK15-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK15-NEXT: store ptr null, ptr [[TMP108]], align 4
+// CHECK15-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK15-NEXT: store i64 [[TMP100]], ptr [[TMP109]], align 8
+// CHECK15-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP110]], align 8
+// CHECK15-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP111]], align 4
+// CHECK15-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP112]], align 4
+// CHECK15-NEXT: [[TMP113:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP113]], align 4
+// CHECK15-NEXT: [[TMP114:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK15-NEXT: [[TMP115:%.*]] = icmp ne i32 [[TMP114]], 0
+// CHECK15-NEXT: br i1 [[TMP115]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK15: omp_offload.failed32:
-// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147(i32 [[TMP68]], i32 [[TMP0]], ptr [[VLA]], i32 [[TMP70]]) #[[ATTR3]]
+// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147(i32 [[TMP76]], i32 [[TMP0]], ptr [[VLA]], i32 [[TMP78]]) #[[ATTR3]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK15: omp_offload.cont33:
-// CHECK15-NEXT: [[TMP104:%.*]] = load i32, ptr [[N]], align 4
-// CHECK15-NEXT: store i32 [[TMP104]], ptr [[N_CASTED34]], align 4
-// CHECK15-NEXT: [[TMP105:%.*]] = load i32, ptr [[N_CASTED34]], align 4
-// CHECK15-NEXT: [[TMP106:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK15-NEXT: [[TMP107:%.*]] = sext i32 [[TMP106]] to i64
+// CHECK15-NEXT: [[TMP116:%.*]] = load i32, ptr [[N]], align 4
+// CHECK15-NEXT: store i32 [[TMP116]], ptr [[N_CASTED34]], align 4
+// CHECK15-NEXT: [[TMP117:%.*]] = load i32, ptr [[N_CASTED34]], align 4
+// CHECK15-NEXT: [[TMP118:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK15-NEXT: [[TMP119:%.*]] = sext i32 [[TMP118]] to i64
// CHECK15-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES38]], ptr align 4 @.offload_sizes.12, i32 24, i1 false)
-// CHECK15-NEXT: [[TMP108:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
-// CHECK15-NEXT: store i32 [[TMP105]], ptr [[TMP108]], align 4
-// CHECK15-NEXT: [[TMP109:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
-// CHECK15-NEXT: store i32 [[TMP105]], ptr [[TMP109]], align 4
-// CHECK15-NEXT: [[TMP110:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 0
-// CHECK15-NEXT: store ptr null, ptr [[TMP110]], align 4
-// CHECK15-NEXT: [[TMP111:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP111]], align 4
-// CHECK15-NEXT: [[TMP112:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP112]], align 4
-// CHECK15-NEXT: [[TMP113:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 1
-// CHECK15-NEXT: store ptr null, ptr [[TMP113]], align 4
-// CHECK15-NEXT: [[TMP114:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP114]], align 4
-// CHECK15-NEXT: [[TMP115:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP115]], align 4
-// CHECK15-NEXT: [[TMP116:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 2
-// CHECK15-NEXT: store i64 [[TMP107]], ptr [[TMP116]], align 4
-// CHECK15-NEXT: [[TMP117:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 2
-// CHECK15-NEXT: store ptr null, ptr [[TMP117]], align 4
-// CHECK15-NEXT: [[TMP118:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP119:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP120:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP121:%.*]] = load i32, ptr [[N]], align 4
-// CHECK15-NEXT: store i32 [[TMP121]], ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK15-NEXT: [[TMP122:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK15-NEXT: [[SUB42:%.*]] = sub nsw i32 [[TMP122]], 0
+// CHECK15-NEXT: [[TMP120:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
+// CHECK15-NEXT: store i32 [[TMP117]], ptr [[TMP120]], align 4
+// CHECK15-NEXT: [[TMP121:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
+// CHECK15-NEXT: store i32 [[TMP117]], ptr [[TMP121]], align 4
+// CHECK15-NEXT: [[TMP122:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 0
+// CHECK15-NEXT: store ptr null, ptr [[TMP122]], align 4
+// CHECK15-NEXT: [[TMP123:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP123]], align 4
+// CHECK15-NEXT: [[TMP124:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP124]], align 4
+// CHECK15-NEXT: [[TMP125:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 1
+// CHECK15-NEXT: store ptr null, ptr [[TMP125]], align 4
+// CHECK15-NEXT: [[TMP126:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP126]], align 4
+// CHECK15-NEXT: [[TMP127:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP127]], align 4
+// CHECK15-NEXT: [[TMP128:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 2
+// CHECK15-NEXT: store i64 [[TMP119]], ptr [[TMP128]], align 4
+// CHECK15-NEXT: [[TMP129:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 2
+// CHECK15-NEXT: store ptr null, ptr [[TMP129]], align 4
+// CHECK15-NEXT: [[TMP130:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP131:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP132:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP133:%.*]] = load i32, ptr [[N]], align 4
+// CHECK15-NEXT: store i32 [[TMP133]], ptr [[DOTCAPTURE_EXPR_40]], align 4
+// CHECK15-NEXT: [[TMP134:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
+// CHECK15-NEXT: [[SUB42:%.*]] = sub nsw i32 [[TMP134]], 0
// CHECK15-NEXT: [[DIV43:%.*]] = sdiv i32 [[SUB42]], 1
// CHECK15-NEXT: [[SUB44:%.*]] = sub nsw i32 [[DIV43]], 1
// CHECK15-NEXT: store i32 [[SUB44]], ptr [[DOTCAPTURE_EXPR_41]], align 4
-// CHECK15-NEXT: [[TMP123:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_41]], align 4
-// CHECK15-NEXT: [[ADD45:%.*]] = add nsw i32 [[TMP123]], 1
-// CHECK15-NEXT: [[TMP124:%.*]] = zext i32 [[ADD45]] to i64
+// CHECK15-NEXT: [[TMP135:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_41]], align 4
+// CHECK15-NEXT: [[ADD45:%.*]] = add nsw i32 [[TMP135]], 1
+// CHECK15-NEXT: [[TMP136:%.*]] = zext i32 [[ADD45]] to i64
// CHECK15-NEXT: [[KERNEL_ARGS46:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK15-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP125]], align 4
-// CHECK15-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 1
-// CHECK15-NEXT: store i32 3, ptr [[TMP126]], align 4
-// CHECK15-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[TMP118]], ptr [[TMP127]], align 4
-// CHECK15-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 3
-// CHECK15-NEXT: store ptr [[TMP119]], ptr [[TMP128]], align 4
-// CHECK15-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 4
-// CHECK15-NEXT: store ptr [[TMP120]], ptr [[TMP129]], align 4
-// CHECK15-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 5
-// CHECK15-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP130]], align 4
-// CHECK15-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 6
-// CHECK15-NEXT: store ptr null, ptr [[TMP131]], align 4
-// CHECK15-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 7
-// CHECK15-NEXT: store ptr null, ptr [[TMP132]], align 4
-// CHECK15-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 8
-// CHECK15-NEXT: store i64 [[TMP124]], ptr [[TMP133]], align 8
-// CHECK15-NEXT: [[TMP134:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.region_id, ptr [[KERNEL_ARGS46]])
-// CHECK15-NEXT: [[TMP135:%.*]] = icmp ne i32 [[TMP134]], 0
-// CHECK15-NEXT: br i1 [[TMP135]], label [[OMP_OFFLOAD_FAILED47:%.*]], label [[OMP_OFFLOAD_CONT48:%.*]]
+// CHECK15-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 0
+// CHECK15-NEXT: store i32 2, ptr [[TMP137]], align 4
+// CHECK15-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 1
+// CHECK15-NEXT: store i32 3, ptr [[TMP138]], align 4
+// CHECK15-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[TMP130]], ptr [[TMP139]], align 4
+// CHECK15-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 3
+// CHECK15-NEXT: store ptr [[TMP131]], ptr [[TMP140]], align 4
+// CHECK15-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 4
+// CHECK15-NEXT: store ptr [[TMP132]], ptr [[TMP141]], align 4
+// CHECK15-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 5
+// CHECK15-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP142]], align 4
+// CHECK15-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 6
+// CHECK15-NEXT: store ptr null, ptr [[TMP143]], align 4
+// CHECK15-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 7
+// CHECK15-NEXT: store ptr null, ptr [[TMP144]], align 4
+// CHECK15-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 8
+// CHECK15-NEXT: store i64 [[TMP136]], ptr [[TMP145]], align 8
+// CHECK15-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP146]], align 8
+// CHECK15-NEXT: [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP147]], align 4
+// CHECK15-NEXT: [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP148]], align 4
+// CHECK15-NEXT: [[TMP149:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP149]], align 4
+// CHECK15-NEXT: [[TMP150:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.region_id, ptr [[KERNEL_ARGS46]])
+// CHECK15-NEXT: [[TMP151:%.*]] = icmp ne i32 [[TMP150]], 0
+// CHECK15-NEXT: br i1 [[TMP151]], label [[OMP_OFFLOAD_FAILED47:%.*]], label [[OMP_OFFLOAD_CONT48:%.*]]
// CHECK15: omp_offload.failed47:
-// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151(i32 [[TMP105]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151(i32 [[TMP117]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT48]]
// CHECK15: omp_offload.cont48:
-// CHECK15-NEXT: [[TMP136:%.*]] = load i32, ptr [[M]], align 4
-// CHECK15-NEXT: store i32 [[TMP136]], ptr [[DOTCAPTURE_EXPR_49]], align 4
-// CHECK15-NEXT: [[TMP137:%.*]] = load i32, ptr [[N]], align 4
-// CHECK15-NEXT: store i32 [[TMP137]], ptr [[N_CASTED50]], align 4
-// CHECK15-NEXT: [[TMP138:%.*]] = load i32, ptr [[N_CASTED50]], align 4
-// CHECK15-NEXT: [[TMP139:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_49]], align 4
-// CHECK15-NEXT: store i32 [[TMP139]], ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4
-// CHECK15-NEXT: [[TMP140:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4
-// CHECK15-NEXT: [[TMP141:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK15-NEXT: [[TMP142:%.*]] = sext i32 [[TMP141]] to i64
+// CHECK15-NEXT: [[TMP152:%.*]] = load i32, ptr [[M]], align 4
+// CHECK15-NEXT: store i32 [[TMP152]], ptr [[DOTCAPTURE_EXPR_49]], align 4
+// CHECK15-NEXT: [[TMP153:%.*]] = load i32, ptr [[N]], align 4
+// CHECK15-NEXT: store i32 [[TMP153]], ptr [[N_CASTED50]], align 4
+// CHECK15-NEXT: [[TMP154:%.*]] = load i32, ptr [[N_CASTED50]], align 4
+// CHECK15-NEXT: [[TMP155:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_49]], align 4
+// CHECK15-NEXT: store i32 [[TMP155]], ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4
+// CHECK15-NEXT: [[TMP156:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4
+// CHECK15-NEXT: [[TMP157:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK15-NEXT: [[TMP158:%.*]] = sext i32 [[TMP157]] to i64
// CHECK15-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES55]], ptr align 4 @.offload_sizes.16, i32 32, i1 false)
-// CHECK15-NEXT: [[TMP143:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
-// CHECK15-NEXT: store i32 [[TMP138]], ptr [[TMP143]], align 4
-// CHECK15-NEXT: [[TMP144:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
-// CHECK15-NEXT: store i32 [[TMP138]], ptr [[TMP144]], align 4
-// CHECK15-NEXT: [[TMP145:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 0
-// CHECK15-NEXT: store ptr null, ptr [[TMP145]], align 4
-// CHECK15-NEXT: [[TMP146:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP146]], align 4
-// CHECK15-NEXT: [[TMP147:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP147]], align 4
-// CHECK15-NEXT: [[TMP148:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 1
-// CHECK15-NEXT: store ptr null, ptr [[TMP148]], align 4
-// CHECK15-NEXT: [[TMP149:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP149]], align 4
-// CHECK15-NEXT: [[TMP150:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP150]], align 4
-// CHECK15-NEXT: [[TMP151:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 2
-// CHECK15-NEXT: store i64 [[TMP142]], ptr [[TMP151]], align 4
-// CHECK15-NEXT: [[TMP152:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 2
-// CHECK15-NEXT: store ptr null, ptr [[TMP152]], align 4
-// CHECK15-NEXT: [[TMP153:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 3
-// CHECK15-NEXT: store i32 [[TMP140]], ptr [[TMP153]], align 4
-// CHECK15-NEXT: [[TMP154:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 3
-// CHECK15-NEXT: store i32 [[TMP140]], ptr [[TMP154]], align 4
-// CHECK15-NEXT: [[TMP155:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 3
-// CHECK15-NEXT: store ptr null, ptr [[TMP155]], align 4
-// CHECK15-NEXT: [[TMP156:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP157:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP158:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP159:%.*]] = load i32, ptr [[N]], align 4
-// CHECK15-NEXT: store i32 [[TMP159]], ptr [[DOTCAPTURE_EXPR_57]], align 4
-// CHECK15-NEXT: [[TMP160:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_57]], align 4
-// CHECK15-NEXT: [[SUB59:%.*]] = sub nsw i32 [[TMP160]], 0
+// CHECK15-NEXT: [[TMP159:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
+// CHECK15-NEXT: store i32 [[TMP154]], ptr [[TMP159]], align 4
+// CHECK15-NEXT: [[TMP160:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
+// CHECK15-NEXT: store i32 [[TMP154]], ptr [[TMP160]], align 4
+// CHECK15-NEXT: [[TMP161:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 0
+// CHECK15-NEXT: store ptr null, ptr [[TMP161]], align 4
+// CHECK15-NEXT: [[TMP162:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP162]], align 4
+// CHECK15-NEXT: [[TMP163:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP163]], align 4
+// CHECK15-NEXT: [[TMP164:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 1
+// CHECK15-NEXT: store ptr null, ptr [[TMP164]], align 4
+// CHECK15-NEXT: [[TMP165:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP165]], align 4
+// CHECK15-NEXT: [[TMP166:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP166]], align 4
+// CHECK15-NEXT: [[TMP167:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 2
+// CHECK15-NEXT: store i64 [[TMP158]], ptr [[TMP167]], align 4
+// CHECK15-NEXT: [[TMP168:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 2
+// CHECK15-NEXT: store ptr null, ptr [[TMP168]], align 4
+// CHECK15-NEXT: [[TMP169:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 3
+// CHECK15-NEXT: store i32 [[TMP156]], ptr [[TMP169]], align 4
+// CHECK15-NEXT: [[TMP170:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 3
+// CHECK15-NEXT: store i32 [[TMP156]], ptr [[TMP170]], align 4
+// CHECK15-NEXT: [[TMP171:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 3
+// CHECK15-NEXT: store ptr null, ptr [[TMP171]], align 4
+// CHECK15-NEXT: [[TMP172:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP173:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP174:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP175:%.*]] = load i32, ptr [[N]], align 4
+// CHECK15-NEXT: store i32 [[TMP175]], ptr [[DOTCAPTURE_EXPR_57]], align 4
+// CHECK15-NEXT: [[TMP176:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_57]], align 4
+// CHECK15-NEXT: [[SUB59:%.*]] = sub nsw i32 [[TMP176]], 0
// CHECK15-NEXT: [[DIV60:%.*]] = sdiv i32 [[SUB59]], 1
// CHECK15-NEXT: [[SUB61:%.*]] = sub nsw i32 [[DIV60]], 1
// CHECK15-NEXT: store i32 [[SUB61]], ptr [[DOTCAPTURE_EXPR_58]], align 4
-// CHECK15-NEXT: [[TMP161:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_58]], align 4
-// CHECK15-NEXT: [[ADD62:%.*]] = add nsw i32 [[TMP161]], 1
-// CHECK15-NEXT: [[TMP162:%.*]] = zext i32 [[ADD62]] to i64
+// CHECK15-NEXT: [[TMP177:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_58]], align 4
+// CHECK15-NEXT: [[ADD62:%.*]] = add nsw i32 [[TMP177]], 1
+// CHECK15-NEXT: [[TMP178:%.*]] = zext i32 [[ADD62]] to i64
// CHECK15-NEXT: [[KERNEL_ARGS63:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK15-NEXT: [[TMP163:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP163]], align 4
-// CHECK15-NEXT: [[TMP164:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 1
-// CHECK15-NEXT: store i32 4, ptr [[TMP164]], align 4
-// CHECK15-NEXT: [[TMP165:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[TMP156]], ptr [[TMP165]], align 4
-// CHECK15-NEXT: [[TMP166:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 3
-// CHECK15-NEXT: store ptr [[TMP157]], ptr [[TMP166]], align 4
-// CHECK15-NEXT: [[TMP167:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 4
-// CHECK15-NEXT: store ptr [[TMP158]], ptr [[TMP167]], align 4
-// CHECK15-NEXT: [[TMP168:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 5
-// CHECK15-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP168]], align 4
-// CHECK15-NEXT: [[TMP169:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 6
-// CHECK15-NEXT: store ptr null, ptr [[TMP169]], align 4
-// CHECK15-NEXT: [[TMP170:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 7
-// CHECK15-NEXT: store ptr null, ptr [[TMP170]], align 4
-// CHECK15-NEXT: [[TMP171:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 8
-// CHECK15-NEXT: store i64 [[TMP162]], ptr [[TMP171]], align 8
-// CHECK15-NEXT: [[TMP172:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.region_id, ptr [[KERNEL_ARGS63]])
-// CHECK15-NEXT: [[TMP173:%.*]] = icmp ne i32 [[TMP172]], 0
-// CHECK15-NEXT: br i1 [[TMP173]], label [[OMP_OFFLOAD_FAILED64:%.*]], label [[OMP_OFFLOAD_CONT65:%.*]]
+// CHECK15-NEXT: [[TMP179:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 0
+// CHECK15-NEXT: store i32 2, ptr [[TMP179]], align 4
+// CHECK15-NEXT: [[TMP180:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 1
+// CHECK15-NEXT: store i32 4, ptr [[TMP180]], align 4
+// CHECK15-NEXT: [[TMP181:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[TMP172]], ptr [[TMP181]], align 4
+// CHECK15-NEXT: [[TMP182:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 3
+// CHECK15-NEXT: store ptr [[TMP173]], ptr [[TMP182]], align 4
+// CHECK15-NEXT: [[TMP183:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 4
+// CHECK15-NEXT: store ptr [[TMP174]], ptr [[TMP183]], align 4
+// CHECK15-NEXT: [[TMP184:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 5
+// CHECK15-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP184]], align 4
+// CHECK15-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 6
+// CHECK15-NEXT: store ptr null, ptr [[TMP185]], align 4
+// CHECK15-NEXT: [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 7
+// CHECK15-NEXT: store ptr null, ptr [[TMP186]], align 4
+// CHECK15-NEXT: [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 8
+// CHECK15-NEXT: store i64 [[TMP178]], ptr [[TMP187]], align 8
+// CHECK15-NEXT: [[TMP188:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP188]], align 8
+// CHECK15-NEXT: [[TMP189:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP189]], align 4
+// CHECK15-NEXT: [[TMP190:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP190]], align 4
+// CHECK15-NEXT: [[TMP191:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP191]], align 4
+// CHECK15-NEXT: [[TMP192:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.region_id, ptr [[KERNEL_ARGS63]])
+// CHECK15-NEXT: [[TMP193:%.*]] = icmp ne i32 [[TMP192]], 0
+// CHECK15-NEXT: br i1 [[TMP193]], label [[OMP_OFFLOAD_FAILED64:%.*]], label [[OMP_OFFLOAD_CONT65:%.*]]
// CHECK15: omp_offload.failed64:
-// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155(i32 [[TMP138]], i32 [[TMP0]], ptr [[VLA]], i32 [[TMP140]]) #[[ATTR3]]
+// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155(i32 [[TMP154]], i32 [[TMP0]], ptr [[VLA]], i32 [[TMP156]]) #[[ATTR3]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT65]]
// CHECK15: omp_offload.cont65:
-// CHECK15-NEXT: [[TMP174:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK15-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP174]])
+// CHECK15-NEXT: [[TMP194:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK15-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP194]])
// CHECK15-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK15-NEXT: [[TMP175:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK15-NEXT: call void @llvm.stackrestore(ptr [[TMP175]])
-// CHECK15-NEXT: [[TMP176:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK15-NEXT: ret i32 [[TMP176]]
+// CHECK15-NEXT: [[TMP195:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK15-NEXT: call void @llvm.stackrestore(ptr [[TMP195]])
+// CHECK15-NEXT: [[TMP196:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK15-NEXT: ret i32 [[TMP196]]
//
//
// CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139
@@ -8262,7 +8542,7 @@ int main (int argc, char **argv) {
// CHECK15-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK15-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK15-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK15-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK15-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK15-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK15-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -8279,169 +8559,209 @@ int main (int argc, char **argv) {
// CHECK15-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK15-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK15-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK15-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.region_id, ptr [[KERNEL_ARGS]])
-// CHECK15-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK15-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK15-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK15-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK15-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK15-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK15-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.region_id, ptr [[KERNEL_ARGS]])
+// CHECK15-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK15-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK15: omp_offload.failed:
// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112(ptr [[A]]) #[[ATTR3]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK15: omp_offload.cont:
-// CHECK15-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK15-NEXT: store ptr [[A]], ptr [[TMP16]], align 4
-// CHECK15-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK15-NEXT: store ptr [[A]], ptr [[TMP17]], align 4
-// CHECK15-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
-// CHECK15-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK15-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK15-NEXT: store ptr [[A]], ptr [[TMP20]], align 4
+// CHECK15-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK15-NEXT: store ptr [[A]], ptr [[TMP21]], align 4
+// CHECK15-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
+// CHECK15-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK15-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK15-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK15-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK15-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK15-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK15-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK15-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK15-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK15-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK15-NEXT: store ptr @.offload_sizes.24, ptr [[TMP25]], align 4
-// CHECK15-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK15-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP26]], align 4
-// CHECK15-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK15-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK15-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK15-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK15-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK15-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK15-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK15-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK15-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK15-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK15-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK15-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK15-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK15-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK15-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK15-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK15-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK15-NEXT: store ptr @.offload_sizes.24, ptr [[TMP29]], align 4
+// CHECK15-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK15-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP30]], align 4
+// CHECK15-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK15-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK15-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK15-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK15-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK15-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK15-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK15-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK15-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK15-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK15-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK15-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK15-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK15: omp_offload.failed6:
// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116(ptr [[A]]) #[[ATTR3]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK15: omp_offload.cont7:
-// CHECK15-NEXT: [[TMP32:%.*]] = load i32, ptr [[M]], align 4
-// CHECK15-NEXT: store i32 [[TMP32]], ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK15-NEXT: [[TMP33:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK15-NEXT: store i32 [[TMP33]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK15-NEXT: [[TMP34:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK15-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK15-NEXT: store ptr [[A]], ptr [[TMP35]], align 4
-// CHECK15-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK15-NEXT: store ptr [[A]], ptr [[TMP36]], align 4
-// CHECK15-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
-// CHECK15-NEXT: store ptr null, ptr [[TMP37]], align 4
-// CHECK15-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP34]], ptr [[TMP38]], align 4
-// CHECK15-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP34]], ptr [[TMP39]], align 4
-// CHECK15-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1
-// CHECK15-NEXT: store ptr null, ptr [[TMP40]], align 4
-// CHECK15-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP40:%.*]] = load i32, ptr [[M]], align 4
+// CHECK15-NEXT: store i32 [[TMP40]], ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK15-NEXT: [[TMP41:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK15-NEXT: store i32 [[TMP41]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK15-NEXT: [[TMP42:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK15-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK15-NEXT: store ptr [[A]], ptr [[TMP43]], align 4
+// CHECK15-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK15-NEXT: store ptr [[A]], ptr [[TMP44]], align 4
+// CHECK15-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
+// CHECK15-NEXT: store ptr null, ptr [[TMP45]], align 4
+// CHECK15-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP42]], ptr [[TMP46]], align 4
+// CHECK15-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP42]], ptr [[TMP47]], align 4
+// CHECK15-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1
+// CHECK15-NEXT: store ptr null, ptr [[TMP48]], align 4
+// CHECK15-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK15-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK15-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP43]], align 4
-// CHECK15-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK15-NEXT: store i32 2, ptr [[TMP44]], align 4
-// CHECK15-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[TMP41]], ptr [[TMP45]], align 4
-// CHECK15-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK15-NEXT: store ptr [[TMP42]], ptr [[TMP46]], align 4
-// CHECK15-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK15-NEXT: store ptr @.offload_sizes.28, ptr [[TMP47]], align 4
-// CHECK15-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK15-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP48]], align 4
-// CHECK15-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK15-NEXT: store ptr null, ptr [[TMP49]], align 4
-// CHECK15-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK15-NEXT: store ptr null, ptr [[TMP50]], align 4
-// CHECK15-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK15-NEXT: store i64 10, ptr [[TMP51]], align 8
-// CHECK15-NEXT: [[TMP52:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK15-NEXT: [[TMP53:%.*]] = icmp ne i32 [[TMP52]], 0
-// CHECK15-NEXT: br i1 [[TMP53]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK15-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK15-NEXT: store i32 2, ptr [[TMP51]], align 4
+// CHECK15-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK15-NEXT: store i32 2, ptr [[TMP52]], align 4
+// CHECK15-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 4
+// CHECK15-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK15-NEXT: store ptr [[TMP50]], ptr [[TMP54]], align 4
+// CHECK15-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK15-NEXT: store ptr @.offload_sizes.28, ptr [[TMP55]], align 4
+// CHECK15-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK15-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP56]], align 4
+// CHECK15-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK15-NEXT: store ptr null, ptr [[TMP57]], align 4
+// CHECK15-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK15-NEXT: store ptr null, ptr [[TMP58]], align 4
+// CHECK15-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK15-NEXT: store i64 10, ptr [[TMP59]], align 8
+// CHECK15-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP60]], align 8
+// CHECK15-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4
+// CHECK15-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP62]], align 4
+// CHECK15-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP63]], align 4
+// CHECK15-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK15-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
+// CHECK15-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK15: omp_offload.failed13:
-// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120(ptr [[A]], i32 [[TMP34]]) #[[ATTR3]]
+// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120(ptr [[A]], i32 [[TMP42]]) #[[ATTR3]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT14]]
// CHECK15: omp_offload.cont14:
-// CHECK15-NEXT: [[TMP54:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK15-NEXT: store ptr [[A]], ptr [[TMP54]], align 4
-// CHECK15-NEXT: [[TMP55:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
-// CHECK15-NEXT: store ptr [[A]], ptr [[TMP55]], align 4
-// CHECK15-NEXT: [[TMP56:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i32 0, i32 0
-// CHECK15-NEXT: store ptr null, ptr [[TMP56]], align 4
-// CHECK15-NEXT: [[TMP57:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP58:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK15-NEXT: store ptr [[A]], ptr [[TMP66]], align 4
+// CHECK15-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK15-NEXT: store ptr [[A]], ptr [[TMP67]], align 4
+// CHECK15-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i32 0, i32 0
+// CHECK15-NEXT: store ptr null, ptr [[TMP68]], align 4
+// CHECK15-NEXT: [[TMP69:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP70:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
// CHECK15-NEXT: [[KERNEL_ARGS19:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK15-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP59]], align 4
-// CHECK15-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
-// CHECK15-NEXT: store i32 1, ptr [[TMP60]], align 4
-// CHECK15-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[TMP57]], ptr [[TMP61]], align 4
-// CHECK15-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
-// CHECK15-NEXT: store ptr [[TMP58]], ptr [[TMP62]], align 4
-// CHECK15-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
-// CHECK15-NEXT: store ptr @.offload_sizes.32, ptr [[TMP63]], align 4
-// CHECK15-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
-// CHECK15-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP64]], align 4
-// CHECK15-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
-// CHECK15-NEXT: store ptr null, ptr [[TMP65]], align 4
-// CHECK15-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
-// CHECK15-NEXT: store ptr null, ptr [[TMP66]], align 4
-// CHECK15-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
-// CHECK15-NEXT: store i64 10, ptr [[TMP67]], align 8
-// CHECK15-NEXT: [[TMP68:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.region_id, ptr [[KERNEL_ARGS19]])
-// CHECK15-NEXT: [[TMP69:%.*]] = icmp ne i32 [[TMP68]], 0
-// CHECK15-NEXT: br i1 [[TMP69]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
+// CHECK15-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
+// CHECK15-NEXT: store i32 2, ptr [[TMP71]], align 4
+// CHECK15-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
+// CHECK15-NEXT: store i32 1, ptr [[TMP72]], align 4
+// CHECK15-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[TMP69]], ptr [[TMP73]], align 4
+// CHECK15-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
+// CHECK15-NEXT: store ptr [[TMP70]], ptr [[TMP74]], align 4
+// CHECK15-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
+// CHECK15-NEXT: store ptr @.offload_sizes.32, ptr [[TMP75]], align 4
+// CHECK15-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
+// CHECK15-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP76]], align 4
+// CHECK15-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
+// CHECK15-NEXT: store ptr null, ptr [[TMP77]], align 4
+// CHECK15-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
+// CHECK15-NEXT: store ptr null, ptr [[TMP78]], align 4
+// CHECK15-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
+// CHECK15-NEXT: store i64 10, ptr [[TMP79]], align 8
+// CHECK15-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP80]], align 8
+// CHECK15-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP81]], align 4
+// CHECK15-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP82]], align 4
+// CHECK15-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP83]], align 4
+// CHECK15-NEXT: [[TMP84:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.region_id, ptr [[KERNEL_ARGS19]])
+// CHECK15-NEXT: [[TMP85:%.*]] = icmp ne i32 [[TMP84]], 0
+// CHECK15-NEXT: br i1 [[TMP85]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
// CHECK15: omp_offload.failed20:
// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124(ptr [[A]]) #[[ATTR3]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT21]]
// CHECK15: omp_offload.cont21:
-// CHECK15-NEXT: [[TMP70:%.*]] = load i32, ptr [[M]], align 4
-// CHECK15-NEXT: store i32 [[TMP70]], ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK15-NEXT: [[TMP71:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK15-NEXT: store i32 [[TMP71]], ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4
-// CHECK15-NEXT: [[TMP72:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4
-// CHECK15-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
-// CHECK15-NEXT: store ptr [[A]], ptr [[TMP73]], align 4
-// CHECK15-NEXT: [[TMP74:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
-// CHECK15-NEXT: store ptr [[A]], ptr [[TMP74]], align 4
-// CHECK15-NEXT: [[TMP75:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 0
-// CHECK15-NEXT: store ptr null, ptr [[TMP75]], align 4
-// CHECK15-NEXT: [[TMP76:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP72]], ptr [[TMP76]], align 4
-// CHECK15-NEXT: [[TMP77:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP72]], ptr [[TMP77]], align 4
-// CHECK15-NEXT: [[TMP78:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 1
-// CHECK15-NEXT: store ptr null, ptr [[TMP78]], align 4
-// CHECK15-NEXT: [[TMP79:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP80:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP86:%.*]] = load i32, ptr [[M]], align 4
+// CHECK15-NEXT: store i32 [[TMP86]], ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK15-NEXT: [[TMP87:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK15-NEXT: store i32 [[TMP87]], ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4
+// CHECK15-NEXT: [[TMP88:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4
+// CHECK15-NEXT: [[TMP89:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
+// CHECK15-NEXT: store ptr [[A]], ptr [[TMP89]], align 4
+// CHECK15-NEXT: [[TMP90:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
+// CHECK15-NEXT: store ptr [[A]], ptr [[TMP90]], align 4
+// CHECK15-NEXT: [[TMP91:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 0
+// CHECK15-NEXT: store ptr null, ptr [[TMP91]], align 4
+// CHECK15-NEXT: [[TMP92:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP88]], ptr [[TMP92]], align 4
+// CHECK15-NEXT: [[TMP93:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP88]], ptr [[TMP93]], align 4
+// CHECK15-NEXT: [[TMP94:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 1
+// CHECK15-NEXT: store ptr null, ptr [[TMP94]], align 4
+// CHECK15-NEXT: [[TMP95:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP96:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
// CHECK15-NEXT: [[KERNEL_ARGS28:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK15-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP81]], align 4
-// CHECK15-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
-// CHECK15-NEXT: store i32 2, ptr [[TMP82]], align 4
-// CHECK15-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[TMP79]], ptr [[TMP83]], align 4
-// CHECK15-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
-// CHECK15-NEXT: store ptr [[TMP80]], ptr [[TMP84]], align 4
-// CHECK15-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
-// CHECK15-NEXT: store ptr @.offload_sizes.36, ptr [[TMP85]], align 4
-// CHECK15-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
-// CHECK15-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP86]], align 4
-// CHECK15-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
-// CHECK15-NEXT: store ptr null, ptr [[TMP87]], align 4
-// CHECK15-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
-// CHECK15-NEXT: store ptr null, ptr [[TMP88]], align 4
-// CHECK15-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
-// CHECK15-NEXT: store i64 10, ptr [[TMP89]], align 8
-// CHECK15-NEXT: [[TMP90:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.region_id, ptr [[KERNEL_ARGS28]])
-// CHECK15-NEXT: [[TMP91:%.*]] = icmp ne i32 [[TMP90]], 0
-// CHECK15-NEXT: br i1 [[TMP91]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
+// CHECK15-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
+// CHECK15-NEXT: store i32 2, ptr [[TMP97]], align 4
+// CHECK15-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
+// CHECK15-NEXT: store i32 2, ptr [[TMP98]], align 4
+// CHECK15-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[TMP95]], ptr [[TMP99]], align 4
+// CHECK15-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
+// CHECK15-NEXT: store ptr [[TMP96]], ptr [[TMP100]], align 4
+// CHECK15-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
+// CHECK15-NEXT: store ptr @.offload_sizes.36, ptr [[TMP101]], align 4
+// CHECK15-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
+// CHECK15-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP102]], align 4
+// CHECK15-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
+// CHECK15-NEXT: store ptr null, ptr [[TMP103]], align 4
+// CHECK15-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
+// CHECK15-NEXT: store ptr null, ptr [[TMP104]], align 4
+// CHECK15-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
+// CHECK15-NEXT: store i64 10, ptr [[TMP105]], align 8
+// CHECK15-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP106]], align 8
+// CHECK15-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP107]], align 4
+// CHECK15-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP108]], align 4
+// CHECK15-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP109]], align 4
+// CHECK15-NEXT: [[TMP110:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.region_id, ptr [[KERNEL_ARGS28]])
+// CHECK15-NEXT: [[TMP111:%.*]] = icmp ne i32 [[TMP110]], 0
+// CHECK15-NEXT: br i1 [[TMP111]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
// CHECK15: omp_offload.failed29:
-// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128(ptr [[A]], i32 [[TMP72]]) #[[ATTR3]]
+// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128(ptr [[A]], i32 [[TMP88]]) #[[ATTR3]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT30]]
// CHECK15: omp_offload.cont30:
// CHECK15-NEXT: ret i32 0
@@ -9329,7 +9649,7 @@ int main (int argc, char **argv) {
// CHECK17-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK17-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK17-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK17-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK17-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK17-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK17-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -9346,298 +9666,338 @@ int main (int argc, char **argv) {
// CHECK17-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK17-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK17-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK17-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.region_id, ptr [[KERNEL_ARGS]])
-// CHECK17-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK17-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK17-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK17-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK17-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK17-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK17-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.region_id, ptr [[KERNEL_ARGS]])
+// CHECK17-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK17-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK17: omp_offload.failed:
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139(i64 [[TMP4]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK17: omp_offload.cont:
-// CHECK17-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK17-NEXT: [[TMP35:%.*]] = load i64, ptr [[N_CASTED3]], align 8
-// CHECK17-NEXT: [[TMP36:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK17-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK17-NEXT: [[TMP39:%.*]] = load i64, ptr [[N_CASTED3]], align 8
+// CHECK17-NEXT: [[TMP40:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES7]], ptr align 8 @.offload_sizes.4, i64 24, i1 false)
-// CHECK17-NEXT: [[TMP37:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP35]], ptr [[TMP37]], align 8
-// CHECK17-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP35]], ptr [[TMP38]], align 8
-// CHECK17-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK17-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP40]], align 8
-// CHECK17-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP41]], align 8
-// CHECK17-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
-// CHECK17-NEXT: store ptr null, ptr [[TMP42]], align 8
-// CHECK17-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP43]], align 8
-// CHECK17-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 8
-// CHECK17-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK17-NEXT: store i64 [[TMP36]], ptr [[TMP45]], align 8
-// CHECK17-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK17-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP39]], ptr [[TMP41]], align 8
+// CHECK17-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP39]], ptr [[TMP42]], align 8
+// CHECK17-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK17-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP44]], align 8
+// CHECK17-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP45]], align 8
+// CHECK17-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
// CHECK17-NEXT: store ptr null, ptr [[TMP46]], align 8
-// CHECK17-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP50:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP50]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK17-NEXT: [[TMP51:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK17-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP51]], 0
+// CHECK17-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP47]], align 8
+// CHECK17-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 8
+// CHECK17-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK17-NEXT: store i64 [[TMP40]], ptr [[TMP49]], align 8
+// CHECK17-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK17-NEXT: store ptr null, ptr [[TMP50]], align 8
+// CHECK17-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP54:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP54]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK17-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK17-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP55]], 0
// CHECK17-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK17-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK17-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK17-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK17-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP52]], 1
-// CHECK17-NEXT: [[TMP53:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK17-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK17-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP56]], 1
+// CHECK17-NEXT: [[TMP57:%.*]] = zext i32 [[ADD14]] to i64
// CHECK17-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK17-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK17-NEXT: store i32 3, ptr [[TMP55]], align 4
-// CHECK17-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP47]], ptr [[TMP56]], align 8
-// CHECK17-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 8
-// CHECK17-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK17-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 8
-// CHECK17-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP59]], align 8
-// CHECK17-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP60]], align 8
-// CHECK17-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP61]], align 8
-// CHECK17-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK17-NEXT: store i64 [[TMP53]], ptr [[TMP62]], align 8
-// CHECK17-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK17-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
-// CHECK17-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK17-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP58]], align 4
+// CHECK17-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK17-NEXT: store i32 3, ptr [[TMP59]], align 4
+// CHECK17-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP51]], ptr [[TMP60]], align 8
+// CHECK17-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 8
+// CHECK17-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK17-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 8
+// CHECK17-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP63]], align 8
+// CHECK17-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP64]], align 8
+// CHECK17-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP65]], align 8
+// CHECK17-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK17-NEXT: store i64 [[TMP57]], ptr [[TMP66]], align 8
+// CHECK17-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP67]], align 8
+// CHECK17-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP68]], align 4
+// CHECK17-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK17-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP70]], align 4
+// CHECK17-NEXT: [[TMP71:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK17-NEXT: [[TMP72:%.*]] = icmp ne i32 [[TMP71]], 0
+// CHECK17-NEXT: br i1 [[TMP72]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK17: omp_offload.failed16:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143(i64 [[TMP35]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143(i64 [[TMP39]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK17: omp_offload.cont17:
-// CHECK17-NEXT: [[TMP65:%.*]] = load i32, ptr [[M]], align 4
-// CHECK17-NEXT: store i32 [[TMP65]], ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK17-NEXT: [[TMP66:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP66]], ptr [[N_CASTED19]], align 4
-// CHECK17-NEXT: [[TMP67:%.*]] = load i64, ptr [[N_CASTED19]], align 8
-// CHECK17-NEXT: [[TMP68:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK17-NEXT: store i32 [[TMP68]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK17-NEXT: [[TMP69:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK17-NEXT: [[TMP70:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK17-NEXT: [[TMP73:%.*]] = load i32, ptr [[M]], align 4
+// CHECK17-NEXT: store i32 [[TMP73]], ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK17-NEXT: [[TMP74:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP74]], ptr [[N_CASTED19]], align 4
+// CHECK17-NEXT: [[TMP75:%.*]] = load i64, ptr [[N_CASTED19]], align 8
+// CHECK17-NEXT: [[TMP76:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK17-NEXT: store i32 [[TMP76]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK17-NEXT: [[TMP77:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK17-NEXT: [[TMP78:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES23]], ptr align 8 @.offload_sizes.8, i64 32, i1 false)
-// CHECK17-NEXT: [[TMP71:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP67]], ptr [[TMP71]], align 8
-// CHECK17-NEXT: [[TMP72:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP67]], ptr [[TMP72]], align 8
-// CHECK17-NEXT: [[TMP73:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP73]], align 8
-// CHECK17-NEXT: [[TMP74:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP74]], align 8
-// CHECK17-NEXT: [[TMP75:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP75]], align 8
-// CHECK17-NEXT: [[TMP76:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 1
-// CHECK17-NEXT: store ptr null, ptr [[TMP76]], align 8
-// CHECK17-NEXT: [[TMP77:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP77]], align 8
-// CHECK17-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP78]], align 8
-// CHECK17-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
-// CHECK17-NEXT: store i64 [[TMP70]], ptr [[TMP79]], align 8
-// CHECK17-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 2
-// CHECK17-NEXT: store ptr null, ptr [[TMP80]], align 8
-// CHECK17-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
-// CHECK17-NEXT: store i64 [[TMP69]], ptr [[TMP81]], align 8
-// CHECK17-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
-// CHECK17-NEXT: store i64 [[TMP69]], ptr [[TMP82]], align 8
-// CHECK17-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 3
-// CHECK17-NEXT: store ptr null, ptr [[TMP83]], align 8
-// CHECK17-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP87:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP87]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK17-NEXT: [[TMP88:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK17-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP88]], 0
+// CHECK17-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP75]], ptr [[TMP79]], align 8
+// CHECK17-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP75]], ptr [[TMP80]], align 8
+// CHECK17-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP81]], align 8
+// CHECK17-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP82]], align 8
+// CHECK17-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP83]], align 8
+// CHECK17-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 1
+// CHECK17-NEXT: store ptr null, ptr [[TMP84]], align 8
+// CHECK17-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP85]], align 8
+// CHECK17-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP86]], align 8
+// CHECK17-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
+// CHECK17-NEXT: store i64 [[TMP78]], ptr [[TMP87]], align 8
+// CHECK17-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 2
+// CHECK17-NEXT: store ptr null, ptr [[TMP88]], align 8
+// CHECK17-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
+// CHECK17-NEXT: store i64 [[TMP77]], ptr [[TMP89]], align 8
+// CHECK17-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
+// CHECK17-NEXT: store i64 [[TMP77]], ptr [[TMP90]], align 8
+// CHECK17-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 3
+// CHECK17-NEXT: store ptr null, ptr [[TMP91]], align 8
+// CHECK17-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP94:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP95:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP95]], ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK17-NEXT: [[TMP96:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK17-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP96]], 0
// CHECK17-NEXT: [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1
// CHECK17-NEXT: [[SUB29:%.*]] = sub nsw i32 [[DIV28]], 1
// CHECK17-NEXT: store i32 [[SUB29]], ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK17-NEXT: [[TMP89:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK17-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP89]], 1
-// CHECK17-NEXT: [[TMP90:%.*]] = zext i32 [[ADD30]] to i64
+// CHECK17-NEXT: [[TMP97:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
+// CHECK17-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP97]], 1
+// CHECK17-NEXT: [[TMP98:%.*]] = zext i32 [[ADD30]] to i64
// CHECK17-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP91]], align 4
-// CHECK17-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK17-NEXT: store i32 4, ptr [[TMP92]], align 4
-// CHECK17-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP84]], ptr [[TMP93]], align 8
-// CHECK17-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP85]], ptr [[TMP94]], align 8
-// CHECK17-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK17-NEXT: store ptr [[TMP86]], ptr [[TMP95]], align 8
-// CHECK17-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP96]], align 8
-// CHECK17-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP97]], align 8
-// CHECK17-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP98]], align 8
-// CHECK17-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK17-NEXT: store i64 [[TMP90]], ptr [[TMP99]], align 8
-// CHECK17-NEXT: [[TMP100:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK17-NEXT: [[TMP101:%.*]] = icmp ne i32 [[TMP100]], 0
-// CHECK17-NEXT: br i1 [[TMP101]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK17-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP99]], align 4
+// CHECK17-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK17-NEXT: store i32 4, ptr [[TMP100]], align 4
+// CHECK17-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP92]], ptr [[TMP101]], align 8
+// CHECK17-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP93]], ptr [[TMP102]], align 8
+// CHECK17-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK17-NEXT: store ptr [[TMP94]], ptr [[TMP103]], align 8
+// CHECK17-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP104]], align 8
+// CHECK17-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP105]], align 8
+// CHECK17-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP106]], align 8
+// CHECK17-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK17-NEXT: store i64 [[TMP98]], ptr [[TMP107]], align 8
+// CHECK17-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP108]], align 8
+// CHECK17-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP109]], align 4
+// CHECK17-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP110]], align 4
+// CHECK17-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP111]], align 4
+// CHECK17-NEXT: [[TMP112:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK17-NEXT: [[TMP113:%.*]] = icmp ne i32 [[TMP112]], 0
+// CHECK17-NEXT: br i1 [[TMP113]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK17: omp_offload.failed32:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147(i64 [[TMP67]], i64 [[TMP1]], ptr [[VLA]], i64 [[TMP69]]) #[[ATTR3]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147(i64 [[TMP75]], i64 [[TMP1]], ptr [[VLA]], i64 [[TMP77]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK17: omp_offload.cont33:
-// CHECK17-NEXT: [[TMP102:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP102]], ptr [[N_CASTED34]], align 4
-// CHECK17-NEXT: [[TMP103:%.*]] = load i64, ptr [[N_CASTED34]], align 8
-// CHECK17-NEXT: [[TMP104:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK17-NEXT: [[TMP114:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP114]], ptr [[N_CASTED34]], align 4
+// CHECK17-NEXT: [[TMP115:%.*]] = load i64, ptr [[N_CASTED34]], align 8
+// CHECK17-NEXT: [[TMP116:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES38]], ptr align 8 @.offload_sizes.12, i64 24, i1 false)
-// CHECK17-NEXT: [[TMP105:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP103]], ptr [[TMP105]], align 8
-// CHECK17-NEXT: [[TMP106:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP103]], ptr [[TMP106]], align 8
-// CHECK17-NEXT: [[TMP107:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP107]], align 8
-// CHECK17-NEXT: [[TMP108:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP108]], align 8
-// CHECK17-NEXT: [[TMP109:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP109]], align 8
-// CHECK17-NEXT: [[TMP110:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 1
-// CHECK17-NEXT: store ptr null, ptr [[TMP110]], align 8
-// CHECK17-NEXT: [[TMP111:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP111]], align 8
-// CHECK17-NEXT: [[TMP112:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP112]], align 8
-// CHECK17-NEXT: [[TMP113:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 2
-// CHECK17-NEXT: store i64 [[TMP104]], ptr [[TMP113]], align 8
-// CHECK17-NEXT: [[TMP114:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 2
-// CHECK17-NEXT: store ptr null, ptr [[TMP114]], align 8
-// CHECK17-NEXT: [[TMP115:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP116:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP117:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP118:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP118]], ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK17-NEXT: [[TMP119:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK17-NEXT: [[SUB42:%.*]] = sub nsw i32 [[TMP119]], 0
+// CHECK17-NEXT: [[TMP117:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP115]], ptr [[TMP117]], align 8
+// CHECK17-NEXT: [[TMP118:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP115]], ptr [[TMP118]], align 8
+// CHECK17-NEXT: [[TMP119:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP119]], align 8
+// CHECK17-NEXT: [[TMP120:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP120]], align 8
+// CHECK17-NEXT: [[TMP121:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP121]], align 8
+// CHECK17-NEXT: [[TMP122:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 1
+// CHECK17-NEXT: store ptr null, ptr [[TMP122]], align 8
+// CHECK17-NEXT: [[TMP123:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP123]], align 8
+// CHECK17-NEXT: [[TMP124:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP124]], align 8
+// CHECK17-NEXT: [[TMP125:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 2
+// CHECK17-NEXT: store i64 [[TMP116]], ptr [[TMP125]], align 8
+// CHECK17-NEXT: [[TMP126:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 2
+// CHECK17-NEXT: store ptr null, ptr [[TMP126]], align 8
+// CHECK17-NEXT: [[TMP127:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP128:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP129:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP130:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP130]], ptr [[DOTCAPTURE_EXPR_40]], align 4
+// CHECK17-NEXT: [[TMP131:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
+// CHECK17-NEXT: [[SUB42:%.*]] = sub nsw i32 [[TMP131]], 0
// CHECK17-NEXT: [[DIV43:%.*]] = sdiv i32 [[SUB42]], 1
// CHECK17-NEXT: [[SUB44:%.*]] = sub nsw i32 [[DIV43]], 1
// CHECK17-NEXT: store i32 [[SUB44]], ptr [[DOTCAPTURE_EXPR_41]], align 4
-// CHECK17-NEXT: [[TMP120:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_41]], align 4
-// CHECK17-NEXT: [[ADD45:%.*]] = add nsw i32 [[TMP120]], 1
-// CHECK17-NEXT: [[TMP121:%.*]] = zext i32 [[ADD45]] to i64
+// CHECK17-NEXT: [[TMP132:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_41]], align 4
+// CHECK17-NEXT: [[ADD45:%.*]] = add nsw i32 [[TMP132]], 1
+// CHECK17-NEXT: [[TMP133:%.*]] = zext i32 [[ADD45]] to i64
// CHECK17-NEXT: [[KERNEL_ARGS46:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP122:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP122]], align 4
-// CHECK17-NEXT: [[TMP123:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 1
-// CHECK17-NEXT: store i32 3, ptr [[TMP123]], align 4
-// CHECK17-NEXT: [[TMP124:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP115]], ptr [[TMP124]], align 8
-// CHECK17-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP116]], ptr [[TMP125]], align 8
-// CHECK17-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 4
-// CHECK17-NEXT: store ptr [[TMP117]], ptr [[TMP126]], align 8
-// CHECK17-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP127]], align 8
-// CHECK17-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP128]], align 8
-// CHECK17-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP129]], align 8
-// CHECK17-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 8
-// CHECK17-NEXT: store i64 [[TMP121]], ptr [[TMP130]], align 8
-// CHECK17-NEXT: [[TMP131:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.region_id, ptr [[KERNEL_ARGS46]])
-// CHECK17-NEXT: [[TMP132:%.*]] = icmp ne i32 [[TMP131]], 0
-// CHECK17-NEXT: br i1 [[TMP132]], label [[OMP_OFFLOAD_FAILED47:%.*]], label [[OMP_OFFLOAD_CONT48:%.*]]
+// CHECK17-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP134]], align 4
+// CHECK17-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 1
+// CHECK17-NEXT: store i32 3, ptr [[TMP135]], align 4
+// CHECK17-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP127]], ptr [[TMP136]], align 8
+// CHECK17-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP128]], ptr [[TMP137]], align 8
+// CHECK17-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 4
+// CHECK17-NEXT: store ptr [[TMP129]], ptr [[TMP138]], align 8
+// CHECK17-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP139]], align 8
+// CHECK17-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP140]], align 8
+// CHECK17-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP141]], align 8
+// CHECK17-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 8
+// CHECK17-NEXT: store i64 [[TMP133]], ptr [[TMP142]], align 8
+// CHECK17-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP143]], align 8
+// CHECK17-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP144]], align 4
+// CHECK17-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP145]], align 4
+// CHECK17-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP146]], align 4
+// CHECK17-NEXT: [[TMP147:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.region_id, ptr [[KERNEL_ARGS46]])
+// CHECK17-NEXT: [[TMP148:%.*]] = icmp ne i32 [[TMP147]], 0
+// CHECK17-NEXT: br i1 [[TMP148]], label [[OMP_OFFLOAD_FAILED47:%.*]], label [[OMP_OFFLOAD_CONT48:%.*]]
// CHECK17: omp_offload.failed47:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151(i64 [[TMP103]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151(i64 [[TMP115]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT48]]
// CHECK17: omp_offload.cont48:
-// CHECK17-NEXT: [[TMP133:%.*]] = load i32, ptr [[M]], align 4
-// CHECK17-NEXT: store i32 [[TMP133]], ptr [[DOTCAPTURE_EXPR_49]], align 4
-// CHECK17-NEXT: [[TMP134:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP134]], ptr [[N_CASTED50]], align 4
-// CHECK17-NEXT: [[TMP135:%.*]] = load i64, ptr [[N_CASTED50]], align 8
-// CHECK17-NEXT: [[TMP136:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_49]], align 4
-// CHECK17-NEXT: store i32 [[TMP136]], ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4
-// CHECK17-NEXT: [[TMP137:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED51]], align 8
-// CHECK17-NEXT: [[TMP138:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK17-NEXT: [[TMP149:%.*]] = load i32, ptr [[M]], align 4
+// CHECK17-NEXT: store i32 [[TMP149]], ptr [[DOTCAPTURE_EXPR_49]], align 4
+// CHECK17-NEXT: [[TMP150:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP150]], ptr [[N_CASTED50]], align 4
+// CHECK17-NEXT: [[TMP151:%.*]] = load i64, ptr [[N_CASTED50]], align 8
+// CHECK17-NEXT: [[TMP152:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_49]], align 4
+// CHECK17-NEXT: store i32 [[TMP152]], ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4
+// CHECK17-NEXT: [[TMP153:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED51]], align 8
+// CHECK17-NEXT: [[TMP154:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES55]], ptr align 8 @.offload_sizes.16, i64 32, i1 false)
-// CHECK17-NEXT: [[TMP139:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP135]], ptr [[TMP139]], align 8
-// CHECK17-NEXT: [[TMP140:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP135]], ptr [[TMP140]], align 8
-// CHECK17-NEXT: [[TMP141:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP141]], align 8
-// CHECK17-NEXT: [[TMP142:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP142]], align 8
-// CHECK17-NEXT: [[TMP143:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP143]], align 8
-// CHECK17-NEXT: [[TMP144:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 1
-// CHECK17-NEXT: store ptr null, ptr [[TMP144]], align 8
-// CHECK17-NEXT: [[TMP145:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP145]], align 8
-// CHECK17-NEXT: [[TMP146:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP146]], align 8
-// CHECK17-NEXT: [[TMP147:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 2
-// CHECK17-NEXT: store i64 [[TMP138]], ptr [[TMP147]], align 8
-// CHECK17-NEXT: [[TMP148:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 2
-// CHECK17-NEXT: store ptr null, ptr [[TMP148]], align 8
-// CHECK17-NEXT: [[TMP149:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 3
-// CHECK17-NEXT: store i64 [[TMP137]], ptr [[TMP149]], align 8
-// CHECK17-NEXT: [[TMP150:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 3
-// CHECK17-NEXT: store i64 [[TMP137]], ptr [[TMP150]], align 8
-// CHECK17-NEXT: [[TMP151:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 3
-// CHECK17-NEXT: store ptr null, ptr [[TMP151]], align 8
-// CHECK17-NEXT: [[TMP152:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP153:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP154:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP155:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP155]], ptr [[DOTCAPTURE_EXPR_57]], align 4
-// CHECK17-NEXT: [[TMP156:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_57]], align 4
-// CHECK17-NEXT: [[SUB59:%.*]] = sub nsw i32 [[TMP156]], 0
+// CHECK17-NEXT: [[TMP155:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP151]], ptr [[TMP155]], align 8
+// CHECK17-NEXT: [[TMP156:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP151]], ptr [[TMP156]], align 8
+// CHECK17-NEXT: [[TMP157:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP157]], align 8
+// CHECK17-NEXT: [[TMP158:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP158]], align 8
+// CHECK17-NEXT: [[TMP159:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP159]], align 8
+// CHECK17-NEXT: [[TMP160:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 1
+// CHECK17-NEXT: store ptr null, ptr [[TMP160]], align 8
+// CHECK17-NEXT: [[TMP161:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP161]], align 8
+// CHECK17-NEXT: [[TMP162:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP162]], align 8
+// CHECK17-NEXT: [[TMP163:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 2
+// CHECK17-NEXT: store i64 [[TMP154]], ptr [[TMP163]], align 8
+// CHECK17-NEXT: [[TMP164:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 2
+// CHECK17-NEXT: store ptr null, ptr [[TMP164]], align 8
+// CHECK17-NEXT: [[TMP165:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 3
+// CHECK17-NEXT: store i64 [[TMP153]], ptr [[TMP165]], align 8
+// CHECK17-NEXT: [[TMP166:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 3
+// CHECK17-NEXT: store i64 [[TMP153]], ptr [[TMP166]], align 8
+// CHECK17-NEXT: [[TMP167:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 3
+// CHECK17-NEXT: store ptr null, ptr [[TMP167]], align 8
+// CHECK17-NEXT: [[TMP168:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP169:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP170:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP171:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP171]], ptr [[DOTCAPTURE_EXPR_57]], align 4
+// CHECK17-NEXT: [[TMP172:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_57]], align 4
+// CHECK17-NEXT: [[SUB59:%.*]] = sub nsw i32 [[TMP172]], 0
// CHECK17-NEXT: [[DIV60:%.*]] = sdiv i32 [[SUB59]], 1
// CHECK17-NEXT: [[SUB61:%.*]] = sub nsw i32 [[DIV60]], 1
// CHECK17-NEXT: store i32 [[SUB61]], ptr [[DOTCAPTURE_EXPR_58]], align 4
-// CHECK17-NEXT: [[TMP157:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_58]], align 4
-// CHECK17-NEXT: [[ADD62:%.*]] = add nsw i32 [[TMP157]], 1
-// CHECK17-NEXT: [[TMP158:%.*]] = zext i32 [[ADD62]] to i64
+// CHECK17-NEXT: [[TMP173:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_58]], align 4
+// CHECK17-NEXT: [[ADD62:%.*]] = add nsw i32 [[TMP173]], 1
+// CHECK17-NEXT: [[TMP174:%.*]] = zext i32 [[ADD62]] to i64
// CHECK17-NEXT: [[KERNEL_ARGS63:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP159:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP159]], align 4
-// CHECK17-NEXT: [[TMP160:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 1
-// CHECK17-NEXT: store i32 4, ptr [[TMP160]], align 4
-// CHECK17-NEXT: [[TMP161:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP152]], ptr [[TMP161]], align 8
-// CHECK17-NEXT: [[TMP162:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP153]], ptr [[TMP162]], align 8
-// CHECK17-NEXT: [[TMP163:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 4
-// CHECK17-NEXT: store ptr [[TMP154]], ptr [[TMP163]], align 8
-// CHECK17-NEXT: [[TMP164:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP164]], align 8
-// CHECK17-NEXT: [[TMP165:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP165]], align 8
-// CHECK17-NEXT: [[TMP166:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP166]], align 8
-// CHECK17-NEXT: [[TMP167:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 8
-// CHECK17-NEXT: store i64 [[TMP158]], ptr [[TMP167]], align 8
-// CHECK17-NEXT: [[TMP168:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.region_id, ptr [[KERNEL_ARGS63]])
-// CHECK17-NEXT: [[TMP169:%.*]] = icmp ne i32 [[TMP168]], 0
-// CHECK17-NEXT: br i1 [[TMP169]], label [[OMP_OFFLOAD_FAILED64:%.*]], label [[OMP_OFFLOAD_CONT65:%.*]]
+// CHECK17-NEXT: [[TMP175:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP175]], align 4
+// CHECK17-NEXT: [[TMP176:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 1
+// CHECK17-NEXT: store i32 4, ptr [[TMP176]], align 4
+// CHECK17-NEXT: [[TMP177:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP168]], ptr [[TMP177]], align 8
+// CHECK17-NEXT: [[TMP178:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP169]], ptr [[TMP178]], align 8
+// CHECK17-NEXT: [[TMP179:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 4
+// CHECK17-NEXT: store ptr [[TMP170]], ptr [[TMP179]], align 8
+// CHECK17-NEXT: [[TMP180:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP180]], align 8
+// CHECK17-NEXT: [[TMP181:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP181]], align 8
+// CHECK17-NEXT: [[TMP182:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP182]], align 8
+// CHECK17-NEXT: [[TMP183:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 8
+// CHECK17-NEXT: store i64 [[TMP174]], ptr [[TMP183]], align 8
+// CHECK17-NEXT: [[TMP184:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP184]], align 8
+// CHECK17-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP185]], align 4
+// CHECK17-NEXT: [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP186]], align 4
+// CHECK17-NEXT: [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP187]], align 4
+// CHECK17-NEXT: [[TMP188:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.region_id, ptr [[KERNEL_ARGS63]])
+// CHECK17-NEXT: [[TMP189:%.*]] = icmp ne i32 [[TMP188]], 0
+// CHECK17-NEXT: br i1 [[TMP189]], label [[OMP_OFFLOAD_FAILED64:%.*]], label [[OMP_OFFLOAD_CONT65:%.*]]
// CHECK17: omp_offload.failed64:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155(i64 [[TMP135]], i64 [[TMP1]], ptr [[VLA]], i64 [[TMP137]]) #[[ATTR3]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155(i64 [[TMP151]], i64 [[TMP1]], ptr [[VLA]], i64 [[TMP153]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT65]]
// CHECK17: omp_offload.cont65:
-// CHECK17-NEXT: [[TMP170:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK17-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP170]])
+// CHECK17-NEXT: [[TMP190:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK17-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP190]])
// CHECK17-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK17-NEXT: [[TMP171:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK17-NEXT: call void @llvm.stackrestore(ptr [[TMP171]])
-// CHECK17-NEXT: [[TMP172:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK17-NEXT: ret i32 [[TMP172]]
+// CHECK17-NEXT: [[TMP191:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK17-NEXT: call void @llvm.stackrestore(ptr [[TMP191]])
+// CHECK17-NEXT: [[TMP192:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK17-NEXT: ret i32 [[TMP192]]
//
//
// CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139
@@ -10813,7 +11173,7 @@ int main (int argc, char **argv) {
// CHECK17-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK17-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK17-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK17-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK17-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK17-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -10830,169 +11190,209 @@ int main (int argc, char **argv) {
// CHECK17-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK17-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK17-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK17-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.region_id, ptr [[KERNEL_ARGS]])
-// CHECK17-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK17-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK17-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK17-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK17-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK17-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK17-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.region_id, ptr [[KERNEL_ARGS]])
+// CHECK17-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK17-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK17: omp_offload.failed:
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112(ptr [[A]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK17: omp_offload.cont:
-// CHECK17-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP16]], align 8
-// CHECK17-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP17]], align 8
-// CHECK17-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK17-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP20]], align 8
+// CHECK17-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP21]], align 8
+// CHECK17-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK17-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK17-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK17-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK17-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK17-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK17-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK17-NEXT: store ptr @.offload_sizes.24, ptr [[TMP25]], align 8
-// CHECK17-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP26]], align 8
-// CHECK17-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK17-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK17-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK17-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK17-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK17-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK17-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK17-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK17-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK17-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK17-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK17-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK17-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK17-NEXT: store ptr @.offload_sizes.24, ptr [[TMP29]], align 8
+// CHECK17-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP30]], align 8
+// CHECK17-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK17-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK17-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK17-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK17-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK17-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK17-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK17-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK17-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK17-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK17-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK17: omp_offload.failed6:
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116(ptr [[A]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK17: omp_offload.cont7:
-// CHECK17-NEXT: [[TMP32:%.*]] = load i32, ptr [[M]], align 4
-// CHECK17-NEXT: store i32 [[TMP32]], ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK17-NEXT: [[TMP33:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK17-NEXT: store i32 [[TMP33]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK17-NEXT: [[TMP34:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK17-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP35]], align 8
-// CHECK17-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP36]], align 8
-// CHECK17-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP37]], align 8
-// CHECK17-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP34]], ptr [[TMP38]], align 8
-// CHECK17-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP34]], ptr [[TMP39]], align 8
-// CHECK17-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
-// CHECK17-NEXT: store ptr null, ptr [[TMP40]], align 8
-// CHECK17-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP40:%.*]] = load i32, ptr [[M]], align 4
+// CHECK17-NEXT: store i32 [[TMP40]], ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK17-NEXT: [[TMP41:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK17-NEXT: store i32 [[TMP41]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK17-NEXT: [[TMP42:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK17-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP43]], align 8
+// CHECK17-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP44]], align 8
+// CHECK17-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP45]], align 8
+// CHECK17-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP42]], ptr [[TMP46]], align 8
+// CHECK17-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP42]], ptr [[TMP47]], align 8
+// CHECK17-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
+// CHECK17-NEXT: store ptr null, ptr [[TMP48]], align 8
+// CHECK17-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP43]], align 4
-// CHECK17-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK17-NEXT: store i32 2, ptr [[TMP44]], align 4
-// CHECK17-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP41]], ptr [[TMP45]], align 8
-// CHECK17-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP42]], ptr [[TMP46]], align 8
-// CHECK17-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK17-NEXT: store ptr @.offload_sizes.28, ptr [[TMP47]], align 8
-// CHECK17-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP48]], align 8
-// CHECK17-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP49]], align 8
-// CHECK17-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP50]], align 8
-// CHECK17-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK17-NEXT: store i64 10, ptr [[TMP51]], align 8
-// CHECK17-NEXT: [[TMP52:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK17-NEXT: [[TMP53:%.*]] = icmp ne i32 [[TMP52]], 0
-// CHECK17-NEXT: br i1 [[TMP53]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK17-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP51]], align 4
+// CHECK17-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK17-NEXT: store i32 2, ptr [[TMP52]], align 4
+// CHECK17-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 8
+// CHECK17-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP50]], ptr [[TMP54]], align 8
+// CHECK17-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK17-NEXT: store ptr @.offload_sizes.28, ptr [[TMP55]], align 8
+// CHECK17-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP56]], align 8
+// CHECK17-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP57]], align 8
+// CHECK17-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP58]], align 8
+// CHECK17-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK17-NEXT: store i64 10, ptr [[TMP59]], align 8
+// CHECK17-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP60]], align 8
+// CHECK17-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4
+// CHECK17-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP62]], align 4
+// CHECK17-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP63]], align 4
+// CHECK17-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK17-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
+// CHECK17-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK17: omp_offload.failed13:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120(ptr [[A]], i64 [[TMP34]]) #[[ATTR3]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120(ptr [[A]], i64 [[TMP42]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT14]]
// CHECK17: omp_offload.cont14:
-// CHECK17-NEXT: [[TMP54:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP54]], align 8
-// CHECK17-NEXT: [[TMP55:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP55]], align 8
-// CHECK17-NEXT: [[TMP56:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i64 0, i64 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP56]], align 8
-// CHECK17-NEXT: [[TMP57:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP58:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP66]], align 8
+// CHECK17-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP67]], align 8
+// CHECK17-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i64 0, i64 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP68]], align 8
+// CHECK17-NEXT: [[TMP69:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP70:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS19:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP59]], align 4
-// CHECK17-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
-// CHECK17-NEXT: store i32 1, ptr [[TMP60]], align 4
-// CHECK17-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP57]], ptr [[TMP61]], align 8
-// CHECK17-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP58]], ptr [[TMP62]], align 8
-// CHECK17-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
-// CHECK17-NEXT: store ptr @.offload_sizes.32, ptr [[TMP63]], align 8
-// CHECK17-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP64]], align 8
-// CHECK17-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP65]], align 8
-// CHECK17-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP66]], align 8
-// CHECK17-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
-// CHECK17-NEXT: store i64 10, ptr [[TMP67]], align 8
-// CHECK17-NEXT: [[TMP68:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.region_id, ptr [[KERNEL_ARGS19]])
-// CHECK17-NEXT: [[TMP69:%.*]] = icmp ne i32 [[TMP68]], 0
-// CHECK17-NEXT: br i1 [[TMP69]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
+// CHECK17-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP71]], align 4
+// CHECK17-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
+// CHECK17-NEXT: store i32 1, ptr [[TMP72]], align 4
+// CHECK17-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP69]], ptr [[TMP73]], align 8
+// CHECK17-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP70]], ptr [[TMP74]], align 8
+// CHECK17-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
+// CHECK17-NEXT: store ptr @.offload_sizes.32, ptr [[TMP75]], align 8
+// CHECK17-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP76]], align 8
+// CHECK17-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP77]], align 8
+// CHECK17-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP78]], align 8
+// CHECK17-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
+// CHECK17-NEXT: store i64 10, ptr [[TMP79]], align 8
+// CHECK17-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP80]], align 8
+// CHECK17-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP81]], align 4
+// CHECK17-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP82]], align 4
+// CHECK17-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP83]], align 4
+// CHECK17-NEXT: [[TMP84:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.region_id, ptr [[KERNEL_ARGS19]])
+// CHECK17-NEXT: [[TMP85:%.*]] = icmp ne i32 [[TMP84]], 0
+// CHECK17-NEXT: br i1 [[TMP85]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
// CHECK17: omp_offload.failed20:
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124(ptr [[A]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT21]]
// CHECK17: omp_offload.cont21:
-// CHECK17-NEXT: [[TMP70:%.*]] = load i32, ptr [[M]], align 4
-// CHECK17-NEXT: store i32 [[TMP70]], ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK17-NEXT: [[TMP71:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK17-NEXT: store i32 [[TMP71]], ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4
-// CHECK17-NEXT: [[TMP72:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED23]], align 8
-// CHECK17-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP73]], align 8
-// CHECK17-NEXT: [[TMP74:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP74]], align 8
-// CHECK17-NEXT: [[TMP75:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP75]], align 8
-// CHECK17-NEXT: [[TMP76:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP72]], ptr [[TMP76]], align 8
-// CHECK17-NEXT: [[TMP77:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP72]], ptr [[TMP77]], align 8
-// CHECK17-NEXT: [[TMP78:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 1
-// CHECK17-NEXT: store ptr null, ptr [[TMP78]], align 8
-// CHECK17-NEXT: [[TMP79:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP80:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP86:%.*]] = load i32, ptr [[M]], align 4
+// CHECK17-NEXT: store i32 [[TMP86]], ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK17-NEXT: [[TMP87:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK17-NEXT: store i32 [[TMP87]], ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4
+// CHECK17-NEXT: [[TMP88:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED23]], align 8
+// CHECK17-NEXT: [[TMP89:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP89]], align 8
+// CHECK17-NEXT: [[TMP90:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP90]], align 8
+// CHECK17-NEXT: [[TMP91:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP91]], align 8
+// CHECK17-NEXT: [[TMP92:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP88]], ptr [[TMP92]], align 8
+// CHECK17-NEXT: [[TMP93:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP88]], ptr [[TMP93]], align 8
+// CHECK17-NEXT: [[TMP94:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 1
+// CHECK17-NEXT: store ptr null, ptr [[TMP94]], align 8
+// CHECK17-NEXT: [[TMP95:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP96:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS28:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP81]], align 4
-// CHECK17-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
-// CHECK17-NEXT: store i32 2, ptr [[TMP82]], align 4
-// CHECK17-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP79]], ptr [[TMP83]], align 8
-// CHECK17-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP80]], ptr [[TMP84]], align 8
-// CHECK17-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
-// CHECK17-NEXT: store ptr @.offload_sizes.36, ptr [[TMP85]], align 8
-// CHECK17-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP86]], align 8
-// CHECK17-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP87]], align 8
-// CHECK17-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP88]], align 8
-// CHECK17-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
-// CHECK17-NEXT: store i64 10, ptr [[TMP89]], align 8
-// CHECK17-NEXT: [[TMP90:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.region_id, ptr [[KERNEL_ARGS28]])
-// CHECK17-NEXT: [[TMP91:%.*]] = icmp ne i32 [[TMP90]], 0
-// CHECK17-NEXT: br i1 [[TMP91]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
+// CHECK17-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP97]], align 4
+// CHECK17-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
+// CHECK17-NEXT: store i32 2, ptr [[TMP98]], align 4
+// CHECK17-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP95]], ptr [[TMP99]], align 8
+// CHECK17-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP96]], ptr [[TMP100]], align 8
+// CHECK17-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
+// CHECK17-NEXT: store ptr @.offload_sizes.36, ptr [[TMP101]], align 8
+// CHECK17-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP102]], align 8
+// CHECK17-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP103]], align 8
+// CHECK17-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP104]], align 8
+// CHECK17-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
+// CHECK17-NEXT: store i64 10, ptr [[TMP105]], align 8
+// CHECK17-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP106]], align 8
+// CHECK17-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP107]], align 4
+// CHECK17-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP108]], align 4
+// CHECK17-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP109]], align 4
+// CHECK17-NEXT: [[TMP110:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.region_id, ptr [[KERNEL_ARGS28]])
+// CHECK17-NEXT: [[TMP111:%.*]] = icmp ne i32 [[TMP110]], 0
+// CHECK17-NEXT: br i1 [[TMP111]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
// CHECK17: omp_offload.failed29:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128(ptr [[A]], i64 [[TMP72]]) #[[ATTR3]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128(ptr [[A]], i64 [[TMP88]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT30]]
// CHECK17: omp_offload.cont30:
// CHECK17-NEXT: ret i32 0
@@ -11907,7 +12307,7 @@ int main (int argc, char **argv) {
// CHECK19-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK19-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK19-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK19-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK19-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK19-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK19-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -11924,302 +12324,342 @@ int main (int argc, char **argv) {
// CHECK19-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK19-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK19-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK19-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.region_id, ptr [[KERNEL_ARGS]])
-// CHECK19-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK19-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK19-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK19-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK19-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK19-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK19-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.region_id, ptr [[KERNEL_ARGS]])
+// CHECK19-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK19-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK19: omp_offload.failed:
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139(i32 [[TMP3]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK19: omp_offload.cont:
-// CHECK19-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK19-NEXT: [[TMP35:%.*]] = load i32, ptr [[N_CASTED3]], align 4
-// CHECK19-NEXT: [[TMP36:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK19-NEXT: [[TMP37:%.*]] = sext i32 [[TMP36]] to i64
+// CHECK19-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK19-NEXT: [[TMP39:%.*]] = load i32, ptr [[N_CASTED3]], align 4
+// CHECK19-NEXT: [[TMP40:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK19-NEXT: [[TMP41:%.*]] = sext i32 [[TMP40]] to i64
// CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES7]], ptr align 4 @.offload_sizes.4, i32 24, i1 false)
-// CHECK19-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP35]], ptr [[TMP38]], align 4
-// CHECK19-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP35]], ptr [[TMP39]], align 4
-// CHECK19-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP40]], align 4
-// CHECK19-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP41]], align 4
-// CHECK19-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP42]], align 4
-// CHECK19-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
-// CHECK19-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK19-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 4
-// CHECK19-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP45]], align 4
-// CHECK19-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK19-NEXT: store i64 [[TMP37]], ptr [[TMP46]], align 4
-// CHECK19-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK19-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP39]], ptr [[TMP42]], align 4
+// CHECK19-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP39]], ptr [[TMP43]], align 4
+// CHECK19-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP44]], align 4
+// CHECK19-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP45]], align 4
+// CHECK19-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP46]], align 4
+// CHECK19-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
// CHECK19-NEXT: store ptr null, ptr [[TMP47]], align 4
-// CHECK19-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP51:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP51]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK19-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK19-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP52]], 0
+// CHECK19-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 4
+// CHECK19-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP49]], align 4
+// CHECK19-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK19-NEXT: store i64 [[TMP41]], ptr [[TMP50]], align 4
+// CHECK19-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK19-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK19-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP54:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP55:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP55]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK19-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK19-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP56]], 0
// CHECK19-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK19-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK19-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK19-NEXT: [[TMP53:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK19-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP53]], 1
-// CHECK19-NEXT: [[TMP54:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK19-NEXT: [[TMP57:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK19-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP57]], 1
+// CHECK19-NEXT: [[TMP58:%.*]] = zext i32 [[ADD14]] to i64
// CHECK19-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP55]], align 4
-// CHECK19-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK19-NEXT: store i32 3, ptr [[TMP56]], align 4
-// CHECK19-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 4
-// CHECK19-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 4
-// CHECK19-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK19-NEXT: store ptr [[TMP50]], ptr [[TMP59]], align 4
-// CHECK19-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP60]], align 4
-// CHECK19-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP61]], align 4
-// CHECK19-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP62]], align 4
-// CHECK19-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK19-NEXT: store i64 [[TMP54]], ptr [[TMP63]], align 8
-// CHECK19-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK19-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
-// CHECK19-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK19-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP59]], align 4
+// CHECK19-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK19-NEXT: store i32 3, ptr [[TMP60]], align 4
+// CHECK19-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 4
+// CHECK19-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 4
+// CHECK19-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK19-NEXT: store ptr [[TMP54]], ptr [[TMP63]], align 4
+// CHECK19-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP64]], align 4
+// CHECK19-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP65]], align 4
+// CHECK19-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP66]], align 4
+// CHECK19-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK19-NEXT: store i64 [[TMP58]], ptr [[TMP67]], align 8
+// CHECK19-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP68]], align 8
+// CHECK19-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK19-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP70]], align 4
+// CHECK19-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP71]], align 4
+// CHECK19-NEXT: [[TMP72:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK19-NEXT: [[TMP73:%.*]] = icmp ne i32 [[TMP72]], 0
+// CHECK19-NEXT: br i1 [[TMP73]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK19: omp_offload.failed16:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143(i32 [[TMP35]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143(i32 [[TMP39]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK19: omp_offload.cont17:
-// CHECK19-NEXT: [[TMP66:%.*]] = load i32, ptr [[M]], align 4
-// CHECK19-NEXT: store i32 [[TMP66]], ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK19-NEXT: [[TMP67:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP67]], ptr [[N_CASTED19]], align 4
-// CHECK19-NEXT: [[TMP68:%.*]] = load i32, ptr [[N_CASTED19]], align 4
-// CHECK19-NEXT: [[TMP69:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK19-NEXT: store i32 [[TMP69]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK19-NEXT: [[TMP70:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK19-NEXT: [[TMP71:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK19-NEXT: [[TMP72:%.*]] = sext i32 [[TMP71]] to i64
+// CHECK19-NEXT: [[TMP74:%.*]] = load i32, ptr [[M]], align 4
+// CHECK19-NEXT: store i32 [[TMP74]], ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK19-NEXT: [[TMP75:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP75]], ptr [[N_CASTED19]], align 4
+// CHECK19-NEXT: [[TMP76:%.*]] = load i32, ptr [[N_CASTED19]], align 4
+// CHECK19-NEXT: [[TMP77:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK19-NEXT: store i32 [[TMP77]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK19-NEXT: [[TMP78:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK19-NEXT: [[TMP79:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK19-NEXT: [[TMP80:%.*]] = sext i32 [[TMP79]] to i64
// CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES23]], ptr align 4 @.offload_sizes.8, i32 32, i1 false)
-// CHECK19-NEXT: [[TMP73:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP68]], ptr [[TMP73]], align 4
-// CHECK19-NEXT: [[TMP74:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP68]], ptr [[TMP74]], align 4
-// CHECK19-NEXT: [[TMP75:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP75]], align 4
-// CHECK19-NEXT: [[TMP76:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP76]], align 4
-// CHECK19-NEXT: [[TMP77:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP77]], align 4
-// CHECK19-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 1
-// CHECK19-NEXT: store ptr null, ptr [[TMP78]], align 4
-// CHECK19-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP79]], align 4
-// CHECK19-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP80]], align 4
-// CHECK19-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
-// CHECK19-NEXT: store i64 [[TMP72]], ptr [[TMP81]], align 4
-// CHECK19-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 2
-// CHECK19-NEXT: store ptr null, ptr [[TMP82]], align 4
-// CHECK19-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
-// CHECK19-NEXT: store i32 [[TMP70]], ptr [[TMP83]], align 4
-// CHECK19-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
-// CHECK19-NEXT: store i32 [[TMP70]], ptr [[TMP84]], align 4
-// CHECK19-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 3
-// CHECK19-NEXT: store ptr null, ptr [[TMP85]], align 4
-// CHECK19-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP89:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP89]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK19-NEXT: [[TMP90:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK19-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP90]], 0
+// CHECK19-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP76]], ptr [[TMP81]], align 4
+// CHECK19-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP76]], ptr [[TMP82]], align 4
+// CHECK19-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP83]], align 4
+// CHECK19-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP84]], align 4
+// CHECK19-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP85]], align 4
+// CHECK19-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 1
+// CHECK19-NEXT: store ptr null, ptr [[TMP86]], align 4
+// CHECK19-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP87]], align 4
+// CHECK19-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP88]], align 4
+// CHECK19-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
+// CHECK19-NEXT: store i64 [[TMP80]], ptr [[TMP89]], align 4
+// CHECK19-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 2
+// CHECK19-NEXT: store ptr null, ptr [[TMP90]], align 4
+// CHECK19-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
+// CHECK19-NEXT: store i32 [[TMP78]], ptr [[TMP91]], align 4
+// CHECK19-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
+// CHECK19-NEXT: store i32 [[TMP78]], ptr [[TMP92]], align 4
+// CHECK19-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 3
+// CHECK19-NEXT: store ptr null, ptr [[TMP93]], align 4
+// CHECK19-NEXT: [[TMP94:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP95:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP96:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP97:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP97]], ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK19-NEXT: [[TMP98:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK19-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP98]], 0
// CHECK19-NEXT: [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1
// CHECK19-NEXT: [[SUB29:%.*]] = sub nsw i32 [[DIV28]], 1
// CHECK19-NEXT: store i32 [[SUB29]], ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK19-NEXT: [[TMP91:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK19-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP91]], 1
-// CHECK19-NEXT: [[TMP92:%.*]] = zext i32 [[ADD30]] to i64
+// CHECK19-NEXT: [[TMP99:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
+// CHECK19-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP99]], 1
+// CHECK19-NEXT: [[TMP100:%.*]] = zext i32 [[ADD30]] to i64
// CHECK19-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP93]], align 4
-// CHECK19-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK19-NEXT: store i32 4, ptr [[TMP94]], align 4
-// CHECK19-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP86]], ptr [[TMP95]], align 4
-// CHECK19-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP87]], ptr [[TMP96]], align 4
-// CHECK19-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK19-NEXT: store ptr [[TMP88]], ptr [[TMP97]], align 4
-// CHECK19-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP98]], align 4
-// CHECK19-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP99]], align 4
-// CHECK19-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP100]], align 4
-// CHECK19-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK19-NEXT: store i64 [[TMP92]], ptr [[TMP101]], align 8
-// CHECK19-NEXT: [[TMP102:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK19-NEXT: [[TMP103:%.*]] = icmp ne i32 [[TMP102]], 0
-// CHECK19-NEXT: br i1 [[TMP103]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK19-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP101]], align 4
+// CHECK19-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK19-NEXT: store i32 4, ptr [[TMP102]], align 4
+// CHECK19-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP94]], ptr [[TMP103]], align 4
+// CHECK19-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP95]], ptr [[TMP104]], align 4
+// CHECK19-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK19-NEXT: store ptr [[TMP96]], ptr [[TMP105]], align 4
+// CHECK19-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP106]], align 4
+// CHECK19-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP107]], align 4
+// CHECK19-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP108]], align 4
+// CHECK19-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK19-NEXT: store i64 [[TMP100]], ptr [[TMP109]], align 8
+// CHECK19-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP110]], align 8
+// CHECK19-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP111]], align 4
+// CHECK19-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP112]], align 4
+// CHECK19-NEXT: [[TMP113:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP113]], align 4
+// CHECK19-NEXT: [[TMP114:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK19-NEXT: [[TMP115:%.*]] = icmp ne i32 [[TMP114]], 0
+// CHECK19-NEXT: br i1 [[TMP115]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK19: omp_offload.failed32:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147(i32 [[TMP68]], i32 [[TMP0]], ptr [[VLA]], i32 [[TMP70]]) #[[ATTR3]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147(i32 [[TMP76]], i32 [[TMP0]], ptr [[VLA]], i32 [[TMP78]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK19: omp_offload.cont33:
-// CHECK19-NEXT: [[TMP104:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP104]], ptr [[N_CASTED34]], align 4
-// CHECK19-NEXT: [[TMP105:%.*]] = load i32, ptr [[N_CASTED34]], align 4
-// CHECK19-NEXT: [[TMP106:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK19-NEXT: [[TMP107:%.*]] = sext i32 [[TMP106]] to i64
+// CHECK19-NEXT: [[TMP116:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP116]], ptr [[N_CASTED34]], align 4
+// CHECK19-NEXT: [[TMP117:%.*]] = load i32, ptr [[N_CASTED34]], align 4
+// CHECK19-NEXT: [[TMP118:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK19-NEXT: [[TMP119:%.*]] = sext i32 [[TMP118]] to i64
// CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES38]], ptr align 4 @.offload_sizes.12, i32 24, i1 false)
-// CHECK19-NEXT: [[TMP108:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP105]], ptr [[TMP108]], align 4
-// CHECK19-NEXT: [[TMP109:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP105]], ptr [[TMP109]], align 4
-// CHECK19-NEXT: [[TMP110:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP110]], align 4
-// CHECK19-NEXT: [[TMP111:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP111]], align 4
-// CHECK19-NEXT: [[TMP112:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP112]], align 4
-// CHECK19-NEXT: [[TMP113:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 1
-// CHECK19-NEXT: store ptr null, ptr [[TMP113]], align 4
-// CHECK19-NEXT: [[TMP114:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP114]], align 4
-// CHECK19-NEXT: [[TMP115:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP115]], align 4
-// CHECK19-NEXT: [[TMP116:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 2
-// CHECK19-NEXT: store i64 [[TMP107]], ptr [[TMP116]], align 4
-// CHECK19-NEXT: [[TMP117:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 2
-// CHECK19-NEXT: store ptr null, ptr [[TMP117]], align 4
-// CHECK19-NEXT: [[TMP118:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP119:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP120:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP121:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP121]], ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK19-NEXT: [[TMP122:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK19-NEXT: [[SUB42:%.*]] = sub nsw i32 [[TMP122]], 0
+// CHECK19-NEXT: [[TMP120:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP117]], ptr [[TMP120]], align 4
+// CHECK19-NEXT: [[TMP121:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP117]], ptr [[TMP121]], align 4
+// CHECK19-NEXT: [[TMP122:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP122]], align 4
+// CHECK19-NEXT: [[TMP123:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP123]], align 4
+// CHECK19-NEXT: [[TMP124:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP124]], align 4
+// CHECK19-NEXT: [[TMP125:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 1
+// CHECK19-NEXT: store ptr null, ptr [[TMP125]], align 4
+// CHECK19-NEXT: [[TMP126:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP126]], align 4
+// CHECK19-NEXT: [[TMP127:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP127]], align 4
+// CHECK19-NEXT: [[TMP128:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 2
+// CHECK19-NEXT: store i64 [[TMP119]], ptr [[TMP128]], align 4
+// CHECK19-NEXT: [[TMP129:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 2
+// CHECK19-NEXT: store ptr null, ptr [[TMP129]], align 4
+// CHECK19-NEXT: [[TMP130:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP131:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP132:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP133:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP133]], ptr [[DOTCAPTURE_EXPR_40]], align 4
+// CHECK19-NEXT: [[TMP134:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
+// CHECK19-NEXT: [[SUB42:%.*]] = sub nsw i32 [[TMP134]], 0
// CHECK19-NEXT: [[DIV43:%.*]] = sdiv i32 [[SUB42]], 1
// CHECK19-NEXT: [[SUB44:%.*]] = sub nsw i32 [[DIV43]], 1
// CHECK19-NEXT: store i32 [[SUB44]], ptr [[DOTCAPTURE_EXPR_41]], align 4
-// CHECK19-NEXT: [[TMP123:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_41]], align 4
-// CHECK19-NEXT: [[ADD45:%.*]] = add nsw i32 [[TMP123]], 1
-// CHECK19-NEXT: [[TMP124:%.*]] = zext i32 [[ADD45]] to i64
+// CHECK19-NEXT: [[TMP135:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_41]], align 4
+// CHECK19-NEXT: [[ADD45:%.*]] = add nsw i32 [[TMP135]], 1
+// CHECK19-NEXT: [[TMP136:%.*]] = zext i32 [[ADD45]] to i64
// CHECK19-NEXT: [[KERNEL_ARGS46:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP125]], align 4
-// CHECK19-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 1
-// CHECK19-NEXT: store i32 3, ptr [[TMP126]], align 4
-// CHECK19-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP118]], ptr [[TMP127]], align 4
-// CHECK19-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP119]], ptr [[TMP128]], align 4
-// CHECK19-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 4
-// CHECK19-NEXT: store ptr [[TMP120]], ptr [[TMP129]], align 4
-// CHECK19-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP130]], align 4
-// CHECK19-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP131]], align 4
-// CHECK19-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP132]], align 4
-// CHECK19-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 8
-// CHECK19-NEXT: store i64 [[TMP124]], ptr [[TMP133]], align 8
-// CHECK19-NEXT: [[TMP134:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.region_id, ptr [[KERNEL_ARGS46]])
-// CHECK19-NEXT: [[TMP135:%.*]] = icmp ne i32 [[TMP134]], 0
-// CHECK19-NEXT: br i1 [[TMP135]], label [[OMP_OFFLOAD_FAILED47:%.*]], label [[OMP_OFFLOAD_CONT48:%.*]]
+// CHECK19-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP137]], align 4
+// CHECK19-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 1
+// CHECK19-NEXT: store i32 3, ptr [[TMP138]], align 4
+// CHECK19-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP130]], ptr [[TMP139]], align 4
+// CHECK19-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP131]], ptr [[TMP140]], align 4
+// CHECK19-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 4
+// CHECK19-NEXT: store ptr [[TMP132]], ptr [[TMP141]], align 4
+// CHECK19-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP142]], align 4
+// CHECK19-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP143]], align 4
+// CHECK19-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP144]], align 4
+// CHECK19-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 8
+// CHECK19-NEXT: store i64 [[TMP136]], ptr [[TMP145]], align 8
+// CHECK19-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP146]], align 8
+// CHECK19-NEXT: [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP147]], align 4
+// CHECK19-NEXT: [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP148]], align 4
+// CHECK19-NEXT: [[TMP149:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP149]], align 4
+// CHECK19-NEXT: [[TMP150:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.region_id, ptr [[KERNEL_ARGS46]])
+// CHECK19-NEXT: [[TMP151:%.*]] = icmp ne i32 [[TMP150]], 0
+// CHECK19-NEXT: br i1 [[TMP151]], label [[OMP_OFFLOAD_FAILED47:%.*]], label [[OMP_OFFLOAD_CONT48:%.*]]
// CHECK19: omp_offload.failed47:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151(i32 [[TMP105]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151(i32 [[TMP117]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT48]]
// CHECK19: omp_offload.cont48:
-// CHECK19-NEXT: [[TMP136:%.*]] = load i32, ptr [[M]], align 4
-// CHECK19-NEXT: store i32 [[TMP136]], ptr [[DOTCAPTURE_EXPR_49]], align 4
-// CHECK19-NEXT: [[TMP137:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP137]], ptr [[N_CASTED50]], align 4
-// CHECK19-NEXT: [[TMP138:%.*]] = load i32, ptr [[N_CASTED50]], align 4
-// CHECK19-NEXT: [[TMP139:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_49]], align 4
-// CHECK19-NEXT: store i32 [[TMP139]], ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4
-// CHECK19-NEXT: [[TMP140:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4
-// CHECK19-NEXT: [[TMP141:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK19-NEXT: [[TMP142:%.*]] = sext i32 [[TMP141]] to i64
+// CHECK19-NEXT: [[TMP152:%.*]] = load i32, ptr [[M]], align 4
+// CHECK19-NEXT: store i32 [[TMP152]], ptr [[DOTCAPTURE_EXPR_49]], align 4
+// CHECK19-NEXT: [[TMP153:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP153]], ptr [[N_CASTED50]], align 4
+// CHECK19-NEXT: [[TMP154:%.*]] = load i32, ptr [[N_CASTED50]], align 4
+// CHECK19-NEXT: [[TMP155:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_49]], align 4
+// CHECK19-NEXT: store i32 [[TMP155]], ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4
+// CHECK19-NEXT: [[TMP156:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4
+// CHECK19-NEXT: [[TMP157:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK19-NEXT: [[TMP158:%.*]] = sext i32 [[TMP157]] to i64
// CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES55]], ptr align 4 @.offload_sizes.16, i32 32, i1 false)
-// CHECK19-NEXT: [[TMP143:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP138]], ptr [[TMP143]], align 4
-// CHECK19-NEXT: [[TMP144:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP138]], ptr [[TMP144]], align 4
-// CHECK19-NEXT: [[TMP145:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP145]], align 4
-// CHECK19-NEXT: [[TMP146:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP146]], align 4
-// CHECK19-NEXT: [[TMP147:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP147]], align 4
-// CHECK19-NEXT: [[TMP148:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 1
-// CHECK19-NEXT: store ptr null, ptr [[TMP148]], align 4
-// CHECK19-NEXT: [[TMP149:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP149]], align 4
-// CHECK19-NEXT: [[TMP150:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP150]], align 4
-// CHECK19-NEXT: [[TMP151:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 2
-// CHECK19-NEXT: store i64 [[TMP142]], ptr [[TMP151]], align 4
-// CHECK19-NEXT: [[TMP152:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 2
-// CHECK19-NEXT: store ptr null, ptr [[TMP152]], align 4
-// CHECK19-NEXT: [[TMP153:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 3
-// CHECK19-NEXT: store i32 [[TMP140]], ptr [[TMP153]], align 4
-// CHECK19-NEXT: [[TMP154:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 3
-// CHECK19-NEXT: store i32 [[TMP140]], ptr [[TMP154]], align 4
-// CHECK19-NEXT: [[TMP155:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 3
-// CHECK19-NEXT: store ptr null, ptr [[TMP155]], align 4
-// CHECK19-NEXT: [[TMP156:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP157:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP158:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP159:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP159]], ptr [[DOTCAPTURE_EXPR_57]], align 4
-// CHECK19-NEXT: [[TMP160:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_57]], align 4
-// CHECK19-NEXT: [[SUB59:%.*]] = sub nsw i32 [[TMP160]], 0
+// CHECK19-NEXT: [[TMP159:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP154]], ptr [[TMP159]], align 4
+// CHECK19-NEXT: [[TMP160:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP154]], ptr [[TMP160]], align 4
+// CHECK19-NEXT: [[TMP161:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP161]], align 4
+// CHECK19-NEXT: [[TMP162:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP162]], align 4
+// CHECK19-NEXT: [[TMP163:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP163]], align 4
+// CHECK19-NEXT: [[TMP164:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 1
+// CHECK19-NEXT: store ptr null, ptr [[TMP164]], align 4
+// CHECK19-NEXT: [[TMP165:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP165]], align 4
+// CHECK19-NEXT: [[TMP166:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP166]], align 4
+// CHECK19-NEXT: [[TMP167:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 2
+// CHECK19-NEXT: store i64 [[TMP158]], ptr [[TMP167]], align 4
+// CHECK19-NEXT: [[TMP168:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 2
+// CHECK19-NEXT: store ptr null, ptr [[TMP168]], align 4
+// CHECK19-NEXT: [[TMP169:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 3
+// CHECK19-NEXT: store i32 [[TMP156]], ptr [[TMP169]], align 4
+// CHECK19-NEXT: [[TMP170:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 3
+// CHECK19-NEXT: store i32 [[TMP156]], ptr [[TMP170]], align 4
+// CHECK19-NEXT: [[TMP171:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 3
+// CHECK19-NEXT: store ptr null, ptr [[TMP171]], align 4
+// CHECK19-NEXT: [[TMP172:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP173:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP174:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP175:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP175]], ptr [[DOTCAPTURE_EXPR_57]], align 4
+// CHECK19-NEXT: [[TMP176:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_57]], align 4
+// CHECK19-NEXT: [[SUB59:%.*]] = sub nsw i32 [[TMP176]], 0
// CHECK19-NEXT: [[DIV60:%.*]] = sdiv i32 [[SUB59]], 1
// CHECK19-NEXT: [[SUB61:%.*]] = sub nsw i32 [[DIV60]], 1
// CHECK19-NEXT: store i32 [[SUB61]], ptr [[DOTCAPTURE_EXPR_58]], align 4
-// CHECK19-NEXT: [[TMP161:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_58]], align 4
-// CHECK19-NEXT: [[ADD62:%.*]] = add nsw i32 [[TMP161]], 1
-// CHECK19-NEXT: [[TMP162:%.*]] = zext i32 [[ADD62]] to i64
+// CHECK19-NEXT: [[TMP177:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_58]], align 4
+// CHECK19-NEXT: [[ADD62:%.*]] = add nsw i32 [[TMP177]], 1
+// CHECK19-NEXT: [[TMP178:%.*]] = zext i32 [[ADD62]] to i64
// CHECK19-NEXT: [[KERNEL_ARGS63:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP163:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP163]], align 4
-// CHECK19-NEXT: [[TMP164:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 1
-// CHECK19-NEXT: store i32 4, ptr [[TMP164]], align 4
-// CHECK19-NEXT: [[TMP165:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP156]], ptr [[TMP165]], align 4
-// CHECK19-NEXT: [[TMP166:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP157]], ptr [[TMP166]], align 4
-// CHECK19-NEXT: [[TMP167:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 4
-// CHECK19-NEXT: store ptr [[TMP158]], ptr [[TMP167]], align 4
-// CHECK19-NEXT: [[TMP168:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP168]], align 4
-// CHECK19-NEXT: [[TMP169:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP169]], align 4
-// CHECK19-NEXT: [[TMP170:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP170]], align 4
-// CHECK19-NEXT: [[TMP171:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 8
-// CHECK19-NEXT: store i64 [[TMP162]], ptr [[TMP171]], align 8
-// CHECK19-NEXT: [[TMP172:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.region_id, ptr [[KERNEL_ARGS63]])
-// CHECK19-NEXT: [[TMP173:%.*]] = icmp ne i32 [[TMP172]], 0
-// CHECK19-NEXT: br i1 [[TMP173]], label [[OMP_OFFLOAD_FAILED64:%.*]], label [[OMP_OFFLOAD_CONT65:%.*]]
+// CHECK19-NEXT: [[TMP179:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP179]], align 4
+// CHECK19-NEXT: [[TMP180:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 1
+// CHECK19-NEXT: store i32 4, ptr [[TMP180]], align 4
+// CHECK19-NEXT: [[TMP181:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP172]], ptr [[TMP181]], align 4
+// CHECK19-NEXT: [[TMP182:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP173]], ptr [[TMP182]], align 4
+// CHECK19-NEXT: [[TMP183:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 4
+// CHECK19-NEXT: store ptr [[TMP174]], ptr [[TMP183]], align 4
+// CHECK19-NEXT: [[TMP184:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP184]], align 4
+// CHECK19-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP185]], align 4
+// CHECK19-NEXT: [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP186]], align 4
+// CHECK19-NEXT: [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 8
+// CHECK19-NEXT: store i64 [[TMP178]], ptr [[TMP187]], align 8
+// CHECK19-NEXT: [[TMP188:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP188]], align 8
+// CHECK19-NEXT: [[TMP189:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP189]], align 4
+// CHECK19-NEXT: [[TMP190:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP190]], align 4
+// CHECK19-NEXT: [[TMP191:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP191]], align 4
+// CHECK19-NEXT: [[TMP192:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.region_id, ptr [[KERNEL_ARGS63]])
+// CHECK19-NEXT: [[TMP193:%.*]] = icmp ne i32 [[TMP192]], 0
+// CHECK19-NEXT: br i1 [[TMP193]], label [[OMP_OFFLOAD_FAILED64:%.*]], label [[OMP_OFFLOAD_CONT65:%.*]]
// CHECK19: omp_offload.failed64:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155(i32 [[TMP138]], i32 [[TMP0]], ptr [[VLA]], i32 [[TMP140]]) #[[ATTR3]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155(i32 [[TMP154]], i32 [[TMP0]], ptr [[VLA]], i32 [[TMP156]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT65]]
// CHECK19: omp_offload.cont65:
-// CHECK19-NEXT: [[TMP174:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK19-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP174]])
+// CHECK19-NEXT: [[TMP194:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK19-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP194]])
// CHECK19-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK19-NEXT: [[TMP175:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK19-NEXT: call void @llvm.stackrestore(ptr [[TMP175]])
-// CHECK19-NEXT: [[TMP176:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK19-NEXT: ret i32 [[TMP176]]
+// CHECK19-NEXT: [[TMP195:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK19-NEXT: call void @llvm.stackrestore(ptr [[TMP195]])
+// CHECK19-NEXT: [[TMP196:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK19-NEXT: ret i32 [[TMP196]]
//
//
// CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139
@@ -13370,7 +13810,7 @@ int main (int argc, char **argv) {
// CHECK19-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK19-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK19-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK19-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK19-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK19-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -13387,169 +13827,209 @@ int main (int argc, char **argv) {
// CHECK19-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK19-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK19-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK19-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.region_id, ptr [[KERNEL_ARGS]])
-// CHECK19-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK19-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK19-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK19-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK19-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK19-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK19-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.region_id, ptr [[KERNEL_ARGS]])
+// CHECK19-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK19-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK19: omp_offload.failed:
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112(ptr [[A]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK19: omp_offload.cont:
-// CHECK19-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP16]], align 4
-// CHECK19-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP17]], align 4
-// CHECK19-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK19-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP20]], align 4
+// CHECK19-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP21]], align 4
+// CHECK19-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK19-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK19-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK19-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK19-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK19-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK19-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK19-NEXT: store ptr @.offload_sizes.24, ptr [[TMP25]], align 4
-// CHECK19-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP26]], align 4
-// CHECK19-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK19-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK19-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK19-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK19-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK19-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK19-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK19-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK19-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK19-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK19-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK19-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK19-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK19-NEXT: store ptr @.offload_sizes.24, ptr [[TMP29]], align 4
+// CHECK19-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP30]], align 4
+// CHECK19-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK19-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK19-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK19-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK19-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK19-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK19-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK19-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK19-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK19-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK19-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK19: omp_offload.failed6:
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116(ptr [[A]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK19: omp_offload.cont7:
-// CHECK19-NEXT: [[TMP32:%.*]] = load i32, ptr [[M]], align 4
-// CHECK19-NEXT: store i32 [[TMP32]], ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK19-NEXT: [[TMP33:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK19-NEXT: store i32 [[TMP33]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK19-NEXT: [[TMP34:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK19-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP35]], align 4
-// CHECK19-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP36]], align 4
-// CHECK19-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP37]], align 4
-// CHECK19-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP34]], ptr [[TMP38]], align 4
-// CHECK19-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP34]], ptr [[TMP39]], align 4
-// CHECK19-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1
-// CHECK19-NEXT: store ptr null, ptr [[TMP40]], align 4
-// CHECK19-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP40:%.*]] = load i32, ptr [[M]], align 4
+// CHECK19-NEXT: store i32 [[TMP40]], ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK19-NEXT: [[TMP41:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK19-NEXT: store i32 [[TMP41]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK19-NEXT: [[TMP42:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK19-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP43]], align 4
+// CHECK19-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP44]], align 4
+// CHECK19-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP45]], align 4
+// CHECK19-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP42]], ptr [[TMP46]], align 4
+// CHECK19-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP42]], ptr [[TMP47]], align 4
+// CHECK19-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1
+// CHECK19-NEXT: store ptr null, ptr [[TMP48]], align 4
+// CHECK19-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP43]], align 4
-// CHECK19-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK19-NEXT: store i32 2, ptr [[TMP44]], align 4
-// CHECK19-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP41]], ptr [[TMP45]], align 4
-// CHECK19-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP42]], ptr [[TMP46]], align 4
-// CHECK19-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK19-NEXT: store ptr @.offload_sizes.28, ptr [[TMP47]], align 4
-// CHECK19-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP48]], align 4
-// CHECK19-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP49]], align 4
-// CHECK19-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP50]], align 4
-// CHECK19-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK19-NEXT: store i64 10, ptr [[TMP51]], align 8
-// CHECK19-NEXT: [[TMP52:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK19-NEXT: [[TMP53:%.*]] = icmp ne i32 [[TMP52]], 0
-// CHECK19-NEXT: br i1 [[TMP53]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK19-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP51]], align 4
+// CHECK19-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK19-NEXT: store i32 2, ptr [[TMP52]], align 4
+// CHECK19-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 4
+// CHECK19-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP50]], ptr [[TMP54]], align 4
+// CHECK19-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK19-NEXT: store ptr @.offload_sizes.28, ptr [[TMP55]], align 4
+// CHECK19-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP56]], align 4
+// CHECK19-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP57]], align 4
+// CHECK19-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP58]], align 4
+// CHECK19-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK19-NEXT: store i64 10, ptr [[TMP59]], align 8
+// CHECK19-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP60]], align 8
+// CHECK19-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4
+// CHECK19-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP62]], align 4
+// CHECK19-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP63]], align 4
+// CHECK19-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK19-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
+// CHECK19-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK19: omp_offload.failed13:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120(ptr [[A]], i32 [[TMP34]]) #[[ATTR3]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120(ptr [[A]], i32 [[TMP42]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT14]]
// CHECK19: omp_offload.cont14:
-// CHECK19-NEXT: [[TMP54:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP54]], align 4
-// CHECK19-NEXT: [[TMP55:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP55]], align 4
-// CHECK19-NEXT: [[TMP56:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP56]], align 4
-// CHECK19-NEXT: [[TMP57:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP58:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP66]], align 4
+// CHECK19-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP67]], align 4
+// CHECK19-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP68]], align 4
+// CHECK19-NEXT: [[TMP69:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP70:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS19:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP59]], align 4
-// CHECK19-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
-// CHECK19-NEXT: store i32 1, ptr [[TMP60]], align 4
-// CHECK19-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP57]], ptr [[TMP61]], align 4
-// CHECK19-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP58]], ptr [[TMP62]], align 4
-// CHECK19-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
-// CHECK19-NEXT: store ptr @.offload_sizes.32, ptr [[TMP63]], align 4
-// CHECK19-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP64]], align 4
-// CHECK19-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP65]], align 4
-// CHECK19-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP66]], align 4
-// CHECK19-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
-// CHECK19-NEXT: store i64 10, ptr [[TMP67]], align 8
-// CHECK19-NEXT: [[TMP68:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.region_id, ptr [[KERNEL_ARGS19]])
-// CHECK19-NEXT: [[TMP69:%.*]] = icmp ne i32 [[TMP68]], 0
-// CHECK19-NEXT: br i1 [[TMP69]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
+// CHECK19-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP71]], align 4
+// CHECK19-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
+// CHECK19-NEXT: store i32 1, ptr [[TMP72]], align 4
+// CHECK19-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP69]], ptr [[TMP73]], align 4
+// CHECK19-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP70]], ptr [[TMP74]], align 4
+// CHECK19-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
+// CHECK19-NEXT: store ptr @.offload_sizes.32, ptr [[TMP75]], align 4
+// CHECK19-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP76]], align 4
+// CHECK19-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP77]], align 4
+// CHECK19-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP78]], align 4
+// CHECK19-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
+// CHECK19-NEXT: store i64 10, ptr [[TMP79]], align 8
+// CHECK19-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP80]], align 8
+// CHECK19-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP81]], align 4
+// CHECK19-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP82]], align 4
+// CHECK19-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP83]], align 4
+// CHECK19-NEXT: [[TMP84:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.region_id, ptr [[KERNEL_ARGS19]])
+// CHECK19-NEXT: [[TMP85:%.*]] = icmp ne i32 [[TMP84]], 0
+// CHECK19-NEXT: br i1 [[TMP85]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
// CHECK19: omp_offload.failed20:
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124(ptr [[A]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT21]]
// CHECK19: omp_offload.cont21:
-// CHECK19-NEXT: [[TMP70:%.*]] = load i32, ptr [[M]], align 4
-// CHECK19-NEXT: store i32 [[TMP70]], ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK19-NEXT: [[TMP71:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK19-NEXT: store i32 [[TMP71]], ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4
-// CHECK19-NEXT: [[TMP72:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4
-// CHECK19-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP73]], align 4
-// CHECK19-NEXT: [[TMP74:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP74]], align 4
-// CHECK19-NEXT: [[TMP75:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP75]], align 4
-// CHECK19-NEXT: [[TMP76:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP72]], ptr [[TMP76]], align 4
-// CHECK19-NEXT: [[TMP77:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP72]], ptr [[TMP77]], align 4
-// CHECK19-NEXT: [[TMP78:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 1
-// CHECK19-NEXT: store ptr null, ptr [[TMP78]], align 4
-// CHECK19-NEXT: [[TMP79:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP80:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP86:%.*]] = load i32, ptr [[M]], align 4
+// CHECK19-NEXT: store i32 [[TMP86]], ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK19-NEXT: [[TMP87:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK19-NEXT: store i32 [[TMP87]], ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4
+// CHECK19-NEXT: [[TMP88:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4
+// CHECK19-NEXT: [[TMP89:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP89]], align 4
+// CHECK19-NEXT: [[TMP90:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP90]], align 4
+// CHECK19-NEXT: [[TMP91:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP91]], align 4
+// CHECK19-NEXT: [[TMP92:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP88]], ptr [[TMP92]], align 4
+// CHECK19-NEXT: [[TMP93:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP88]], ptr [[TMP93]], align 4
+// CHECK19-NEXT: [[TMP94:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 1
+// CHECK19-NEXT: store ptr null, ptr [[TMP94]], align 4
+// CHECK19-NEXT: [[TMP95:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP96:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS28:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP81]], align 4
-// CHECK19-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
-// CHECK19-NEXT: store i32 2, ptr [[TMP82]], align 4
-// CHECK19-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP79]], ptr [[TMP83]], align 4
-// CHECK19-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP80]], ptr [[TMP84]], align 4
-// CHECK19-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
-// CHECK19-NEXT: store ptr @.offload_sizes.36, ptr [[TMP85]], align 4
-// CHECK19-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP86]], align 4
-// CHECK19-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP87]], align 4
-// CHECK19-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP88]], align 4
-// CHECK19-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
-// CHECK19-NEXT: store i64 10, ptr [[TMP89]], align 8
-// CHECK19-NEXT: [[TMP90:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.region_id, ptr [[KERNEL_ARGS28]])
-// CHECK19-NEXT: [[TMP91:%.*]] = icmp ne i32 [[TMP90]], 0
-// CHECK19-NEXT: br i1 [[TMP91]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
+// CHECK19-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP97]], align 4
+// CHECK19-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
+// CHECK19-NEXT: store i32 2, ptr [[TMP98]], align 4
+// CHECK19-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP95]], ptr [[TMP99]], align 4
+// CHECK19-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP96]], ptr [[TMP100]], align 4
+// CHECK19-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
+// CHECK19-NEXT: store ptr @.offload_sizes.36, ptr [[TMP101]], align 4
+// CHECK19-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP102]], align 4
+// CHECK19-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP103]], align 4
+// CHECK19-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP104]], align 4
+// CHECK19-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
+// CHECK19-NEXT: store i64 10, ptr [[TMP105]], align 8
+// CHECK19-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP106]], align 8
+// CHECK19-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP107]], align 4
+// CHECK19-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP108]], align 4
+// CHECK19-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP109]], align 4
+// CHECK19-NEXT: [[TMP110:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.region_id, ptr [[KERNEL_ARGS28]])
+// CHECK19-NEXT: [[TMP111:%.*]] = icmp ne i32 [[TMP110]], 0
+// CHECK19-NEXT: br i1 [[TMP111]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
// CHECK19: omp_offload.failed29:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128(ptr [[A]], i32 [[TMP72]]) #[[ATTR3]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128(ptr [[A]], i32 [[TMP88]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT30]]
// CHECK19: omp_offload.cont30:
// CHECK19-NEXT: ret i32 0
diff --git a/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_codegen.cpp
index c4bd3098b768d..89bb85a3f3e86 100644
--- a/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_codegen.cpp
@@ -164,95 +164,113 @@ int target_teams_fun(int *g){
// CHECK1-NEXT: [[TMP32:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_4]], align 4
// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP32]], 1
// CHECK1-NEXT: [[TMP33:%.*]] = zext i32 [[ADD]] to i64
+// CHECK1-NEXT: [[TMP34:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP28]], 0
+// CHECK1-NEXT: [[TMP35:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP29]], 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP34]], align 4
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK1-NEXT: store i32 5, ptr [[TMP35]], align 4
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP26]], ptr [[TMP36]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP27]], ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP38]], align 8
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP41]], align 8
-// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK1-NEXT: store i64 [[TMP33]], ptr [[TMP42]], align 8
-// CHECK1-NEXT: [[TMP43:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 [[TMP28]], i32 [[TMP29]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l50.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP44:%.*]] = icmp ne i32 [[TMP43]], 0
-// CHECK1-NEXT: br i1 [[TMP44]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP36]], align 4
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK1-NEXT: store i32 5, ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP26]], ptr [[TMP38]], align 8
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP27]], ptr [[TMP39]], align 8
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK1-NEXT: store i64 [[TMP33]], ptr [[TMP44]], align 8
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP45]], align 8
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [[TMP34]], ptr [[TMP46]], align 4
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP35]], ptr [[TMP47]], align 4
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP48]], align 4
+// CHECK1-NEXT: [[TMP49:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 [[TMP28]], i32 [[TMP29]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l50.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP50:%.*]] = icmp ne i32 [[TMP49]], 0
+// CHECK1-NEXT: br i1 [[TMP50]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l50(i64 [[TMP4]], i64 [[TMP6]], ptr [[A]], i64 [[TMP8]], i64 [[TMP10]]) #[[ATTR4:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
-// CHECK1-NEXT: [[TMP45:%.*]] = load i32, ptr [[N]], align 4
-// CHECK1-NEXT: store i32 [[TMP45]], ptr [[N_CASTED7]], align 4
-// CHECK1-NEXT: [[TMP46:%.*]] = load i64, ptr [[N_CASTED7]], align 8
-// CHECK1-NEXT: [[TMP47:%.*]] = load ptr, ptr [[G_ADDR]], align 8
-// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP46]], ptr [[TMP48]], align 8
-// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP46]], ptr [[TMP49]], align 8
-// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP50]], align 8
-// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
-// CHECK1-NEXT: store ptr [[A]], ptr [[TMP51]], align 8
-// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
-// CHECK1-NEXT: store ptr [[A]], ptr [[TMP52]], align 8
-// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP53]], align 8
-// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP47]], ptr [[TMP54]], align 8
-// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP47]], ptr [[TMP55]], align 8
-// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 2
+// CHECK1-NEXT: [[TMP51:%.*]] = load i32, ptr [[N]], align 4
+// CHECK1-NEXT: store i32 [[TMP51]], ptr [[N_CASTED7]], align 4
+// CHECK1-NEXT: [[TMP52:%.*]] = load i64, ptr [[N_CASTED7]], align 8
+// CHECK1-NEXT: [[TMP53:%.*]] = load ptr, ptr [[G_ADDR]], align 8
+// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP52]], ptr [[TMP54]], align 8
+// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP52]], ptr [[TMP55]], align 8
+// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
// CHECK1-NEXT: store ptr null, ptr [[TMP56]], align 8
-// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP58:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP59:%.*]] = load i32, ptr [[N]], align 4
-// CHECK1-NEXT: store i32 [[TMP59]], ptr [[DOTCAPTURE_EXPR_12]], align 4
-// CHECK1-NEXT: [[TMP60:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_12]], align 4
-// CHECK1-NEXT: [[SUB14:%.*]] = sub nsw i32 [[TMP60]], 0
+// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
+// CHECK1-NEXT: store ptr [[A]], ptr [[TMP57]], align 8
+// CHECK1-NEXT: [[TMP58:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
+// CHECK1-NEXT: store ptr [[A]], ptr [[TMP58]], align 8
+// CHECK1-NEXT: [[TMP59:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP59]], align 8
+// CHECK1-NEXT: [[TMP60:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP53]], ptr [[TMP60]], align 8
+// CHECK1-NEXT: [[TMP61:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP53]], ptr [[TMP61]], align 8
+// CHECK1-NEXT: [[TMP62:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 2
+// CHECK1-NEXT: store ptr null, ptr [[TMP62]], align 8
+// CHECK1-NEXT: [[TMP63:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP64:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP65:%.*]] = load i32, ptr [[N]], align 4
+// CHECK1-NEXT: store i32 [[TMP65]], ptr [[DOTCAPTURE_EXPR_12]], align 4
+// CHECK1-NEXT: [[TMP66:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_12]], align 4
+// CHECK1-NEXT: [[SUB14:%.*]] = sub nsw i32 [[TMP66]], 0
// CHECK1-NEXT: [[DIV15:%.*]] = sdiv i32 [[SUB14]], 1
// CHECK1-NEXT: [[SUB16:%.*]] = sub nsw i32 [[DIV15]], 1
// CHECK1-NEXT: store i32 [[SUB16]], ptr [[DOTCAPTURE_EXPR_13]], align 4
-// CHECK1-NEXT: [[TMP61:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_13]], align 4
-// CHECK1-NEXT: [[ADD17:%.*]] = add nsw i32 [[TMP61]], 1
-// CHECK1-NEXT: [[TMP62:%.*]] = zext i32 [[ADD17]] to i64
+// CHECK1-NEXT: [[TMP67:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_13]], align 4
+// CHECK1-NEXT: [[ADD17:%.*]] = add nsw i32 [[TMP67]], 1
+// CHECK1-NEXT: [[TMP68:%.*]] = zext i32 [[ADD17]] to i64
// CHECK1-NEXT: [[KERNEL_ARGS18:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP63]], align 4
-// CHECK1-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 1
-// CHECK1-NEXT: store i32 3, ptr [[TMP64]], align 4
-// CHECK1-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP57]], ptr [[TMP65]], align 8
-// CHECK1-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP58]], ptr [[TMP66]], align 8
-// CHECK1-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.4, ptr [[TMP67]], align 8
-// CHECK1-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP68]], align 8
-// CHECK1-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP69]], align 8
-// CHECK1-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP70]], align 8
-// CHECK1-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 8
-// CHECK1-NEXT: store i64 [[TMP62]], ptr [[TMP71]], align 8
-// CHECK1-NEXT: [[TMP72:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l56.region_id, ptr [[KERNEL_ARGS18]])
-// CHECK1-NEXT: [[TMP73:%.*]] = icmp ne i32 [[TMP72]], 0
-// CHECK1-NEXT: br i1 [[TMP73]], label [[OMP_OFFLOAD_FAILED19:%.*]], label [[OMP_OFFLOAD_CONT20:%.*]]
+// CHECK1-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP69]], align 4
+// CHECK1-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 1
+// CHECK1-NEXT: store i32 3, ptr [[TMP70]], align 4
+// CHECK1-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP63]], ptr [[TMP71]], align 8
+// CHECK1-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP64]], ptr [[TMP72]], align 8
+// CHECK1-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.4, ptr [[TMP73]], align 8
+// CHECK1-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP74]], align 8
+// CHECK1-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP75]], align 8
+// CHECK1-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP76]], align 8
+// CHECK1-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 8
+// CHECK1-NEXT: store i64 [[TMP68]], ptr [[TMP77]], align 8
+// CHECK1-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP78]], align 8
+// CHECK1-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP79]], align 4
+// CHECK1-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP80]], align 4
+// CHECK1-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP81]], align 4
+// CHECK1-NEXT: [[TMP82:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l56.region_id, ptr [[KERNEL_ARGS18]])
+// CHECK1-NEXT: [[TMP83:%.*]] = icmp ne i32 [[TMP82]], 0
+// CHECK1-NEXT: br i1 [[TMP83]], label [[OMP_OFFLOAD_FAILED19:%.*]], label [[OMP_OFFLOAD_CONT20:%.*]]
// CHECK1: omp_offload.failed19:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l56(i64 [[TMP46]], ptr [[A]], ptr [[TMP47]]) #[[ATTR4]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l56(i64 [[TMP52]], ptr [[A]], ptr [[TMP53]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT20]]
// CHECK1: omp_offload.cont20:
// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i32], ptr [[A]], i64 0, i64 0
-// CHECK1-NEXT: [[TMP74:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK1-NEXT: ret i32 [[TMP74]]
+// CHECK1-NEXT: [[TMP84:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK1-NEXT: ret i32 [[TMP84]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l50
@@ -873,95 +891,113 @@ int target_teams_fun(int *g){
// CHECK3-NEXT: [[TMP32:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_4]], align 4
// CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP32]], 1
// CHECK3-NEXT: [[TMP33:%.*]] = zext i32 [[ADD]] to i64
+// CHECK3-NEXT: [[TMP34:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP28]], 0
+// CHECK3-NEXT: [[TMP35:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP29]], 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP34]], align 4
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK3-NEXT: store i32 5, ptr [[TMP35]], align 4
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP26]], ptr [[TMP36]], align 4
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP27]], ptr [[TMP37]], align 4
-// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes, ptr [[TMP38]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP39]], align 4
-// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP40]], align 4
-// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP41]], align 4
-// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK3-NEXT: store i64 [[TMP33]], ptr [[TMP42]], align 8
-// CHECK3-NEXT: [[TMP43:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 [[TMP28]], i32 [[TMP29]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l50.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP44:%.*]] = icmp ne i32 [[TMP43]], 0
-// CHECK3-NEXT: br i1 [[TMP44]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP36]], align 4
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK3-NEXT: store i32 5, ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP26]], ptr [[TMP38]], align 4
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP27]], ptr [[TMP39]], align 4
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes, ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP42]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP43]], align 4
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK3-NEXT: store i64 [[TMP33]], ptr [[TMP44]], align 8
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP45]], align 8
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [[TMP34]], ptr [[TMP46]], align 4
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [[TMP35]], ptr [[TMP47]], align 4
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP48]], align 4
+// CHECK3-NEXT: [[TMP49:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 [[TMP28]], i32 [[TMP29]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l50.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP50:%.*]] = icmp ne i32 [[TMP49]], 0
+// CHECK3-NEXT: br i1 [[TMP50]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l50(i32 [[TMP4]], i32 [[TMP6]], ptr [[A]], i32 [[TMP8]], i32 [[TMP10]]) #[[ATTR4:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
-// CHECK3-NEXT: [[TMP45:%.*]] = load i32, ptr [[N]], align 4
-// CHECK3-NEXT: store i32 [[TMP45]], ptr [[N_CASTED7]], align 4
-// CHECK3-NEXT: [[TMP46:%.*]] = load i32, ptr [[N_CASTED7]], align 4
-// CHECK3-NEXT: [[TMP47:%.*]] = load ptr, ptr [[G_ADDR]], align 4
-// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP46]], ptr [[TMP48]], align 4
-// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP46]], ptr [[TMP49]], align 4
-// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP50]], align 4
-// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
-// CHECK3-NEXT: store ptr [[A]], ptr [[TMP51]], align 4
-// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
-// CHECK3-NEXT: store ptr [[A]], ptr [[TMP52]], align 4
-// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1
-// CHECK3-NEXT: store ptr null, ptr [[TMP53]], align 4
-// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP47]], ptr [[TMP54]], align 4
-// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP47]], ptr [[TMP55]], align 4
-// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 2
+// CHECK3-NEXT: [[TMP51:%.*]] = load i32, ptr [[N]], align 4
+// CHECK3-NEXT: store i32 [[TMP51]], ptr [[N_CASTED7]], align 4
+// CHECK3-NEXT: [[TMP52:%.*]] = load i32, ptr [[N_CASTED7]], align 4
+// CHECK3-NEXT: [[TMP53:%.*]] = load ptr, ptr [[G_ADDR]], align 4
+// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP52]], ptr [[TMP54]], align 4
+// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP52]], ptr [[TMP55]], align 4
+// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
// CHECK3-NEXT: store ptr null, ptr [[TMP56]], align 4
-// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP58:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP59:%.*]] = load i32, ptr [[N]], align 4
-// CHECK3-NEXT: store i32 [[TMP59]], ptr [[DOTCAPTURE_EXPR_12]], align 4
-// CHECK3-NEXT: [[TMP60:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_12]], align 4
-// CHECK3-NEXT: [[SUB14:%.*]] = sub nsw i32 [[TMP60]], 0
+// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
+// CHECK3-NEXT: store ptr [[A]], ptr [[TMP57]], align 4
+// CHECK3-NEXT: [[TMP58:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
+// CHECK3-NEXT: store ptr [[A]], ptr [[TMP58]], align 4
+// CHECK3-NEXT: [[TMP59:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1
+// CHECK3-NEXT: store ptr null, ptr [[TMP59]], align 4
+// CHECK3-NEXT: [[TMP60:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP53]], ptr [[TMP60]], align 4
+// CHECK3-NEXT: [[TMP61:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP53]], ptr [[TMP61]], align 4
+// CHECK3-NEXT: [[TMP62:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 2
+// CHECK3-NEXT: store ptr null, ptr [[TMP62]], align 4
+// CHECK3-NEXT: [[TMP63:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP64:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP65:%.*]] = load i32, ptr [[N]], align 4
+// CHECK3-NEXT: store i32 [[TMP65]], ptr [[DOTCAPTURE_EXPR_12]], align 4
+// CHECK3-NEXT: [[TMP66:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_12]], align 4
+// CHECK3-NEXT: [[SUB14:%.*]] = sub nsw i32 [[TMP66]], 0
// CHECK3-NEXT: [[DIV15:%.*]] = sdiv i32 [[SUB14]], 1
// CHECK3-NEXT: [[SUB16:%.*]] = sub nsw i32 [[DIV15]], 1
// CHECK3-NEXT: store i32 [[SUB16]], ptr [[DOTCAPTURE_EXPR_13]], align 4
-// CHECK3-NEXT: [[TMP61:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_13]], align 4
-// CHECK3-NEXT: [[ADD17:%.*]] = add nsw i32 [[TMP61]], 1
-// CHECK3-NEXT: [[TMP62:%.*]] = zext i32 [[ADD17]] to i64
+// CHECK3-NEXT: [[TMP67:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_13]], align 4
+// CHECK3-NEXT: [[ADD17:%.*]] = add nsw i32 [[TMP67]], 1
+// CHECK3-NEXT: [[TMP68:%.*]] = zext i32 [[ADD17]] to i64
// CHECK3-NEXT: [[KERNEL_ARGS18:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP63]], align 4
-// CHECK3-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 1
-// CHECK3-NEXT: store i32 3, ptr [[TMP64]], align 4
-// CHECK3-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP57]], ptr [[TMP65]], align 4
-// CHECK3-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP58]], ptr [[TMP66]], align 4
-// CHECK3-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.4, ptr [[TMP67]], align 4
-// CHECK3-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP68]], align 4
-// CHECK3-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP69]], align 4
-// CHECK3-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP70]], align 4
-// CHECK3-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 8
-// CHECK3-NEXT: store i64 [[TMP62]], ptr [[TMP71]], align 8
-// CHECK3-NEXT: [[TMP72:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l56.region_id, ptr [[KERNEL_ARGS18]])
-// CHECK3-NEXT: [[TMP73:%.*]] = icmp ne i32 [[TMP72]], 0
-// CHECK3-NEXT: br i1 [[TMP73]], label [[OMP_OFFLOAD_FAILED19:%.*]], label [[OMP_OFFLOAD_CONT20:%.*]]
+// CHECK3-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP69]], align 4
+// CHECK3-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 1
+// CHECK3-NEXT: store i32 3, ptr [[TMP70]], align 4
+// CHECK3-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP63]], ptr [[TMP71]], align 4
+// CHECK3-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP64]], ptr [[TMP72]], align 4
+// CHECK3-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.4, ptr [[TMP73]], align 4
+// CHECK3-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP74]], align 4
+// CHECK3-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP75]], align 4
+// CHECK3-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP76]], align 4
+// CHECK3-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 8
+// CHECK3-NEXT: store i64 [[TMP68]], ptr [[TMP77]], align 8
+// CHECK3-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP78]], align 8
+// CHECK3-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP79]], align 4
+// CHECK3-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP80]], align 4
+// CHECK3-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS18]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP81]], align 4
+// CHECK3-NEXT: [[TMP82:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l56.region_id, ptr [[KERNEL_ARGS18]])
+// CHECK3-NEXT: [[TMP83:%.*]] = icmp ne i32 [[TMP82]], 0
+// CHECK3-NEXT: br i1 [[TMP83]], label [[OMP_OFFLOAD_FAILED19:%.*]], label [[OMP_OFFLOAD_CONT20:%.*]]
// CHECK3: omp_offload.failed19:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l56(i32 [[TMP46]], ptr [[A]], ptr [[TMP47]]) #[[ATTR4]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l56(i32 [[TMP52]], ptr [[A]], ptr [[TMP53]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT20]]
// CHECK3: omp_offload.cont20:
// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i32], ptr [[A]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP74:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK3-NEXT: ret i32 [[TMP74]]
+// CHECK3-NEXT: [[TMP84:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK3-NEXT: ret i32 [[TMP84]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16target_teams_funPi_l50
diff --git a/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_collapse_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_collapse_codegen.cpp
index dbc26d92daf34..6b0847203d835 100644
--- a/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_collapse_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_collapse_codegen.cpp
@@ -129,7 +129,7 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -146,9 +146,17 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 56088, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -156,8 +164,8 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[A3:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x [456 x i32]], ptr [[A3]], i64 0, i64 0
// CHECK1-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds [456 x i32], ptr [[ARRAYIDX]], i64 0, i64 0
-// CHECK1-NEXT: [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
-// CHECK1-NEXT: ret i32 [[TMP16]]
+// CHECK1-NEXT: [[TMP20:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
+// CHECK1-NEXT: ret i32 [[TMP20]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -380,7 +388,7 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -397,9 +405,17 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 56088, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -407,8 +423,8 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[A3:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x [456 x i32]], ptr [[A3]], i32 0, i32 0
// CHECK3-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds [456 x i32], ptr [[ARRAYIDX]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
-// CHECK3-NEXT: ret i32 [[TMP16]]
+// CHECK3-NEXT: [[TMP20:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
+// CHECK3-NEXT: ret i32 [[TMP20]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -835,7 +851,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[ADD:%.*]] = add nsw i64 [[TMP35]], 1
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP36]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP36]], align 4
// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 5, ptr [[TMP37]], align 4
// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -852,20 +868,28 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP43]], align 8
// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 [[ADD]], ptr [[TMP44]], align 8
-// CHECK9-NEXT: [[TMP45:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP46:%.*]] = icmp ne i32 [[TMP45]], 0
-// CHECK9-NEXT: br i1 [[TMP46]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP46]], align 4
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP47]], align 4
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP48]], align 4
+// CHECK9-NEXT: [[TMP49:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP50:%.*]] = icmp ne i32 [[TMP49]], 0
+// CHECK9-NEXT: br i1 [[TMP50]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81(i64 [[TMP7]], i64 [[TMP9]], i64 [[TMP1]], i64 [[TMP3]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP47:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef signext [[TMP47]])
+// CHECK9-NEXT: [[TMP51:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef signext [[TMP51]])
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK9-NEXT: [[TMP48:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP48]])
-// CHECK9-NEXT: [[TMP49:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP49]]
+// CHECK9-NEXT: [[TMP52:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP52]])
+// CHECK9-NEXT: [[TMP53:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP53]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81
@@ -1225,7 +1249,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK9-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK9-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1242,9 +1266,17 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 20, ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l68.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK9-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK9-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l68.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK9-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l68(ptr [[A]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1540,7 +1572,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[ADD:%.*]] = add nsw i64 [[TMP34]], 1
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP35]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP35]], align 4
// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 5, ptr [[TMP36]], align 4
// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1557,20 +1589,28 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP42]], align 4
// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 [[ADD]], ptr [[TMP43]], align 8
-// CHECK11-NEXT: [[TMP44:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP45:%.*]] = icmp ne i32 [[TMP44]], 0
-// CHECK11-NEXT: br i1 [[TMP45]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP44]], align 8
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP47]], align 4
+// CHECK11-NEXT: [[TMP48:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP49:%.*]] = icmp ne i32 [[TMP48]], 0
+// CHECK11-NEXT: br i1 [[TMP49]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81(i32 [[TMP5]], i32 [[TMP7]], i32 [[TMP0]], i32 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP46:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef [[TMP46]])
+// CHECK11-NEXT: [[TMP50:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef [[TMP50]])
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK11-NEXT: [[TMP47:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP47]])
-// CHECK11-NEXT: [[TMP48:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP48]]
+// CHECK11-NEXT: [[TMP51:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP51]])
+// CHECK11-NEXT: [[TMP52:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP52]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81
@@ -1932,7 +1972,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK11-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK11-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1949,9 +1989,17 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK11-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 20, ptr [[TMP13]], align 8
-// CHECK11-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l68.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK11-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK11-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l68.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK11-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l68(ptr [[A]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_depend_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_depend_codegen.cpp
index 1c9ebfaea8592..0deadaeaab039 100644
--- a/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_depend_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_depend_codegen.cpp
@@ -200,7 +200,7 @@ int foo(int n) {
// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2
// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]],
// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64
-// CHECK: [[RET:%.+]] = call i32 @__tgt_target_kernel_nowait(%struct.ident_t* @{{.+}}, i64 [[DEVICE]], i32 0, i32 0, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* %{{.+}})
+// CHECK: [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 [[DEVICE]], i32 0, i32 0, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* %{{.+}})
// CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]]
diff --git a/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_dist_schedule_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_dist_schedule_codegen.cpp
index 768713944d934..1ea96dfdce7df 100644
--- a/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_dist_schedule_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_dist_schedule_codegen.cpp
@@ -171,7 +171,7 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -188,87 +188,111 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.4, ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l32.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.4, ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l32.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK1-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK1-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK1: omp_offload.failed8:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l32(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK1: omp_offload.cont9:
// CHECK1-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A10]], ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A10]], ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.8, ptr [[TMP41]], align 8
-// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP42]], align 8
-// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP43]], align 8
-// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP44]], align 8
-// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK1-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK1-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK1-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 8
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 8
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.8, ptr [[TMP49]], align 8
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP50]], align 8
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP51]], align 8
+// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP52]], align 8
+// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK1-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK1-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK1-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK1: omp_offload.failed16:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK1: omp_offload.cont17:
// CHECK1-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A18]], i64 0, i64 0
-// CHECK1-NEXT: [[TMP48:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK1-NEXT: ret i32 [[TMP48]]
+// CHECK1-NEXT: [[TMP60:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK1-NEXT: ret i32 [[TMP60]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -830,7 +854,7 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -847,87 +871,111 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 4
-// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A2]], ptr [[TMP17]], align 4
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 4
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A2]], ptr [[TMP21]], align 4
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.4, ptr [[TMP25]], align 4
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP26]], align 4
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK3-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l32.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK3-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK3-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.4, ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK3-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l32.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK3-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK3-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK3: omp_offload.failed8:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l32(ptr [[THIS1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK3: omp_offload.cont9:
// CHECK3-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 4
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A10]], ptr [[TMP33]], align 4
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP34]], align 4
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A10]], ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP42]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 4
-// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 4
-// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.8, ptr [[TMP41]], align 4
-// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP42]], align 4
-// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP44]], align 4
-// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK3-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK3-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK3-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK3-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 4
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 4
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.8, ptr [[TMP49]], align 4
+// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP50]], align 4
+// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP52]], align 4
+// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK3-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK3-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK3-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK3-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK3: omp_offload.failed16:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36(ptr [[THIS1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK3: omp_offload.cont17:
// CHECK3-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A18]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP48:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK3-NEXT: ret i32 [[TMP48]]
+// CHECK3-NEXT: [[TMP60:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK3-NEXT: ret i32 [[TMP60]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -1771,7 +1819,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1788,159 +1836,183 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK9-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK9-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK9-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100(i64 [[TMP4]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK9-NEXT: [[TMP35:%.*]] = load i64, ptr [[N_CASTED3]], align 8
-// CHECK9-NEXT: [[TMP36:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK9-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK9-NEXT: [[TMP39:%.*]] = load i64, ptr [[N_CASTED3]], align 8
+// CHECK9-NEXT: [[TMP40:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK9-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES7]], ptr align 8 @.offload_sizes.4, i64 24, i1 false)
-// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP35]], ptr [[TMP37]], align 8
-// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP35]], ptr [[TMP38]], align 8
-// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP40]], align 8
-// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP41]], align 8
-// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP42]], align 8
-// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP43]], align 8
-// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 8
-// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK9-NEXT: store i64 [[TMP36]], ptr [[TMP45]], align 8
-// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP39]], ptr [[TMP41]], align 8
+// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP39]], ptr [[TMP42]], align 8
+// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP44]], align 8
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
// CHECK9-NEXT: store ptr null, ptr [[TMP46]], align 8
-// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP50:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP50]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK9-NEXT: [[TMP51:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK9-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP51]], 0
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP47]], align 8
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 8
+// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK9-NEXT: store i64 [[TMP40]], ptr [[TMP49]], align 8
+// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP50]], align 8
+// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP54:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP54]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK9-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK9-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP55]], 0
// CHECK9-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK9-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK9-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK9-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK9-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP52]], 1
-// CHECK9-NEXT: [[TMP53:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK9-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK9-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP56]], 1
+// CHECK9-NEXT: [[TMP57:%.*]] = zext i32 [[ADD14]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK9-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK9-NEXT: store i32 3, ptr [[TMP55]], align 4
-// CHECK9-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP47]], ptr [[TMP56]], align 8
-// CHECK9-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 8
-// CHECK9-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 8
-// CHECK9-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP59]], align 8
-// CHECK9-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP60]], align 8
-// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP61]], align 8
-// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP53]], ptr [[TMP62]], align 8
-// CHECK9-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l104.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK9-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
-// CHECK9-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK9-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP58]], align 4
+// CHECK9-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK9-NEXT: store i32 3, ptr [[TMP59]], align 4
+// CHECK9-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP51]], ptr [[TMP60]], align 8
+// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 8
+// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 8
+// CHECK9-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP63]], align 8
+// CHECK9-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP64]], align 8
+// CHECK9-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP65]], align 8
+// CHECK9-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP57]], ptr [[TMP66]], align 8
+// CHECK9-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP67]], align 8
+// CHECK9-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP68]], align 4
+// CHECK9-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK9-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP70]], align 4
+// CHECK9-NEXT: [[TMP71:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l104.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK9-NEXT: [[TMP72:%.*]] = icmp ne i32 [[TMP71]], 0
+// CHECK9-NEXT: br i1 [[TMP72]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK9: omp_offload.failed16:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l104(i64 [[TMP35]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l104(i64 [[TMP39]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK9: omp_offload.cont17:
-// CHECK9-NEXT: [[TMP65:%.*]] = load i32, ptr [[M]], align 4
-// CHECK9-NEXT: store i32 [[TMP65]], ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK9-NEXT: [[TMP66:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP66]], ptr [[N_CASTED19]], align 4
-// CHECK9-NEXT: [[TMP67:%.*]] = load i64, ptr [[N_CASTED19]], align 8
-// CHECK9-NEXT: [[TMP68:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK9-NEXT: store i32 [[TMP68]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK9-NEXT: [[TMP69:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK9-NEXT: [[TMP70:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK9-NEXT: [[TMP73:%.*]] = load i32, ptr [[M]], align 4
+// CHECK9-NEXT: store i32 [[TMP73]], ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK9-NEXT: [[TMP74:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP74]], ptr [[N_CASTED19]], align 4
+// CHECK9-NEXT: [[TMP75:%.*]] = load i64, ptr [[N_CASTED19]], align 8
+// CHECK9-NEXT: [[TMP76:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK9-NEXT: store i32 [[TMP76]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK9-NEXT: [[TMP77:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK9-NEXT: [[TMP78:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK9-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES23]], ptr align 8 @.offload_sizes.8, i64 32, i1 false)
-// CHECK9-NEXT: [[TMP71:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP67]], ptr [[TMP71]], align 8
-// CHECK9-NEXT: [[TMP72:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP67]], ptr [[TMP72]], align 8
-// CHECK9-NEXT: [[TMP73:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP73]], align 8
-// CHECK9-NEXT: [[TMP74:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP74]], align 8
-// CHECK9-NEXT: [[TMP75:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP75]], align 8
-// CHECK9-NEXT: [[TMP76:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP76]], align 8
-// CHECK9-NEXT: [[TMP77:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP77]], align 8
-// CHECK9-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP78]], align 8
-// CHECK9-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
-// CHECK9-NEXT: store i64 [[TMP70]], ptr [[TMP79]], align 8
-// CHECK9-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP80]], align 8
-// CHECK9-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
-// CHECK9-NEXT: store i64 [[TMP69]], ptr [[TMP81]], align 8
-// CHECK9-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
-// CHECK9-NEXT: store i64 [[TMP69]], ptr [[TMP82]], align 8
-// CHECK9-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP83]], align 8
-// CHECK9-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP87:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP87]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK9-NEXT: [[TMP88:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK9-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP88]], 0
+// CHECK9-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP75]], ptr [[TMP79]], align 8
+// CHECK9-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP75]], ptr [[TMP80]], align 8
+// CHECK9-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP81]], align 8
+// CHECK9-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP82]], align 8
+// CHECK9-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP83]], align 8
+// CHECK9-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP84]], align 8
+// CHECK9-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP85]], align 8
+// CHECK9-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP86]], align 8
+// CHECK9-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
+// CHECK9-NEXT: store i64 [[TMP78]], ptr [[TMP87]], align 8
+// CHECK9-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP88]], align 8
+// CHECK9-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
+// CHECK9-NEXT: store i64 [[TMP77]], ptr [[TMP89]], align 8
+// CHECK9-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
+// CHECK9-NEXT: store i64 [[TMP77]], ptr [[TMP90]], align 8
+// CHECK9-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP91]], align 8
+// CHECK9-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP94:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP95:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP95]], ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK9-NEXT: [[TMP96:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK9-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP96]], 0
// CHECK9-NEXT: [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1
// CHECK9-NEXT: [[SUB29:%.*]] = sub nsw i32 [[DIV28]], 1
// CHECK9-NEXT: store i32 [[SUB29]], ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK9-NEXT: [[TMP89:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK9-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP89]], 1
-// CHECK9-NEXT: [[TMP90:%.*]] = zext i32 [[ADD30]] to i64
+// CHECK9-NEXT: [[TMP97:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
+// CHECK9-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP97]], 1
+// CHECK9-NEXT: [[TMP98:%.*]] = zext i32 [[ADD30]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP91]], align 4
-// CHECK9-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK9-NEXT: store i32 4, ptr [[TMP92]], align 4
-// CHECK9-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP84]], ptr [[TMP93]], align 8
-// CHECK9-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP85]], ptr [[TMP94]], align 8
-// CHECK9-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP86]], ptr [[TMP95]], align 8
-// CHECK9-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP96]], align 8
-// CHECK9-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP97]], align 8
-// CHECK9-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP98]], align 8
-// CHECK9-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP90]], ptr [[TMP99]], align 8
-// CHECK9-NEXT: [[TMP100:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l108.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK9-NEXT: [[TMP101:%.*]] = icmp ne i32 [[TMP100]], 0
-// CHECK9-NEXT: br i1 [[TMP101]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK9-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP99]], align 4
+// CHECK9-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK9-NEXT: store i32 4, ptr [[TMP100]], align 4
+// CHECK9-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP92]], ptr [[TMP101]], align 8
+// CHECK9-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP93]], ptr [[TMP102]], align 8
+// CHECK9-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP94]], ptr [[TMP103]], align 8
+// CHECK9-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP104]], align 8
+// CHECK9-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP105]], align 8
+// CHECK9-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP106]], align 8
+// CHECK9-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP98]], ptr [[TMP107]], align 8
+// CHECK9-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP108]], align 8
+// CHECK9-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP109]], align 4
+// CHECK9-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP110]], align 4
+// CHECK9-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP111]], align 4
+// CHECK9-NEXT: [[TMP112:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l108.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK9-NEXT: [[TMP113:%.*]] = icmp ne i32 [[TMP112]], 0
+// CHECK9-NEXT: br i1 [[TMP113]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK9: omp_offload.failed32:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l108(i64 [[TMP67]], i64 [[TMP1]], ptr [[VLA]], i64 [[TMP69]]) #[[ATTR3]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l108(i64 [[TMP75]], i64 [[TMP1]], ptr [[VLA]], i64 [[TMP77]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK9: omp_offload.cont33:
-// CHECK9-NEXT: [[TMP102:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP102]])
+// CHECK9-NEXT: [[TMP114:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP114]])
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK9-NEXT: [[TMP103:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP103]])
-// CHECK9-NEXT: [[TMP104:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP104]]
+// CHECK9-NEXT: [[TMP115:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP115]])
+// CHECK9-NEXT: [[TMP116:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP116]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100
@@ -2737,7 +2809,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK9-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK9-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2754,90 +2826,114 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l81.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK9-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK9-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l81.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK9-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l81(ptr [[A]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP16]], align 8
-// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP17]], align 8
-// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP20]], align 8
+// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP21]], align 8
+// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK9-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.16, ptr [[TMP25]], align 8
-// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP26]], align 8
-// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK9-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK9-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l85.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK9-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK9-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK9-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.16, ptr [[TMP29]], align 8
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP30]], align 8
+// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK9-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK9-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l85.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK9-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK9-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK9: omp_offload.failed6:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l85(ptr [[A]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK9: omp_offload.cont7:
-// CHECK9-NEXT: [[TMP32:%.*]] = load i32, ptr [[M]], align 4
-// CHECK9-NEXT: store i32 [[TMP32]], ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK9-NEXT: [[TMP33:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK9-NEXT: store i32 [[TMP33]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK9-NEXT: [[TMP34:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP35]], align 8
-// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP36]], align 8
-// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP37]], align 8
-// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP34]], ptr [[TMP38]], align 8
-// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP34]], ptr [[TMP39]], align 8
-// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP40]], align 8
-// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP40:%.*]] = load i32, ptr [[M]], align 4
+// CHECK9-NEXT: store i32 [[TMP40]], ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK9-NEXT: [[TMP41:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK9-NEXT: store i32 [[TMP41]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK9-NEXT: [[TMP42:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP43]], align 8
+// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP44]], align 8
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP42]], ptr [[TMP46]], align 8
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP42]], ptr [[TMP47]], align 8
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP48]], align 8
+// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP43]], align 4
-// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK9-NEXT: store i32 2, ptr [[TMP44]], align 4
-// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP41]], ptr [[TMP45]], align 8
-// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP42]], ptr [[TMP46]], align 8
-// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.20, ptr [[TMP47]], align 8
-// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.21, ptr [[TMP48]], align 8
-// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP49]], align 8
-// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP50]], align 8
-// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK9-NEXT: store i64 10, ptr [[TMP51]], align 8
-// CHECK9-NEXT: [[TMP52:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK9-NEXT: [[TMP53:%.*]] = icmp ne i32 [[TMP52]], 0
-// CHECK9-NEXT: br i1 [[TMP53]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP51]], align 4
+// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK9-NEXT: store i32 2, ptr [[TMP52]], align 4
+// CHECK9-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 8
+// CHECK9-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP50]], ptr [[TMP54]], align 8
+// CHECK9-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.20, ptr [[TMP55]], align 8
+// CHECK9-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.21, ptr [[TMP56]], align 8
+// CHECK9-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP57]], align 8
+// CHECK9-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP58]], align 8
+// CHECK9-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK9-NEXT: store i64 10, ptr [[TMP59]], align 8
+// CHECK9-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP60]], align 8
+// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4
+// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP62]], align 4
+// CHECK9-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP63]], align 4
+// CHECK9-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK9-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
+// CHECK9-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK9: omp_offload.failed13:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89(ptr [[A]], i64 [[TMP34]]) #[[ATTR3]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89(ptr [[A]], i64 [[TMP42]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT14]]
// CHECK9: omp_offload.cont14:
// CHECK9-NEXT: ret i32 0
@@ -3464,7 +3560,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3481,161 +3577,185 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK11-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK11-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK11-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100(i32 [[TMP3]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK11-NEXT: [[TMP35:%.*]] = load i32, ptr [[N_CASTED3]], align 4
-// CHECK11-NEXT: [[TMP36:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK11-NEXT: [[TMP37:%.*]] = sext i32 [[TMP36]] to i64
+// CHECK11-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK11-NEXT: [[TMP39:%.*]] = load i32, ptr [[N_CASTED3]], align 4
+// CHECK11-NEXT: [[TMP40:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK11-NEXT: [[TMP41:%.*]] = sext i32 [[TMP40]] to i64
// CHECK11-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES7]], ptr align 4 @.offload_sizes.4, i32 24, i1 false)
-// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP35]], ptr [[TMP38]], align 4
-// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP35]], ptr [[TMP39]], align 4
-// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP40]], align 4
-// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP41]], align 4
-// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP42]], align 4
-// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 4
-// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP45]], align 4
-// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK11-NEXT: store i64 [[TMP37]], ptr [[TMP46]], align 4
-// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP39]], ptr [[TMP42]], align 4
+// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP39]], ptr [[TMP43]], align 4
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP44]], align 4
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
// CHECK11-NEXT: store ptr null, ptr [[TMP47]], align 4
-// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP51:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP51]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK11-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK11-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP52]], 0
+// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 4
+// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP49]], align 4
+// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK11-NEXT: store i64 [[TMP41]], ptr [[TMP50]], align 4
+// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP54:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP55:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP55]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK11-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK11-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP56]], 0
// CHECK11-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK11-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK11-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK11-NEXT: [[TMP53:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK11-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP53]], 1
-// CHECK11-NEXT: [[TMP54:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK11-NEXT: [[TMP57:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK11-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP57]], 1
+// CHECK11-NEXT: [[TMP58:%.*]] = zext i32 [[ADD14]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP55]], align 4
-// CHECK11-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK11-NEXT: store i32 3, ptr [[TMP56]], align 4
-// CHECK11-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 4
-// CHECK11-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 4
-// CHECK11-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP50]], ptr [[TMP59]], align 4
-// CHECK11-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP60]], align 4
-// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP61]], align 4
-// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP62]], align 4
-// CHECK11-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP54]], ptr [[TMP63]], align 8
-// CHECK11-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l104.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK11-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
-// CHECK11-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK11-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP59]], align 4
+// CHECK11-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK11-NEXT: store i32 3, ptr [[TMP60]], align 4
+// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 4
+// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 4
+// CHECK11-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP54]], ptr [[TMP63]], align 4
+// CHECK11-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP64]], align 4
+// CHECK11-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP65]], align 4
+// CHECK11-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP66]], align 4
+// CHECK11-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP58]], ptr [[TMP67]], align 8
+// CHECK11-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP68]], align 8
+// CHECK11-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK11-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP70]], align 4
+// CHECK11-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP71]], align 4
+// CHECK11-NEXT: [[TMP72:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l104.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK11-NEXT: [[TMP73:%.*]] = icmp ne i32 [[TMP72]], 0
+// CHECK11-NEXT: br i1 [[TMP73]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK11: omp_offload.failed16:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l104(i32 [[TMP35]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l104(i32 [[TMP39]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK11: omp_offload.cont17:
-// CHECK11-NEXT: [[TMP66:%.*]] = load i32, ptr [[M]], align 4
-// CHECK11-NEXT: store i32 [[TMP66]], ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK11-NEXT: [[TMP67:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP67]], ptr [[N_CASTED19]], align 4
-// CHECK11-NEXT: [[TMP68:%.*]] = load i32, ptr [[N_CASTED19]], align 4
-// CHECK11-NEXT: [[TMP69:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK11-NEXT: store i32 [[TMP69]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK11-NEXT: [[TMP70:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK11-NEXT: [[TMP71:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK11-NEXT: [[TMP72:%.*]] = sext i32 [[TMP71]] to i64
+// CHECK11-NEXT: [[TMP74:%.*]] = load i32, ptr [[M]], align 4
+// CHECK11-NEXT: store i32 [[TMP74]], ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK11-NEXT: [[TMP75:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP75]], ptr [[N_CASTED19]], align 4
+// CHECK11-NEXT: [[TMP76:%.*]] = load i32, ptr [[N_CASTED19]], align 4
+// CHECK11-NEXT: [[TMP77:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK11-NEXT: store i32 [[TMP77]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK11-NEXT: [[TMP78:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK11-NEXT: [[TMP79:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK11-NEXT: [[TMP80:%.*]] = sext i32 [[TMP79]] to i64
// CHECK11-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES23]], ptr align 4 @.offload_sizes.8, i32 32, i1 false)
-// CHECK11-NEXT: [[TMP73:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP68]], ptr [[TMP73]], align 4
-// CHECK11-NEXT: [[TMP74:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP68]], ptr [[TMP74]], align 4
-// CHECK11-NEXT: [[TMP75:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP75]], align 4
-// CHECK11-NEXT: [[TMP76:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP76]], align 4
-// CHECK11-NEXT: [[TMP77:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP77]], align 4
-// CHECK11-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP78]], align 4
-// CHECK11-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP79]], align 4
-// CHECK11-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP80]], align 4
-// CHECK11-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
-// CHECK11-NEXT: store i64 [[TMP72]], ptr [[TMP81]], align 4
-// CHECK11-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP82]], align 4
-// CHECK11-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
-// CHECK11-NEXT: store i32 [[TMP70]], ptr [[TMP83]], align 4
-// CHECK11-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
-// CHECK11-NEXT: store i32 [[TMP70]], ptr [[TMP84]], align 4
-// CHECK11-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP85]], align 4
-// CHECK11-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP89:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP89]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK11-NEXT: [[TMP90:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK11-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP90]], 0
+// CHECK11-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP76]], ptr [[TMP81]], align 4
+// CHECK11-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP76]], ptr [[TMP82]], align 4
+// CHECK11-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP83]], align 4
+// CHECK11-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP84]], align 4
+// CHECK11-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP85]], align 4
+// CHECK11-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP86]], align 4
+// CHECK11-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP87]], align 4
+// CHECK11-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP88]], align 4
+// CHECK11-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
+// CHECK11-NEXT: store i64 [[TMP80]], ptr [[TMP89]], align 4
+// CHECK11-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP90]], align 4
+// CHECK11-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
+// CHECK11-NEXT: store i32 [[TMP78]], ptr [[TMP91]], align 4
+// CHECK11-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
+// CHECK11-NEXT: store i32 [[TMP78]], ptr [[TMP92]], align 4
+// CHECK11-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP93]], align 4
+// CHECK11-NEXT: [[TMP94:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP95:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP96:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP97:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP97]], ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK11-NEXT: [[TMP98:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK11-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP98]], 0
// CHECK11-NEXT: [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1
// CHECK11-NEXT: [[SUB29:%.*]] = sub nsw i32 [[DIV28]], 1
// CHECK11-NEXT: store i32 [[SUB29]], ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK11-NEXT: [[TMP91:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK11-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP91]], 1
-// CHECK11-NEXT: [[TMP92:%.*]] = zext i32 [[ADD30]] to i64
+// CHECK11-NEXT: [[TMP99:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
+// CHECK11-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP99]], 1
+// CHECK11-NEXT: [[TMP100:%.*]] = zext i32 [[ADD30]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP93]], align 4
-// CHECK11-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK11-NEXT: store i32 4, ptr [[TMP94]], align 4
-// CHECK11-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP86]], ptr [[TMP95]], align 4
-// CHECK11-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP87]], ptr [[TMP96]], align 4
-// CHECK11-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP88]], ptr [[TMP97]], align 4
-// CHECK11-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP98]], align 4
-// CHECK11-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP99]], align 4
-// CHECK11-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP100]], align 4
-// CHECK11-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP92]], ptr [[TMP101]], align 8
-// CHECK11-NEXT: [[TMP102:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l108.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK11-NEXT: [[TMP103:%.*]] = icmp ne i32 [[TMP102]], 0
-// CHECK11-NEXT: br i1 [[TMP103]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK11-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP101]], align 4
+// CHECK11-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK11-NEXT: store i32 4, ptr [[TMP102]], align 4
+// CHECK11-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP94]], ptr [[TMP103]], align 4
+// CHECK11-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP95]], ptr [[TMP104]], align 4
+// CHECK11-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP96]], ptr [[TMP105]], align 4
+// CHECK11-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP106]], align 4
+// CHECK11-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP107]], align 4
+// CHECK11-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP108]], align 4
+// CHECK11-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP100]], ptr [[TMP109]], align 8
+// CHECK11-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP110]], align 8
+// CHECK11-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP111]], align 4
+// CHECK11-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP112]], align 4
+// CHECK11-NEXT: [[TMP113:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP113]], align 4
+// CHECK11-NEXT: [[TMP114:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l108.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK11-NEXT: [[TMP115:%.*]] = icmp ne i32 [[TMP114]], 0
+// CHECK11-NEXT: br i1 [[TMP115]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK11: omp_offload.failed32:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l108(i32 [[TMP68]], i32 [[TMP0]], ptr [[VLA]], i32 [[TMP70]]) #[[ATTR3]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l108(i32 [[TMP76]], i32 [[TMP0]], ptr [[VLA]], i32 [[TMP78]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK11: omp_offload.cont33:
-// CHECK11-NEXT: [[TMP104:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP104]])
+// CHECK11-NEXT: [[TMP116:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP116]])
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK11-NEXT: [[TMP105:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP105]])
-// CHECK11-NEXT: [[TMP106:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP106]]
+// CHECK11-NEXT: [[TMP117:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP117]])
+// CHECK11-NEXT: [[TMP118:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP118]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100
@@ -4417,7 +4537,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK11-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK11-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -4434,90 +4554,114 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK11-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK11-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l81.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK11-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK11-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l81.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK11-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l81(ptr [[A]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP16]], align 4
-// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP17]], align 4
-// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP20]], align 4
+// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP21]], align 4
+// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK11-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.16, ptr [[TMP25]], align 4
-// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP26]], align 4
-// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK11-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK11-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l85.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK11-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK11-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK11-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.16, ptr [[TMP29]], align 4
+// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP30]], align 4
+// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK11-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK11-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l85.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK11-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK11-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK11: omp_offload.failed6:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l85(ptr [[A]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK11: omp_offload.cont7:
-// CHECK11-NEXT: [[TMP32:%.*]] = load i32, ptr [[M]], align 4
-// CHECK11-NEXT: store i32 [[TMP32]], ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK11-NEXT: [[TMP33:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK11-NEXT: store i32 [[TMP33]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK11-NEXT: [[TMP34:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP35]], align 4
-// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP36]], align 4
-// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP37]], align 4
-// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP34]], ptr [[TMP38]], align 4
-// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP34]], ptr [[TMP39]], align 4
-// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP40]], align 4
-// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP40:%.*]] = load i32, ptr [[M]], align 4
+// CHECK11-NEXT: store i32 [[TMP40]], ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK11-NEXT: [[TMP41:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK11-NEXT: store i32 [[TMP41]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK11-NEXT: [[TMP42:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP43]], align 4
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP44]], align 4
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP42]], ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP42]], ptr [[TMP47]], align 4
+// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP48]], align 4
+// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP43]], align 4
-// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK11-NEXT: store i32 2, ptr [[TMP44]], align 4
-// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP41]], ptr [[TMP45]], align 4
-// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP42]], ptr [[TMP46]], align 4
-// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.20, ptr [[TMP47]], align 4
-// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.21, ptr [[TMP48]], align 4
-// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP49]], align 4
-// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP50]], align 4
-// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK11-NEXT: store i64 10, ptr [[TMP51]], align 8
-// CHECK11-NEXT: [[TMP52:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK11-NEXT: [[TMP53:%.*]] = icmp ne i32 [[TMP52]], 0
-// CHECK11-NEXT: br i1 [[TMP53]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP51]], align 4
+// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK11-NEXT: store i32 2, ptr [[TMP52]], align 4
+// CHECK11-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 4
+// CHECK11-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP50]], ptr [[TMP54]], align 4
+// CHECK11-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.20, ptr [[TMP55]], align 4
+// CHECK11-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.21, ptr [[TMP56]], align 4
+// CHECK11-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP57]], align 4
+// CHECK11-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP58]], align 4
+// CHECK11-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK11-NEXT: store i64 10, ptr [[TMP59]], align 8
+// CHECK11-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP60]], align 8
+// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4
+// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP62]], align 4
+// CHECK11-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP63]], align 4
+// CHECK11-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK11-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
+// CHECK11-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK11: omp_offload.failed13:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89(ptr [[A]], i32 [[TMP34]]) #[[ATTR3]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89(ptr [[A]], i32 [[TMP42]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT14]]
// CHECK11: omp_offload.cont14:
// CHECK11-NEXT: ret i32 0
diff --git a/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_firstprivate_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_firstprivate_codegen.cpp
index 74e3cb89da70b..35bd6fa8bffe5 100644
--- a/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_firstprivate_codegen.cpp
@@ -362,7 +362,7 @@ int main() {
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 5, ptr [[TMP22]], align 4
// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -379,9 +379,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l122.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK1-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l122.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK1-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l122(ptr @vec, i64 [[TMP1]], ptr @s_arr, ptr @var, i64 [[TMP3]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -777,7 +785,7 @@ int main() {
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -794,27 +802,35 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP25]], align 8
// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l81.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK1-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP29]], align 4
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l81.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK1-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l81(ptr [[VEC]], i64 [[TMP2]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK1-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK1: arraydestroy.body:
-// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP29]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP33]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK1-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK1-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK1: arraydestroy.done2:
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK1-NEXT: [[TMP30:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK1-NEXT: ret i32 [[TMP30]]
+// CHECK1-NEXT: [[TMP34:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK1-NEXT: ret i32 [[TMP34]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN2StC2Ev
@@ -1435,7 +1451,7 @@ int main() {
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 5, ptr [[TMP22]], align 4
// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1452,9 +1468,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4
// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP29]], align 8
-// CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l122.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK3-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK3-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l122.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK3-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l122(ptr @vec, i32 [[TMP1]], ptr @s_arr, ptr @var, i32 [[TMP3]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1844,7 +1868,7 @@ int main() {
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1861,27 +1885,35 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP25]], align 4
// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP26]], align 8
-// CHECK3-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l81.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK3-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l81.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK3-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l81(ptr [[VEC]], i32 [[TMP2]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK3-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK3-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK3: arraydestroy.body:
-// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP29]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP33]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK3-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK3-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK3: arraydestroy.done2:
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK3-NEXT: [[TMP30:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK3-NEXT: ret i32 [[TMP30]]
+// CHECK3-NEXT: [[TMP34:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK3-NEXT: ret i32 [[TMP34]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN2StC2Ev
diff --git a/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_if_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_if_codegen.cpp
index 633a19002267a..ee3efd3820572 100644
--- a/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_if_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_if_codegen.cpp
@@ -130,7 +130,7 @@ int main() {
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -147,35 +147,51 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l47.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK1-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l47.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l47(i64 [[TMP1]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP18]], align 4
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP19]], align 4
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP22]], align 4
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP23]], align 4
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK1-NEXT: store ptr null, ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l53.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK1-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP26]], align 8
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP33]], align 4
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP34]], align 4
+// CHECK1-NEXT: [[TMP35:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l53.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP36:%.*]] = icmp ne i32 [[TMP35]], 0
+// CHECK1-NEXT: br i1 [[TMP36]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l53() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -520,7 +536,7 @@ int main() {
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -537,71 +553,88 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l78.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l78.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l78() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l85() #[[ATTR2]]
-// CHECK1-NEXT: [[TMP11:%.*]] = load i32, ptr @Arg, align 4
-// CHECK1-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP11]], 0
+// CHECK1-NEXT: [[TMP15:%.*]] = load i32, ptr @Arg, align 4
+// CHECK1-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP15]], 0
// CHECK1-NEXT: [[FROMBOOL:%.*]] = zext i1 [[TOBOOL]] to i8
// CHECK1-NEXT: store i8 [[FROMBOOL]], ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TMP12:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TOBOOL1:%.*]] = trunc i8 [[TMP12]] to i1
+// CHECK1-NEXT: [[TMP16:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK1-NEXT: [[TOBOOL1:%.*]] = trunc i8 [[TMP16]] to i1
// CHECK1-NEXT: [[FROMBOOL2:%.*]] = zext i1 [[TOBOOL1]] to i8
// CHECK1-NEXT: store i8 [[FROMBOOL2]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 1
-// CHECK1-NEXT: [[TMP13:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TOBOOL3:%.*]] = trunc i8 [[TMP14]] to i1
+// CHECK1-NEXT: [[TMP17:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK1-NEXT: [[TMP18:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK1-NEXT: [[TOBOOL3:%.*]] = trunc i8 [[TMP18]] to i1
// CHECK1-NEXT: br i1 [[TOBOOL3]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK1: omp_if.then:
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP13]], ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP13]], ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP20:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TOBOOL4:%.*]] = trunc i8 [[TMP20]] to i1
-// CHECK1-NEXT: [[TMP21:%.*]] = select i1 [[TOBOOL4]], i32 0, i32 1
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP17]], ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP17]], ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP24:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK1-NEXT: [[TOBOOL4:%.*]] = trunc i8 [[TMP24]] to i1
+// CHECK1-NEXT: [[TMP25:%.*]] = select i1 [[TOBOOL4]], i32 0, i32 1
+// CHECK1-NEXT: [[TMP26:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP25]], 0
// CHECK1-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP23]], align 4
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP18]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP19]], ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.10, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.11, ptr [[TMP27]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP30]], align 8
-// CHECK1-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP21]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92.region_id, ptr [[KERNEL_ARGS6]])
-// CHECK1-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
-// CHECK1-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP28]], align 4
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP22]], ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP23]], ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.10, ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.11, ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP35]], align 8
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP36]], align 8
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP26]], ptr [[TMP38]], align 4
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP39]], align 4
+// CHECK1-NEXT: [[TMP40:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP25]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92.region_id, ptr [[KERNEL_ARGS6]])
+// CHECK1-NEXT: [[TMP41:%.*]] = icmp ne i32 [[TMP40]], 0
+// CHECK1-NEXT: br i1 [[TMP41]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK1: omp_offload.failed7:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92(i64 [[TMP13]]) #[[ATTR2]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92(i64 [[TMP17]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK1: omp_offload.cont8:
// CHECK1-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK1: omp_if.else:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92(i64 [[TMP13]]) #[[ATTR2]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92(i64 [[TMP17]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
-// CHECK1-NEXT: [[TMP33:%.*]] = load i32, ptr @Arg, align 4
-// CHECK1-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP33]])
+// CHECK1-NEXT: [[TMP42:%.*]] = load i32, ptr @Arg, align 4
+// CHECK1-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP42]])
// CHECK1-NEXT: ret i32 [[CALL]]
//
//
@@ -1104,7 +1137,7 @@ int main() {
// CHECK1-NEXT: store i32 [[ARG]], ptr [[ARG_ADDR]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1121,58 +1154,75 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l62.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l62.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l62() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l66() #[[ATTR2]]
-// CHECK1-NEXT: [[TMP11:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
-// CHECK1-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP11]], 0
+// CHECK1-NEXT: [[TMP15:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
+// CHECK1-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP15]], 0
// CHECK1-NEXT: [[FROMBOOL:%.*]] = zext i1 [[TOBOOL]] to i8
// CHECK1-NEXT: store i8 [[FROMBOOL]], ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TMP12:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TOBOOL1:%.*]] = trunc i8 [[TMP12]] to i1
+// CHECK1-NEXT: [[TMP16:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK1-NEXT: [[TOBOOL1:%.*]] = trunc i8 [[TMP16]] to i1
// CHECK1-NEXT: [[FROMBOOL2:%.*]] = zext i1 [[TOBOOL1]] to i8
// CHECK1-NEXT: store i8 [[FROMBOOL2]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 1
-// CHECK1-NEXT: [[TMP13:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP13]], ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP13]], ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP19:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TOBOOL3:%.*]] = trunc i8 [[TMP19]] to i1
-// CHECK1-NEXT: [[TMP20:%.*]] = select i1 [[TOBOOL3]], i32 0, i32 1
+// CHECK1-NEXT: [[TMP17:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP17]], ptr [[TMP18]], align 8
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP17]], ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP23:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK1-NEXT: [[TOBOOL3:%.*]] = trunc i8 [[TMP23]] to i1
+// CHECK1-NEXT: [[TMP24:%.*]] = select i1 [[TOBOOL3]], i32 0, i32 1
+// CHECK1-NEXT: [[TMP25:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP24]], 0
// CHECK1-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP17]], ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP18]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.18, ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.19, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP20]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l70.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP21]], ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP22]], ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.18, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.19, ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP35]], align 8
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP25]], ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP38]], align 4
+// CHECK1-NEXT: [[TMP39:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP24]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l70.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK1-NEXT: [[TMP40:%.*]] = icmp ne i32 [[TMP39]], 0
+// CHECK1-NEXT: br i1 [[TMP40]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK1: omp_offload.failed6:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l70(i64 [[TMP13]]) #[[ATTR2]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l70(i64 [[TMP17]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK1: omp_offload.cont7:
// CHECK1-NEXT: ret i32 0
@@ -1692,7 +1742,7 @@ int main() {
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1709,35 +1759,51 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP14]], align 8
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 100, ptr [[TMP15]], align 8
-// CHECK3-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l45.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK3-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK3-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l45.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK3-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l45(i64 [[TMP1]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP18]], align 4
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK3-NEXT: store i32 0, ptr [[TMP19]], align 4
-// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK3-NEXT: store ptr null, ptr [[TMP20]], align 8
-// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK3-NEXT: store ptr null, ptr [[TMP21]], align 8
-// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK3-NEXT: store ptr null, ptr [[TMP22]], align 8
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK3-NEXT: store ptr null, ptr [[TMP23]], align 8
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP22]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK3-NEXT: store i32 0, ptr [[TMP23]], align 4
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK3-NEXT: store ptr null, ptr [[TMP24]], align 8
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK3-NEXT: store ptr null, ptr [[TMP25]], align 8
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK3-NEXT: store i64 100, ptr [[TMP26]], align 8
-// CHECK3-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l53.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK3-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK3-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK3-NEXT: store ptr null, ptr [[TMP26]], align 8
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK3-NEXT: store ptr null, ptr [[TMP27]], align 8
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 8
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP29]], align 8
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK3-NEXT: store i64 100, ptr [[TMP30]], align 8
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP31]], align 8
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP33]], align 4
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP34]], align 4
+// CHECK3-NEXT: [[TMP35:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l53.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK3-NEXT: [[TMP36:%.*]] = icmp ne i32 [[TMP35]], 0
+// CHECK3-NEXT: br i1 [[TMP36]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK3: omp_offload.failed3:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l53() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -2082,7 +2148,7 @@ int main() {
// CHECK3-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2099,71 +2165,88 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l78.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK3-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l78.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK3-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l78() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l85() #[[ATTR2]]
-// CHECK3-NEXT: [[TMP11:%.*]] = load i32, ptr @Arg, align 4
-// CHECK3-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP11]], 0
+// CHECK3-NEXT: [[TMP15:%.*]] = load i32, ptr @Arg, align 4
+// CHECK3-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP15]], 0
// CHECK3-NEXT: [[FROMBOOL:%.*]] = zext i1 [[TOBOOL]] to i8
// CHECK3-NEXT: store i8 [[FROMBOOL]], ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK3-NEXT: [[TMP12:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK3-NEXT: [[TOBOOL1:%.*]] = trunc i8 [[TMP12]] to i1
+// CHECK3-NEXT: [[TMP16:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK3-NEXT: [[TOBOOL1:%.*]] = trunc i8 [[TMP16]] to i1
// CHECK3-NEXT: [[FROMBOOL2:%.*]] = zext i1 [[TOBOOL1]] to i8
// CHECK3-NEXT: store i8 [[FROMBOOL2]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 1
-// CHECK3-NEXT: [[TMP13:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK3-NEXT: [[TOBOOL3:%.*]] = trunc i8 [[TMP14]] to i1
+// CHECK3-NEXT: [[TMP17:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK3-NEXT: [[TMP18:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK3-NEXT: [[TOBOOL3:%.*]] = trunc i8 [[TMP18]] to i1
// CHECK3-NEXT: br i1 [[TOBOOL3]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK3: omp_if.then:
-// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK3-NEXT: store i64 [[TMP13]], ptr [[TMP15]], align 8
-// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK3-NEXT: store i64 [[TMP13]], ptr [[TMP16]], align 8
-// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP20:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK3-NEXT: [[TOBOOL4:%.*]] = trunc i8 [[TMP20]] to i1
-// CHECK3-NEXT: [[TMP21:%.*]] = select i1 [[TOBOOL4]], i32 0, i32 1
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK3-NEXT: store i64 [[TMP17]], ptr [[TMP19]], align 8
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK3-NEXT: store i64 [[TMP17]], ptr [[TMP20]], align 8
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP24:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK3-NEXT: [[TOBOOL4:%.*]] = trunc i8 [[TMP24]] to i1
+// CHECK3-NEXT: [[TMP25:%.*]] = select i1 [[TOBOOL4]], i32 0, i32 1
+// CHECK3-NEXT: [[TMP26:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP25]], 0
// CHECK3-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP23]], align 4
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP18]], ptr [[TMP24]], align 8
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP19]], ptr [[TMP25]], align 8
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.11, ptr [[TMP26]], align 8
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP27]], align 8
-// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP29]], align 8
-// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK3-NEXT: store i64 100, ptr [[TMP30]], align 8
-// CHECK3-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP21]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92.region_id, ptr [[KERNEL_ARGS6]])
-// CHECK3-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
-// CHECK3-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP22]], ptr [[TMP29]], align 8
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP23]], ptr [[TMP30]], align 8
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.11, ptr [[TMP31]], align 8
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP32]], align 8
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP33]], align 8
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK3-NEXT: store i64 100, ptr [[TMP35]], align 8
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP36]], align 8
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [[TMP26]], ptr [[TMP38]], align 4
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP39]], align 4
+// CHECK3-NEXT: [[TMP40:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP25]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92.region_id, ptr [[KERNEL_ARGS6]])
+// CHECK3-NEXT: [[TMP41:%.*]] = icmp ne i32 [[TMP40]], 0
+// CHECK3-NEXT: br i1 [[TMP41]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK3: omp_offload.failed7:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92(i64 [[TMP13]]) #[[ATTR2]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92(i64 [[TMP17]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK3: omp_offload.cont8:
// CHECK3-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK3: omp_if.else:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92(i64 [[TMP13]]) #[[ATTR2]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92(i64 [[TMP17]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
-// CHECK3-NEXT: [[TMP33:%.*]] = load i32, ptr @Arg, align 4
-// CHECK3-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP33]])
+// CHECK3-NEXT: [[TMP42:%.*]] = load i32, ptr @Arg, align 4
+// CHECK3-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP42]])
// CHECK3-NEXT: ret i32 [[CALL]]
//
//
@@ -2896,7 +2979,7 @@ int main() {
// CHECK3-NEXT: store i32 [[ARG]], ptr [[ARG_ADDR]], align 4
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2913,58 +2996,75 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l62.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK3-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l62.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK3-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l62() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l66() #[[ATTR2]]
-// CHECK3-NEXT: [[TMP11:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
-// CHECK3-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP11]], 0
+// CHECK3-NEXT: [[TMP15:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
+// CHECK3-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP15]], 0
// CHECK3-NEXT: [[FROMBOOL:%.*]] = zext i1 [[TOBOOL]] to i8
// CHECK3-NEXT: store i8 [[FROMBOOL]], ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK3-NEXT: [[TMP12:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK3-NEXT: [[TOBOOL1:%.*]] = trunc i8 [[TMP12]] to i1
+// CHECK3-NEXT: [[TMP16:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK3-NEXT: [[TOBOOL1:%.*]] = trunc i8 [[TMP16]] to i1
// CHECK3-NEXT: [[FROMBOOL2:%.*]] = zext i1 [[TOBOOL1]] to i8
// CHECK3-NEXT: store i8 [[FROMBOOL2]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 1
-// CHECK3-NEXT: [[TMP13:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK3-NEXT: store i64 [[TMP13]], ptr [[TMP14]], align 8
-// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK3-NEXT: store i64 [[TMP13]], ptr [[TMP15]], align 8
-// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP19:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK3-NEXT: [[TOBOOL3:%.*]] = trunc i8 [[TMP19]] to i1
-// CHECK3-NEXT: [[TMP20:%.*]] = select i1 [[TOBOOL3]], i32 0, i32 1
+// CHECK3-NEXT: [[TMP17:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK3-NEXT: store i64 [[TMP17]], ptr [[TMP18]], align 8
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK3-NEXT: store i64 [[TMP17]], ptr [[TMP19]], align 8
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP23:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK3-NEXT: [[TOBOOL3:%.*]] = trunc i8 [[TMP23]] to i1
+// CHECK3-NEXT: [[TMP24:%.*]] = select i1 [[TOBOOL3]], i32 0, i32 1
+// CHECK3-NEXT: [[TMP25:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP24]], 0
// CHECK3-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP17]], ptr [[TMP23]], align 8
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP18]], ptr [[TMP24]], align 8
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.19, ptr [[TMP25]], align 8
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.20, ptr [[TMP26]], align 8
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK3-NEXT: store i64 100, ptr [[TMP29]], align 8
-// CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP20]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l70.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK3-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK3-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP21]], ptr [[TMP28]], align 8
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP22]], ptr [[TMP29]], align 8
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.19, ptr [[TMP30]], align 8
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.20, ptr [[TMP31]], align 8
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP33]], align 8
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK3-NEXT: store i64 100, ptr [[TMP34]], align 8
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP35]], align 8
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [[TMP25]], ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP38]], align 4
+// CHECK3-NEXT: [[TMP39:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP24]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l70.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK3-NEXT: [[TMP40:%.*]] = icmp ne i32 [[TMP39]], 0
+// CHECK3-NEXT: br i1 [[TMP40]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK3: omp_offload.failed6:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l70(i64 [[TMP13]]) #[[ATTR2]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l70(i64 [[TMP17]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK3: omp_offload.cont7:
// CHECK3-NEXT: ret i32 0
@@ -4077,7 +4177,7 @@ int main() {
// CHECK9-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK9-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK9-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -4094,35 +4194,51 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP14]], align 8
// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 100, ptr [[TMP15]], align 8
-// CHECK9-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l47.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK9-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK9-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l47.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK9-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l47(i64 [[TMP1]]) #[[ATTR2:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP18]], align 4
-// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK9-NEXT: store i32 0, ptr [[TMP19]], align 4
-// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP20]], align 8
-// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP21]], align 8
-// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK9-NEXT: store ptr null, ptr [[TMP22]], align 8
-// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK9-NEXT: store ptr null, ptr [[TMP23]], align 8
-// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP22]], align 4
+// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK9-NEXT: store i32 0, ptr [[TMP23]], align 4
+// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK9-NEXT: store ptr null, ptr [[TMP24]], align 8
-// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK9-NEXT: store ptr null, ptr [[TMP25]], align 8
-// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK9-NEXT: store i64 100, ptr [[TMP26]], align 8
-// CHECK9-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l53.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK9-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK9-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK9-NEXT: store ptr null, ptr [[TMP26]], align 8
+// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK9-NEXT: store ptr null, ptr [[TMP27]], align 8
+// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP28]], align 8
+// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP29]], align 8
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK9-NEXT: store i64 100, ptr [[TMP30]], align 8
+// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP31]], align 8
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP33]], align 4
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP34]], align 4
+// CHECK9-NEXT: [[TMP35:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l53.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK9-NEXT: [[TMP36:%.*]] = icmp ne i32 [[TMP35]], 0
+// CHECK9-NEXT: br i1 [[TMP36]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK9: omp_offload.failed3:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l53() #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -4467,7 +4583,7 @@ int main() {
// CHECK9-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK9-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK9-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -4484,71 +4600,88 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK9-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK9-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l78.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK9-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK9-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK9-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l78.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK9-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l78() #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l85() #[[ATTR2]]
-// CHECK9-NEXT: [[TMP11:%.*]] = load i32, ptr @Arg, align 4
-// CHECK9-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP11]], 0
+// CHECK9-NEXT: [[TMP15:%.*]] = load i32, ptr @Arg, align 4
+// CHECK9-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP15]], 0
// CHECK9-NEXT: [[FROMBOOL:%.*]] = zext i1 [[TOBOOL]] to i8
// CHECK9-NEXT: store i8 [[FROMBOOL]], ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK9-NEXT: [[TMP12:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK9-NEXT: [[TOBOOL1:%.*]] = trunc i8 [[TMP12]] to i1
+// CHECK9-NEXT: [[TMP16:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK9-NEXT: [[TOBOOL1:%.*]] = trunc i8 [[TMP16]] to i1
// CHECK9-NEXT: [[FROMBOOL2:%.*]] = zext i1 [[TOBOOL1]] to i8
// CHECK9-NEXT: store i8 [[FROMBOOL2]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 1
-// CHECK9-NEXT: [[TMP13:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK9-NEXT: [[TOBOOL3:%.*]] = trunc i8 [[TMP14]] to i1
+// CHECK9-NEXT: [[TMP17:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK9-NEXT: [[TMP18:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK9-NEXT: [[TOBOOL3:%.*]] = trunc i8 [[TMP18]] to i1
// CHECK9-NEXT: br i1 [[TOBOOL3]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK9: omp_if.then:
-// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP13]], ptr [[TMP15]], align 8
-// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP13]], ptr [[TMP16]], align 8
-// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP20:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK9-NEXT: [[TOBOOL4:%.*]] = trunc i8 [[TMP20]] to i1
-// CHECK9-NEXT: [[TMP21:%.*]] = select i1 [[TOBOOL4]], i32 0, i32 1
+// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP17]], ptr [[TMP19]], align 8
+// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP17]], ptr [[TMP20]], align 8
+// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP24:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK9-NEXT: [[TOBOOL4:%.*]] = trunc i8 [[TMP24]] to i1
+// CHECK9-NEXT: [[TMP25:%.*]] = select i1 [[TOBOOL4]], i32 0, i32 1
+// CHECK9-NEXT: [[TMP26:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP25]], 0
// CHECK9-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK9-NEXT: store i32 1, ptr [[TMP23]], align 4
-// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP18]], ptr [[TMP24]], align 8
-// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP19]], ptr [[TMP25]], align 8
-// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.10, ptr [[TMP26]], align 8
-// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.11, ptr [[TMP27]], align 8
-// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP29]], align 8
-// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK9-NEXT: store i64 100, ptr [[TMP30]], align 8
-// CHECK9-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP21]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92.region_id, ptr [[KERNEL_ARGS6]])
-// CHECK9-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
-// CHECK9-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP27]], align 4
+// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK9-NEXT: store i32 1, ptr [[TMP28]], align 4
+// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP22]], ptr [[TMP29]], align 8
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP23]], ptr [[TMP30]], align 8
+// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.10, ptr [[TMP31]], align 8
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.11, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP33]], align 8
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK9-NEXT: store i64 100, ptr [[TMP35]], align 8
+// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP36]], align 8
+// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP37]], align 4
+// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [[TMP26]], ptr [[TMP38]], align 4
+// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP39]], align 4
+// CHECK9-NEXT: [[TMP40:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP25]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92.region_id, ptr [[KERNEL_ARGS6]])
+// CHECK9-NEXT: [[TMP41:%.*]] = icmp ne i32 [[TMP40]], 0
+// CHECK9-NEXT: br i1 [[TMP41]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK9: omp_offload.failed7:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92(i64 [[TMP13]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92(i64 [[TMP17]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK9: omp_offload.cont8:
// CHECK9-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK9: omp_if.else:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92(i64 [[TMP13]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92(i64 [[TMP17]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_IF_END]]
// CHECK9: omp_if.end:
-// CHECK9-NEXT: [[TMP33:%.*]] = load i32, ptr @Arg, align 4
-// CHECK9-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP33]])
+// CHECK9-NEXT: [[TMP42:%.*]] = load i32, ptr @Arg, align 4
+// CHECK9-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP42]])
// CHECK9-NEXT: ret i32 [[CALL]]
//
//
@@ -5051,7 +5184,7 @@ int main() {
// CHECK9-NEXT: store i32 [[ARG]], ptr [[ARG_ADDR]], align 4
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK9-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK9-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -5068,58 +5201,75 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK9-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK9-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l62.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK9-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK9-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK9-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l62.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK9-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l62() #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l66() #[[ATTR2]]
-// CHECK9-NEXT: [[TMP11:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
-// CHECK9-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP11]], 0
+// CHECK9-NEXT: [[TMP15:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
+// CHECK9-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP15]], 0
// CHECK9-NEXT: [[FROMBOOL:%.*]] = zext i1 [[TOBOOL]] to i8
// CHECK9-NEXT: store i8 [[FROMBOOL]], ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK9-NEXT: [[TMP12:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK9-NEXT: [[TOBOOL1:%.*]] = trunc i8 [[TMP12]] to i1
+// CHECK9-NEXT: [[TMP16:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK9-NEXT: [[TOBOOL1:%.*]] = trunc i8 [[TMP16]] to i1
// CHECK9-NEXT: [[FROMBOOL2:%.*]] = zext i1 [[TOBOOL1]] to i8
// CHECK9-NEXT: store i8 [[FROMBOOL2]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 1
-// CHECK9-NEXT: [[TMP13:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP13]], ptr [[TMP14]], align 8
-// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP13]], ptr [[TMP15]], align 8
-// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP19:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK9-NEXT: [[TOBOOL3:%.*]] = trunc i8 [[TMP19]] to i1
-// CHECK9-NEXT: [[TMP20:%.*]] = select i1 [[TOBOOL3]], i32 0, i32 1
+// CHECK9-NEXT: [[TMP17:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP17]], ptr [[TMP18]], align 8
+// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP17]], ptr [[TMP19]], align 8
+// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP23:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK9-NEXT: [[TOBOOL3:%.*]] = trunc i8 [[TMP23]] to i1
+// CHECK9-NEXT: [[TMP24:%.*]] = select i1 [[TOBOOL3]], i32 0, i32 1
+// CHECK9-NEXT: [[TMP25:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP24]], 0
// CHECK9-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK9-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP17]], ptr [[TMP23]], align 8
-// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP18]], ptr [[TMP24]], align 8
-// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.18, ptr [[TMP25]], align 8
-// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.19, ptr [[TMP26]], align 8
-// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK9-NEXT: store i64 100, ptr [[TMP29]], align 8
-// CHECK9-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP20]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l70.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK9-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK9-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP26]], align 4
+// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK9-NEXT: store i32 1, ptr [[TMP27]], align 4
+// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP21]], ptr [[TMP28]], align 8
+// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP22]], ptr [[TMP29]], align 8
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.18, ptr [[TMP30]], align 8
+// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.19, ptr [[TMP31]], align 8
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP33]], align 8
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK9-NEXT: store i64 100, ptr [[TMP34]], align 8
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP35]], align 8
+// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [[TMP25]], ptr [[TMP37]], align 4
+// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP38]], align 4
+// CHECK9-NEXT: [[TMP39:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP24]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l70.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK9-NEXT: [[TMP40:%.*]] = icmp ne i32 [[TMP39]], 0
+// CHECK9-NEXT: br i1 [[TMP40]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK9: omp_offload.failed6:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l70(i64 [[TMP13]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l70(i64 [[TMP17]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK9: omp_offload.cont7:
// CHECK9-NEXT: ret i32 0
@@ -5639,7 +5789,7 @@ int main() {
// CHECK11-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK11-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK11-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -5656,35 +5806,51 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP14]], align 8
// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 100, ptr [[TMP15]], align 8
-// CHECK11-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l45.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK11-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK11-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l45.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK11-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l45(i64 [[TMP1]]) #[[ATTR2:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP18]], align 4
-// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK11-NEXT: store i32 0, ptr [[TMP19]], align 4
-// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP20]], align 8
-// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP21]], align 8
-// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK11-NEXT: store ptr null, ptr [[TMP22]], align 8
-// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK11-NEXT: store ptr null, ptr [[TMP23]], align 8
-// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP22]], align 4
+// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK11-NEXT: store i32 0, ptr [[TMP23]], align 4
+// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK11-NEXT: store ptr null, ptr [[TMP24]], align 8
-// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK11-NEXT: store ptr null, ptr [[TMP25]], align 8
-// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK11-NEXT: store i64 100, ptr [[TMP26]], align 8
-// CHECK11-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l53.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK11-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK11-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK11-NEXT: store ptr null, ptr [[TMP26]], align 8
+// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK11-NEXT: store ptr null, ptr [[TMP27]], align 8
+// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP28]], align 8
+// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP29]], align 8
+// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK11-NEXT: store i64 100, ptr [[TMP30]], align 8
+// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP31]], align 8
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP33]], align 4
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP34]], align 4
+// CHECK11-NEXT: [[TMP35:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l53.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK11-NEXT: [[TMP36:%.*]] = icmp ne i32 [[TMP35]], 0
+// CHECK11-NEXT: br i1 [[TMP36]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK11: omp_offload.failed3:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l53() #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -6029,7 +6195,7 @@ int main() {
// CHECK11-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK11-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK11-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -6046,71 +6212,88 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK11-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK11-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l78.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK11-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK11-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK11-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK11-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l78.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK11-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l78() #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l85() #[[ATTR2]]
-// CHECK11-NEXT: [[TMP11:%.*]] = load i32, ptr @Arg, align 4
-// CHECK11-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP11]], 0
+// CHECK11-NEXT: [[TMP15:%.*]] = load i32, ptr @Arg, align 4
+// CHECK11-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP15]], 0
// CHECK11-NEXT: [[FROMBOOL:%.*]] = zext i1 [[TOBOOL]] to i8
// CHECK11-NEXT: store i8 [[FROMBOOL]], ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK11-NEXT: [[TMP12:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK11-NEXT: [[TOBOOL1:%.*]] = trunc i8 [[TMP12]] to i1
+// CHECK11-NEXT: [[TMP16:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK11-NEXT: [[TOBOOL1:%.*]] = trunc i8 [[TMP16]] to i1
// CHECK11-NEXT: [[FROMBOOL2:%.*]] = zext i1 [[TOBOOL1]] to i8
// CHECK11-NEXT: store i8 [[FROMBOOL2]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 1
-// CHECK11-NEXT: [[TMP13:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK11-NEXT: [[TMP14:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK11-NEXT: [[TOBOOL3:%.*]] = trunc i8 [[TMP14]] to i1
+// CHECK11-NEXT: [[TMP17:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK11-NEXT: [[TMP18:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK11-NEXT: [[TOBOOL3:%.*]] = trunc i8 [[TMP18]] to i1
// CHECK11-NEXT: br i1 [[TOBOOL3]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK11: omp_if.then:
-// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK11-NEXT: store i64 [[TMP13]], ptr [[TMP15]], align 8
-// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK11-NEXT: store i64 [[TMP13]], ptr [[TMP16]], align 8
-// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP20:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK11-NEXT: [[TOBOOL4:%.*]] = trunc i8 [[TMP20]] to i1
-// CHECK11-NEXT: [[TMP21:%.*]] = select i1 [[TOBOOL4]], i32 0, i32 1
+// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK11-NEXT: store i64 [[TMP17]], ptr [[TMP19]], align 8
+// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK11-NEXT: store i64 [[TMP17]], ptr [[TMP20]], align 8
+// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP24:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK11-NEXT: [[TOBOOL4:%.*]] = trunc i8 [[TMP24]] to i1
+// CHECK11-NEXT: [[TMP25:%.*]] = select i1 [[TOBOOL4]], i32 0, i32 1
+// CHECK11-NEXT: [[TMP26:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP25]], 0
// CHECK11-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK11-NEXT: store i32 1, ptr [[TMP23]], align 4
-// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP18]], ptr [[TMP24]], align 8
-// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP19]], ptr [[TMP25]], align 8
-// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.11, ptr [[TMP26]], align 8
-// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP27]], align 8
-// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP29]], align 8
-// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK11-NEXT: store i64 100, ptr [[TMP30]], align 8
-// CHECK11-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP21]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92.region_id, ptr [[KERNEL_ARGS6]])
-// CHECK11-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
-// CHECK11-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP27]], align 4
+// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK11-NEXT: store i32 1, ptr [[TMP28]], align 4
+// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP22]], ptr [[TMP29]], align 8
+// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP23]], ptr [[TMP30]], align 8
+// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.11, ptr [[TMP31]], align 8
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP32]], align 8
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP33]], align 8
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK11-NEXT: store i64 100, ptr [[TMP35]], align 8
+// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP36]], align 8
+// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP37]], align 4
+// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [[TMP26]], ptr [[TMP38]], align 4
+// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP39]], align 4
+// CHECK11-NEXT: [[TMP40:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP25]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92.region_id, ptr [[KERNEL_ARGS6]])
+// CHECK11-NEXT: [[TMP41:%.*]] = icmp ne i32 [[TMP40]], 0
+// CHECK11-NEXT: br i1 [[TMP41]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK11: omp_offload.failed7:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92(i64 [[TMP13]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92(i64 [[TMP17]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK11: omp_offload.cont8:
// CHECK11-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK11: omp_if.else:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92(i64 [[TMP13]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92(i64 [[TMP17]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_IF_END]]
// CHECK11: omp_if.end:
-// CHECK11-NEXT: [[TMP33:%.*]] = load i32, ptr @Arg, align 4
-// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP33]])
+// CHECK11-NEXT: [[TMP42:%.*]] = load i32, ptr @Arg, align 4
+// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP42]])
// CHECK11-NEXT: ret i32 [[CALL]]
//
//
@@ -6843,7 +7026,7 @@ int main() {
// CHECK11-NEXT: store i32 [[ARG]], ptr [[ARG_ADDR]], align 4
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK11-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK11-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -6860,58 +7043,75 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK11-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK11-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l62.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK11-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK11-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK11-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK11-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l62.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK11-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l62() #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l66() #[[ATTR2]]
-// CHECK11-NEXT: [[TMP11:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
-// CHECK11-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP11]], 0
+// CHECK11-NEXT: [[TMP15:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
+// CHECK11-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP15]], 0
// CHECK11-NEXT: [[FROMBOOL:%.*]] = zext i1 [[TOBOOL]] to i8
// CHECK11-NEXT: store i8 [[FROMBOOL]], ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK11-NEXT: [[TMP12:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK11-NEXT: [[TOBOOL1:%.*]] = trunc i8 [[TMP12]] to i1
+// CHECK11-NEXT: [[TMP16:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK11-NEXT: [[TOBOOL1:%.*]] = trunc i8 [[TMP16]] to i1
// CHECK11-NEXT: [[FROMBOOL2:%.*]] = zext i1 [[TOBOOL1]] to i8
// CHECK11-NEXT: store i8 [[FROMBOOL2]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 1
-// CHECK11-NEXT: [[TMP13:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK11-NEXT: [[TMP14:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK11-NEXT: store i64 [[TMP13]], ptr [[TMP14]], align 8
-// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK11-NEXT: store i64 [[TMP13]], ptr [[TMP15]], align 8
-// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP19:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK11-NEXT: [[TOBOOL3:%.*]] = trunc i8 [[TMP19]] to i1
-// CHECK11-NEXT: [[TMP20:%.*]] = select i1 [[TOBOOL3]], i32 0, i32 1
+// CHECK11-NEXT: [[TMP17:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK11-NEXT: store i64 [[TMP17]], ptr [[TMP18]], align 8
+// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK11-NEXT: store i64 [[TMP17]], ptr [[TMP19]], align 8
+// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP23:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK11-NEXT: [[TOBOOL3:%.*]] = trunc i8 [[TMP23]] to i1
+// CHECK11-NEXT: [[TMP24:%.*]] = select i1 [[TOBOOL3]], i32 0, i32 1
+// CHECK11-NEXT: [[TMP25:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP24]], 0
// CHECK11-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK11-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP17]], ptr [[TMP23]], align 8
-// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP18]], ptr [[TMP24]], align 8
-// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.19, ptr [[TMP25]], align 8
-// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.20, ptr [[TMP26]], align 8
-// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK11-NEXT: store i64 100, ptr [[TMP29]], align 8
-// CHECK11-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP20]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l70.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK11-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK11-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP26]], align 4
+// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK11-NEXT: store i32 1, ptr [[TMP27]], align 4
+// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP21]], ptr [[TMP28]], align 8
+// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP22]], ptr [[TMP29]], align 8
+// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.19, ptr [[TMP30]], align 8
+// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.20, ptr [[TMP31]], align 8
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP33]], align 8
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK11-NEXT: store i64 100, ptr [[TMP34]], align 8
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP35]], align 8
+// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [[TMP25]], ptr [[TMP37]], align 4
+// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP38]], align 4
+// CHECK11-NEXT: [[TMP39:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP24]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l70.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK11-NEXT: [[TMP40:%.*]] = icmp ne i32 [[TMP39]], 0
+// CHECK11-NEXT: br i1 [[TMP40]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK11: omp_offload.failed6:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l70(i64 [[TMP13]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l70(i64 [[TMP17]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK11: omp_offload.cont7:
// CHECK11-NEXT: ret i32 0
diff --git a/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_lastprivate_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_lastprivate_codegen.cpp
index 7c512d933760f..ea51af89e47f5 100644
--- a/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_lastprivate_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_lastprivate_codegen.cpp
@@ -825,7 +825,7 @@ int main() {
// CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 5, ptr [[TMP24]], align 4
// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -842,9 +842,17 @@ int main() {
// CHECK5-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK5-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 2, ptr [[TMP31]], align 8
-// CHECK5-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l106.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK5-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK5-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK5-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK5-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK5-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l106.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK5-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l106(ptr [[VEC]], i64 [[TMP2]], ptr [[S_ARR]], ptr [[TMP3]], i64 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -852,18 +860,18 @@ int main() {
// CHECK5-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiET_v()
// CHECK5-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK5-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK5-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK5-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK5: arraydestroy.body:
-// CHECK5-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP34]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK5-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP38]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK5-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK5-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK5-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK5-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK5: arraydestroy.done2:
// CHECK5-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK5-NEXT: [[TMP35:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK5-NEXT: ret i32 [[TMP35]]
+// CHECK5-NEXT: [[TMP39:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK5-NEXT: ret i32 [[TMP39]]
//
//
// CHECK5-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -1300,7 +1308,7 @@ int main() {
// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1317,27 +1325,35 @@ int main() {
// CHECK5-NEXT: store ptr null, ptr [[TMP25]], align 8
// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 2, ptr [[TMP26]], align 8
-// CHECK5-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l50.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK5-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK5-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK5-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP29]], align 4
+// CHECK5-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK5-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l50.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK5-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l50(ptr [[VEC]], i64 [[TMP2]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK5: omp_offload.cont:
// CHECK5-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK5-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK5-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK5-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK5: arraydestroy.body:
-// CHECK5-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP29]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK5-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP33]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK5-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK5-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK5-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK5-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK5: arraydestroy.done2:
// CHECK5-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK5-NEXT: [[TMP30:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK5-NEXT: ret i32 [[TMP30]]
+// CHECK5-NEXT: [[TMP34:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK5-NEXT: ret i32 [[TMP34]]
//
//
// CHECK5-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
@@ -1844,7 +1860,7 @@ int main() {
// CHECK7-NEXT: [[TMP22:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK7-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK7-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK7-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK7-NEXT: store i32 5, ptr [[TMP24]], align 4
// CHECK7-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1861,9 +1877,17 @@ int main() {
// CHECK7-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK7-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK7-NEXT: store i64 2, ptr [[TMP31]], align 8
-// CHECK7-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l106.region_id, ptr [[KERNEL_ARGS]])
-// CHECK7-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK7-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK7-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK7-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK7-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK7-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK7-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l106.region_id, ptr [[KERNEL_ARGS]])
+// CHECK7-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK7-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK7: omp_offload.failed:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l106(ptr [[VEC]], i32 [[TMP2]], ptr [[S_ARR]], ptr [[TMP3]], i32 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1871,18 +1895,18 @@ int main() {
// CHECK7-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiET_v()
// CHECK7-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK7-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK7-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK7-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK7: arraydestroy.body:
-// CHECK7-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP34]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK7-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP38]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK7-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK7-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK7-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK7-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK7: arraydestroy.done2:
// CHECK7-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK7-NEXT: [[TMP35:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK7-NEXT: ret i32 [[TMP35]]
+// CHECK7-NEXT: [[TMP39:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK7-NEXT: ret i32 [[TMP39]]
//
//
// CHECK7-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -2313,7 +2337,7 @@ int main() {
// CHECK7-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK7-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK7-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK7-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK7-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK7-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2330,27 +2354,35 @@ int main() {
// CHECK7-NEXT: store ptr null, ptr [[TMP25]], align 4
// CHECK7-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK7-NEXT: store i64 2, ptr [[TMP26]], align 8
-// CHECK7-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l50.region_id, ptr [[KERNEL_ARGS]])
-// CHECK7-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK7-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK7-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK7-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK7-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP29]], align 4
+// CHECK7-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK7-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l50.region_id, ptr [[KERNEL_ARGS]])
+// CHECK7-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK7-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK7: omp_offload.failed:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l50(ptr [[VEC]], i32 [[TMP2]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK7: omp_offload.cont:
// CHECK7-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK7-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK7-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK7-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK7: arraydestroy.body:
-// CHECK7-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP29]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK7-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP33]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK7-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK7-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK7-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK7-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK7: arraydestroy.done2:
// CHECK7-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK7-NEXT: [[TMP30:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK7-NEXT: ret i32 [[TMP30]]
+// CHECK7-NEXT: [[TMP34:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK7-NEXT: ret i32 [[TMP34]]
//
//
// CHECK7-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
diff --git a/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_private_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_private_codegen.cpp
index 90f939352eafa..a7d9e847e2f67 100644
--- a/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_private_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_private_codegen.cpp
@@ -301,7 +301,7 @@ int main() {
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -318,9 +318,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l124.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l124.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l124() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -577,7 +585,7 @@ int main() {
// CHECK1-NEXT: store ptr undef, ptr [[_TMP1]], align 8
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -594,27 +602,35 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l80.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l80.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l80() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK1-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK1: arraydestroy.body:
-// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK1-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK1-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK1: arraydestroy.done2:
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK1-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK1-NEXT: ret i32 [[TMP12]]
+// CHECK1-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK1-NEXT: ret i32 [[TMP16]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN1SIiEC1Ev
@@ -1053,7 +1069,7 @@ int main() {
// CHECK3-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1070,9 +1086,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l124.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK3-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l124.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK3-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l124() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1323,7 +1347,7 @@ int main() {
// CHECK3-NEXT: store ptr undef, ptr [[_TMP1]], align 4
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1340,27 +1364,35 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l80.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK3-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l80.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK3-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l80() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK3-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK3-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK3: arraydestroy.body:
-// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK3-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK3-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK3: arraydestroy.done2:
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK3-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK3-NEXT: ret i32 [[TMP12]]
+// CHECK3-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK3-NEXT: ret i32 [[TMP16]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN1SIiEC1Ev
diff --git a/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_proc_bind_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_proc_bind_codegen.cpp
index 78b0062e10134..a953d41542673 100644
--- a/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_proc_bind_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_proc_bind_codegen.cpp
@@ -57,7 +57,7 @@ int main() {
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -74,35 +74,51 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 1000, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l36.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l36.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l36() #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 1000, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l38.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 1000, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l38.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK1-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l38() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -421,7 +437,7 @@ int main() {
// CHECK1-NEXT: [[TMP:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -438,9 +454,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 1000, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l30.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l30.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l30() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_reduction_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_reduction_codegen.cpp
index 6127ba076d96f..652e0ce08a90b 100644
--- a/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_reduction_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_reduction_codegen.cpp
@@ -109,7 +109,7 @@ int main() {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -126,9 +126,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66(ptr @_ZZ4mainE5sivar) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -406,7 +414,7 @@ int main() {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -423,9 +431,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32(ptr [[T_VAR]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -707,7 +723,7 @@ int main() {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -724,9 +740,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66(ptr @_ZZ4mainE5sivar) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1000,7 +1024,7 @@ int main() {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1017,9 +1041,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32(ptr [[T_VAR]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_schedule_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_schedule_codegen.cpp
index 363d2828c4f36..c4f504bf066a7 100644
--- a/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_schedule_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_schedule_codegen.cpp
@@ -239,7 +239,7 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -256,157 +256,197 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.4, ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.4, ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK1-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK1-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK1: omp_offload.failed8:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK1: omp_offload.cont9:
// CHECK1-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A10]], ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A10]], ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.8, ptr [[TMP41]], align 8
-// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP42]], align 8
-// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP43]], align 8
-// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP44]], align 8
-// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK1-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK1-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK1-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 8
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 8
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.8, ptr [[TMP49]], align 8
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP50]], align 8
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP51]], align 8
+// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP52]], align 8
+// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK1-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK1-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK1-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK1: omp_offload.failed16:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK1: omp_offload.cont17:
// CHECK1-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP48]], align 8
-// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A18]], ptr [[TMP49]], align 8
-// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP50]], align 8
-// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP60:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP60]], align 8
+// CHECK1-NEXT: [[TMP61:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A18]], ptr [[TMP61]], align 8
+// CHECK1-NEXT: [[TMP62:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP62]], align 8
+// CHECK1-NEXT: [[TMP63:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS23:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP53]], align 4
-// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP51]], ptr [[TMP55]], align 8
-// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP52]], ptr [[TMP56]], align 8
-// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.12, ptr [[TMP57]], align 8
-// CHECK1-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP58]], align 8
-// CHECK1-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP59]], align 8
-// CHECK1-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP60]], align 8
-// CHECK1-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP61]], align 8
-// CHECK1-NEXT: [[TMP62:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.region_id, ptr [[KERNEL_ARGS23]])
-// CHECK1-NEXT: [[TMP63:%.*]] = icmp ne i32 [[TMP62]], 0
-// CHECK1-NEXT: br i1 [[TMP63]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
+// CHECK1-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP65]], align 4
+// CHECK1-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP66]], align 4
+// CHECK1-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP63]], ptr [[TMP67]], align 8
+// CHECK1-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP64]], ptr [[TMP68]], align 8
+// CHECK1-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.12, ptr [[TMP69]], align 8
+// CHECK1-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP70]], align 8
+// CHECK1-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP71]], align 8
+// CHECK1-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP72]], align 8
+// CHECK1-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP73]], align 8
+// CHECK1-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP74]], align 8
+// CHECK1-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP75]], align 4
+// CHECK1-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP76]], align 4
+// CHECK1-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP77]], align 4
+// CHECK1-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.region_id, ptr [[KERNEL_ARGS23]])
+// CHECK1-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
+// CHECK1-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
// CHECK1: omp_offload.failed24:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT25]]
// CHECK1: omp_offload.cont25:
// CHECK1-NEXT: [[A26:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP64]], align 8
-// CHECK1-NEXT: [[TMP65:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A26]], ptr [[TMP65]], align 8
-// CHECK1-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP66]], align 8
-// CHECK1-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP80:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP80]], align 8
+// CHECK1-NEXT: [[TMP81:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A26]], ptr [[TMP81]], align 8
+// CHECK1-NEXT: [[TMP82:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP82]], align 8
+// CHECK1-NEXT: [[TMP83:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP84:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP69]], align 4
-// CHECK1-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP70]], align 4
-// CHECK1-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP67]], ptr [[TMP71]], align 8
-// CHECK1-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 8
-// CHECK1-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.16, ptr [[TMP73]], align 8
-// CHECK1-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP74]], align 8
-// CHECK1-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP75]], align 8
-// CHECK1-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP76]], align 8
-// CHECK1-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP77]], align 8
-// CHECK1-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK1-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
-// CHECK1-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK1-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP85]], align 4
+// CHECK1-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP86]], align 4
+// CHECK1-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP83]], ptr [[TMP87]], align 8
+// CHECK1-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP84]], ptr [[TMP88]], align 8
+// CHECK1-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.16, ptr [[TMP89]], align 8
+// CHECK1-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP90]], align 8
+// CHECK1-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP91]], align 8
+// CHECK1-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP92]], align 8
+// CHECK1-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP93]], align 8
+// CHECK1-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP94]], align 8
+// CHECK1-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP95]], align 4
+// CHECK1-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP96]], align 4
+// CHECK1-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP97]], align 4
+// CHECK1-NEXT: [[TMP98:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK1-NEXT: [[TMP99:%.*]] = icmp ne i32 [[TMP98]], 0
+// CHECK1-NEXT: br i1 [[TMP99]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK1: omp_offload.failed32:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK1: omp_offload.cont33:
// CHECK1-NEXT: [[A34:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A34]], i64 0, i64 0
-// CHECK1-NEXT: [[TMP80:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK1-NEXT: ret i32 [[TMP80]]
+// CHECK1-NEXT: [[TMP100:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK1-NEXT: ret i32 [[TMP100]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36
@@ -1303,7 +1343,7 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1320,157 +1360,197 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 4
-// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A2]], ptr [[TMP17]], align 4
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 4
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A2]], ptr [[TMP21]], align 4
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.4, ptr [[TMP25]], align 4
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP26]], align 4
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK3-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK3-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK3-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.4, ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK3-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK3-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK3-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK3: omp_offload.failed8:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40(ptr [[THIS1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK3: omp_offload.cont9:
// CHECK3-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 4
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A10]], ptr [[TMP33]], align 4
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP34]], align 4
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A10]], ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP42]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 4
-// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 4
-// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.8, ptr [[TMP41]], align 4
-// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP42]], align 4
-// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP44]], align 4
-// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK3-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK3-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK3-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK3-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 4
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 4
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.8, ptr [[TMP49]], align 4
+// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP50]], align 4
+// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP52]], align 4
+// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK3-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK3-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK3-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK3-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK3: omp_offload.failed16:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44(ptr [[THIS1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK3: omp_offload.cont17:
// CHECK3-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP48]], align 4
-// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A18]], ptr [[TMP49]], align 4
-// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP50]], align 4
-// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP60:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP60]], align 4
+// CHECK3-NEXT: [[TMP61:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A18]], ptr [[TMP61]], align 4
+// CHECK3-NEXT: [[TMP62:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP62]], align 4
+// CHECK3-NEXT: [[TMP63:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS23:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP53]], align 4
-// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP51]], ptr [[TMP55]], align 4
-// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP52]], ptr [[TMP56]], align 4
-// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.12, ptr [[TMP57]], align 4
-// CHECK3-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP58]], align 4
-// CHECK3-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP59]], align 4
-// CHECK3-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP60]], align 4
-// CHECK3-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
-// CHECK3-NEXT: store i64 123, ptr [[TMP61]], align 8
-// CHECK3-NEXT: [[TMP62:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.region_id, ptr [[KERNEL_ARGS23]])
-// CHECK3-NEXT: [[TMP63:%.*]] = icmp ne i32 [[TMP62]], 0
-// CHECK3-NEXT: br i1 [[TMP63]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
+// CHECK3-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP65]], align 4
+// CHECK3-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP66]], align 4
+// CHECK3-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP63]], ptr [[TMP67]], align 4
+// CHECK3-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP64]], ptr [[TMP68]], align 4
+// CHECK3-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.12, ptr [[TMP69]], align 4
+// CHECK3-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP70]], align 4
+// CHECK3-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP71]], align 4
+// CHECK3-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP72]], align 4
+// CHECK3-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
+// CHECK3-NEXT: store i64 123, ptr [[TMP73]], align 8
+// CHECK3-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP74]], align 8
+// CHECK3-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP75]], align 4
+// CHECK3-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP76]], align 4
+// CHECK3-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP77]], align 4
+// CHECK3-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.region_id, ptr [[KERNEL_ARGS23]])
+// CHECK3-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
+// CHECK3-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
// CHECK3: omp_offload.failed24:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49(ptr [[THIS1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT25]]
// CHECK3: omp_offload.cont25:
// CHECK3-NEXT: [[A26:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP64]], align 4
-// CHECK3-NEXT: [[TMP65:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A26]], ptr [[TMP65]], align 4
-// CHECK3-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP66]], align 4
-// CHECK3-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP80:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP80]], align 4
+// CHECK3-NEXT: [[TMP81:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A26]], ptr [[TMP81]], align 4
+// CHECK3-NEXT: [[TMP82:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP82]], align 4
+// CHECK3-NEXT: [[TMP83:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP84:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP69]], align 4
-// CHECK3-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP70]], align 4
-// CHECK3-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP67]], ptr [[TMP71]], align 4
-// CHECK3-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 4
-// CHECK3-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.16, ptr [[TMP73]], align 4
-// CHECK3-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP74]], align 4
-// CHECK3-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP75]], align 4
-// CHECK3-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP76]], align 4
-// CHECK3-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK3-NEXT: store i64 123, ptr [[TMP77]], align 8
-// CHECK3-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK3-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
-// CHECK3-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK3-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP85]], align 4
+// CHECK3-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP86]], align 4
+// CHECK3-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP83]], ptr [[TMP87]], align 4
+// CHECK3-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP84]], ptr [[TMP88]], align 4
+// CHECK3-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.16, ptr [[TMP89]], align 4
+// CHECK3-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP90]], align 4
+// CHECK3-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP91]], align 4
+// CHECK3-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP92]], align 4
+// CHECK3-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK3-NEXT: store i64 123, ptr [[TMP93]], align 8
+// CHECK3-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP94]], align 8
+// CHECK3-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP95]], align 4
+// CHECK3-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP96]], align 4
+// CHECK3-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP97]], align 4
+// CHECK3-NEXT: [[TMP98:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK3-NEXT: [[TMP99:%.*]] = icmp ne i32 [[TMP98]], 0
+// CHECK3-NEXT: br i1 [[TMP99]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK3: omp_offload.failed32:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54(ptr [[THIS1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK3: omp_offload.cont33:
// CHECK3-NEXT: [[A34:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A34]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP80:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK3-NEXT: ret i32 [[TMP80]]
+// CHECK3-NEXT: [[TMP100:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK3-NEXT: ret i32 [[TMP100]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36
@@ -2340,7 +2420,7 @@ int main (int argc, char **argv) {
// CHECK5-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK5-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK5-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2357,157 +2437,197 @@ int main (int argc, char **argv) {
// CHECK5-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK5-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK5-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK5-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK5-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK5-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK5-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK5: omp_offload.cont:
// CHECK5-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 8
-// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[A2]], ptr [[TMP17]], align 8
-// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
-// CHECK5-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 8
+// CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[A2]], ptr [[TMP21]], align 8
+// CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
+// CHECK5-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK5-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK5-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK5-NEXT: store ptr @.offload_sizes.4, ptr [[TMP25]], align 8
-// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK5-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP26]], align 8
-// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK5-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK5-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK5-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK5-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK5-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK5-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK5-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK5-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK5-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK5-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK5-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK5-NEXT: store ptr @.offload_sizes.4, ptr [[TMP29]], align 8
+// CHECK5-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK5-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP30]], align 8
+// CHECK5-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK5-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK5-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK5-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK5-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK5-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK5-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK5-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK5-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK5-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK5-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK5: omp_offload.failed8:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40(ptr [[THIS1]]) #[[ATTR2]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK5: omp_offload.cont9:
// CHECK5-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 8
-// CHECK5-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[A10]], ptr [[TMP33]], align 8
-// CHECK5-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
-// CHECK5-NEXT: store ptr null, ptr [[TMP34]], align 8
-// CHECK5-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 8
+// CHECK5-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[A10]], ptr [[TMP41]], align 8
+// CHECK5-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
+// CHECK5-NEXT: store ptr null, ptr [[TMP42]], align 8
+// CHECK5-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK5-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK5-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK5-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK5-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
-// CHECK5-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
-// CHECK5-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK5-NEXT: store ptr @.offload_sizes.8, ptr [[TMP41]], align 8
-// CHECK5-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK5-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP42]], align 8
-// CHECK5-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP43]], align 8
-// CHECK5-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK5-NEXT: store ptr null, ptr [[TMP44]], align 8
-// CHECK5-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK5-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK5-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK5-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK5-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK5-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK5-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK5-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK5-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK5-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 8
+// CHECK5-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 8
+// CHECK5-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK5-NEXT: store ptr @.offload_sizes.8, ptr [[TMP49]], align 8
+// CHECK5-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK5-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP50]], align 8
+// CHECK5-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP51]], align 8
+// CHECK5-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP52]], align 8
+// CHECK5-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK5-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK5-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK5-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK5-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK5-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK5-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK5-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK5-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK5: omp_offload.failed16:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44(ptr [[THIS1]]) #[[ATTR2]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK5: omp_offload.cont17:
// CHECK5-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP48:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP48]], align 8
-// CHECK5-NEXT: [[TMP49:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[A18]], ptr [[TMP49]], align 8
-// CHECK5-NEXT: [[TMP50:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
-// CHECK5-NEXT: store ptr null, ptr [[TMP50]], align 8
-// CHECK5-NEXT: [[TMP51:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP52:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP60:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP60]], align 8
+// CHECK5-NEXT: [[TMP61:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[A18]], ptr [[TMP61]], align 8
+// CHECK5-NEXT: [[TMP62:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
+// CHECK5-NEXT: store ptr null, ptr [[TMP62]], align 8
+// CHECK5-NEXT: [[TMP63:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS23:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP53]], align 4
-// CHECK5-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
-// CHECK5-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK5-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
-// CHECK5-NEXT: store ptr [[TMP51]], ptr [[TMP55]], align 8
-// CHECK5-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[TMP52]], ptr [[TMP56]], align 8
-// CHECK5-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
-// CHECK5-NEXT: store ptr @.offload_sizes.12, ptr [[TMP57]], align 8
-// CHECK5-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
-// CHECK5-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP58]], align 8
-// CHECK5-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP59]], align 8
-// CHECK5-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
-// CHECK5-NEXT: store ptr null, ptr [[TMP60]], align 8
-// CHECK5-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
-// CHECK5-NEXT: store i64 123, ptr [[TMP61]], align 8
-// CHECK5-NEXT: [[TMP62:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.region_id, ptr [[KERNEL_ARGS23]])
-// CHECK5-NEXT: [[TMP63:%.*]] = icmp ne i32 [[TMP62]], 0
-// CHECK5-NEXT: br i1 [[TMP63]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
+// CHECK5-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP65]], align 4
+// CHECK5-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
+// CHECK5-NEXT: store i32 1, ptr [[TMP66]], align 4
+// CHECK5-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
+// CHECK5-NEXT: store ptr [[TMP63]], ptr [[TMP67]], align 8
+// CHECK5-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[TMP64]], ptr [[TMP68]], align 8
+// CHECK5-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
+// CHECK5-NEXT: store ptr @.offload_sizes.12, ptr [[TMP69]], align 8
+// CHECK5-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
+// CHECK5-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP70]], align 8
+// CHECK5-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP71]], align 8
+// CHECK5-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP72]], align 8
+// CHECK5-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
+// CHECK5-NEXT: store i64 123, ptr [[TMP73]], align 8
+// CHECK5-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP74]], align 8
+// CHECK5-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP75]], align 4
+// CHECK5-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP76]], align 4
+// CHECK5-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP77]], align 4
+// CHECK5-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.region_id, ptr [[KERNEL_ARGS23]])
+// CHECK5-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
+// CHECK5-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
// CHECK5: omp_offload.failed24:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49(ptr [[THIS1]]) #[[ATTR2]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT25]]
// CHECK5: omp_offload.cont25:
// CHECK5-NEXT: [[A26:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP64]], align 8
-// CHECK5-NEXT: [[TMP65:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[A26]], ptr [[TMP65]], align 8
-// CHECK5-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i64 0, i64 0
-// CHECK5-NEXT: store ptr null, ptr [[TMP66]], align 8
-// CHECK5-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP80:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP80]], align 8
+// CHECK5-NEXT: [[TMP81:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[A26]], ptr [[TMP81]], align 8
+// CHECK5-NEXT: [[TMP82:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i64 0, i64 0
+// CHECK5-NEXT: store ptr null, ptr [[TMP82]], align 8
+// CHECK5-NEXT: [[TMP83:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP84:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP69]], align 4
-// CHECK5-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK5-NEXT: store i32 1, ptr [[TMP70]], align 4
-// CHECK5-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK5-NEXT: store ptr [[TMP67]], ptr [[TMP71]], align 8
-// CHECK5-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 8
-// CHECK5-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK5-NEXT: store ptr @.offload_sizes.16, ptr [[TMP73]], align 8
-// CHECK5-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK5-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP74]], align 8
-// CHECK5-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP75]], align 8
-// CHECK5-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK5-NEXT: store ptr null, ptr [[TMP76]], align 8
-// CHECK5-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK5-NEXT: store i64 123, ptr [[TMP77]], align 8
-// CHECK5-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK5-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
-// CHECK5-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK5-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP85]], align 4
+// CHECK5-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK5-NEXT: store i32 1, ptr [[TMP86]], align 4
+// CHECK5-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK5-NEXT: store ptr [[TMP83]], ptr [[TMP87]], align 8
+// CHECK5-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[TMP84]], ptr [[TMP88]], align 8
+// CHECK5-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK5-NEXT: store ptr @.offload_sizes.16, ptr [[TMP89]], align 8
+// CHECK5-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK5-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP90]], align 8
+// CHECK5-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP91]], align 8
+// CHECK5-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP92]], align 8
+// CHECK5-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK5-NEXT: store i64 123, ptr [[TMP93]], align 8
+// CHECK5-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP94]], align 8
+// CHECK5-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP95]], align 4
+// CHECK5-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP96]], align 4
+// CHECK5-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP97]], align 4
+// CHECK5-NEXT: [[TMP98:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK5-NEXT: [[TMP99:%.*]] = icmp ne i32 [[TMP98]], 0
+// CHECK5-NEXT: br i1 [[TMP99]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK5: omp_offload.failed32:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54(ptr [[THIS1]]) #[[ATTR2]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK5: omp_offload.cont33:
// CHECK5-NEXT: [[A34:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK5-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A34]], i64 0, i64 0
-// CHECK5-NEXT: [[TMP80:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK5-NEXT: ret i32 [[TMP80]]
+// CHECK5-NEXT: [[TMP100:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK5-NEXT: ret i32 [[TMP100]]
//
//
// CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36
@@ -3404,7 +3524,7 @@ int main (int argc, char **argv) {
// CHECK7-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK7-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK7-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK7-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK7-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK7-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3421,157 +3541,197 @@ int main (int argc, char **argv) {
// CHECK7-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK7-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK7-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK7-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS]])
-// CHECK7-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK7-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK7-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK7-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK7-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK7-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK7-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS]])
+// CHECK7-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK7-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK7: omp_offload.failed:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK7: omp_offload.cont:
// CHECK7-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 4
-// CHECK7-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK7-NEXT: store ptr [[A2]], ptr [[TMP17]], align 4
-// CHECK7-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
-// CHECK7-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK7-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 4
+// CHECK7-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK7-NEXT: store ptr [[A2]], ptr [[TMP21]], align 4
+// CHECK7-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
+// CHECK7-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK7-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK7-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK7-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK7-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK7-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK7-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK7-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK7-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK7-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK7-NEXT: store ptr @.offload_sizes.4, ptr [[TMP25]], align 4
-// CHECK7-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK7-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP26]], align 4
-// CHECK7-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK7-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK7-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK7-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK7-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK7-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK7-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK7-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK7-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK7-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK7-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK7-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK7-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK7-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK7-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK7-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK7-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK7-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK7-NEXT: store ptr @.offload_sizes.4, ptr [[TMP29]], align 4
+// CHECK7-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK7-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP30]], align 4
+// CHECK7-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK7-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK7-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK7-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK7-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK7-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK7-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK7-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK7-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK7-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK7-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK7-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK7-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK7: omp_offload.failed8:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40(ptr [[THIS1]]) #[[ATTR2]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK7: omp_offload.cont9:
// CHECK7-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 4
-// CHECK7-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK7-NEXT: store ptr [[A10]], ptr [[TMP33]], align 4
-// CHECK7-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
-// CHECK7-NEXT: store ptr null, ptr [[TMP34]], align 4
-// CHECK7-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 4
+// CHECK7-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK7-NEXT: store ptr [[A10]], ptr [[TMP41]], align 4
+// CHECK7-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
+// CHECK7-NEXT: store ptr null, ptr [[TMP42]], align 4
+// CHECK7-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK7-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK7-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK7-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK7-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK7-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 4
-// CHECK7-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK7-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 4
-// CHECK7-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK7-NEXT: store ptr @.offload_sizes.8, ptr [[TMP41]], align 4
-// CHECK7-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK7-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP42]], align 4
-// CHECK7-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK7-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK7-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK7-NEXT: store ptr null, ptr [[TMP44]], align 4
-// CHECK7-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK7-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK7-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK7-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK7-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK7-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK7-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK7-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK7-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK7-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK7-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 4
+// CHECK7-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK7-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 4
+// CHECK7-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK7-NEXT: store ptr @.offload_sizes.8, ptr [[TMP49]], align 4
+// CHECK7-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK7-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP50]], align 4
+// CHECK7-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK7-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK7-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK7-NEXT: store ptr null, ptr [[TMP52]], align 4
+// CHECK7-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK7-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK7-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK7-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK7-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK7-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK7-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK7-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK7-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK7: omp_offload.failed16:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44(ptr [[THIS1]]) #[[ATTR2]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK7: omp_offload.cont17:
// CHECK7-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP48:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP48]], align 4
-// CHECK7-NEXT: [[TMP49:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK7-NEXT: store ptr [[A18]], ptr [[TMP49]], align 4
-// CHECK7-NEXT: [[TMP50:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
-// CHECK7-NEXT: store ptr null, ptr [[TMP50]], align 4
-// CHECK7-NEXT: [[TMP51:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP52:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP60:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP60]], align 4
+// CHECK7-NEXT: [[TMP61:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK7-NEXT: store ptr [[A18]], ptr [[TMP61]], align 4
+// CHECK7-NEXT: [[TMP62:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
+// CHECK7-NEXT: store ptr null, ptr [[TMP62]], align 4
+// CHECK7-NEXT: [[TMP63:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS23:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK7-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP53]], align 4
-// CHECK7-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
-// CHECK7-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK7-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
-// CHECK7-NEXT: store ptr [[TMP51]], ptr [[TMP55]], align 4
-// CHECK7-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
-// CHECK7-NEXT: store ptr [[TMP52]], ptr [[TMP56]], align 4
-// CHECK7-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
-// CHECK7-NEXT: store ptr @.offload_sizes.12, ptr [[TMP57]], align 4
-// CHECK7-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
-// CHECK7-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP58]], align 4
-// CHECK7-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
-// CHECK7-NEXT: store ptr null, ptr [[TMP59]], align 4
-// CHECK7-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
-// CHECK7-NEXT: store ptr null, ptr [[TMP60]], align 4
-// CHECK7-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
-// CHECK7-NEXT: store i64 123, ptr [[TMP61]], align 8
-// CHECK7-NEXT: [[TMP62:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.region_id, ptr [[KERNEL_ARGS23]])
-// CHECK7-NEXT: [[TMP63:%.*]] = icmp ne i32 [[TMP62]], 0
-// CHECK7-NEXT: br i1 [[TMP63]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
+// CHECK7-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
+// CHECK7-NEXT: store i32 2, ptr [[TMP65]], align 4
+// CHECK7-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
+// CHECK7-NEXT: store i32 1, ptr [[TMP66]], align 4
+// CHECK7-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
+// CHECK7-NEXT: store ptr [[TMP63]], ptr [[TMP67]], align 4
+// CHECK7-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
+// CHECK7-NEXT: store ptr [[TMP64]], ptr [[TMP68]], align 4
+// CHECK7-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
+// CHECK7-NEXT: store ptr @.offload_sizes.12, ptr [[TMP69]], align 4
+// CHECK7-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
+// CHECK7-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP70]], align 4
+// CHECK7-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
+// CHECK7-NEXT: store ptr null, ptr [[TMP71]], align 4
+// CHECK7-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
+// CHECK7-NEXT: store ptr null, ptr [[TMP72]], align 4
+// CHECK7-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
+// CHECK7-NEXT: store i64 123, ptr [[TMP73]], align 8
+// CHECK7-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP74]], align 8
+// CHECK7-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP75]], align 4
+// CHECK7-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP76]], align 4
+// CHECK7-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP77]], align 4
+// CHECK7-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.region_id, ptr [[KERNEL_ARGS23]])
+// CHECK7-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
+// CHECK7-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
// CHECK7: omp_offload.failed24:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49(ptr [[THIS1]]) #[[ATTR2]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT25]]
// CHECK7: omp_offload.cont25:
// CHECK7-NEXT: [[A26:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP64]], align 4
-// CHECK7-NEXT: [[TMP65:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
-// CHECK7-NEXT: store ptr [[A26]], ptr [[TMP65]], align 4
-// CHECK7-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i32 0, i32 0
-// CHECK7-NEXT: store ptr null, ptr [[TMP66]], align 4
-// CHECK7-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP80:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP80]], align 4
+// CHECK7-NEXT: [[TMP81:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK7-NEXT: store ptr [[A26]], ptr [[TMP81]], align 4
+// CHECK7-NEXT: [[TMP82:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i32 0, i32 0
+// CHECK7-NEXT: store ptr null, ptr [[TMP82]], align 4
+// CHECK7-NEXT: [[TMP83:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP84:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK7-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP69]], align 4
-// CHECK7-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK7-NEXT: store i32 1, ptr [[TMP70]], align 4
-// CHECK7-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK7-NEXT: store ptr [[TMP67]], ptr [[TMP71]], align 4
-// CHECK7-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK7-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 4
-// CHECK7-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK7-NEXT: store ptr @.offload_sizes.16, ptr [[TMP73]], align 4
-// CHECK7-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK7-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP74]], align 4
-// CHECK7-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK7-NEXT: store ptr null, ptr [[TMP75]], align 4
-// CHECK7-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK7-NEXT: store ptr null, ptr [[TMP76]], align 4
-// CHECK7-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK7-NEXT: store i64 123, ptr [[TMP77]], align 8
-// CHECK7-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK7-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
-// CHECK7-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK7-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK7-NEXT: store i32 2, ptr [[TMP85]], align 4
+// CHECK7-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK7-NEXT: store i32 1, ptr [[TMP86]], align 4
+// CHECK7-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK7-NEXT: store ptr [[TMP83]], ptr [[TMP87]], align 4
+// CHECK7-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK7-NEXT: store ptr [[TMP84]], ptr [[TMP88]], align 4
+// CHECK7-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK7-NEXT: store ptr @.offload_sizes.16, ptr [[TMP89]], align 4
+// CHECK7-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK7-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP90]], align 4
+// CHECK7-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK7-NEXT: store ptr null, ptr [[TMP91]], align 4
+// CHECK7-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK7-NEXT: store ptr null, ptr [[TMP92]], align 4
+// CHECK7-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK7-NEXT: store i64 123, ptr [[TMP93]], align 8
+// CHECK7-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP94]], align 8
+// CHECK7-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP95]], align 4
+// CHECK7-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP96]], align 4
+// CHECK7-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP97]], align 4
+// CHECK7-NEXT: [[TMP98:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK7-NEXT: [[TMP99:%.*]] = icmp ne i32 [[TMP98]], 0
+// CHECK7-NEXT: br i1 [[TMP99]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK7: omp_offload.failed32:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54(ptr [[THIS1]]) #[[ATTR2]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK7: omp_offload.cont33:
// CHECK7-NEXT: [[A34:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK7-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A34]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP80:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK7-NEXT: ret i32 [[TMP80]]
+// CHECK7-NEXT: [[TMP100:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK7-NEXT: ret i32 [[TMP100]]
//
//
// CHECK7-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36
@@ -4886,7 +5046,7 @@ int main (int argc, char **argv) {
// CHECK13-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK13-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK13-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK13-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK13-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK13-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK13-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -4903,298 +5063,338 @@ int main (int argc, char **argv) {
// CHECK13-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK13-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK13-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK13-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.region_id, ptr [[KERNEL_ARGS]])
-// CHECK13-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK13-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK13-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK13-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK13-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK13-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK13-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.region_id, ptr [[KERNEL_ARGS]])
+// CHECK13-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK13-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK13: omp_offload.failed:
// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139(i64 [[TMP4]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK13: omp_offload.cont:
-// CHECK13-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK13-NEXT: [[TMP35:%.*]] = load i64, ptr [[N_CASTED3]], align 8
-// CHECK13-NEXT: [[TMP36:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK13-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK13-NEXT: [[TMP39:%.*]] = load i64, ptr [[N_CASTED3]], align 8
+// CHECK13-NEXT: [[TMP40:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK13-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES7]], ptr align 8 @.offload_sizes.4, i64 24, i1 false)
-// CHECK13-NEXT: [[TMP37:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP35]], ptr [[TMP37]], align 8
-// CHECK13-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP35]], ptr [[TMP38]], align 8
-// CHECK13-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK13-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP40]], align 8
-// CHECK13-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP41]], align 8
-// CHECK13-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
-// CHECK13-NEXT: store ptr null, ptr [[TMP42]], align 8
-// CHECK13-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP43]], align 8
-// CHECK13-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 8
-// CHECK13-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK13-NEXT: store i64 [[TMP36]], ptr [[TMP45]], align 8
-// CHECK13-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK13-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP39]], ptr [[TMP41]], align 8
+// CHECK13-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP39]], ptr [[TMP42]], align 8
+// CHECK13-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK13-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP44]], align 8
+// CHECK13-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP45]], align 8
+// CHECK13-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
// CHECK13-NEXT: store ptr null, ptr [[TMP46]], align 8
-// CHECK13-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP50:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP50]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK13-NEXT: [[TMP51:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK13-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP51]], 0
+// CHECK13-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP47]], align 8
+// CHECK13-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 8
+// CHECK13-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK13-NEXT: store i64 [[TMP40]], ptr [[TMP49]], align 8
+// CHECK13-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK13-NEXT: store ptr null, ptr [[TMP50]], align 8
+// CHECK13-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP54:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP54]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK13-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK13-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP55]], 0
// CHECK13-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK13-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK13-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK13-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK13-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP52]], 1
-// CHECK13-NEXT: [[TMP53:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK13-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK13-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP56]], 1
+// CHECK13-NEXT: [[TMP57:%.*]] = zext i32 [[ADD14]] to i64
// CHECK13-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK13-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK13-NEXT: store i32 3, ptr [[TMP55]], align 4
-// CHECK13-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP47]], ptr [[TMP56]], align 8
-// CHECK13-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 8
-// CHECK13-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK13-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 8
-// CHECK13-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP59]], align 8
-// CHECK13-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP60]], align 8
-// CHECK13-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP61]], align 8
-// CHECK13-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK13-NEXT: store i64 [[TMP53]], ptr [[TMP62]], align 8
-// CHECK13-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK13-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
-// CHECK13-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK13-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP58]], align 4
+// CHECK13-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK13-NEXT: store i32 3, ptr [[TMP59]], align 4
+// CHECK13-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP51]], ptr [[TMP60]], align 8
+// CHECK13-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 8
+// CHECK13-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK13-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 8
+// CHECK13-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP63]], align 8
+// CHECK13-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP64]], align 8
+// CHECK13-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP65]], align 8
+// CHECK13-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK13-NEXT: store i64 [[TMP57]], ptr [[TMP66]], align 8
+// CHECK13-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP67]], align 8
+// CHECK13-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP68]], align 4
+// CHECK13-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK13-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP70]], align 4
+// CHECK13-NEXT: [[TMP71:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK13-NEXT: [[TMP72:%.*]] = icmp ne i32 [[TMP71]], 0
+// CHECK13-NEXT: br i1 [[TMP72]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK13: omp_offload.failed16:
-// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143(i64 [[TMP35]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143(i64 [[TMP39]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK13: omp_offload.cont17:
-// CHECK13-NEXT: [[TMP65:%.*]] = load i32, ptr [[M]], align 4
-// CHECK13-NEXT: store i32 [[TMP65]], ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK13-NEXT: [[TMP66:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP66]], ptr [[N_CASTED19]], align 4
-// CHECK13-NEXT: [[TMP67:%.*]] = load i64, ptr [[N_CASTED19]], align 8
-// CHECK13-NEXT: [[TMP68:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK13-NEXT: store i32 [[TMP68]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK13-NEXT: [[TMP69:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK13-NEXT: [[TMP70:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK13-NEXT: [[TMP73:%.*]] = load i32, ptr [[M]], align 4
+// CHECK13-NEXT: store i32 [[TMP73]], ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK13-NEXT: [[TMP74:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP74]], ptr [[N_CASTED19]], align 4
+// CHECK13-NEXT: [[TMP75:%.*]] = load i64, ptr [[N_CASTED19]], align 8
+// CHECK13-NEXT: [[TMP76:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK13-NEXT: store i32 [[TMP76]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK13-NEXT: [[TMP77:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK13-NEXT: [[TMP78:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK13-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES23]], ptr align 8 @.offload_sizes.8, i64 32, i1 false)
-// CHECK13-NEXT: [[TMP71:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP67]], ptr [[TMP71]], align 8
-// CHECK13-NEXT: [[TMP72:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP67]], ptr [[TMP72]], align 8
-// CHECK13-NEXT: [[TMP73:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP73]], align 8
-// CHECK13-NEXT: [[TMP74:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP74]], align 8
-// CHECK13-NEXT: [[TMP75:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP75]], align 8
-// CHECK13-NEXT: [[TMP76:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 1
-// CHECK13-NEXT: store ptr null, ptr [[TMP76]], align 8
-// CHECK13-NEXT: [[TMP77:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP77]], align 8
-// CHECK13-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP78]], align 8
-// CHECK13-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
-// CHECK13-NEXT: store i64 [[TMP70]], ptr [[TMP79]], align 8
-// CHECK13-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 2
-// CHECK13-NEXT: store ptr null, ptr [[TMP80]], align 8
-// CHECK13-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
-// CHECK13-NEXT: store i64 [[TMP69]], ptr [[TMP81]], align 8
-// CHECK13-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
-// CHECK13-NEXT: store i64 [[TMP69]], ptr [[TMP82]], align 8
-// CHECK13-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 3
-// CHECK13-NEXT: store ptr null, ptr [[TMP83]], align 8
-// CHECK13-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP87:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP87]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK13-NEXT: [[TMP88:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK13-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP88]], 0
+// CHECK13-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP75]], ptr [[TMP79]], align 8
+// CHECK13-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP75]], ptr [[TMP80]], align 8
+// CHECK13-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP81]], align 8
+// CHECK13-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP82]], align 8
+// CHECK13-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP83]], align 8
+// CHECK13-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 1
+// CHECK13-NEXT: store ptr null, ptr [[TMP84]], align 8
+// CHECK13-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP85]], align 8
+// CHECK13-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP86]], align 8
+// CHECK13-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
+// CHECK13-NEXT: store i64 [[TMP78]], ptr [[TMP87]], align 8
+// CHECK13-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 2
+// CHECK13-NEXT: store ptr null, ptr [[TMP88]], align 8
+// CHECK13-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
+// CHECK13-NEXT: store i64 [[TMP77]], ptr [[TMP89]], align 8
+// CHECK13-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
+// CHECK13-NEXT: store i64 [[TMP77]], ptr [[TMP90]], align 8
+// CHECK13-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 3
+// CHECK13-NEXT: store ptr null, ptr [[TMP91]], align 8
+// CHECK13-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP94:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP95:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP95]], ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK13-NEXT: [[TMP96:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK13-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP96]], 0
// CHECK13-NEXT: [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1
// CHECK13-NEXT: [[SUB29:%.*]] = sub nsw i32 [[DIV28]], 1
// CHECK13-NEXT: store i32 [[SUB29]], ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK13-NEXT: [[TMP89:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK13-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP89]], 1
-// CHECK13-NEXT: [[TMP90:%.*]] = zext i32 [[ADD30]] to i64
+// CHECK13-NEXT: [[TMP97:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
+// CHECK13-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP97]], 1
+// CHECK13-NEXT: [[TMP98:%.*]] = zext i32 [[ADD30]] to i64
// CHECK13-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP91]], align 4
-// CHECK13-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK13-NEXT: store i32 4, ptr [[TMP92]], align 4
-// CHECK13-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP84]], ptr [[TMP93]], align 8
-// CHECK13-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP85]], ptr [[TMP94]], align 8
-// CHECK13-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK13-NEXT: store ptr [[TMP86]], ptr [[TMP95]], align 8
-// CHECK13-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP96]], align 8
-// CHECK13-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP97]], align 8
-// CHECK13-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP98]], align 8
-// CHECK13-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK13-NEXT: store i64 [[TMP90]], ptr [[TMP99]], align 8
-// CHECK13-NEXT: [[TMP100:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK13-NEXT: [[TMP101:%.*]] = icmp ne i32 [[TMP100]], 0
-// CHECK13-NEXT: br i1 [[TMP101]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK13-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP99]], align 4
+// CHECK13-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK13-NEXT: store i32 4, ptr [[TMP100]], align 4
+// CHECK13-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP92]], ptr [[TMP101]], align 8
+// CHECK13-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP93]], ptr [[TMP102]], align 8
+// CHECK13-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK13-NEXT: store ptr [[TMP94]], ptr [[TMP103]], align 8
+// CHECK13-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP104]], align 8
+// CHECK13-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP105]], align 8
+// CHECK13-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP106]], align 8
+// CHECK13-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK13-NEXT: store i64 [[TMP98]], ptr [[TMP107]], align 8
+// CHECK13-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP108]], align 8
+// CHECK13-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP109]], align 4
+// CHECK13-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP110]], align 4
+// CHECK13-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP111]], align 4
+// CHECK13-NEXT: [[TMP112:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK13-NEXT: [[TMP113:%.*]] = icmp ne i32 [[TMP112]], 0
+// CHECK13-NEXT: br i1 [[TMP113]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK13: omp_offload.failed32:
-// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147(i64 [[TMP67]], i64 [[TMP1]], ptr [[VLA]], i64 [[TMP69]]) #[[ATTR3]]
+// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147(i64 [[TMP75]], i64 [[TMP1]], ptr [[VLA]], i64 [[TMP77]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK13: omp_offload.cont33:
-// CHECK13-NEXT: [[TMP102:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP102]], ptr [[N_CASTED34]], align 4
-// CHECK13-NEXT: [[TMP103:%.*]] = load i64, ptr [[N_CASTED34]], align 8
-// CHECK13-NEXT: [[TMP104:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK13-NEXT: [[TMP114:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP114]], ptr [[N_CASTED34]], align 4
+// CHECK13-NEXT: [[TMP115:%.*]] = load i64, ptr [[N_CASTED34]], align 8
+// CHECK13-NEXT: [[TMP116:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK13-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES38]], ptr align 8 @.offload_sizes.12, i64 24, i1 false)
-// CHECK13-NEXT: [[TMP105:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP103]], ptr [[TMP105]], align 8
-// CHECK13-NEXT: [[TMP106:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP103]], ptr [[TMP106]], align 8
-// CHECK13-NEXT: [[TMP107:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP107]], align 8
-// CHECK13-NEXT: [[TMP108:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP108]], align 8
-// CHECK13-NEXT: [[TMP109:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP109]], align 8
-// CHECK13-NEXT: [[TMP110:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 1
-// CHECK13-NEXT: store ptr null, ptr [[TMP110]], align 8
-// CHECK13-NEXT: [[TMP111:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP111]], align 8
-// CHECK13-NEXT: [[TMP112:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP112]], align 8
-// CHECK13-NEXT: [[TMP113:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 2
-// CHECK13-NEXT: store i64 [[TMP104]], ptr [[TMP113]], align 8
-// CHECK13-NEXT: [[TMP114:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 2
-// CHECK13-NEXT: store ptr null, ptr [[TMP114]], align 8
-// CHECK13-NEXT: [[TMP115:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP116:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP117:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP118:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP118]], ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK13-NEXT: [[TMP119:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK13-NEXT: [[SUB42:%.*]] = sub nsw i32 [[TMP119]], 0
+// CHECK13-NEXT: [[TMP117:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP115]], ptr [[TMP117]], align 8
+// CHECK13-NEXT: [[TMP118:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP115]], ptr [[TMP118]], align 8
+// CHECK13-NEXT: [[TMP119:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP119]], align 8
+// CHECK13-NEXT: [[TMP120:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP120]], align 8
+// CHECK13-NEXT: [[TMP121:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP121]], align 8
+// CHECK13-NEXT: [[TMP122:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 1
+// CHECK13-NEXT: store ptr null, ptr [[TMP122]], align 8
+// CHECK13-NEXT: [[TMP123:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP123]], align 8
+// CHECK13-NEXT: [[TMP124:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP124]], align 8
+// CHECK13-NEXT: [[TMP125:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 2
+// CHECK13-NEXT: store i64 [[TMP116]], ptr [[TMP125]], align 8
+// CHECK13-NEXT: [[TMP126:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 2
+// CHECK13-NEXT: store ptr null, ptr [[TMP126]], align 8
+// CHECK13-NEXT: [[TMP127:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP128:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP129:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP130:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP130]], ptr [[DOTCAPTURE_EXPR_40]], align 4
+// CHECK13-NEXT: [[TMP131:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
+// CHECK13-NEXT: [[SUB42:%.*]] = sub nsw i32 [[TMP131]], 0
// CHECK13-NEXT: [[DIV43:%.*]] = sdiv i32 [[SUB42]], 1
// CHECK13-NEXT: [[SUB44:%.*]] = sub nsw i32 [[DIV43]], 1
// CHECK13-NEXT: store i32 [[SUB44]], ptr [[DOTCAPTURE_EXPR_41]], align 4
-// CHECK13-NEXT: [[TMP120:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_41]], align 4
-// CHECK13-NEXT: [[ADD45:%.*]] = add nsw i32 [[TMP120]], 1
-// CHECK13-NEXT: [[TMP121:%.*]] = zext i32 [[ADD45]] to i64
+// CHECK13-NEXT: [[TMP132:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_41]], align 4
+// CHECK13-NEXT: [[ADD45:%.*]] = add nsw i32 [[TMP132]], 1
+// CHECK13-NEXT: [[TMP133:%.*]] = zext i32 [[ADD45]] to i64
// CHECK13-NEXT: [[KERNEL_ARGS46:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP122:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP122]], align 4
-// CHECK13-NEXT: [[TMP123:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 1
-// CHECK13-NEXT: store i32 3, ptr [[TMP123]], align 4
-// CHECK13-NEXT: [[TMP124:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP115]], ptr [[TMP124]], align 8
-// CHECK13-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP116]], ptr [[TMP125]], align 8
-// CHECK13-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 4
-// CHECK13-NEXT: store ptr [[TMP117]], ptr [[TMP126]], align 8
-// CHECK13-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP127]], align 8
-// CHECK13-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP128]], align 8
-// CHECK13-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP129]], align 8
-// CHECK13-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 8
-// CHECK13-NEXT: store i64 [[TMP121]], ptr [[TMP130]], align 8
-// CHECK13-NEXT: [[TMP131:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.region_id, ptr [[KERNEL_ARGS46]])
-// CHECK13-NEXT: [[TMP132:%.*]] = icmp ne i32 [[TMP131]], 0
-// CHECK13-NEXT: br i1 [[TMP132]], label [[OMP_OFFLOAD_FAILED47:%.*]], label [[OMP_OFFLOAD_CONT48:%.*]]
+// CHECK13-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP134]], align 4
+// CHECK13-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 1
+// CHECK13-NEXT: store i32 3, ptr [[TMP135]], align 4
+// CHECK13-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP127]], ptr [[TMP136]], align 8
+// CHECK13-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP128]], ptr [[TMP137]], align 8
+// CHECK13-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 4
+// CHECK13-NEXT: store ptr [[TMP129]], ptr [[TMP138]], align 8
+// CHECK13-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP139]], align 8
+// CHECK13-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP140]], align 8
+// CHECK13-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP141]], align 8
+// CHECK13-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 8
+// CHECK13-NEXT: store i64 [[TMP133]], ptr [[TMP142]], align 8
+// CHECK13-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP143]], align 8
+// CHECK13-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP144]], align 4
+// CHECK13-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP145]], align 4
+// CHECK13-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP146]], align 4
+// CHECK13-NEXT: [[TMP147:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.region_id, ptr [[KERNEL_ARGS46]])
+// CHECK13-NEXT: [[TMP148:%.*]] = icmp ne i32 [[TMP147]], 0
+// CHECK13-NEXT: br i1 [[TMP148]], label [[OMP_OFFLOAD_FAILED47:%.*]], label [[OMP_OFFLOAD_CONT48:%.*]]
// CHECK13: omp_offload.failed47:
-// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151(i64 [[TMP103]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151(i64 [[TMP115]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT48]]
// CHECK13: omp_offload.cont48:
-// CHECK13-NEXT: [[TMP133:%.*]] = load i32, ptr [[M]], align 4
-// CHECK13-NEXT: store i32 [[TMP133]], ptr [[DOTCAPTURE_EXPR_49]], align 4
-// CHECK13-NEXT: [[TMP134:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP134]], ptr [[N_CASTED50]], align 4
-// CHECK13-NEXT: [[TMP135:%.*]] = load i64, ptr [[N_CASTED50]], align 8
-// CHECK13-NEXT: [[TMP136:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_49]], align 4
-// CHECK13-NEXT: store i32 [[TMP136]], ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4
-// CHECK13-NEXT: [[TMP137:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED51]], align 8
-// CHECK13-NEXT: [[TMP138:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK13-NEXT: [[TMP149:%.*]] = load i32, ptr [[M]], align 4
+// CHECK13-NEXT: store i32 [[TMP149]], ptr [[DOTCAPTURE_EXPR_49]], align 4
+// CHECK13-NEXT: [[TMP150:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP150]], ptr [[N_CASTED50]], align 4
+// CHECK13-NEXT: [[TMP151:%.*]] = load i64, ptr [[N_CASTED50]], align 8
+// CHECK13-NEXT: [[TMP152:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_49]], align 4
+// CHECK13-NEXT: store i32 [[TMP152]], ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4
+// CHECK13-NEXT: [[TMP153:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED51]], align 8
+// CHECK13-NEXT: [[TMP154:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK13-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES55]], ptr align 8 @.offload_sizes.16, i64 32, i1 false)
-// CHECK13-NEXT: [[TMP139:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP135]], ptr [[TMP139]], align 8
-// CHECK13-NEXT: [[TMP140:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP135]], ptr [[TMP140]], align 8
-// CHECK13-NEXT: [[TMP141:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP141]], align 8
-// CHECK13-NEXT: [[TMP142:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP142]], align 8
-// CHECK13-NEXT: [[TMP143:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP143]], align 8
-// CHECK13-NEXT: [[TMP144:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 1
-// CHECK13-NEXT: store ptr null, ptr [[TMP144]], align 8
-// CHECK13-NEXT: [[TMP145:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP145]], align 8
-// CHECK13-NEXT: [[TMP146:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP146]], align 8
-// CHECK13-NEXT: [[TMP147:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 2
-// CHECK13-NEXT: store i64 [[TMP138]], ptr [[TMP147]], align 8
-// CHECK13-NEXT: [[TMP148:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 2
-// CHECK13-NEXT: store ptr null, ptr [[TMP148]], align 8
-// CHECK13-NEXT: [[TMP149:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 3
-// CHECK13-NEXT: store i64 [[TMP137]], ptr [[TMP149]], align 8
-// CHECK13-NEXT: [[TMP150:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 3
-// CHECK13-NEXT: store i64 [[TMP137]], ptr [[TMP150]], align 8
-// CHECK13-NEXT: [[TMP151:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 3
-// CHECK13-NEXT: store ptr null, ptr [[TMP151]], align 8
-// CHECK13-NEXT: [[TMP152:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP153:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP154:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP155:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP155]], ptr [[DOTCAPTURE_EXPR_57]], align 4
-// CHECK13-NEXT: [[TMP156:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_57]], align 4
-// CHECK13-NEXT: [[SUB59:%.*]] = sub nsw i32 [[TMP156]], 0
+// CHECK13-NEXT: [[TMP155:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP151]], ptr [[TMP155]], align 8
+// CHECK13-NEXT: [[TMP156:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP151]], ptr [[TMP156]], align 8
+// CHECK13-NEXT: [[TMP157:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP157]], align 8
+// CHECK13-NEXT: [[TMP158:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP158]], align 8
+// CHECK13-NEXT: [[TMP159:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP159]], align 8
+// CHECK13-NEXT: [[TMP160:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 1
+// CHECK13-NEXT: store ptr null, ptr [[TMP160]], align 8
+// CHECK13-NEXT: [[TMP161:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP161]], align 8
+// CHECK13-NEXT: [[TMP162:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP162]], align 8
+// CHECK13-NEXT: [[TMP163:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 2
+// CHECK13-NEXT: store i64 [[TMP154]], ptr [[TMP163]], align 8
+// CHECK13-NEXT: [[TMP164:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 2
+// CHECK13-NEXT: store ptr null, ptr [[TMP164]], align 8
+// CHECK13-NEXT: [[TMP165:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 3
+// CHECK13-NEXT: store i64 [[TMP153]], ptr [[TMP165]], align 8
+// CHECK13-NEXT: [[TMP166:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 3
+// CHECK13-NEXT: store i64 [[TMP153]], ptr [[TMP166]], align 8
+// CHECK13-NEXT: [[TMP167:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 3
+// CHECK13-NEXT: store ptr null, ptr [[TMP167]], align 8
+// CHECK13-NEXT: [[TMP168:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP169:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP170:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP171:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP171]], ptr [[DOTCAPTURE_EXPR_57]], align 4
+// CHECK13-NEXT: [[TMP172:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_57]], align 4
+// CHECK13-NEXT: [[SUB59:%.*]] = sub nsw i32 [[TMP172]], 0
// CHECK13-NEXT: [[DIV60:%.*]] = sdiv i32 [[SUB59]], 1
// CHECK13-NEXT: [[SUB61:%.*]] = sub nsw i32 [[DIV60]], 1
// CHECK13-NEXT: store i32 [[SUB61]], ptr [[DOTCAPTURE_EXPR_58]], align 4
-// CHECK13-NEXT: [[TMP157:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_58]], align 4
-// CHECK13-NEXT: [[ADD62:%.*]] = add nsw i32 [[TMP157]], 1
-// CHECK13-NEXT: [[TMP158:%.*]] = zext i32 [[ADD62]] to i64
+// CHECK13-NEXT: [[TMP173:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_58]], align 4
+// CHECK13-NEXT: [[ADD62:%.*]] = add nsw i32 [[TMP173]], 1
+// CHECK13-NEXT: [[TMP174:%.*]] = zext i32 [[ADD62]] to i64
// CHECK13-NEXT: [[KERNEL_ARGS63:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP159:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP159]], align 4
-// CHECK13-NEXT: [[TMP160:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 1
-// CHECK13-NEXT: store i32 4, ptr [[TMP160]], align 4
-// CHECK13-NEXT: [[TMP161:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP152]], ptr [[TMP161]], align 8
-// CHECK13-NEXT: [[TMP162:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP153]], ptr [[TMP162]], align 8
-// CHECK13-NEXT: [[TMP163:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 4
-// CHECK13-NEXT: store ptr [[TMP154]], ptr [[TMP163]], align 8
-// CHECK13-NEXT: [[TMP164:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP164]], align 8
-// CHECK13-NEXT: [[TMP165:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP165]], align 8
-// CHECK13-NEXT: [[TMP166:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP166]], align 8
-// CHECK13-NEXT: [[TMP167:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 8
-// CHECK13-NEXT: store i64 [[TMP158]], ptr [[TMP167]], align 8
-// CHECK13-NEXT: [[TMP168:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.region_id, ptr [[KERNEL_ARGS63]])
-// CHECK13-NEXT: [[TMP169:%.*]] = icmp ne i32 [[TMP168]], 0
-// CHECK13-NEXT: br i1 [[TMP169]], label [[OMP_OFFLOAD_FAILED64:%.*]], label [[OMP_OFFLOAD_CONT65:%.*]]
+// CHECK13-NEXT: [[TMP175:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP175]], align 4
+// CHECK13-NEXT: [[TMP176:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 1
+// CHECK13-NEXT: store i32 4, ptr [[TMP176]], align 4
+// CHECK13-NEXT: [[TMP177:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP168]], ptr [[TMP177]], align 8
+// CHECK13-NEXT: [[TMP178:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP169]], ptr [[TMP178]], align 8
+// CHECK13-NEXT: [[TMP179:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 4
+// CHECK13-NEXT: store ptr [[TMP170]], ptr [[TMP179]], align 8
+// CHECK13-NEXT: [[TMP180:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP180]], align 8
+// CHECK13-NEXT: [[TMP181:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP181]], align 8
+// CHECK13-NEXT: [[TMP182:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP182]], align 8
+// CHECK13-NEXT: [[TMP183:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 8
+// CHECK13-NEXT: store i64 [[TMP174]], ptr [[TMP183]], align 8
+// CHECK13-NEXT: [[TMP184:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP184]], align 8
+// CHECK13-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP185]], align 4
+// CHECK13-NEXT: [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP186]], align 4
+// CHECK13-NEXT: [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP187]], align 4
+// CHECK13-NEXT: [[TMP188:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.region_id, ptr [[KERNEL_ARGS63]])
+// CHECK13-NEXT: [[TMP189:%.*]] = icmp ne i32 [[TMP188]], 0
+// CHECK13-NEXT: br i1 [[TMP189]], label [[OMP_OFFLOAD_FAILED64:%.*]], label [[OMP_OFFLOAD_CONT65:%.*]]
// CHECK13: omp_offload.failed64:
-// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155(i64 [[TMP135]], i64 [[TMP1]], ptr [[VLA]], i64 [[TMP137]]) #[[ATTR3]]
+// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155(i64 [[TMP151]], i64 [[TMP1]], ptr [[VLA]], i64 [[TMP153]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT65]]
// CHECK13: omp_offload.cont65:
-// CHECK13-NEXT: [[TMP170:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK13-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP170]])
+// CHECK13-NEXT: [[TMP190:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK13-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP190]])
// CHECK13-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK13-NEXT: [[TMP171:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK13-NEXT: call void @llvm.stackrestore(ptr [[TMP171]])
-// CHECK13-NEXT: [[TMP172:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK13-NEXT: ret i32 [[TMP172]]
+// CHECK13-NEXT: [[TMP191:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK13-NEXT: call void @llvm.stackrestore(ptr [[TMP191]])
+// CHECK13-NEXT: [[TMP192:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK13-NEXT: ret i32 [[TMP192]]
//
//
// CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139
@@ -6490,7 +6690,7 @@ int main (int argc, char **argv) {
// CHECK13-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK13-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK13-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK13-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK13-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK13-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK13-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -6507,169 +6707,209 @@ int main (int argc, char **argv) {
// CHECK13-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK13-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK13-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK13-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.region_id, ptr [[KERNEL_ARGS]])
-// CHECK13-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK13-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK13-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK13-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK13-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK13-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK13-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.region_id, ptr [[KERNEL_ARGS]])
+// CHECK13-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK13-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK13: omp_offload.failed:
// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112(ptr [[A]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK13: omp_offload.cont:
-// CHECK13-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP16]], align 8
-// CHECK13-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP17]], align 8
-// CHECK13-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK13-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP20]], align 8
+// CHECK13-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP21]], align 8
+// CHECK13-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK13-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK13-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK13-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK13-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK13-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK13-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK13-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK13-NEXT: store ptr @.offload_sizes.24, ptr [[TMP25]], align 8
-// CHECK13-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP26]], align 8
-// CHECK13-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK13-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK13-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK13-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK13-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK13-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK13-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK13-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK13-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK13-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK13-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK13-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK13-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK13-NEXT: store ptr @.offload_sizes.24, ptr [[TMP29]], align 8
+// CHECK13-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP30]], align 8
+// CHECK13-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK13-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK13-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK13-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK13-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK13-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK13-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK13-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK13-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK13-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK13-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK13: omp_offload.failed6:
// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116(ptr [[A]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK13: omp_offload.cont7:
-// CHECK13-NEXT: [[TMP32:%.*]] = load i32, ptr [[M]], align 4
-// CHECK13-NEXT: store i32 [[TMP32]], ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK13-NEXT: [[TMP33:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK13-NEXT: store i32 [[TMP33]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK13-NEXT: [[TMP34:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK13-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP35]], align 8
-// CHECK13-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP36]], align 8
-// CHECK13-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP37]], align 8
-// CHECK13-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP34]], ptr [[TMP38]], align 8
-// CHECK13-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP34]], ptr [[TMP39]], align 8
-// CHECK13-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
-// CHECK13-NEXT: store ptr null, ptr [[TMP40]], align 8
-// CHECK13-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP40:%.*]] = load i32, ptr [[M]], align 4
+// CHECK13-NEXT: store i32 [[TMP40]], ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK13-NEXT: [[TMP41:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK13-NEXT: store i32 [[TMP41]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK13-NEXT: [[TMP42:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK13-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP43]], align 8
+// CHECK13-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP44]], align 8
+// CHECK13-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP45]], align 8
+// CHECK13-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP42]], ptr [[TMP46]], align 8
+// CHECK13-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP42]], ptr [[TMP47]], align 8
+// CHECK13-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
+// CHECK13-NEXT: store ptr null, ptr [[TMP48]], align 8
+// CHECK13-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK13-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP43]], align 4
-// CHECK13-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK13-NEXT: store i32 2, ptr [[TMP44]], align 4
-// CHECK13-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP41]], ptr [[TMP45]], align 8
-// CHECK13-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP42]], ptr [[TMP46]], align 8
-// CHECK13-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK13-NEXT: store ptr @.offload_sizes.28, ptr [[TMP47]], align 8
-// CHECK13-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP48]], align 8
-// CHECK13-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP49]], align 8
-// CHECK13-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP50]], align 8
-// CHECK13-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK13-NEXT: store i64 10, ptr [[TMP51]], align 8
-// CHECK13-NEXT: [[TMP52:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK13-NEXT: [[TMP53:%.*]] = icmp ne i32 [[TMP52]], 0
-// CHECK13-NEXT: br i1 [[TMP53]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK13-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP51]], align 4
+// CHECK13-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK13-NEXT: store i32 2, ptr [[TMP52]], align 4
+// CHECK13-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 8
+// CHECK13-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP50]], ptr [[TMP54]], align 8
+// CHECK13-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK13-NEXT: store ptr @.offload_sizes.28, ptr [[TMP55]], align 8
+// CHECK13-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP56]], align 8
+// CHECK13-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP57]], align 8
+// CHECK13-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP58]], align 8
+// CHECK13-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK13-NEXT: store i64 10, ptr [[TMP59]], align 8
+// CHECK13-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP60]], align 8
+// CHECK13-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4
+// CHECK13-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP62]], align 4
+// CHECK13-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP63]], align 4
+// CHECK13-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK13-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
+// CHECK13-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK13: omp_offload.failed13:
-// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120(ptr [[A]], i64 [[TMP34]]) #[[ATTR3]]
+// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120(ptr [[A]], i64 [[TMP42]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT14]]
// CHECK13: omp_offload.cont14:
-// CHECK13-NEXT: [[TMP54:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP54]], align 8
-// CHECK13-NEXT: [[TMP55:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP55]], align 8
-// CHECK13-NEXT: [[TMP56:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP56]], align 8
-// CHECK13-NEXT: [[TMP57:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP58:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP66]], align 8
+// CHECK13-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP67]], align 8
+// CHECK13-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP68]], align 8
+// CHECK13-NEXT: [[TMP69:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP70:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
// CHECK13-NEXT: [[KERNEL_ARGS19:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP59]], align 4
-// CHECK13-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
-// CHECK13-NEXT: store i32 1, ptr [[TMP60]], align 4
-// CHECK13-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP57]], ptr [[TMP61]], align 8
-// CHECK13-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP58]], ptr [[TMP62]], align 8
-// CHECK13-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
-// CHECK13-NEXT: store ptr @.offload_sizes.32, ptr [[TMP63]], align 8
-// CHECK13-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP64]], align 8
-// CHECK13-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP65]], align 8
-// CHECK13-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP66]], align 8
-// CHECK13-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
-// CHECK13-NEXT: store i64 10, ptr [[TMP67]], align 8
-// CHECK13-NEXT: [[TMP68:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.region_id, ptr [[KERNEL_ARGS19]])
-// CHECK13-NEXT: [[TMP69:%.*]] = icmp ne i32 [[TMP68]], 0
-// CHECK13-NEXT: br i1 [[TMP69]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
+// CHECK13-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP71]], align 4
+// CHECK13-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
+// CHECK13-NEXT: store i32 1, ptr [[TMP72]], align 4
+// CHECK13-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP69]], ptr [[TMP73]], align 8
+// CHECK13-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP70]], ptr [[TMP74]], align 8
+// CHECK13-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
+// CHECK13-NEXT: store ptr @.offload_sizes.32, ptr [[TMP75]], align 8
+// CHECK13-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP76]], align 8
+// CHECK13-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP77]], align 8
+// CHECK13-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP78]], align 8
+// CHECK13-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
+// CHECK13-NEXT: store i64 10, ptr [[TMP79]], align 8
+// CHECK13-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP80]], align 8
+// CHECK13-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP81]], align 4
+// CHECK13-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP82]], align 4
+// CHECK13-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP83]], align 4
+// CHECK13-NEXT: [[TMP84:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.region_id, ptr [[KERNEL_ARGS19]])
+// CHECK13-NEXT: [[TMP85:%.*]] = icmp ne i32 [[TMP84]], 0
+// CHECK13-NEXT: br i1 [[TMP85]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
// CHECK13: omp_offload.failed20:
// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124(ptr [[A]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT21]]
// CHECK13: omp_offload.cont21:
-// CHECK13-NEXT: [[TMP70:%.*]] = load i32, ptr [[M]], align 4
-// CHECK13-NEXT: store i32 [[TMP70]], ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK13-NEXT: [[TMP71:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK13-NEXT: store i32 [[TMP71]], ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4
-// CHECK13-NEXT: [[TMP72:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED23]], align 8
-// CHECK13-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP73]], align 8
-// CHECK13-NEXT: [[TMP74:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP74]], align 8
-// CHECK13-NEXT: [[TMP75:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP75]], align 8
-// CHECK13-NEXT: [[TMP76:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP72]], ptr [[TMP76]], align 8
-// CHECK13-NEXT: [[TMP77:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP72]], ptr [[TMP77]], align 8
-// CHECK13-NEXT: [[TMP78:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 1
-// CHECK13-NEXT: store ptr null, ptr [[TMP78]], align 8
-// CHECK13-NEXT: [[TMP79:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP80:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP86:%.*]] = load i32, ptr [[M]], align 4
+// CHECK13-NEXT: store i32 [[TMP86]], ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK13-NEXT: [[TMP87:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK13-NEXT: store i32 [[TMP87]], ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4
+// CHECK13-NEXT: [[TMP88:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED23]], align 8
+// CHECK13-NEXT: [[TMP89:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP89]], align 8
+// CHECK13-NEXT: [[TMP90:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP90]], align 8
+// CHECK13-NEXT: [[TMP91:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP91]], align 8
+// CHECK13-NEXT: [[TMP92:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP88]], ptr [[TMP92]], align 8
+// CHECK13-NEXT: [[TMP93:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP88]], ptr [[TMP93]], align 8
+// CHECK13-NEXT: [[TMP94:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 1
+// CHECK13-NEXT: store ptr null, ptr [[TMP94]], align 8
+// CHECK13-NEXT: [[TMP95:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP96:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
// CHECK13-NEXT: [[KERNEL_ARGS28:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP81]], align 4
-// CHECK13-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
-// CHECK13-NEXT: store i32 2, ptr [[TMP82]], align 4
-// CHECK13-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP79]], ptr [[TMP83]], align 8
-// CHECK13-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP80]], ptr [[TMP84]], align 8
-// CHECK13-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
-// CHECK13-NEXT: store ptr @.offload_sizes.36, ptr [[TMP85]], align 8
-// CHECK13-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP86]], align 8
-// CHECK13-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP87]], align 8
-// CHECK13-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP88]], align 8
-// CHECK13-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
-// CHECK13-NEXT: store i64 10, ptr [[TMP89]], align 8
-// CHECK13-NEXT: [[TMP90:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.region_id, ptr [[KERNEL_ARGS28]])
-// CHECK13-NEXT: [[TMP91:%.*]] = icmp ne i32 [[TMP90]], 0
-// CHECK13-NEXT: br i1 [[TMP91]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
+// CHECK13-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP97]], align 4
+// CHECK13-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
+// CHECK13-NEXT: store i32 2, ptr [[TMP98]], align 4
+// CHECK13-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP95]], ptr [[TMP99]], align 8
+// CHECK13-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP96]], ptr [[TMP100]], align 8
+// CHECK13-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
+// CHECK13-NEXT: store ptr @.offload_sizes.36, ptr [[TMP101]], align 8
+// CHECK13-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP102]], align 8
+// CHECK13-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP103]], align 8
+// CHECK13-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP104]], align 8
+// CHECK13-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
+// CHECK13-NEXT: store i64 10, ptr [[TMP105]], align 8
+// CHECK13-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP106]], align 8
+// CHECK13-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP107]], align 4
+// CHECK13-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP108]], align 4
+// CHECK13-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP109]], align 4
+// CHECK13-NEXT: [[TMP110:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.region_id, ptr [[KERNEL_ARGS28]])
+// CHECK13-NEXT: [[TMP111:%.*]] = icmp ne i32 [[TMP110]], 0
+// CHECK13-NEXT: br i1 [[TMP111]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
// CHECK13: omp_offload.failed29:
-// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128(ptr [[A]], i64 [[TMP72]]) #[[ATTR3]]
+// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128(ptr [[A]], i64 [[TMP88]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT30]]
// CHECK13: omp_offload.cont30:
// CHECK13-NEXT: ret i32 0
@@ -7654,7 +7894,7 @@ int main (int argc, char **argv) {
// CHECK15-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK15-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK15-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK15-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK15-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK15-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK15-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -7671,302 +7911,342 @@ int main (int argc, char **argv) {
// CHECK15-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK15-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK15-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK15-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.region_id, ptr [[KERNEL_ARGS]])
-// CHECK15-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK15-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK15-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK15-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK15-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK15-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK15-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.region_id, ptr [[KERNEL_ARGS]])
+// CHECK15-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK15-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK15: omp_offload.failed:
// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139(i32 [[TMP3]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK15: omp_offload.cont:
-// CHECK15-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK15-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK15-NEXT: [[TMP35:%.*]] = load i32, ptr [[N_CASTED3]], align 4
-// CHECK15-NEXT: [[TMP36:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK15-NEXT: [[TMP37:%.*]] = sext i32 [[TMP36]] to i64
+// CHECK15-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK15-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK15-NEXT: [[TMP39:%.*]] = load i32, ptr [[N_CASTED3]], align 4
+// CHECK15-NEXT: [[TMP40:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK15-NEXT: [[TMP41:%.*]] = sext i32 [[TMP40]] to i64
// CHECK15-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES7]], ptr align 4 @.offload_sizes.4, i32 24, i1 false)
-// CHECK15-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK15-NEXT: store i32 [[TMP35]], ptr [[TMP38]], align 4
-// CHECK15-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK15-NEXT: store i32 [[TMP35]], ptr [[TMP39]], align 4
-// CHECK15-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
-// CHECK15-NEXT: store ptr null, ptr [[TMP40]], align 4
-// CHECK15-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP41]], align 4
-// CHECK15-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP42]], align 4
-// CHECK15-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
-// CHECK15-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK15-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 4
-// CHECK15-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP45]], align 4
-// CHECK15-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK15-NEXT: store i64 [[TMP37]], ptr [[TMP46]], align 4
-// CHECK15-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK15-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK15-NEXT: store i32 [[TMP39]], ptr [[TMP42]], align 4
+// CHECK15-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK15-NEXT: store i32 [[TMP39]], ptr [[TMP43]], align 4
+// CHECK15-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
+// CHECK15-NEXT: store ptr null, ptr [[TMP44]], align 4
+// CHECK15-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP45]], align 4
+// CHECK15-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP46]], align 4
+// CHECK15-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
// CHECK15-NEXT: store ptr null, ptr [[TMP47]], align 4
-// CHECK15-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP51:%.*]] = load i32, ptr [[N]], align 4
-// CHECK15-NEXT: store i32 [[TMP51]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK15-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK15-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP52]], 0
+// CHECK15-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 4
+// CHECK15-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP49]], align 4
+// CHECK15-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK15-NEXT: store i64 [[TMP41]], ptr [[TMP50]], align 4
+// CHECK15-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK15-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK15-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP54:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP55:%.*]] = load i32, ptr [[N]], align 4
+// CHECK15-NEXT: store i32 [[TMP55]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK15-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK15-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP56]], 0
// CHECK15-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK15-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK15-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK15-NEXT: [[TMP53:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK15-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP53]], 1
-// CHECK15-NEXT: [[TMP54:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK15-NEXT: [[TMP57:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK15-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP57]], 1
+// CHECK15-NEXT: [[TMP58:%.*]] = zext i32 [[ADD14]] to i64
// CHECK15-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK15-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP55]], align 4
-// CHECK15-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK15-NEXT: store i32 3, ptr [[TMP56]], align 4
-// CHECK15-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 4
-// CHECK15-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK15-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 4
-// CHECK15-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK15-NEXT: store ptr [[TMP50]], ptr [[TMP59]], align 4
-// CHECK15-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK15-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP60]], align 4
-// CHECK15-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK15-NEXT: store ptr null, ptr [[TMP61]], align 4
-// CHECK15-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK15-NEXT: store ptr null, ptr [[TMP62]], align 4
-// CHECK15-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK15-NEXT: store i64 [[TMP54]], ptr [[TMP63]], align 8
-// CHECK15-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK15-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
-// CHECK15-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK15-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK15-NEXT: store i32 2, ptr [[TMP59]], align 4
+// CHECK15-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK15-NEXT: store i32 3, ptr [[TMP60]], align 4
+// CHECK15-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 4
+// CHECK15-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK15-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 4
+// CHECK15-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK15-NEXT: store ptr [[TMP54]], ptr [[TMP63]], align 4
+// CHECK15-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK15-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP64]], align 4
+// CHECK15-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK15-NEXT: store ptr null, ptr [[TMP65]], align 4
+// CHECK15-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK15-NEXT: store ptr null, ptr [[TMP66]], align 4
+// CHECK15-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK15-NEXT: store i64 [[TMP58]], ptr [[TMP67]], align 8
+// CHECK15-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP68]], align 8
+// CHECK15-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK15-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP70]], align 4
+// CHECK15-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP71]], align 4
+// CHECK15-NEXT: [[TMP72:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK15-NEXT: [[TMP73:%.*]] = icmp ne i32 [[TMP72]], 0
+// CHECK15-NEXT: br i1 [[TMP73]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK15: omp_offload.failed16:
-// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143(i32 [[TMP35]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143(i32 [[TMP39]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK15: omp_offload.cont17:
-// CHECK15-NEXT: [[TMP66:%.*]] = load i32, ptr [[M]], align 4
-// CHECK15-NEXT: store i32 [[TMP66]], ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK15-NEXT: [[TMP67:%.*]] = load i32, ptr [[N]], align 4
-// CHECK15-NEXT: store i32 [[TMP67]], ptr [[N_CASTED19]], align 4
-// CHECK15-NEXT: [[TMP68:%.*]] = load i32, ptr [[N_CASTED19]], align 4
-// CHECK15-NEXT: [[TMP69:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK15-NEXT: store i32 [[TMP69]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK15-NEXT: [[TMP70:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK15-NEXT: [[TMP71:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK15-NEXT: [[TMP72:%.*]] = sext i32 [[TMP71]] to i64
+// CHECK15-NEXT: [[TMP74:%.*]] = load i32, ptr [[M]], align 4
+// CHECK15-NEXT: store i32 [[TMP74]], ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK15-NEXT: [[TMP75:%.*]] = load i32, ptr [[N]], align 4
+// CHECK15-NEXT: store i32 [[TMP75]], ptr [[N_CASTED19]], align 4
+// CHECK15-NEXT: [[TMP76:%.*]] = load i32, ptr [[N_CASTED19]], align 4
+// CHECK15-NEXT: [[TMP77:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK15-NEXT: store i32 [[TMP77]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK15-NEXT: [[TMP78:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK15-NEXT: [[TMP79:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK15-NEXT: [[TMP80:%.*]] = sext i32 [[TMP79]] to i64
// CHECK15-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES23]], ptr align 4 @.offload_sizes.8, i32 32, i1 false)
-// CHECK15-NEXT: [[TMP73:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK15-NEXT: store i32 [[TMP68]], ptr [[TMP73]], align 4
-// CHECK15-NEXT: [[TMP74:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK15-NEXT: store i32 [[TMP68]], ptr [[TMP74]], align 4
-// CHECK15-NEXT: [[TMP75:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 0
-// CHECK15-NEXT: store ptr null, ptr [[TMP75]], align 4
-// CHECK15-NEXT: [[TMP76:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP76]], align 4
-// CHECK15-NEXT: [[TMP77:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP77]], align 4
-// CHECK15-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 1
-// CHECK15-NEXT: store ptr null, ptr [[TMP78]], align 4
-// CHECK15-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP79]], align 4
-// CHECK15-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP80]], align 4
-// CHECK15-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
-// CHECK15-NEXT: store i64 [[TMP72]], ptr [[TMP81]], align 4
-// CHECK15-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 2
-// CHECK15-NEXT: store ptr null, ptr [[TMP82]], align 4
-// CHECK15-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
-// CHECK15-NEXT: store i32 [[TMP70]], ptr [[TMP83]], align 4
-// CHECK15-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
-// CHECK15-NEXT: store i32 [[TMP70]], ptr [[TMP84]], align 4
-// CHECK15-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 3
-// CHECK15-NEXT: store ptr null, ptr [[TMP85]], align 4
-// CHECK15-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP89:%.*]] = load i32, ptr [[N]], align 4
-// CHECK15-NEXT: store i32 [[TMP89]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK15-NEXT: [[TMP90:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK15-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP90]], 0
+// CHECK15-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK15-NEXT: store i32 [[TMP76]], ptr [[TMP81]], align 4
+// CHECK15-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK15-NEXT: store i32 [[TMP76]], ptr [[TMP82]], align 4
+// CHECK15-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 0
+// CHECK15-NEXT: store ptr null, ptr [[TMP83]], align 4
+// CHECK15-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP84]], align 4
+// CHECK15-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP85]], align 4
+// CHECK15-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 1
+// CHECK15-NEXT: store ptr null, ptr [[TMP86]], align 4
+// CHECK15-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP87]], align 4
+// CHECK15-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP88]], align 4
+// CHECK15-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
+// CHECK15-NEXT: store i64 [[TMP80]], ptr [[TMP89]], align 4
+// CHECK15-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 2
+// CHECK15-NEXT: store ptr null, ptr [[TMP90]], align 4
+// CHECK15-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
+// CHECK15-NEXT: store i32 [[TMP78]], ptr [[TMP91]], align 4
+// CHECK15-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
+// CHECK15-NEXT: store i32 [[TMP78]], ptr [[TMP92]], align 4
+// CHECK15-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 3
+// CHECK15-NEXT: store ptr null, ptr [[TMP93]], align 4
+// CHECK15-NEXT: [[TMP94:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP95:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP96:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP97:%.*]] = load i32, ptr [[N]], align 4
+// CHECK15-NEXT: store i32 [[TMP97]], ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK15-NEXT: [[TMP98:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK15-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP98]], 0
// CHECK15-NEXT: [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1
// CHECK15-NEXT: [[SUB29:%.*]] = sub nsw i32 [[DIV28]], 1
// CHECK15-NEXT: store i32 [[SUB29]], ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK15-NEXT: [[TMP91:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK15-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP91]], 1
-// CHECK15-NEXT: [[TMP92:%.*]] = zext i32 [[ADD30]] to i64
+// CHECK15-NEXT: [[TMP99:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
+// CHECK15-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP99]], 1
+// CHECK15-NEXT: [[TMP100:%.*]] = zext i32 [[ADD30]] to i64
// CHECK15-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK15-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP93]], align 4
-// CHECK15-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK15-NEXT: store i32 4, ptr [[TMP94]], align 4
-// CHECK15-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[TMP86]], ptr [[TMP95]], align 4
-// CHECK15-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK15-NEXT: store ptr [[TMP87]], ptr [[TMP96]], align 4
-// CHECK15-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK15-NEXT: store ptr [[TMP88]], ptr [[TMP97]], align 4
-// CHECK15-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK15-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP98]], align 4
-// CHECK15-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK15-NEXT: store ptr null, ptr [[TMP99]], align 4
-// CHECK15-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK15-NEXT: store ptr null, ptr [[TMP100]], align 4
-// CHECK15-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK15-NEXT: store i64 [[TMP92]], ptr [[TMP101]], align 8
-// CHECK15-NEXT: [[TMP102:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK15-NEXT: [[TMP103:%.*]] = icmp ne i32 [[TMP102]], 0
-// CHECK15-NEXT: br i1 [[TMP103]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK15-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK15-NEXT: store i32 2, ptr [[TMP101]], align 4
+// CHECK15-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK15-NEXT: store i32 4, ptr [[TMP102]], align 4
+// CHECK15-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[TMP94]], ptr [[TMP103]], align 4
+// CHECK15-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK15-NEXT: store ptr [[TMP95]], ptr [[TMP104]], align 4
+// CHECK15-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK15-NEXT: store ptr [[TMP96]], ptr [[TMP105]], align 4
+// CHECK15-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK15-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP106]], align 4
+// CHECK15-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK15-NEXT: store ptr null, ptr [[TMP107]], align 4
+// CHECK15-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK15-NEXT: store ptr null, ptr [[TMP108]], align 4
+// CHECK15-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK15-NEXT: store i64 [[TMP100]], ptr [[TMP109]], align 8
+// CHECK15-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP110]], align 8
+// CHECK15-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP111]], align 4
+// CHECK15-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP112]], align 4
+// CHECK15-NEXT: [[TMP113:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP113]], align 4
+// CHECK15-NEXT: [[TMP114:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK15-NEXT: [[TMP115:%.*]] = icmp ne i32 [[TMP114]], 0
+// CHECK15-NEXT: br i1 [[TMP115]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK15: omp_offload.failed32:
-// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147(i32 [[TMP68]], i32 [[TMP0]], ptr [[VLA]], i32 [[TMP70]]) #[[ATTR3]]
+// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147(i32 [[TMP76]], i32 [[TMP0]], ptr [[VLA]], i32 [[TMP78]]) #[[ATTR3]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK15: omp_offload.cont33:
-// CHECK15-NEXT: [[TMP104:%.*]] = load i32, ptr [[N]], align 4
-// CHECK15-NEXT: store i32 [[TMP104]], ptr [[N_CASTED34]], align 4
-// CHECK15-NEXT: [[TMP105:%.*]] = load i32, ptr [[N_CASTED34]], align 4
-// CHECK15-NEXT: [[TMP106:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK15-NEXT: [[TMP107:%.*]] = sext i32 [[TMP106]] to i64
+// CHECK15-NEXT: [[TMP116:%.*]] = load i32, ptr [[N]], align 4
+// CHECK15-NEXT: store i32 [[TMP116]], ptr [[N_CASTED34]], align 4
+// CHECK15-NEXT: [[TMP117:%.*]] = load i32, ptr [[N_CASTED34]], align 4
+// CHECK15-NEXT: [[TMP118:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK15-NEXT: [[TMP119:%.*]] = sext i32 [[TMP118]] to i64
// CHECK15-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES38]], ptr align 4 @.offload_sizes.12, i32 24, i1 false)
-// CHECK15-NEXT: [[TMP108:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
-// CHECK15-NEXT: store i32 [[TMP105]], ptr [[TMP108]], align 4
-// CHECK15-NEXT: [[TMP109:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
-// CHECK15-NEXT: store i32 [[TMP105]], ptr [[TMP109]], align 4
-// CHECK15-NEXT: [[TMP110:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 0
-// CHECK15-NEXT: store ptr null, ptr [[TMP110]], align 4
-// CHECK15-NEXT: [[TMP111:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP111]], align 4
-// CHECK15-NEXT: [[TMP112:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP112]], align 4
-// CHECK15-NEXT: [[TMP113:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 1
-// CHECK15-NEXT: store ptr null, ptr [[TMP113]], align 4
-// CHECK15-NEXT: [[TMP114:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP114]], align 4
-// CHECK15-NEXT: [[TMP115:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP115]], align 4
-// CHECK15-NEXT: [[TMP116:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 2
-// CHECK15-NEXT: store i64 [[TMP107]], ptr [[TMP116]], align 4
-// CHECK15-NEXT: [[TMP117:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 2
-// CHECK15-NEXT: store ptr null, ptr [[TMP117]], align 4
-// CHECK15-NEXT: [[TMP118:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP119:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP120:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP121:%.*]] = load i32, ptr [[N]], align 4
-// CHECK15-NEXT: store i32 [[TMP121]], ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK15-NEXT: [[TMP122:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK15-NEXT: [[SUB42:%.*]] = sub nsw i32 [[TMP122]], 0
+// CHECK15-NEXT: [[TMP120:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
+// CHECK15-NEXT: store i32 [[TMP117]], ptr [[TMP120]], align 4
+// CHECK15-NEXT: [[TMP121:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
+// CHECK15-NEXT: store i32 [[TMP117]], ptr [[TMP121]], align 4
+// CHECK15-NEXT: [[TMP122:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 0
+// CHECK15-NEXT: store ptr null, ptr [[TMP122]], align 4
+// CHECK15-NEXT: [[TMP123:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP123]], align 4
+// CHECK15-NEXT: [[TMP124:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP124]], align 4
+// CHECK15-NEXT: [[TMP125:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 1
+// CHECK15-NEXT: store ptr null, ptr [[TMP125]], align 4
+// CHECK15-NEXT: [[TMP126:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP126]], align 4
+// CHECK15-NEXT: [[TMP127:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP127]], align 4
+// CHECK15-NEXT: [[TMP128:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 2
+// CHECK15-NEXT: store i64 [[TMP119]], ptr [[TMP128]], align 4
+// CHECK15-NEXT: [[TMP129:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 2
+// CHECK15-NEXT: store ptr null, ptr [[TMP129]], align 4
+// CHECK15-NEXT: [[TMP130:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP131:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP132:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP133:%.*]] = load i32, ptr [[N]], align 4
+// CHECK15-NEXT: store i32 [[TMP133]], ptr [[DOTCAPTURE_EXPR_40]], align 4
+// CHECK15-NEXT: [[TMP134:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
+// CHECK15-NEXT: [[SUB42:%.*]] = sub nsw i32 [[TMP134]], 0
// CHECK15-NEXT: [[DIV43:%.*]] = sdiv i32 [[SUB42]], 1
// CHECK15-NEXT: [[SUB44:%.*]] = sub nsw i32 [[DIV43]], 1
// CHECK15-NEXT: store i32 [[SUB44]], ptr [[DOTCAPTURE_EXPR_41]], align 4
-// CHECK15-NEXT: [[TMP123:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_41]], align 4
-// CHECK15-NEXT: [[ADD45:%.*]] = add nsw i32 [[TMP123]], 1
-// CHECK15-NEXT: [[TMP124:%.*]] = zext i32 [[ADD45]] to i64
+// CHECK15-NEXT: [[TMP135:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_41]], align 4
+// CHECK15-NEXT: [[ADD45:%.*]] = add nsw i32 [[TMP135]], 1
+// CHECK15-NEXT: [[TMP136:%.*]] = zext i32 [[ADD45]] to i64
// CHECK15-NEXT: [[KERNEL_ARGS46:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK15-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP125]], align 4
-// CHECK15-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 1
-// CHECK15-NEXT: store i32 3, ptr [[TMP126]], align 4
-// CHECK15-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[TMP118]], ptr [[TMP127]], align 4
-// CHECK15-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 3
-// CHECK15-NEXT: store ptr [[TMP119]], ptr [[TMP128]], align 4
-// CHECK15-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 4
-// CHECK15-NEXT: store ptr [[TMP120]], ptr [[TMP129]], align 4
-// CHECK15-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 5
-// CHECK15-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP130]], align 4
-// CHECK15-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 6
-// CHECK15-NEXT: store ptr null, ptr [[TMP131]], align 4
-// CHECK15-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 7
-// CHECK15-NEXT: store ptr null, ptr [[TMP132]], align 4
-// CHECK15-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 8
-// CHECK15-NEXT: store i64 [[TMP124]], ptr [[TMP133]], align 8
-// CHECK15-NEXT: [[TMP134:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.region_id, ptr [[KERNEL_ARGS46]])
-// CHECK15-NEXT: [[TMP135:%.*]] = icmp ne i32 [[TMP134]], 0
-// CHECK15-NEXT: br i1 [[TMP135]], label [[OMP_OFFLOAD_FAILED47:%.*]], label [[OMP_OFFLOAD_CONT48:%.*]]
+// CHECK15-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 0
+// CHECK15-NEXT: store i32 2, ptr [[TMP137]], align 4
+// CHECK15-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 1
+// CHECK15-NEXT: store i32 3, ptr [[TMP138]], align 4
+// CHECK15-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[TMP130]], ptr [[TMP139]], align 4
+// CHECK15-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 3
+// CHECK15-NEXT: store ptr [[TMP131]], ptr [[TMP140]], align 4
+// CHECK15-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 4
+// CHECK15-NEXT: store ptr [[TMP132]], ptr [[TMP141]], align 4
+// CHECK15-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 5
+// CHECK15-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP142]], align 4
+// CHECK15-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 6
+// CHECK15-NEXT: store ptr null, ptr [[TMP143]], align 4
+// CHECK15-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 7
+// CHECK15-NEXT: store ptr null, ptr [[TMP144]], align 4
+// CHECK15-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 8
+// CHECK15-NEXT: store i64 [[TMP136]], ptr [[TMP145]], align 8
+// CHECK15-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP146]], align 8
+// CHECK15-NEXT: [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP147]], align 4
+// CHECK15-NEXT: [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP148]], align 4
+// CHECK15-NEXT: [[TMP149:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP149]], align 4
+// CHECK15-NEXT: [[TMP150:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.region_id, ptr [[KERNEL_ARGS46]])
+// CHECK15-NEXT: [[TMP151:%.*]] = icmp ne i32 [[TMP150]], 0
+// CHECK15-NEXT: br i1 [[TMP151]], label [[OMP_OFFLOAD_FAILED47:%.*]], label [[OMP_OFFLOAD_CONT48:%.*]]
// CHECK15: omp_offload.failed47:
-// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151(i32 [[TMP105]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151(i32 [[TMP117]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT48]]
// CHECK15: omp_offload.cont48:
-// CHECK15-NEXT: [[TMP136:%.*]] = load i32, ptr [[M]], align 4
-// CHECK15-NEXT: store i32 [[TMP136]], ptr [[DOTCAPTURE_EXPR_49]], align 4
-// CHECK15-NEXT: [[TMP137:%.*]] = load i32, ptr [[N]], align 4
-// CHECK15-NEXT: store i32 [[TMP137]], ptr [[N_CASTED50]], align 4
-// CHECK15-NEXT: [[TMP138:%.*]] = load i32, ptr [[N_CASTED50]], align 4
-// CHECK15-NEXT: [[TMP139:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_49]], align 4
-// CHECK15-NEXT: store i32 [[TMP139]], ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4
-// CHECK15-NEXT: [[TMP140:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4
-// CHECK15-NEXT: [[TMP141:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK15-NEXT: [[TMP142:%.*]] = sext i32 [[TMP141]] to i64
+// CHECK15-NEXT: [[TMP152:%.*]] = load i32, ptr [[M]], align 4
+// CHECK15-NEXT: store i32 [[TMP152]], ptr [[DOTCAPTURE_EXPR_49]], align 4
+// CHECK15-NEXT: [[TMP153:%.*]] = load i32, ptr [[N]], align 4
+// CHECK15-NEXT: store i32 [[TMP153]], ptr [[N_CASTED50]], align 4
+// CHECK15-NEXT: [[TMP154:%.*]] = load i32, ptr [[N_CASTED50]], align 4
+// CHECK15-NEXT: [[TMP155:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_49]], align 4
+// CHECK15-NEXT: store i32 [[TMP155]], ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4
+// CHECK15-NEXT: [[TMP156:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4
+// CHECK15-NEXT: [[TMP157:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK15-NEXT: [[TMP158:%.*]] = sext i32 [[TMP157]] to i64
// CHECK15-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES55]], ptr align 4 @.offload_sizes.16, i32 32, i1 false)
-// CHECK15-NEXT: [[TMP143:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
-// CHECK15-NEXT: store i32 [[TMP138]], ptr [[TMP143]], align 4
-// CHECK15-NEXT: [[TMP144:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
-// CHECK15-NEXT: store i32 [[TMP138]], ptr [[TMP144]], align 4
-// CHECK15-NEXT: [[TMP145:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 0
-// CHECK15-NEXT: store ptr null, ptr [[TMP145]], align 4
-// CHECK15-NEXT: [[TMP146:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP146]], align 4
-// CHECK15-NEXT: [[TMP147:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP147]], align 4
-// CHECK15-NEXT: [[TMP148:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 1
-// CHECK15-NEXT: store ptr null, ptr [[TMP148]], align 4
-// CHECK15-NEXT: [[TMP149:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP149]], align 4
-// CHECK15-NEXT: [[TMP150:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP150]], align 4
-// CHECK15-NEXT: [[TMP151:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 2
-// CHECK15-NEXT: store i64 [[TMP142]], ptr [[TMP151]], align 4
-// CHECK15-NEXT: [[TMP152:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 2
-// CHECK15-NEXT: store ptr null, ptr [[TMP152]], align 4
-// CHECK15-NEXT: [[TMP153:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 3
-// CHECK15-NEXT: store i32 [[TMP140]], ptr [[TMP153]], align 4
-// CHECK15-NEXT: [[TMP154:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 3
-// CHECK15-NEXT: store i32 [[TMP140]], ptr [[TMP154]], align 4
-// CHECK15-NEXT: [[TMP155:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 3
-// CHECK15-NEXT: store ptr null, ptr [[TMP155]], align 4
-// CHECK15-NEXT: [[TMP156:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP157:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP158:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP159:%.*]] = load i32, ptr [[N]], align 4
-// CHECK15-NEXT: store i32 [[TMP159]], ptr [[DOTCAPTURE_EXPR_57]], align 4
-// CHECK15-NEXT: [[TMP160:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_57]], align 4
-// CHECK15-NEXT: [[SUB59:%.*]] = sub nsw i32 [[TMP160]], 0
+// CHECK15-NEXT: [[TMP159:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
+// CHECK15-NEXT: store i32 [[TMP154]], ptr [[TMP159]], align 4
+// CHECK15-NEXT: [[TMP160:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
+// CHECK15-NEXT: store i32 [[TMP154]], ptr [[TMP160]], align 4
+// CHECK15-NEXT: [[TMP161:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 0
+// CHECK15-NEXT: store ptr null, ptr [[TMP161]], align 4
+// CHECK15-NEXT: [[TMP162:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP162]], align 4
+// CHECK15-NEXT: [[TMP163:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP163]], align 4
+// CHECK15-NEXT: [[TMP164:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 1
+// CHECK15-NEXT: store ptr null, ptr [[TMP164]], align 4
+// CHECK15-NEXT: [[TMP165:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP165]], align 4
+// CHECK15-NEXT: [[TMP166:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP166]], align 4
+// CHECK15-NEXT: [[TMP167:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 2
+// CHECK15-NEXT: store i64 [[TMP158]], ptr [[TMP167]], align 4
+// CHECK15-NEXT: [[TMP168:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 2
+// CHECK15-NEXT: store ptr null, ptr [[TMP168]], align 4
+// CHECK15-NEXT: [[TMP169:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 3
+// CHECK15-NEXT: store i32 [[TMP156]], ptr [[TMP169]], align 4
+// CHECK15-NEXT: [[TMP170:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 3
+// CHECK15-NEXT: store i32 [[TMP156]], ptr [[TMP170]], align 4
+// CHECK15-NEXT: [[TMP171:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 3
+// CHECK15-NEXT: store ptr null, ptr [[TMP171]], align 4
+// CHECK15-NEXT: [[TMP172:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP173:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP174:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP175:%.*]] = load i32, ptr [[N]], align 4
+// CHECK15-NEXT: store i32 [[TMP175]], ptr [[DOTCAPTURE_EXPR_57]], align 4
+// CHECK15-NEXT: [[TMP176:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_57]], align 4
+// CHECK15-NEXT: [[SUB59:%.*]] = sub nsw i32 [[TMP176]], 0
// CHECK15-NEXT: [[DIV60:%.*]] = sdiv i32 [[SUB59]], 1
// CHECK15-NEXT: [[SUB61:%.*]] = sub nsw i32 [[DIV60]], 1
// CHECK15-NEXT: store i32 [[SUB61]], ptr [[DOTCAPTURE_EXPR_58]], align 4
-// CHECK15-NEXT: [[TMP161:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_58]], align 4
-// CHECK15-NEXT: [[ADD62:%.*]] = add nsw i32 [[TMP161]], 1
-// CHECK15-NEXT: [[TMP162:%.*]] = zext i32 [[ADD62]] to i64
+// CHECK15-NEXT: [[TMP177:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_58]], align 4
+// CHECK15-NEXT: [[ADD62:%.*]] = add nsw i32 [[TMP177]], 1
+// CHECK15-NEXT: [[TMP178:%.*]] = zext i32 [[ADD62]] to i64
// CHECK15-NEXT: [[KERNEL_ARGS63:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK15-NEXT: [[TMP163:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP163]], align 4
-// CHECK15-NEXT: [[TMP164:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 1
-// CHECK15-NEXT: store i32 4, ptr [[TMP164]], align 4
-// CHECK15-NEXT: [[TMP165:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[TMP156]], ptr [[TMP165]], align 4
-// CHECK15-NEXT: [[TMP166:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 3
-// CHECK15-NEXT: store ptr [[TMP157]], ptr [[TMP166]], align 4
-// CHECK15-NEXT: [[TMP167:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 4
-// CHECK15-NEXT: store ptr [[TMP158]], ptr [[TMP167]], align 4
-// CHECK15-NEXT: [[TMP168:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 5
-// CHECK15-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP168]], align 4
-// CHECK15-NEXT: [[TMP169:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 6
-// CHECK15-NEXT: store ptr null, ptr [[TMP169]], align 4
-// CHECK15-NEXT: [[TMP170:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 7
-// CHECK15-NEXT: store ptr null, ptr [[TMP170]], align 4
-// CHECK15-NEXT: [[TMP171:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 8
-// CHECK15-NEXT: store i64 [[TMP162]], ptr [[TMP171]], align 8
-// CHECK15-NEXT: [[TMP172:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.region_id, ptr [[KERNEL_ARGS63]])
-// CHECK15-NEXT: [[TMP173:%.*]] = icmp ne i32 [[TMP172]], 0
-// CHECK15-NEXT: br i1 [[TMP173]], label [[OMP_OFFLOAD_FAILED64:%.*]], label [[OMP_OFFLOAD_CONT65:%.*]]
+// CHECK15-NEXT: [[TMP179:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 0
+// CHECK15-NEXT: store i32 2, ptr [[TMP179]], align 4
+// CHECK15-NEXT: [[TMP180:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 1
+// CHECK15-NEXT: store i32 4, ptr [[TMP180]], align 4
+// CHECK15-NEXT: [[TMP181:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[TMP172]], ptr [[TMP181]], align 4
+// CHECK15-NEXT: [[TMP182:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 3
+// CHECK15-NEXT: store ptr [[TMP173]], ptr [[TMP182]], align 4
+// CHECK15-NEXT: [[TMP183:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 4
+// CHECK15-NEXT: store ptr [[TMP174]], ptr [[TMP183]], align 4
+// CHECK15-NEXT: [[TMP184:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 5
+// CHECK15-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP184]], align 4
+// CHECK15-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 6
+// CHECK15-NEXT: store ptr null, ptr [[TMP185]], align 4
+// CHECK15-NEXT: [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 7
+// CHECK15-NEXT: store ptr null, ptr [[TMP186]], align 4
+// CHECK15-NEXT: [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 8
+// CHECK15-NEXT: store i64 [[TMP178]], ptr [[TMP187]], align 8
+// CHECK15-NEXT: [[TMP188:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP188]], align 8
+// CHECK15-NEXT: [[TMP189:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP189]], align 4
+// CHECK15-NEXT: [[TMP190:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP190]], align 4
+// CHECK15-NEXT: [[TMP191:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP191]], align 4
+// CHECK15-NEXT: [[TMP192:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.region_id, ptr [[KERNEL_ARGS63]])
+// CHECK15-NEXT: [[TMP193:%.*]] = icmp ne i32 [[TMP192]], 0
+// CHECK15-NEXT: br i1 [[TMP193]], label [[OMP_OFFLOAD_FAILED64:%.*]], label [[OMP_OFFLOAD_CONT65:%.*]]
// CHECK15: omp_offload.failed64:
-// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155(i32 [[TMP138]], i32 [[TMP0]], ptr [[VLA]], i32 [[TMP140]]) #[[ATTR3]]
+// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155(i32 [[TMP154]], i32 [[TMP0]], ptr [[VLA]], i32 [[TMP156]]) #[[ATTR3]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT65]]
// CHECK15: omp_offload.cont65:
-// CHECK15-NEXT: [[TMP174:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK15-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP174]])
+// CHECK15-NEXT: [[TMP194:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK15-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP194]])
// CHECK15-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK15-NEXT: [[TMP175:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK15-NEXT: call void @llvm.stackrestore(ptr [[TMP175]])
-// CHECK15-NEXT: [[TMP176:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK15-NEXT: ret i32 [[TMP176]]
+// CHECK15-NEXT: [[TMP195:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK15-NEXT: call void @llvm.stackrestore(ptr [[TMP195]])
+// CHECK15-NEXT: [[TMP196:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK15-NEXT: ret i32 [[TMP196]]
//
//
// CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139
@@ -9237,7 +9517,7 @@ int main (int argc, char **argv) {
// CHECK15-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK15-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK15-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK15-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK15-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK15-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK15-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -9254,169 +9534,209 @@ int main (int argc, char **argv) {
// CHECK15-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK15-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK15-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK15-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.region_id, ptr [[KERNEL_ARGS]])
-// CHECK15-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK15-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK15-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK15-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK15-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK15-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK15-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.region_id, ptr [[KERNEL_ARGS]])
+// CHECK15-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK15-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK15: omp_offload.failed:
// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112(ptr [[A]]) #[[ATTR3]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK15: omp_offload.cont:
-// CHECK15-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK15-NEXT: store ptr [[A]], ptr [[TMP16]], align 4
-// CHECK15-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK15-NEXT: store ptr [[A]], ptr [[TMP17]], align 4
-// CHECK15-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
-// CHECK15-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK15-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK15-NEXT: store ptr [[A]], ptr [[TMP20]], align 4
+// CHECK15-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK15-NEXT: store ptr [[A]], ptr [[TMP21]], align 4
+// CHECK15-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
+// CHECK15-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK15-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK15-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK15-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK15-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK15-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK15-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK15-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK15-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK15-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK15-NEXT: store ptr @.offload_sizes.24, ptr [[TMP25]], align 4
-// CHECK15-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK15-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP26]], align 4
-// CHECK15-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK15-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK15-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK15-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK15-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK15-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK15-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK15-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK15-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK15-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK15-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK15-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK15-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK15-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK15-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK15-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK15-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK15-NEXT: store ptr @.offload_sizes.24, ptr [[TMP29]], align 4
+// CHECK15-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK15-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP30]], align 4
+// CHECK15-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK15-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK15-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK15-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK15-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK15-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK15-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK15-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK15-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK15-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK15-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK15-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK15-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK15: omp_offload.failed6:
// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116(ptr [[A]]) #[[ATTR3]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK15: omp_offload.cont7:
-// CHECK15-NEXT: [[TMP32:%.*]] = load i32, ptr [[M]], align 4
-// CHECK15-NEXT: store i32 [[TMP32]], ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK15-NEXT: [[TMP33:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK15-NEXT: store i32 [[TMP33]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK15-NEXT: [[TMP34:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK15-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK15-NEXT: store ptr [[A]], ptr [[TMP35]], align 4
-// CHECK15-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK15-NEXT: store ptr [[A]], ptr [[TMP36]], align 4
-// CHECK15-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
-// CHECK15-NEXT: store ptr null, ptr [[TMP37]], align 4
-// CHECK15-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP34]], ptr [[TMP38]], align 4
-// CHECK15-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP34]], ptr [[TMP39]], align 4
-// CHECK15-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1
-// CHECK15-NEXT: store ptr null, ptr [[TMP40]], align 4
-// CHECK15-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP40:%.*]] = load i32, ptr [[M]], align 4
+// CHECK15-NEXT: store i32 [[TMP40]], ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK15-NEXT: [[TMP41:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK15-NEXT: store i32 [[TMP41]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK15-NEXT: [[TMP42:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK15-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK15-NEXT: store ptr [[A]], ptr [[TMP43]], align 4
+// CHECK15-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK15-NEXT: store ptr [[A]], ptr [[TMP44]], align 4
+// CHECK15-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
+// CHECK15-NEXT: store ptr null, ptr [[TMP45]], align 4
+// CHECK15-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP42]], ptr [[TMP46]], align 4
+// CHECK15-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP42]], ptr [[TMP47]], align 4
+// CHECK15-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1
+// CHECK15-NEXT: store ptr null, ptr [[TMP48]], align 4
+// CHECK15-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK15-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK15-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP43]], align 4
-// CHECK15-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK15-NEXT: store i32 2, ptr [[TMP44]], align 4
-// CHECK15-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[TMP41]], ptr [[TMP45]], align 4
-// CHECK15-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK15-NEXT: store ptr [[TMP42]], ptr [[TMP46]], align 4
-// CHECK15-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK15-NEXT: store ptr @.offload_sizes.28, ptr [[TMP47]], align 4
-// CHECK15-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK15-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP48]], align 4
-// CHECK15-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK15-NEXT: store ptr null, ptr [[TMP49]], align 4
-// CHECK15-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK15-NEXT: store ptr null, ptr [[TMP50]], align 4
-// CHECK15-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK15-NEXT: store i64 10, ptr [[TMP51]], align 8
-// CHECK15-NEXT: [[TMP52:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK15-NEXT: [[TMP53:%.*]] = icmp ne i32 [[TMP52]], 0
-// CHECK15-NEXT: br i1 [[TMP53]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK15-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK15-NEXT: store i32 2, ptr [[TMP51]], align 4
+// CHECK15-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK15-NEXT: store i32 2, ptr [[TMP52]], align 4
+// CHECK15-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 4
+// CHECK15-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK15-NEXT: store ptr [[TMP50]], ptr [[TMP54]], align 4
+// CHECK15-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK15-NEXT: store ptr @.offload_sizes.28, ptr [[TMP55]], align 4
+// CHECK15-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK15-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP56]], align 4
+// CHECK15-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK15-NEXT: store ptr null, ptr [[TMP57]], align 4
+// CHECK15-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK15-NEXT: store ptr null, ptr [[TMP58]], align 4
+// CHECK15-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK15-NEXT: store i64 10, ptr [[TMP59]], align 8
+// CHECK15-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP60]], align 8
+// CHECK15-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4
+// CHECK15-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP62]], align 4
+// CHECK15-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP63]], align 4
+// CHECK15-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK15-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
+// CHECK15-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK15: omp_offload.failed13:
-// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120(ptr [[A]], i32 [[TMP34]]) #[[ATTR3]]
+// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120(ptr [[A]], i32 [[TMP42]]) #[[ATTR3]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT14]]
// CHECK15: omp_offload.cont14:
-// CHECK15-NEXT: [[TMP54:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK15-NEXT: store ptr [[A]], ptr [[TMP54]], align 4
-// CHECK15-NEXT: [[TMP55:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
-// CHECK15-NEXT: store ptr [[A]], ptr [[TMP55]], align 4
-// CHECK15-NEXT: [[TMP56:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i32 0, i32 0
-// CHECK15-NEXT: store ptr null, ptr [[TMP56]], align 4
-// CHECK15-NEXT: [[TMP57:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP58:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK15-NEXT: store ptr [[A]], ptr [[TMP66]], align 4
+// CHECK15-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK15-NEXT: store ptr [[A]], ptr [[TMP67]], align 4
+// CHECK15-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i32 0, i32 0
+// CHECK15-NEXT: store ptr null, ptr [[TMP68]], align 4
+// CHECK15-NEXT: [[TMP69:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP70:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
// CHECK15-NEXT: [[KERNEL_ARGS19:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK15-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP59]], align 4
-// CHECK15-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
-// CHECK15-NEXT: store i32 1, ptr [[TMP60]], align 4
-// CHECK15-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[TMP57]], ptr [[TMP61]], align 4
-// CHECK15-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
-// CHECK15-NEXT: store ptr [[TMP58]], ptr [[TMP62]], align 4
-// CHECK15-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
-// CHECK15-NEXT: store ptr @.offload_sizes.32, ptr [[TMP63]], align 4
-// CHECK15-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
-// CHECK15-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP64]], align 4
-// CHECK15-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
-// CHECK15-NEXT: store ptr null, ptr [[TMP65]], align 4
-// CHECK15-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
-// CHECK15-NEXT: store ptr null, ptr [[TMP66]], align 4
-// CHECK15-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
-// CHECK15-NEXT: store i64 10, ptr [[TMP67]], align 8
-// CHECK15-NEXT: [[TMP68:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.region_id, ptr [[KERNEL_ARGS19]])
-// CHECK15-NEXT: [[TMP69:%.*]] = icmp ne i32 [[TMP68]], 0
-// CHECK15-NEXT: br i1 [[TMP69]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
+// CHECK15-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
+// CHECK15-NEXT: store i32 2, ptr [[TMP71]], align 4
+// CHECK15-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
+// CHECK15-NEXT: store i32 1, ptr [[TMP72]], align 4
+// CHECK15-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[TMP69]], ptr [[TMP73]], align 4
+// CHECK15-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
+// CHECK15-NEXT: store ptr [[TMP70]], ptr [[TMP74]], align 4
+// CHECK15-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
+// CHECK15-NEXT: store ptr @.offload_sizes.32, ptr [[TMP75]], align 4
+// CHECK15-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
+// CHECK15-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP76]], align 4
+// CHECK15-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
+// CHECK15-NEXT: store ptr null, ptr [[TMP77]], align 4
+// CHECK15-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
+// CHECK15-NEXT: store ptr null, ptr [[TMP78]], align 4
+// CHECK15-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
+// CHECK15-NEXT: store i64 10, ptr [[TMP79]], align 8
+// CHECK15-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP80]], align 8
+// CHECK15-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP81]], align 4
+// CHECK15-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP82]], align 4
+// CHECK15-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP83]], align 4
+// CHECK15-NEXT: [[TMP84:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.region_id, ptr [[KERNEL_ARGS19]])
+// CHECK15-NEXT: [[TMP85:%.*]] = icmp ne i32 [[TMP84]], 0
+// CHECK15-NEXT: br i1 [[TMP85]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
// CHECK15: omp_offload.failed20:
// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124(ptr [[A]]) #[[ATTR3]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT21]]
// CHECK15: omp_offload.cont21:
-// CHECK15-NEXT: [[TMP70:%.*]] = load i32, ptr [[M]], align 4
-// CHECK15-NEXT: store i32 [[TMP70]], ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK15-NEXT: [[TMP71:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK15-NEXT: store i32 [[TMP71]], ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4
-// CHECK15-NEXT: [[TMP72:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4
-// CHECK15-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
-// CHECK15-NEXT: store ptr [[A]], ptr [[TMP73]], align 4
-// CHECK15-NEXT: [[TMP74:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
-// CHECK15-NEXT: store ptr [[A]], ptr [[TMP74]], align 4
-// CHECK15-NEXT: [[TMP75:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 0
-// CHECK15-NEXT: store ptr null, ptr [[TMP75]], align 4
-// CHECK15-NEXT: [[TMP76:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP72]], ptr [[TMP76]], align 4
-// CHECK15-NEXT: [[TMP77:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP72]], ptr [[TMP77]], align 4
-// CHECK15-NEXT: [[TMP78:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 1
-// CHECK15-NEXT: store ptr null, ptr [[TMP78]], align 4
-// CHECK15-NEXT: [[TMP79:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP80:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP86:%.*]] = load i32, ptr [[M]], align 4
+// CHECK15-NEXT: store i32 [[TMP86]], ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK15-NEXT: [[TMP87:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK15-NEXT: store i32 [[TMP87]], ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4
+// CHECK15-NEXT: [[TMP88:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4
+// CHECK15-NEXT: [[TMP89:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
+// CHECK15-NEXT: store ptr [[A]], ptr [[TMP89]], align 4
+// CHECK15-NEXT: [[TMP90:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
+// CHECK15-NEXT: store ptr [[A]], ptr [[TMP90]], align 4
+// CHECK15-NEXT: [[TMP91:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 0
+// CHECK15-NEXT: store ptr null, ptr [[TMP91]], align 4
+// CHECK15-NEXT: [[TMP92:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP88]], ptr [[TMP92]], align 4
+// CHECK15-NEXT: [[TMP93:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP88]], ptr [[TMP93]], align 4
+// CHECK15-NEXT: [[TMP94:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 1
+// CHECK15-NEXT: store ptr null, ptr [[TMP94]], align 4
+// CHECK15-NEXT: [[TMP95:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP96:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
// CHECK15-NEXT: [[KERNEL_ARGS28:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK15-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP81]], align 4
-// CHECK15-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
-// CHECK15-NEXT: store i32 2, ptr [[TMP82]], align 4
-// CHECK15-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[TMP79]], ptr [[TMP83]], align 4
-// CHECK15-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
-// CHECK15-NEXT: store ptr [[TMP80]], ptr [[TMP84]], align 4
-// CHECK15-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
-// CHECK15-NEXT: store ptr @.offload_sizes.36, ptr [[TMP85]], align 4
-// CHECK15-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
-// CHECK15-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP86]], align 4
-// CHECK15-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
-// CHECK15-NEXT: store ptr null, ptr [[TMP87]], align 4
-// CHECK15-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
-// CHECK15-NEXT: store ptr null, ptr [[TMP88]], align 4
-// CHECK15-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
-// CHECK15-NEXT: store i64 10, ptr [[TMP89]], align 8
-// CHECK15-NEXT: [[TMP90:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.region_id, ptr [[KERNEL_ARGS28]])
-// CHECK15-NEXT: [[TMP91:%.*]] = icmp ne i32 [[TMP90]], 0
-// CHECK15-NEXT: br i1 [[TMP91]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
+// CHECK15-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
+// CHECK15-NEXT: store i32 2, ptr [[TMP97]], align 4
+// CHECK15-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
+// CHECK15-NEXT: store i32 2, ptr [[TMP98]], align 4
+// CHECK15-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[TMP95]], ptr [[TMP99]], align 4
+// CHECK15-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
+// CHECK15-NEXT: store ptr [[TMP96]], ptr [[TMP100]], align 4
+// CHECK15-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
+// CHECK15-NEXT: store ptr @.offload_sizes.36, ptr [[TMP101]], align 4
+// CHECK15-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
+// CHECK15-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP102]], align 4
+// CHECK15-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
+// CHECK15-NEXT: store ptr null, ptr [[TMP103]], align 4
+// CHECK15-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
+// CHECK15-NEXT: store ptr null, ptr [[TMP104]], align 4
+// CHECK15-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
+// CHECK15-NEXT: store i64 10, ptr [[TMP105]], align 8
+// CHECK15-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP106]], align 8
+// CHECK15-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP107]], align 4
+// CHECK15-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP108]], align 4
+// CHECK15-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP109]], align 4
+// CHECK15-NEXT: [[TMP110:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.region_id, ptr [[KERNEL_ARGS28]])
+// CHECK15-NEXT: [[TMP111:%.*]] = icmp ne i32 [[TMP110]], 0
+// CHECK15-NEXT: br i1 [[TMP111]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
// CHECK15: omp_offload.failed29:
-// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128(ptr [[A]], i32 [[TMP72]]) #[[ATTR3]]
+// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128(ptr [[A]], i32 [[TMP88]]) #[[ATTR3]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT30]]
// CHECK15: omp_offload.cont30:
// CHECK15-NEXT: ret i32 0
@@ -10374,7 +10694,7 @@ int main (int argc, char **argv) {
// CHECK17-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK17-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK17-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK17-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK17-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK17-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK17-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -10391,298 +10711,338 @@ int main (int argc, char **argv) {
// CHECK17-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK17-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK17-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK17-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.region_id, ptr [[KERNEL_ARGS]])
-// CHECK17-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK17-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK17-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK17-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK17-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK17-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK17-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.region_id, ptr [[KERNEL_ARGS]])
+// CHECK17-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK17-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK17: omp_offload.failed:
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139(i64 [[TMP4]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK17: omp_offload.cont:
-// CHECK17-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK17-NEXT: [[TMP35:%.*]] = load i64, ptr [[N_CASTED3]], align 8
-// CHECK17-NEXT: [[TMP36:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK17-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK17-NEXT: [[TMP39:%.*]] = load i64, ptr [[N_CASTED3]], align 8
+// CHECK17-NEXT: [[TMP40:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES7]], ptr align 8 @.offload_sizes.4, i64 24, i1 false)
-// CHECK17-NEXT: [[TMP37:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP35]], ptr [[TMP37]], align 8
-// CHECK17-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP35]], ptr [[TMP38]], align 8
-// CHECK17-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK17-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP40]], align 8
-// CHECK17-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP41]], align 8
-// CHECK17-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
-// CHECK17-NEXT: store ptr null, ptr [[TMP42]], align 8
-// CHECK17-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP43]], align 8
-// CHECK17-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 8
-// CHECK17-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK17-NEXT: store i64 [[TMP36]], ptr [[TMP45]], align 8
-// CHECK17-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK17-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP39]], ptr [[TMP41]], align 8
+// CHECK17-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP39]], ptr [[TMP42]], align 8
+// CHECK17-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK17-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP44]], align 8
+// CHECK17-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP45]], align 8
+// CHECK17-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
// CHECK17-NEXT: store ptr null, ptr [[TMP46]], align 8
-// CHECK17-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP50:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP50]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK17-NEXT: [[TMP51:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK17-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP51]], 0
+// CHECK17-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP47]], align 8
+// CHECK17-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 8
+// CHECK17-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK17-NEXT: store i64 [[TMP40]], ptr [[TMP49]], align 8
+// CHECK17-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK17-NEXT: store ptr null, ptr [[TMP50]], align 8
+// CHECK17-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP54:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP54]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK17-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK17-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP55]], 0
// CHECK17-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK17-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK17-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK17-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK17-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP52]], 1
-// CHECK17-NEXT: [[TMP53:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK17-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK17-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP56]], 1
+// CHECK17-NEXT: [[TMP57:%.*]] = zext i32 [[ADD14]] to i64
// CHECK17-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK17-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK17-NEXT: store i32 3, ptr [[TMP55]], align 4
-// CHECK17-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP47]], ptr [[TMP56]], align 8
-// CHECK17-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 8
-// CHECK17-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK17-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 8
-// CHECK17-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP59]], align 8
-// CHECK17-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP60]], align 8
-// CHECK17-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP61]], align 8
-// CHECK17-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK17-NEXT: store i64 [[TMP53]], ptr [[TMP62]], align 8
-// CHECK17-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK17-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
-// CHECK17-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK17-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP58]], align 4
+// CHECK17-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK17-NEXT: store i32 3, ptr [[TMP59]], align 4
+// CHECK17-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP51]], ptr [[TMP60]], align 8
+// CHECK17-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 8
+// CHECK17-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK17-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 8
+// CHECK17-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP63]], align 8
+// CHECK17-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP64]], align 8
+// CHECK17-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP65]], align 8
+// CHECK17-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK17-NEXT: store i64 [[TMP57]], ptr [[TMP66]], align 8
+// CHECK17-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP67]], align 8
+// CHECK17-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP68]], align 4
+// CHECK17-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK17-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP70]], align 4
+// CHECK17-NEXT: [[TMP71:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK17-NEXT: [[TMP72:%.*]] = icmp ne i32 [[TMP71]], 0
+// CHECK17-NEXT: br i1 [[TMP72]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK17: omp_offload.failed16:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143(i64 [[TMP35]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143(i64 [[TMP39]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK17: omp_offload.cont17:
-// CHECK17-NEXT: [[TMP65:%.*]] = load i32, ptr [[M]], align 4
-// CHECK17-NEXT: store i32 [[TMP65]], ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK17-NEXT: [[TMP66:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP66]], ptr [[N_CASTED19]], align 4
-// CHECK17-NEXT: [[TMP67:%.*]] = load i64, ptr [[N_CASTED19]], align 8
-// CHECK17-NEXT: [[TMP68:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK17-NEXT: store i32 [[TMP68]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK17-NEXT: [[TMP69:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK17-NEXT: [[TMP70:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK17-NEXT: [[TMP73:%.*]] = load i32, ptr [[M]], align 4
+// CHECK17-NEXT: store i32 [[TMP73]], ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK17-NEXT: [[TMP74:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP74]], ptr [[N_CASTED19]], align 4
+// CHECK17-NEXT: [[TMP75:%.*]] = load i64, ptr [[N_CASTED19]], align 8
+// CHECK17-NEXT: [[TMP76:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK17-NEXT: store i32 [[TMP76]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK17-NEXT: [[TMP77:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK17-NEXT: [[TMP78:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES23]], ptr align 8 @.offload_sizes.8, i64 32, i1 false)
-// CHECK17-NEXT: [[TMP71:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP67]], ptr [[TMP71]], align 8
-// CHECK17-NEXT: [[TMP72:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP67]], ptr [[TMP72]], align 8
-// CHECK17-NEXT: [[TMP73:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP73]], align 8
-// CHECK17-NEXT: [[TMP74:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP74]], align 8
-// CHECK17-NEXT: [[TMP75:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP75]], align 8
-// CHECK17-NEXT: [[TMP76:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 1
-// CHECK17-NEXT: store ptr null, ptr [[TMP76]], align 8
-// CHECK17-NEXT: [[TMP77:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP77]], align 8
-// CHECK17-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP78]], align 8
-// CHECK17-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
-// CHECK17-NEXT: store i64 [[TMP70]], ptr [[TMP79]], align 8
-// CHECK17-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 2
-// CHECK17-NEXT: store ptr null, ptr [[TMP80]], align 8
-// CHECK17-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
-// CHECK17-NEXT: store i64 [[TMP69]], ptr [[TMP81]], align 8
-// CHECK17-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
-// CHECK17-NEXT: store i64 [[TMP69]], ptr [[TMP82]], align 8
-// CHECK17-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 3
-// CHECK17-NEXT: store ptr null, ptr [[TMP83]], align 8
-// CHECK17-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP87:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP87]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK17-NEXT: [[TMP88:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK17-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP88]], 0
+// CHECK17-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP75]], ptr [[TMP79]], align 8
+// CHECK17-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP75]], ptr [[TMP80]], align 8
+// CHECK17-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP81]], align 8
+// CHECK17-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP82]], align 8
+// CHECK17-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP83]], align 8
+// CHECK17-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 1
+// CHECK17-NEXT: store ptr null, ptr [[TMP84]], align 8
+// CHECK17-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP85]], align 8
+// CHECK17-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP86]], align 8
+// CHECK17-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
+// CHECK17-NEXT: store i64 [[TMP78]], ptr [[TMP87]], align 8
+// CHECK17-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 2
+// CHECK17-NEXT: store ptr null, ptr [[TMP88]], align 8
+// CHECK17-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
+// CHECK17-NEXT: store i64 [[TMP77]], ptr [[TMP89]], align 8
+// CHECK17-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
+// CHECK17-NEXT: store i64 [[TMP77]], ptr [[TMP90]], align 8
+// CHECK17-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 3
+// CHECK17-NEXT: store ptr null, ptr [[TMP91]], align 8
+// CHECK17-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP94:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP95:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP95]], ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK17-NEXT: [[TMP96:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK17-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP96]], 0
// CHECK17-NEXT: [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1
// CHECK17-NEXT: [[SUB29:%.*]] = sub nsw i32 [[DIV28]], 1
// CHECK17-NEXT: store i32 [[SUB29]], ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK17-NEXT: [[TMP89:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK17-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP89]], 1
-// CHECK17-NEXT: [[TMP90:%.*]] = zext i32 [[ADD30]] to i64
+// CHECK17-NEXT: [[TMP97:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
+// CHECK17-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP97]], 1
+// CHECK17-NEXT: [[TMP98:%.*]] = zext i32 [[ADD30]] to i64
// CHECK17-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP91]], align 4
-// CHECK17-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK17-NEXT: store i32 4, ptr [[TMP92]], align 4
-// CHECK17-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP84]], ptr [[TMP93]], align 8
-// CHECK17-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP85]], ptr [[TMP94]], align 8
-// CHECK17-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK17-NEXT: store ptr [[TMP86]], ptr [[TMP95]], align 8
-// CHECK17-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP96]], align 8
-// CHECK17-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP97]], align 8
-// CHECK17-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP98]], align 8
-// CHECK17-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK17-NEXT: store i64 [[TMP90]], ptr [[TMP99]], align 8
-// CHECK17-NEXT: [[TMP100:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK17-NEXT: [[TMP101:%.*]] = icmp ne i32 [[TMP100]], 0
-// CHECK17-NEXT: br i1 [[TMP101]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK17-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP99]], align 4
+// CHECK17-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK17-NEXT: store i32 4, ptr [[TMP100]], align 4
+// CHECK17-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP92]], ptr [[TMP101]], align 8
+// CHECK17-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP93]], ptr [[TMP102]], align 8
+// CHECK17-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK17-NEXT: store ptr [[TMP94]], ptr [[TMP103]], align 8
+// CHECK17-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP104]], align 8
+// CHECK17-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP105]], align 8
+// CHECK17-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP106]], align 8
+// CHECK17-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK17-NEXT: store i64 [[TMP98]], ptr [[TMP107]], align 8
+// CHECK17-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP108]], align 8
+// CHECK17-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP109]], align 4
+// CHECK17-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP110]], align 4
+// CHECK17-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP111]], align 4
+// CHECK17-NEXT: [[TMP112:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK17-NEXT: [[TMP113:%.*]] = icmp ne i32 [[TMP112]], 0
+// CHECK17-NEXT: br i1 [[TMP113]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK17: omp_offload.failed32:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147(i64 [[TMP67]], i64 [[TMP1]], ptr [[VLA]], i64 [[TMP69]]) #[[ATTR3]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147(i64 [[TMP75]], i64 [[TMP1]], ptr [[VLA]], i64 [[TMP77]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK17: omp_offload.cont33:
-// CHECK17-NEXT: [[TMP102:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP102]], ptr [[N_CASTED34]], align 4
-// CHECK17-NEXT: [[TMP103:%.*]] = load i64, ptr [[N_CASTED34]], align 8
-// CHECK17-NEXT: [[TMP104:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK17-NEXT: [[TMP114:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP114]], ptr [[N_CASTED34]], align 4
+// CHECK17-NEXT: [[TMP115:%.*]] = load i64, ptr [[N_CASTED34]], align 8
+// CHECK17-NEXT: [[TMP116:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES38]], ptr align 8 @.offload_sizes.12, i64 24, i1 false)
-// CHECK17-NEXT: [[TMP105:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP103]], ptr [[TMP105]], align 8
-// CHECK17-NEXT: [[TMP106:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP103]], ptr [[TMP106]], align 8
-// CHECK17-NEXT: [[TMP107:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP107]], align 8
-// CHECK17-NEXT: [[TMP108:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP108]], align 8
-// CHECK17-NEXT: [[TMP109:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP109]], align 8
-// CHECK17-NEXT: [[TMP110:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 1
-// CHECK17-NEXT: store ptr null, ptr [[TMP110]], align 8
-// CHECK17-NEXT: [[TMP111:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP111]], align 8
-// CHECK17-NEXT: [[TMP112:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP112]], align 8
-// CHECK17-NEXT: [[TMP113:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 2
-// CHECK17-NEXT: store i64 [[TMP104]], ptr [[TMP113]], align 8
-// CHECK17-NEXT: [[TMP114:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 2
-// CHECK17-NEXT: store ptr null, ptr [[TMP114]], align 8
-// CHECK17-NEXT: [[TMP115:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP116:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP117:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP118:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP118]], ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK17-NEXT: [[TMP119:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK17-NEXT: [[SUB42:%.*]] = sub nsw i32 [[TMP119]], 0
+// CHECK17-NEXT: [[TMP117:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP115]], ptr [[TMP117]], align 8
+// CHECK17-NEXT: [[TMP118:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP115]], ptr [[TMP118]], align 8
+// CHECK17-NEXT: [[TMP119:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP119]], align 8
+// CHECK17-NEXT: [[TMP120:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP120]], align 8
+// CHECK17-NEXT: [[TMP121:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP121]], align 8
+// CHECK17-NEXT: [[TMP122:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 1
+// CHECK17-NEXT: store ptr null, ptr [[TMP122]], align 8
+// CHECK17-NEXT: [[TMP123:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP123]], align 8
+// CHECK17-NEXT: [[TMP124:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP124]], align 8
+// CHECK17-NEXT: [[TMP125:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 2
+// CHECK17-NEXT: store i64 [[TMP116]], ptr [[TMP125]], align 8
+// CHECK17-NEXT: [[TMP126:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 2
+// CHECK17-NEXT: store ptr null, ptr [[TMP126]], align 8
+// CHECK17-NEXT: [[TMP127:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP128:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP129:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP130:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP130]], ptr [[DOTCAPTURE_EXPR_40]], align 4
+// CHECK17-NEXT: [[TMP131:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
+// CHECK17-NEXT: [[SUB42:%.*]] = sub nsw i32 [[TMP131]], 0
// CHECK17-NEXT: [[DIV43:%.*]] = sdiv i32 [[SUB42]], 1
// CHECK17-NEXT: [[SUB44:%.*]] = sub nsw i32 [[DIV43]], 1
// CHECK17-NEXT: store i32 [[SUB44]], ptr [[DOTCAPTURE_EXPR_41]], align 4
-// CHECK17-NEXT: [[TMP120:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_41]], align 4
-// CHECK17-NEXT: [[ADD45:%.*]] = add nsw i32 [[TMP120]], 1
-// CHECK17-NEXT: [[TMP121:%.*]] = zext i32 [[ADD45]] to i64
+// CHECK17-NEXT: [[TMP132:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_41]], align 4
+// CHECK17-NEXT: [[ADD45:%.*]] = add nsw i32 [[TMP132]], 1
+// CHECK17-NEXT: [[TMP133:%.*]] = zext i32 [[ADD45]] to i64
// CHECK17-NEXT: [[KERNEL_ARGS46:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP122:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP122]], align 4
-// CHECK17-NEXT: [[TMP123:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 1
-// CHECK17-NEXT: store i32 3, ptr [[TMP123]], align 4
-// CHECK17-NEXT: [[TMP124:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP115]], ptr [[TMP124]], align 8
-// CHECK17-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP116]], ptr [[TMP125]], align 8
-// CHECK17-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 4
-// CHECK17-NEXT: store ptr [[TMP117]], ptr [[TMP126]], align 8
-// CHECK17-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP127]], align 8
-// CHECK17-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP128]], align 8
-// CHECK17-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP129]], align 8
-// CHECK17-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 8
-// CHECK17-NEXT: store i64 [[TMP121]], ptr [[TMP130]], align 8
-// CHECK17-NEXT: [[TMP131:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.region_id, ptr [[KERNEL_ARGS46]])
-// CHECK17-NEXT: [[TMP132:%.*]] = icmp ne i32 [[TMP131]], 0
-// CHECK17-NEXT: br i1 [[TMP132]], label [[OMP_OFFLOAD_FAILED47:%.*]], label [[OMP_OFFLOAD_CONT48:%.*]]
+// CHECK17-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP134]], align 4
+// CHECK17-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 1
+// CHECK17-NEXT: store i32 3, ptr [[TMP135]], align 4
+// CHECK17-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP127]], ptr [[TMP136]], align 8
+// CHECK17-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP128]], ptr [[TMP137]], align 8
+// CHECK17-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 4
+// CHECK17-NEXT: store ptr [[TMP129]], ptr [[TMP138]], align 8
+// CHECK17-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP139]], align 8
+// CHECK17-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP140]], align 8
+// CHECK17-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP141]], align 8
+// CHECK17-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 8
+// CHECK17-NEXT: store i64 [[TMP133]], ptr [[TMP142]], align 8
+// CHECK17-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP143]], align 8
+// CHECK17-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP144]], align 4
+// CHECK17-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP145]], align 4
+// CHECK17-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP146]], align 4
+// CHECK17-NEXT: [[TMP147:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.region_id, ptr [[KERNEL_ARGS46]])
+// CHECK17-NEXT: [[TMP148:%.*]] = icmp ne i32 [[TMP147]], 0
+// CHECK17-NEXT: br i1 [[TMP148]], label [[OMP_OFFLOAD_FAILED47:%.*]], label [[OMP_OFFLOAD_CONT48:%.*]]
// CHECK17: omp_offload.failed47:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151(i64 [[TMP103]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151(i64 [[TMP115]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT48]]
// CHECK17: omp_offload.cont48:
-// CHECK17-NEXT: [[TMP133:%.*]] = load i32, ptr [[M]], align 4
-// CHECK17-NEXT: store i32 [[TMP133]], ptr [[DOTCAPTURE_EXPR_49]], align 4
-// CHECK17-NEXT: [[TMP134:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP134]], ptr [[N_CASTED50]], align 4
-// CHECK17-NEXT: [[TMP135:%.*]] = load i64, ptr [[N_CASTED50]], align 8
-// CHECK17-NEXT: [[TMP136:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_49]], align 4
-// CHECK17-NEXT: store i32 [[TMP136]], ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4
-// CHECK17-NEXT: [[TMP137:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED51]], align 8
-// CHECK17-NEXT: [[TMP138:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK17-NEXT: [[TMP149:%.*]] = load i32, ptr [[M]], align 4
+// CHECK17-NEXT: store i32 [[TMP149]], ptr [[DOTCAPTURE_EXPR_49]], align 4
+// CHECK17-NEXT: [[TMP150:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP150]], ptr [[N_CASTED50]], align 4
+// CHECK17-NEXT: [[TMP151:%.*]] = load i64, ptr [[N_CASTED50]], align 8
+// CHECK17-NEXT: [[TMP152:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_49]], align 4
+// CHECK17-NEXT: store i32 [[TMP152]], ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4
+// CHECK17-NEXT: [[TMP153:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED51]], align 8
+// CHECK17-NEXT: [[TMP154:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES55]], ptr align 8 @.offload_sizes.16, i64 32, i1 false)
-// CHECK17-NEXT: [[TMP139:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP135]], ptr [[TMP139]], align 8
-// CHECK17-NEXT: [[TMP140:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP135]], ptr [[TMP140]], align 8
-// CHECK17-NEXT: [[TMP141:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP141]], align 8
-// CHECK17-NEXT: [[TMP142:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP142]], align 8
-// CHECK17-NEXT: [[TMP143:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP143]], align 8
-// CHECK17-NEXT: [[TMP144:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 1
-// CHECK17-NEXT: store ptr null, ptr [[TMP144]], align 8
-// CHECK17-NEXT: [[TMP145:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP145]], align 8
-// CHECK17-NEXT: [[TMP146:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP146]], align 8
-// CHECK17-NEXT: [[TMP147:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 2
-// CHECK17-NEXT: store i64 [[TMP138]], ptr [[TMP147]], align 8
-// CHECK17-NEXT: [[TMP148:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 2
-// CHECK17-NEXT: store ptr null, ptr [[TMP148]], align 8
-// CHECK17-NEXT: [[TMP149:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 3
-// CHECK17-NEXT: store i64 [[TMP137]], ptr [[TMP149]], align 8
-// CHECK17-NEXT: [[TMP150:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 3
-// CHECK17-NEXT: store i64 [[TMP137]], ptr [[TMP150]], align 8
-// CHECK17-NEXT: [[TMP151:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 3
-// CHECK17-NEXT: store ptr null, ptr [[TMP151]], align 8
-// CHECK17-NEXT: [[TMP152:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP153:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP154:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP155:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP155]], ptr [[DOTCAPTURE_EXPR_57]], align 4
-// CHECK17-NEXT: [[TMP156:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_57]], align 4
-// CHECK17-NEXT: [[SUB59:%.*]] = sub nsw i32 [[TMP156]], 0
+// CHECK17-NEXT: [[TMP155:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP151]], ptr [[TMP155]], align 8
+// CHECK17-NEXT: [[TMP156:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP151]], ptr [[TMP156]], align 8
+// CHECK17-NEXT: [[TMP157:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP157]], align 8
+// CHECK17-NEXT: [[TMP158:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP158]], align 8
+// CHECK17-NEXT: [[TMP159:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP159]], align 8
+// CHECK17-NEXT: [[TMP160:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 1
+// CHECK17-NEXT: store ptr null, ptr [[TMP160]], align 8
+// CHECK17-NEXT: [[TMP161:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP161]], align 8
+// CHECK17-NEXT: [[TMP162:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP162]], align 8
+// CHECK17-NEXT: [[TMP163:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 2
+// CHECK17-NEXT: store i64 [[TMP154]], ptr [[TMP163]], align 8
+// CHECK17-NEXT: [[TMP164:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 2
+// CHECK17-NEXT: store ptr null, ptr [[TMP164]], align 8
+// CHECK17-NEXT: [[TMP165:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 3
+// CHECK17-NEXT: store i64 [[TMP153]], ptr [[TMP165]], align 8
+// CHECK17-NEXT: [[TMP166:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 3
+// CHECK17-NEXT: store i64 [[TMP153]], ptr [[TMP166]], align 8
+// CHECK17-NEXT: [[TMP167:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 3
+// CHECK17-NEXT: store ptr null, ptr [[TMP167]], align 8
+// CHECK17-NEXT: [[TMP168:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP169:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP170:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP171:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP171]], ptr [[DOTCAPTURE_EXPR_57]], align 4
+// CHECK17-NEXT: [[TMP172:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_57]], align 4
+// CHECK17-NEXT: [[SUB59:%.*]] = sub nsw i32 [[TMP172]], 0
// CHECK17-NEXT: [[DIV60:%.*]] = sdiv i32 [[SUB59]], 1
// CHECK17-NEXT: [[SUB61:%.*]] = sub nsw i32 [[DIV60]], 1
// CHECK17-NEXT: store i32 [[SUB61]], ptr [[DOTCAPTURE_EXPR_58]], align 4
-// CHECK17-NEXT: [[TMP157:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_58]], align 4
-// CHECK17-NEXT: [[ADD62:%.*]] = add nsw i32 [[TMP157]], 1
-// CHECK17-NEXT: [[TMP158:%.*]] = zext i32 [[ADD62]] to i64
+// CHECK17-NEXT: [[TMP173:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_58]], align 4
+// CHECK17-NEXT: [[ADD62:%.*]] = add nsw i32 [[TMP173]], 1
+// CHECK17-NEXT: [[TMP174:%.*]] = zext i32 [[ADD62]] to i64
// CHECK17-NEXT: [[KERNEL_ARGS63:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP159:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP159]], align 4
-// CHECK17-NEXT: [[TMP160:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 1
-// CHECK17-NEXT: store i32 4, ptr [[TMP160]], align 4
-// CHECK17-NEXT: [[TMP161:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP152]], ptr [[TMP161]], align 8
-// CHECK17-NEXT: [[TMP162:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP153]], ptr [[TMP162]], align 8
-// CHECK17-NEXT: [[TMP163:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 4
-// CHECK17-NEXT: store ptr [[TMP154]], ptr [[TMP163]], align 8
-// CHECK17-NEXT: [[TMP164:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP164]], align 8
-// CHECK17-NEXT: [[TMP165:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP165]], align 8
-// CHECK17-NEXT: [[TMP166:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP166]], align 8
-// CHECK17-NEXT: [[TMP167:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 8
-// CHECK17-NEXT: store i64 [[TMP158]], ptr [[TMP167]], align 8
-// CHECK17-NEXT: [[TMP168:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.region_id, ptr [[KERNEL_ARGS63]])
-// CHECK17-NEXT: [[TMP169:%.*]] = icmp ne i32 [[TMP168]], 0
-// CHECK17-NEXT: br i1 [[TMP169]], label [[OMP_OFFLOAD_FAILED64:%.*]], label [[OMP_OFFLOAD_CONT65:%.*]]
+// CHECK17-NEXT: [[TMP175:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP175]], align 4
+// CHECK17-NEXT: [[TMP176:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 1
+// CHECK17-NEXT: store i32 4, ptr [[TMP176]], align 4
+// CHECK17-NEXT: [[TMP177:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP168]], ptr [[TMP177]], align 8
+// CHECK17-NEXT: [[TMP178:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP169]], ptr [[TMP178]], align 8
+// CHECK17-NEXT: [[TMP179:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 4
+// CHECK17-NEXT: store ptr [[TMP170]], ptr [[TMP179]], align 8
+// CHECK17-NEXT: [[TMP180:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP180]], align 8
+// CHECK17-NEXT: [[TMP181:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP181]], align 8
+// CHECK17-NEXT: [[TMP182:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP182]], align 8
+// CHECK17-NEXT: [[TMP183:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 8
+// CHECK17-NEXT: store i64 [[TMP174]], ptr [[TMP183]], align 8
+// CHECK17-NEXT: [[TMP184:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP184]], align 8
+// CHECK17-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP185]], align 4
+// CHECK17-NEXT: [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP186]], align 4
+// CHECK17-NEXT: [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP187]], align 4
+// CHECK17-NEXT: [[TMP188:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.region_id, ptr [[KERNEL_ARGS63]])
+// CHECK17-NEXT: [[TMP189:%.*]] = icmp ne i32 [[TMP188]], 0
+// CHECK17-NEXT: br i1 [[TMP189]], label [[OMP_OFFLOAD_FAILED64:%.*]], label [[OMP_OFFLOAD_CONT65:%.*]]
// CHECK17: omp_offload.failed64:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155(i64 [[TMP135]], i64 [[TMP1]], ptr [[VLA]], i64 [[TMP137]]) #[[ATTR3]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155(i64 [[TMP151]], i64 [[TMP1]], ptr [[VLA]], i64 [[TMP153]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT65]]
// CHECK17: omp_offload.cont65:
-// CHECK17-NEXT: [[TMP170:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK17-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP170]])
+// CHECK17-NEXT: [[TMP190:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK17-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP190]])
// CHECK17-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK17-NEXT: [[TMP171:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK17-NEXT: call void @llvm.stackrestore(ptr [[TMP171]])
-// CHECK17-NEXT: [[TMP172:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK17-NEXT: ret i32 [[TMP172]]
+// CHECK17-NEXT: [[TMP191:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK17-NEXT: call void @llvm.stackrestore(ptr [[TMP191]])
+// CHECK17-NEXT: [[TMP192:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK17-NEXT: ret i32 [[TMP192]]
//
//
// CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139
@@ -11978,7 +12338,7 @@ int main (int argc, char **argv) {
// CHECK17-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK17-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK17-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK17-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK17-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK17-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -11995,169 +12355,209 @@ int main (int argc, char **argv) {
// CHECK17-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK17-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK17-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK17-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.region_id, ptr [[KERNEL_ARGS]])
-// CHECK17-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK17-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK17-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK17-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK17-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK17-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK17-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.region_id, ptr [[KERNEL_ARGS]])
+// CHECK17-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK17-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK17: omp_offload.failed:
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112(ptr [[A]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK17: omp_offload.cont:
-// CHECK17-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP16]], align 8
-// CHECK17-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP17]], align 8
-// CHECK17-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK17-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP20]], align 8
+// CHECK17-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP21]], align 8
+// CHECK17-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK17-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK17-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK17-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK17-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK17-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK17-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK17-NEXT: store ptr @.offload_sizes.24, ptr [[TMP25]], align 8
-// CHECK17-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP26]], align 8
-// CHECK17-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK17-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK17-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK17-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK17-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK17-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK17-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK17-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK17-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK17-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK17-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK17-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK17-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK17-NEXT: store ptr @.offload_sizes.24, ptr [[TMP29]], align 8
+// CHECK17-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP30]], align 8
+// CHECK17-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK17-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK17-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK17-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK17-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK17-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK17-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK17-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK17-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK17-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK17-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK17: omp_offload.failed6:
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116(ptr [[A]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK17: omp_offload.cont7:
-// CHECK17-NEXT: [[TMP32:%.*]] = load i32, ptr [[M]], align 4
-// CHECK17-NEXT: store i32 [[TMP32]], ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK17-NEXT: [[TMP33:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK17-NEXT: store i32 [[TMP33]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK17-NEXT: [[TMP34:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK17-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP35]], align 8
-// CHECK17-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP36]], align 8
-// CHECK17-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP37]], align 8
-// CHECK17-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP34]], ptr [[TMP38]], align 8
-// CHECK17-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP34]], ptr [[TMP39]], align 8
-// CHECK17-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
-// CHECK17-NEXT: store ptr null, ptr [[TMP40]], align 8
-// CHECK17-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP40:%.*]] = load i32, ptr [[M]], align 4
+// CHECK17-NEXT: store i32 [[TMP40]], ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK17-NEXT: [[TMP41:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK17-NEXT: store i32 [[TMP41]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK17-NEXT: [[TMP42:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK17-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP43]], align 8
+// CHECK17-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP44]], align 8
+// CHECK17-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP45]], align 8
+// CHECK17-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP42]], ptr [[TMP46]], align 8
+// CHECK17-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP42]], ptr [[TMP47]], align 8
+// CHECK17-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
+// CHECK17-NEXT: store ptr null, ptr [[TMP48]], align 8
+// CHECK17-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP43]], align 4
-// CHECK17-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK17-NEXT: store i32 2, ptr [[TMP44]], align 4
-// CHECK17-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP41]], ptr [[TMP45]], align 8
-// CHECK17-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP42]], ptr [[TMP46]], align 8
-// CHECK17-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK17-NEXT: store ptr @.offload_sizes.28, ptr [[TMP47]], align 8
-// CHECK17-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP48]], align 8
-// CHECK17-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP49]], align 8
-// CHECK17-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP50]], align 8
-// CHECK17-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK17-NEXT: store i64 10, ptr [[TMP51]], align 8
-// CHECK17-NEXT: [[TMP52:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK17-NEXT: [[TMP53:%.*]] = icmp ne i32 [[TMP52]], 0
-// CHECK17-NEXT: br i1 [[TMP53]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK17-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP51]], align 4
+// CHECK17-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK17-NEXT: store i32 2, ptr [[TMP52]], align 4
+// CHECK17-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 8
+// CHECK17-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP50]], ptr [[TMP54]], align 8
+// CHECK17-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK17-NEXT: store ptr @.offload_sizes.28, ptr [[TMP55]], align 8
+// CHECK17-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP56]], align 8
+// CHECK17-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP57]], align 8
+// CHECK17-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP58]], align 8
+// CHECK17-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK17-NEXT: store i64 10, ptr [[TMP59]], align 8
+// CHECK17-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP60]], align 8
+// CHECK17-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4
+// CHECK17-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP62]], align 4
+// CHECK17-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP63]], align 4
+// CHECK17-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK17-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
+// CHECK17-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK17: omp_offload.failed13:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120(ptr [[A]], i64 [[TMP34]]) #[[ATTR3]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120(ptr [[A]], i64 [[TMP42]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT14]]
// CHECK17: omp_offload.cont14:
-// CHECK17-NEXT: [[TMP54:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP54]], align 8
-// CHECK17-NEXT: [[TMP55:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP55]], align 8
-// CHECK17-NEXT: [[TMP56:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i64 0, i64 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP56]], align 8
-// CHECK17-NEXT: [[TMP57:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP58:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP66]], align 8
+// CHECK17-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP67]], align 8
+// CHECK17-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i64 0, i64 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP68]], align 8
+// CHECK17-NEXT: [[TMP69:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP70:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS19:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP59]], align 4
-// CHECK17-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
-// CHECK17-NEXT: store i32 1, ptr [[TMP60]], align 4
-// CHECK17-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP57]], ptr [[TMP61]], align 8
-// CHECK17-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP58]], ptr [[TMP62]], align 8
-// CHECK17-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
-// CHECK17-NEXT: store ptr @.offload_sizes.32, ptr [[TMP63]], align 8
-// CHECK17-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP64]], align 8
-// CHECK17-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP65]], align 8
-// CHECK17-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP66]], align 8
-// CHECK17-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
-// CHECK17-NEXT: store i64 10, ptr [[TMP67]], align 8
-// CHECK17-NEXT: [[TMP68:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.region_id, ptr [[KERNEL_ARGS19]])
-// CHECK17-NEXT: [[TMP69:%.*]] = icmp ne i32 [[TMP68]], 0
-// CHECK17-NEXT: br i1 [[TMP69]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
+// CHECK17-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP71]], align 4
+// CHECK17-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
+// CHECK17-NEXT: store i32 1, ptr [[TMP72]], align 4
+// CHECK17-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP69]], ptr [[TMP73]], align 8
+// CHECK17-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP70]], ptr [[TMP74]], align 8
+// CHECK17-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
+// CHECK17-NEXT: store ptr @.offload_sizes.32, ptr [[TMP75]], align 8
+// CHECK17-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP76]], align 8
+// CHECK17-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP77]], align 8
+// CHECK17-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP78]], align 8
+// CHECK17-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
+// CHECK17-NEXT: store i64 10, ptr [[TMP79]], align 8
+// CHECK17-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP80]], align 8
+// CHECK17-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP81]], align 4
+// CHECK17-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP82]], align 4
+// CHECK17-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP83]], align 4
+// CHECK17-NEXT: [[TMP84:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.region_id, ptr [[KERNEL_ARGS19]])
+// CHECK17-NEXT: [[TMP85:%.*]] = icmp ne i32 [[TMP84]], 0
+// CHECK17-NEXT: br i1 [[TMP85]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
// CHECK17: omp_offload.failed20:
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124(ptr [[A]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT21]]
// CHECK17: omp_offload.cont21:
-// CHECK17-NEXT: [[TMP70:%.*]] = load i32, ptr [[M]], align 4
-// CHECK17-NEXT: store i32 [[TMP70]], ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK17-NEXT: [[TMP71:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK17-NEXT: store i32 [[TMP71]], ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4
-// CHECK17-NEXT: [[TMP72:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED23]], align 8
-// CHECK17-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP73]], align 8
-// CHECK17-NEXT: [[TMP74:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP74]], align 8
-// CHECK17-NEXT: [[TMP75:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP75]], align 8
-// CHECK17-NEXT: [[TMP76:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP72]], ptr [[TMP76]], align 8
-// CHECK17-NEXT: [[TMP77:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP72]], ptr [[TMP77]], align 8
-// CHECK17-NEXT: [[TMP78:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 1
-// CHECK17-NEXT: store ptr null, ptr [[TMP78]], align 8
-// CHECK17-NEXT: [[TMP79:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP80:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP86:%.*]] = load i32, ptr [[M]], align 4
+// CHECK17-NEXT: store i32 [[TMP86]], ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK17-NEXT: [[TMP87:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK17-NEXT: store i32 [[TMP87]], ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4
+// CHECK17-NEXT: [[TMP88:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED23]], align 8
+// CHECK17-NEXT: [[TMP89:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP89]], align 8
+// CHECK17-NEXT: [[TMP90:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP90]], align 8
+// CHECK17-NEXT: [[TMP91:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP91]], align 8
+// CHECK17-NEXT: [[TMP92:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP88]], ptr [[TMP92]], align 8
+// CHECK17-NEXT: [[TMP93:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP88]], ptr [[TMP93]], align 8
+// CHECK17-NEXT: [[TMP94:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 1
+// CHECK17-NEXT: store ptr null, ptr [[TMP94]], align 8
+// CHECK17-NEXT: [[TMP95:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP96:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS28:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP81]], align 4
-// CHECK17-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
-// CHECK17-NEXT: store i32 2, ptr [[TMP82]], align 4
-// CHECK17-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP79]], ptr [[TMP83]], align 8
-// CHECK17-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP80]], ptr [[TMP84]], align 8
-// CHECK17-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
-// CHECK17-NEXT: store ptr @.offload_sizes.36, ptr [[TMP85]], align 8
-// CHECK17-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP86]], align 8
-// CHECK17-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP87]], align 8
-// CHECK17-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP88]], align 8
-// CHECK17-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
-// CHECK17-NEXT: store i64 10, ptr [[TMP89]], align 8
-// CHECK17-NEXT: [[TMP90:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.region_id, ptr [[KERNEL_ARGS28]])
-// CHECK17-NEXT: [[TMP91:%.*]] = icmp ne i32 [[TMP90]], 0
-// CHECK17-NEXT: br i1 [[TMP91]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
+// CHECK17-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP97]], align 4
+// CHECK17-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
+// CHECK17-NEXT: store i32 2, ptr [[TMP98]], align 4
+// CHECK17-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP95]], ptr [[TMP99]], align 8
+// CHECK17-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP96]], ptr [[TMP100]], align 8
+// CHECK17-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
+// CHECK17-NEXT: store ptr @.offload_sizes.36, ptr [[TMP101]], align 8
+// CHECK17-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP102]], align 8
+// CHECK17-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP103]], align 8
+// CHECK17-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP104]], align 8
+// CHECK17-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
+// CHECK17-NEXT: store i64 10, ptr [[TMP105]], align 8
+// CHECK17-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP106]], align 8
+// CHECK17-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP107]], align 4
+// CHECK17-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP108]], align 4
+// CHECK17-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP109]], align 4
+// CHECK17-NEXT: [[TMP110:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.region_id, ptr [[KERNEL_ARGS28]])
+// CHECK17-NEXT: [[TMP111:%.*]] = icmp ne i32 [[TMP110]], 0
+// CHECK17-NEXT: br i1 [[TMP111]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
// CHECK17: omp_offload.failed29:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128(ptr [[A]], i64 [[TMP72]]) #[[ATTR3]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128(ptr [[A]], i64 [[TMP88]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT30]]
// CHECK17: omp_offload.cont30:
// CHECK17-NEXT: ret i32 0
@@ -13142,7 +13542,7 @@ int main (int argc, char **argv) {
// CHECK19-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK19-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK19-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK19-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK19-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK19-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK19-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -13159,302 +13559,342 @@ int main (int argc, char **argv) {
// CHECK19-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK19-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK19-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK19-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.region_id, ptr [[KERNEL_ARGS]])
-// CHECK19-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK19-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK19-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK19-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK19-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK19-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK19-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.region_id, ptr [[KERNEL_ARGS]])
+// CHECK19-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK19-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK19: omp_offload.failed:
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139(i32 [[TMP3]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK19: omp_offload.cont:
-// CHECK19-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK19-NEXT: [[TMP35:%.*]] = load i32, ptr [[N_CASTED3]], align 4
-// CHECK19-NEXT: [[TMP36:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK19-NEXT: [[TMP37:%.*]] = sext i32 [[TMP36]] to i64
+// CHECK19-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK19-NEXT: [[TMP39:%.*]] = load i32, ptr [[N_CASTED3]], align 4
+// CHECK19-NEXT: [[TMP40:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK19-NEXT: [[TMP41:%.*]] = sext i32 [[TMP40]] to i64
// CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES7]], ptr align 4 @.offload_sizes.4, i32 24, i1 false)
-// CHECK19-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP35]], ptr [[TMP38]], align 4
-// CHECK19-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP35]], ptr [[TMP39]], align 4
-// CHECK19-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP40]], align 4
-// CHECK19-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP41]], align 4
-// CHECK19-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP42]], align 4
-// CHECK19-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
-// CHECK19-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK19-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 4
-// CHECK19-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP45]], align 4
-// CHECK19-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK19-NEXT: store i64 [[TMP37]], ptr [[TMP46]], align 4
-// CHECK19-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK19-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP39]], ptr [[TMP42]], align 4
+// CHECK19-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP39]], ptr [[TMP43]], align 4
+// CHECK19-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP44]], align 4
+// CHECK19-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP45]], align 4
+// CHECK19-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP46]], align 4
+// CHECK19-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
// CHECK19-NEXT: store ptr null, ptr [[TMP47]], align 4
-// CHECK19-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP51:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP51]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK19-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK19-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP52]], 0
+// CHECK19-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 4
+// CHECK19-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP49]], align 4
+// CHECK19-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK19-NEXT: store i64 [[TMP41]], ptr [[TMP50]], align 4
+// CHECK19-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK19-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK19-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP54:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP55:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP55]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK19-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK19-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP56]], 0
// CHECK19-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK19-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK19-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK19-NEXT: [[TMP53:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK19-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP53]], 1
-// CHECK19-NEXT: [[TMP54:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK19-NEXT: [[TMP57:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK19-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP57]], 1
+// CHECK19-NEXT: [[TMP58:%.*]] = zext i32 [[ADD14]] to i64
// CHECK19-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP55]], align 4
-// CHECK19-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK19-NEXT: store i32 3, ptr [[TMP56]], align 4
-// CHECK19-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 4
-// CHECK19-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 4
-// CHECK19-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK19-NEXT: store ptr [[TMP50]], ptr [[TMP59]], align 4
-// CHECK19-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP60]], align 4
-// CHECK19-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP61]], align 4
-// CHECK19-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP62]], align 4
-// CHECK19-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK19-NEXT: store i64 [[TMP54]], ptr [[TMP63]], align 8
-// CHECK19-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK19-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
-// CHECK19-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK19-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP59]], align 4
+// CHECK19-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK19-NEXT: store i32 3, ptr [[TMP60]], align 4
+// CHECK19-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 4
+// CHECK19-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 4
+// CHECK19-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK19-NEXT: store ptr [[TMP54]], ptr [[TMP63]], align 4
+// CHECK19-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP64]], align 4
+// CHECK19-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP65]], align 4
+// CHECK19-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP66]], align 4
+// CHECK19-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK19-NEXT: store i64 [[TMP58]], ptr [[TMP67]], align 8
+// CHECK19-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP68]], align 8
+// CHECK19-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK19-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP70]], align 4
+// CHECK19-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP71]], align 4
+// CHECK19-NEXT: [[TMP72:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK19-NEXT: [[TMP73:%.*]] = icmp ne i32 [[TMP72]], 0
+// CHECK19-NEXT: br i1 [[TMP73]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK19: omp_offload.failed16:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143(i32 [[TMP35]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143(i32 [[TMP39]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK19: omp_offload.cont17:
-// CHECK19-NEXT: [[TMP66:%.*]] = load i32, ptr [[M]], align 4
-// CHECK19-NEXT: store i32 [[TMP66]], ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK19-NEXT: [[TMP67:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP67]], ptr [[N_CASTED19]], align 4
-// CHECK19-NEXT: [[TMP68:%.*]] = load i32, ptr [[N_CASTED19]], align 4
-// CHECK19-NEXT: [[TMP69:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK19-NEXT: store i32 [[TMP69]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK19-NEXT: [[TMP70:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK19-NEXT: [[TMP71:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK19-NEXT: [[TMP72:%.*]] = sext i32 [[TMP71]] to i64
+// CHECK19-NEXT: [[TMP74:%.*]] = load i32, ptr [[M]], align 4
+// CHECK19-NEXT: store i32 [[TMP74]], ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK19-NEXT: [[TMP75:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP75]], ptr [[N_CASTED19]], align 4
+// CHECK19-NEXT: [[TMP76:%.*]] = load i32, ptr [[N_CASTED19]], align 4
+// CHECK19-NEXT: [[TMP77:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK19-NEXT: store i32 [[TMP77]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK19-NEXT: [[TMP78:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK19-NEXT: [[TMP79:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK19-NEXT: [[TMP80:%.*]] = sext i32 [[TMP79]] to i64
// CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES23]], ptr align 4 @.offload_sizes.8, i32 32, i1 false)
-// CHECK19-NEXT: [[TMP73:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP68]], ptr [[TMP73]], align 4
-// CHECK19-NEXT: [[TMP74:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP68]], ptr [[TMP74]], align 4
-// CHECK19-NEXT: [[TMP75:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP75]], align 4
-// CHECK19-NEXT: [[TMP76:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP76]], align 4
-// CHECK19-NEXT: [[TMP77:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP77]], align 4
-// CHECK19-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 1
-// CHECK19-NEXT: store ptr null, ptr [[TMP78]], align 4
-// CHECK19-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP79]], align 4
-// CHECK19-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP80]], align 4
-// CHECK19-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
-// CHECK19-NEXT: store i64 [[TMP72]], ptr [[TMP81]], align 4
-// CHECK19-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 2
-// CHECK19-NEXT: store ptr null, ptr [[TMP82]], align 4
-// CHECK19-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
-// CHECK19-NEXT: store i32 [[TMP70]], ptr [[TMP83]], align 4
-// CHECK19-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
-// CHECK19-NEXT: store i32 [[TMP70]], ptr [[TMP84]], align 4
-// CHECK19-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 3
-// CHECK19-NEXT: store ptr null, ptr [[TMP85]], align 4
-// CHECK19-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP89:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP89]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK19-NEXT: [[TMP90:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK19-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP90]], 0
+// CHECK19-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP76]], ptr [[TMP81]], align 4
+// CHECK19-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP76]], ptr [[TMP82]], align 4
+// CHECK19-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP83]], align 4
+// CHECK19-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP84]], align 4
+// CHECK19-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP85]], align 4
+// CHECK19-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 1
+// CHECK19-NEXT: store ptr null, ptr [[TMP86]], align 4
+// CHECK19-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP87]], align 4
+// CHECK19-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP88]], align 4
+// CHECK19-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
+// CHECK19-NEXT: store i64 [[TMP80]], ptr [[TMP89]], align 4
+// CHECK19-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 2
+// CHECK19-NEXT: store ptr null, ptr [[TMP90]], align 4
+// CHECK19-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
+// CHECK19-NEXT: store i32 [[TMP78]], ptr [[TMP91]], align 4
+// CHECK19-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
+// CHECK19-NEXT: store i32 [[TMP78]], ptr [[TMP92]], align 4
+// CHECK19-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 3
+// CHECK19-NEXT: store ptr null, ptr [[TMP93]], align 4
+// CHECK19-NEXT: [[TMP94:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP95:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP96:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP97:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP97]], ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK19-NEXT: [[TMP98:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK19-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP98]], 0
// CHECK19-NEXT: [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1
// CHECK19-NEXT: [[SUB29:%.*]] = sub nsw i32 [[DIV28]], 1
// CHECK19-NEXT: store i32 [[SUB29]], ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK19-NEXT: [[TMP91:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK19-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP91]], 1
-// CHECK19-NEXT: [[TMP92:%.*]] = zext i32 [[ADD30]] to i64
+// CHECK19-NEXT: [[TMP99:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
+// CHECK19-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP99]], 1
+// CHECK19-NEXT: [[TMP100:%.*]] = zext i32 [[ADD30]] to i64
// CHECK19-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP93]], align 4
-// CHECK19-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK19-NEXT: store i32 4, ptr [[TMP94]], align 4
-// CHECK19-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP86]], ptr [[TMP95]], align 4
-// CHECK19-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP87]], ptr [[TMP96]], align 4
-// CHECK19-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK19-NEXT: store ptr [[TMP88]], ptr [[TMP97]], align 4
-// CHECK19-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP98]], align 4
-// CHECK19-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP99]], align 4
-// CHECK19-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP100]], align 4
-// CHECK19-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK19-NEXT: store i64 [[TMP92]], ptr [[TMP101]], align 8
-// CHECK19-NEXT: [[TMP102:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK19-NEXT: [[TMP103:%.*]] = icmp ne i32 [[TMP102]], 0
-// CHECK19-NEXT: br i1 [[TMP103]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK19-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP101]], align 4
+// CHECK19-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK19-NEXT: store i32 4, ptr [[TMP102]], align 4
+// CHECK19-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP94]], ptr [[TMP103]], align 4
+// CHECK19-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP95]], ptr [[TMP104]], align 4
+// CHECK19-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK19-NEXT: store ptr [[TMP96]], ptr [[TMP105]], align 4
+// CHECK19-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP106]], align 4
+// CHECK19-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP107]], align 4
+// CHECK19-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP108]], align 4
+// CHECK19-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK19-NEXT: store i64 [[TMP100]], ptr [[TMP109]], align 8
+// CHECK19-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP110]], align 8
+// CHECK19-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP111]], align 4
+// CHECK19-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP112]], align 4
+// CHECK19-NEXT: [[TMP113:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP113]], align 4
+// CHECK19-NEXT: [[TMP114:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK19-NEXT: [[TMP115:%.*]] = icmp ne i32 [[TMP114]], 0
+// CHECK19-NEXT: br i1 [[TMP115]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK19: omp_offload.failed32:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147(i32 [[TMP68]], i32 [[TMP0]], ptr [[VLA]], i32 [[TMP70]]) #[[ATTR3]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147(i32 [[TMP76]], i32 [[TMP0]], ptr [[VLA]], i32 [[TMP78]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK19: omp_offload.cont33:
-// CHECK19-NEXT: [[TMP104:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP104]], ptr [[N_CASTED34]], align 4
-// CHECK19-NEXT: [[TMP105:%.*]] = load i32, ptr [[N_CASTED34]], align 4
-// CHECK19-NEXT: [[TMP106:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK19-NEXT: [[TMP107:%.*]] = sext i32 [[TMP106]] to i64
+// CHECK19-NEXT: [[TMP116:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP116]], ptr [[N_CASTED34]], align 4
+// CHECK19-NEXT: [[TMP117:%.*]] = load i32, ptr [[N_CASTED34]], align 4
+// CHECK19-NEXT: [[TMP118:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK19-NEXT: [[TMP119:%.*]] = sext i32 [[TMP118]] to i64
// CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES38]], ptr align 4 @.offload_sizes.12, i32 24, i1 false)
-// CHECK19-NEXT: [[TMP108:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP105]], ptr [[TMP108]], align 4
-// CHECK19-NEXT: [[TMP109:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP105]], ptr [[TMP109]], align 4
-// CHECK19-NEXT: [[TMP110:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP110]], align 4
-// CHECK19-NEXT: [[TMP111:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP111]], align 4
-// CHECK19-NEXT: [[TMP112:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP112]], align 4
-// CHECK19-NEXT: [[TMP113:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 1
-// CHECK19-NEXT: store ptr null, ptr [[TMP113]], align 4
-// CHECK19-NEXT: [[TMP114:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP114]], align 4
-// CHECK19-NEXT: [[TMP115:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP115]], align 4
-// CHECK19-NEXT: [[TMP116:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 2
-// CHECK19-NEXT: store i64 [[TMP107]], ptr [[TMP116]], align 4
-// CHECK19-NEXT: [[TMP117:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 2
-// CHECK19-NEXT: store ptr null, ptr [[TMP117]], align 4
-// CHECK19-NEXT: [[TMP118:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP119:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP120:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP121:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP121]], ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK19-NEXT: [[TMP122:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK19-NEXT: [[SUB42:%.*]] = sub nsw i32 [[TMP122]], 0
+// CHECK19-NEXT: [[TMP120:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP117]], ptr [[TMP120]], align 4
+// CHECK19-NEXT: [[TMP121:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP117]], ptr [[TMP121]], align 4
+// CHECK19-NEXT: [[TMP122:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP122]], align 4
+// CHECK19-NEXT: [[TMP123:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP123]], align 4
+// CHECK19-NEXT: [[TMP124:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP124]], align 4
+// CHECK19-NEXT: [[TMP125:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 1
+// CHECK19-NEXT: store ptr null, ptr [[TMP125]], align 4
+// CHECK19-NEXT: [[TMP126:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP126]], align 4
+// CHECK19-NEXT: [[TMP127:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP127]], align 4
+// CHECK19-NEXT: [[TMP128:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 2
+// CHECK19-NEXT: store i64 [[TMP119]], ptr [[TMP128]], align 4
+// CHECK19-NEXT: [[TMP129:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 2
+// CHECK19-NEXT: store ptr null, ptr [[TMP129]], align 4
+// CHECK19-NEXT: [[TMP130:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP131:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP132:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP133:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP133]], ptr [[DOTCAPTURE_EXPR_40]], align 4
+// CHECK19-NEXT: [[TMP134:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
+// CHECK19-NEXT: [[SUB42:%.*]] = sub nsw i32 [[TMP134]], 0
// CHECK19-NEXT: [[DIV43:%.*]] = sdiv i32 [[SUB42]], 1
// CHECK19-NEXT: [[SUB44:%.*]] = sub nsw i32 [[DIV43]], 1
// CHECK19-NEXT: store i32 [[SUB44]], ptr [[DOTCAPTURE_EXPR_41]], align 4
-// CHECK19-NEXT: [[TMP123:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_41]], align 4
-// CHECK19-NEXT: [[ADD45:%.*]] = add nsw i32 [[TMP123]], 1
-// CHECK19-NEXT: [[TMP124:%.*]] = zext i32 [[ADD45]] to i64
+// CHECK19-NEXT: [[TMP135:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_41]], align 4
+// CHECK19-NEXT: [[ADD45:%.*]] = add nsw i32 [[TMP135]], 1
+// CHECK19-NEXT: [[TMP136:%.*]] = zext i32 [[ADD45]] to i64
// CHECK19-NEXT: [[KERNEL_ARGS46:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP125]], align 4
-// CHECK19-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 1
-// CHECK19-NEXT: store i32 3, ptr [[TMP126]], align 4
-// CHECK19-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP118]], ptr [[TMP127]], align 4
-// CHECK19-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP119]], ptr [[TMP128]], align 4
-// CHECK19-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 4
-// CHECK19-NEXT: store ptr [[TMP120]], ptr [[TMP129]], align 4
-// CHECK19-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP130]], align 4
-// CHECK19-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP131]], align 4
-// CHECK19-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP132]], align 4
-// CHECK19-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 8
-// CHECK19-NEXT: store i64 [[TMP124]], ptr [[TMP133]], align 8
-// CHECK19-NEXT: [[TMP134:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.region_id, ptr [[KERNEL_ARGS46]])
-// CHECK19-NEXT: [[TMP135:%.*]] = icmp ne i32 [[TMP134]], 0
-// CHECK19-NEXT: br i1 [[TMP135]], label [[OMP_OFFLOAD_FAILED47:%.*]], label [[OMP_OFFLOAD_CONT48:%.*]]
+// CHECK19-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP137]], align 4
+// CHECK19-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 1
+// CHECK19-NEXT: store i32 3, ptr [[TMP138]], align 4
+// CHECK19-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP130]], ptr [[TMP139]], align 4
+// CHECK19-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP131]], ptr [[TMP140]], align 4
+// CHECK19-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 4
+// CHECK19-NEXT: store ptr [[TMP132]], ptr [[TMP141]], align 4
+// CHECK19-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP142]], align 4
+// CHECK19-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP143]], align 4
+// CHECK19-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP144]], align 4
+// CHECK19-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 8
+// CHECK19-NEXT: store i64 [[TMP136]], ptr [[TMP145]], align 8
+// CHECK19-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP146]], align 8
+// CHECK19-NEXT: [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP147]], align 4
+// CHECK19-NEXT: [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP148]], align 4
+// CHECK19-NEXT: [[TMP149:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP149]], align 4
+// CHECK19-NEXT: [[TMP150:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.region_id, ptr [[KERNEL_ARGS46]])
+// CHECK19-NEXT: [[TMP151:%.*]] = icmp ne i32 [[TMP150]], 0
+// CHECK19-NEXT: br i1 [[TMP151]], label [[OMP_OFFLOAD_FAILED47:%.*]], label [[OMP_OFFLOAD_CONT48:%.*]]
// CHECK19: omp_offload.failed47:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151(i32 [[TMP105]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151(i32 [[TMP117]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT48]]
// CHECK19: omp_offload.cont48:
-// CHECK19-NEXT: [[TMP136:%.*]] = load i32, ptr [[M]], align 4
-// CHECK19-NEXT: store i32 [[TMP136]], ptr [[DOTCAPTURE_EXPR_49]], align 4
-// CHECK19-NEXT: [[TMP137:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP137]], ptr [[N_CASTED50]], align 4
-// CHECK19-NEXT: [[TMP138:%.*]] = load i32, ptr [[N_CASTED50]], align 4
-// CHECK19-NEXT: [[TMP139:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_49]], align 4
-// CHECK19-NEXT: store i32 [[TMP139]], ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4
-// CHECK19-NEXT: [[TMP140:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4
-// CHECK19-NEXT: [[TMP141:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK19-NEXT: [[TMP142:%.*]] = sext i32 [[TMP141]] to i64
+// CHECK19-NEXT: [[TMP152:%.*]] = load i32, ptr [[M]], align 4
+// CHECK19-NEXT: store i32 [[TMP152]], ptr [[DOTCAPTURE_EXPR_49]], align 4
+// CHECK19-NEXT: [[TMP153:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP153]], ptr [[N_CASTED50]], align 4
+// CHECK19-NEXT: [[TMP154:%.*]] = load i32, ptr [[N_CASTED50]], align 4
+// CHECK19-NEXT: [[TMP155:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_49]], align 4
+// CHECK19-NEXT: store i32 [[TMP155]], ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4
+// CHECK19-NEXT: [[TMP156:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4
+// CHECK19-NEXT: [[TMP157:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK19-NEXT: [[TMP158:%.*]] = sext i32 [[TMP157]] to i64
// CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES55]], ptr align 4 @.offload_sizes.16, i32 32, i1 false)
-// CHECK19-NEXT: [[TMP143:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP138]], ptr [[TMP143]], align 4
-// CHECK19-NEXT: [[TMP144:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP138]], ptr [[TMP144]], align 4
-// CHECK19-NEXT: [[TMP145:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP145]], align 4
-// CHECK19-NEXT: [[TMP146:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP146]], align 4
-// CHECK19-NEXT: [[TMP147:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP147]], align 4
-// CHECK19-NEXT: [[TMP148:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 1
-// CHECK19-NEXT: store ptr null, ptr [[TMP148]], align 4
-// CHECK19-NEXT: [[TMP149:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP149]], align 4
-// CHECK19-NEXT: [[TMP150:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP150]], align 4
-// CHECK19-NEXT: [[TMP151:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 2
-// CHECK19-NEXT: store i64 [[TMP142]], ptr [[TMP151]], align 4
-// CHECK19-NEXT: [[TMP152:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 2
-// CHECK19-NEXT: store ptr null, ptr [[TMP152]], align 4
-// CHECK19-NEXT: [[TMP153:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 3
-// CHECK19-NEXT: store i32 [[TMP140]], ptr [[TMP153]], align 4
-// CHECK19-NEXT: [[TMP154:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 3
-// CHECK19-NEXT: store i32 [[TMP140]], ptr [[TMP154]], align 4
-// CHECK19-NEXT: [[TMP155:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 3
-// CHECK19-NEXT: store ptr null, ptr [[TMP155]], align 4
-// CHECK19-NEXT: [[TMP156:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP157:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP158:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP159:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP159]], ptr [[DOTCAPTURE_EXPR_57]], align 4
-// CHECK19-NEXT: [[TMP160:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_57]], align 4
-// CHECK19-NEXT: [[SUB59:%.*]] = sub nsw i32 [[TMP160]], 0
+// CHECK19-NEXT: [[TMP159:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP154]], ptr [[TMP159]], align 4
+// CHECK19-NEXT: [[TMP160:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP154]], ptr [[TMP160]], align 4
+// CHECK19-NEXT: [[TMP161:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP161]], align 4
+// CHECK19-NEXT: [[TMP162:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP162]], align 4
+// CHECK19-NEXT: [[TMP163:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP163]], align 4
+// CHECK19-NEXT: [[TMP164:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 1
+// CHECK19-NEXT: store ptr null, ptr [[TMP164]], align 4
+// CHECK19-NEXT: [[TMP165:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP165]], align 4
+// CHECK19-NEXT: [[TMP166:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP166]], align 4
+// CHECK19-NEXT: [[TMP167:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 2
+// CHECK19-NEXT: store i64 [[TMP158]], ptr [[TMP167]], align 4
+// CHECK19-NEXT: [[TMP168:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 2
+// CHECK19-NEXT: store ptr null, ptr [[TMP168]], align 4
+// CHECK19-NEXT: [[TMP169:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 3
+// CHECK19-NEXT: store i32 [[TMP156]], ptr [[TMP169]], align 4
+// CHECK19-NEXT: [[TMP170:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 3
+// CHECK19-NEXT: store i32 [[TMP156]], ptr [[TMP170]], align 4
+// CHECK19-NEXT: [[TMP171:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 3
+// CHECK19-NEXT: store ptr null, ptr [[TMP171]], align 4
+// CHECK19-NEXT: [[TMP172:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP173:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP174:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP175:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP175]], ptr [[DOTCAPTURE_EXPR_57]], align 4
+// CHECK19-NEXT: [[TMP176:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_57]], align 4
+// CHECK19-NEXT: [[SUB59:%.*]] = sub nsw i32 [[TMP176]], 0
// CHECK19-NEXT: [[DIV60:%.*]] = sdiv i32 [[SUB59]], 1
// CHECK19-NEXT: [[SUB61:%.*]] = sub nsw i32 [[DIV60]], 1
// CHECK19-NEXT: store i32 [[SUB61]], ptr [[DOTCAPTURE_EXPR_58]], align 4
-// CHECK19-NEXT: [[TMP161:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_58]], align 4
-// CHECK19-NEXT: [[ADD62:%.*]] = add nsw i32 [[TMP161]], 1
-// CHECK19-NEXT: [[TMP162:%.*]] = zext i32 [[ADD62]] to i64
+// CHECK19-NEXT: [[TMP177:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_58]], align 4
+// CHECK19-NEXT: [[ADD62:%.*]] = add nsw i32 [[TMP177]], 1
+// CHECK19-NEXT: [[TMP178:%.*]] = zext i32 [[ADD62]] to i64
// CHECK19-NEXT: [[KERNEL_ARGS63:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP163:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP163]], align 4
-// CHECK19-NEXT: [[TMP164:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 1
-// CHECK19-NEXT: store i32 4, ptr [[TMP164]], align 4
-// CHECK19-NEXT: [[TMP165:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP156]], ptr [[TMP165]], align 4
-// CHECK19-NEXT: [[TMP166:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP157]], ptr [[TMP166]], align 4
-// CHECK19-NEXT: [[TMP167:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 4
-// CHECK19-NEXT: store ptr [[TMP158]], ptr [[TMP167]], align 4
-// CHECK19-NEXT: [[TMP168:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP168]], align 4
-// CHECK19-NEXT: [[TMP169:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP169]], align 4
-// CHECK19-NEXT: [[TMP170:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP170]], align 4
-// CHECK19-NEXT: [[TMP171:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 8
-// CHECK19-NEXT: store i64 [[TMP162]], ptr [[TMP171]], align 8
-// CHECK19-NEXT: [[TMP172:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.region_id, ptr [[KERNEL_ARGS63]])
-// CHECK19-NEXT: [[TMP173:%.*]] = icmp ne i32 [[TMP172]], 0
-// CHECK19-NEXT: br i1 [[TMP173]], label [[OMP_OFFLOAD_FAILED64:%.*]], label [[OMP_OFFLOAD_CONT65:%.*]]
+// CHECK19-NEXT: [[TMP179:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP179]], align 4
+// CHECK19-NEXT: [[TMP180:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 1
+// CHECK19-NEXT: store i32 4, ptr [[TMP180]], align 4
+// CHECK19-NEXT: [[TMP181:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP172]], ptr [[TMP181]], align 4
+// CHECK19-NEXT: [[TMP182:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP173]], ptr [[TMP182]], align 4
+// CHECK19-NEXT: [[TMP183:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 4
+// CHECK19-NEXT: store ptr [[TMP174]], ptr [[TMP183]], align 4
+// CHECK19-NEXT: [[TMP184:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP184]], align 4
+// CHECK19-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP185]], align 4
+// CHECK19-NEXT: [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP186]], align 4
+// CHECK19-NEXT: [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 8
+// CHECK19-NEXT: store i64 [[TMP178]], ptr [[TMP187]], align 8
+// CHECK19-NEXT: [[TMP188:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP188]], align 8
+// CHECK19-NEXT: [[TMP189:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP189]], align 4
+// CHECK19-NEXT: [[TMP190:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP190]], align 4
+// CHECK19-NEXT: [[TMP191:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP191]], align 4
+// CHECK19-NEXT: [[TMP192:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.region_id, ptr [[KERNEL_ARGS63]])
+// CHECK19-NEXT: [[TMP193:%.*]] = icmp ne i32 [[TMP192]], 0
+// CHECK19-NEXT: br i1 [[TMP193]], label [[OMP_OFFLOAD_FAILED64:%.*]], label [[OMP_OFFLOAD_CONT65:%.*]]
// CHECK19: omp_offload.failed64:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155(i32 [[TMP138]], i32 [[TMP0]], ptr [[VLA]], i32 [[TMP140]]) #[[ATTR3]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155(i32 [[TMP154]], i32 [[TMP0]], ptr [[VLA]], i32 [[TMP156]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT65]]
// CHECK19: omp_offload.cont65:
-// CHECK19-NEXT: [[TMP174:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK19-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP174]])
+// CHECK19-NEXT: [[TMP194:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK19-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP194]])
// CHECK19-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK19-NEXT: [[TMP175:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK19-NEXT: call void @llvm.stackrestore(ptr [[TMP175]])
-// CHECK19-NEXT: [[TMP176:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK19-NEXT: ret i32 [[TMP176]]
+// CHECK19-NEXT: [[TMP195:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK19-NEXT: call void @llvm.stackrestore(ptr [[TMP195]])
+// CHECK19-NEXT: [[TMP196:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK19-NEXT: ret i32 [[TMP196]]
//
//
// CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139
@@ -14725,7 +15165,7 @@ int main (int argc, char **argv) {
// CHECK19-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK19-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK19-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK19-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK19-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK19-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -14742,169 +15182,209 @@ int main (int argc, char **argv) {
// CHECK19-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK19-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK19-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK19-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.region_id, ptr [[KERNEL_ARGS]])
-// CHECK19-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK19-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK19-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK19-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK19-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK19-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK19-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.region_id, ptr [[KERNEL_ARGS]])
+// CHECK19-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK19-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK19: omp_offload.failed:
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112(ptr [[A]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK19: omp_offload.cont:
-// CHECK19-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP16]], align 4
-// CHECK19-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP17]], align 4
-// CHECK19-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK19-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP20]], align 4
+// CHECK19-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP21]], align 4
+// CHECK19-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK19-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK19-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK19-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK19-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK19-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK19-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK19-NEXT: store ptr @.offload_sizes.24, ptr [[TMP25]], align 4
-// CHECK19-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP26]], align 4
-// CHECK19-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK19-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK19-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK19-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK19-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK19-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK19-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK19-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK19-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK19-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK19-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK19-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK19-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK19-NEXT: store ptr @.offload_sizes.24, ptr [[TMP29]], align 4
+// CHECK19-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP30]], align 4
+// CHECK19-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK19-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK19-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK19-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK19-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK19-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK19-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK19-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK19-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK19-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK19-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK19: omp_offload.failed6:
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116(ptr [[A]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK19: omp_offload.cont7:
-// CHECK19-NEXT: [[TMP32:%.*]] = load i32, ptr [[M]], align 4
-// CHECK19-NEXT: store i32 [[TMP32]], ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK19-NEXT: [[TMP33:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK19-NEXT: store i32 [[TMP33]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK19-NEXT: [[TMP34:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK19-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP35]], align 4
-// CHECK19-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP36]], align 4
-// CHECK19-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP37]], align 4
-// CHECK19-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP34]], ptr [[TMP38]], align 4
-// CHECK19-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP34]], ptr [[TMP39]], align 4
-// CHECK19-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1
-// CHECK19-NEXT: store ptr null, ptr [[TMP40]], align 4
-// CHECK19-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP40:%.*]] = load i32, ptr [[M]], align 4
+// CHECK19-NEXT: store i32 [[TMP40]], ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK19-NEXT: [[TMP41:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK19-NEXT: store i32 [[TMP41]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK19-NEXT: [[TMP42:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK19-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP43]], align 4
+// CHECK19-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP44]], align 4
+// CHECK19-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP45]], align 4
+// CHECK19-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP42]], ptr [[TMP46]], align 4
+// CHECK19-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP42]], ptr [[TMP47]], align 4
+// CHECK19-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1
+// CHECK19-NEXT: store ptr null, ptr [[TMP48]], align 4
+// CHECK19-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP43]], align 4
-// CHECK19-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK19-NEXT: store i32 2, ptr [[TMP44]], align 4
-// CHECK19-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP41]], ptr [[TMP45]], align 4
-// CHECK19-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP42]], ptr [[TMP46]], align 4
-// CHECK19-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK19-NEXT: store ptr @.offload_sizes.28, ptr [[TMP47]], align 4
-// CHECK19-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP48]], align 4
-// CHECK19-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP49]], align 4
-// CHECK19-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP50]], align 4
-// CHECK19-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK19-NEXT: store i64 10, ptr [[TMP51]], align 8
-// CHECK19-NEXT: [[TMP52:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK19-NEXT: [[TMP53:%.*]] = icmp ne i32 [[TMP52]], 0
-// CHECK19-NEXT: br i1 [[TMP53]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK19-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP51]], align 4
+// CHECK19-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK19-NEXT: store i32 2, ptr [[TMP52]], align 4
+// CHECK19-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 4
+// CHECK19-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP50]], ptr [[TMP54]], align 4
+// CHECK19-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK19-NEXT: store ptr @.offload_sizes.28, ptr [[TMP55]], align 4
+// CHECK19-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP56]], align 4
+// CHECK19-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP57]], align 4
+// CHECK19-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP58]], align 4
+// CHECK19-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK19-NEXT: store i64 10, ptr [[TMP59]], align 8
+// CHECK19-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP60]], align 8
+// CHECK19-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4
+// CHECK19-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP62]], align 4
+// CHECK19-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP63]], align 4
+// CHECK19-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK19-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
+// CHECK19-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK19: omp_offload.failed13:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120(ptr [[A]], i32 [[TMP34]]) #[[ATTR3]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120(ptr [[A]], i32 [[TMP42]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT14]]
// CHECK19: omp_offload.cont14:
-// CHECK19-NEXT: [[TMP54:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP54]], align 4
-// CHECK19-NEXT: [[TMP55:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP55]], align 4
-// CHECK19-NEXT: [[TMP56:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP56]], align 4
-// CHECK19-NEXT: [[TMP57:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP58:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP66]], align 4
+// CHECK19-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP67]], align 4
+// CHECK19-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP68]], align 4
+// CHECK19-NEXT: [[TMP69:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP70:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS19:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP59]], align 4
-// CHECK19-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
-// CHECK19-NEXT: store i32 1, ptr [[TMP60]], align 4
-// CHECK19-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP57]], ptr [[TMP61]], align 4
-// CHECK19-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP58]], ptr [[TMP62]], align 4
-// CHECK19-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
-// CHECK19-NEXT: store ptr @.offload_sizes.32, ptr [[TMP63]], align 4
-// CHECK19-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP64]], align 4
-// CHECK19-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP65]], align 4
-// CHECK19-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP66]], align 4
-// CHECK19-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
-// CHECK19-NEXT: store i64 10, ptr [[TMP67]], align 8
-// CHECK19-NEXT: [[TMP68:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.region_id, ptr [[KERNEL_ARGS19]])
-// CHECK19-NEXT: [[TMP69:%.*]] = icmp ne i32 [[TMP68]], 0
-// CHECK19-NEXT: br i1 [[TMP69]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
+// CHECK19-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP71]], align 4
+// CHECK19-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
+// CHECK19-NEXT: store i32 1, ptr [[TMP72]], align 4
+// CHECK19-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP69]], ptr [[TMP73]], align 4
+// CHECK19-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP70]], ptr [[TMP74]], align 4
+// CHECK19-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
+// CHECK19-NEXT: store ptr @.offload_sizes.32, ptr [[TMP75]], align 4
+// CHECK19-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP76]], align 4
+// CHECK19-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP77]], align 4
+// CHECK19-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP78]], align 4
+// CHECK19-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
+// CHECK19-NEXT: store i64 10, ptr [[TMP79]], align 8
+// CHECK19-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP80]], align 8
+// CHECK19-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP81]], align 4
+// CHECK19-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP82]], align 4
+// CHECK19-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP83]], align 4
+// CHECK19-NEXT: [[TMP84:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.region_id, ptr [[KERNEL_ARGS19]])
+// CHECK19-NEXT: [[TMP85:%.*]] = icmp ne i32 [[TMP84]], 0
+// CHECK19-NEXT: br i1 [[TMP85]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
// CHECK19: omp_offload.failed20:
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124(ptr [[A]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT21]]
// CHECK19: omp_offload.cont21:
-// CHECK19-NEXT: [[TMP70:%.*]] = load i32, ptr [[M]], align 4
-// CHECK19-NEXT: store i32 [[TMP70]], ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK19-NEXT: [[TMP71:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
-// CHECK19-NEXT: store i32 [[TMP71]], ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4
-// CHECK19-NEXT: [[TMP72:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4
-// CHECK19-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP73]], align 4
-// CHECK19-NEXT: [[TMP74:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP74]], align 4
-// CHECK19-NEXT: [[TMP75:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP75]], align 4
-// CHECK19-NEXT: [[TMP76:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP72]], ptr [[TMP76]], align 4
-// CHECK19-NEXT: [[TMP77:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP72]], ptr [[TMP77]], align 4
-// CHECK19-NEXT: [[TMP78:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 1
-// CHECK19-NEXT: store ptr null, ptr [[TMP78]], align 4
-// CHECK19-NEXT: [[TMP79:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP80:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP86:%.*]] = load i32, ptr [[M]], align 4
+// CHECK19-NEXT: store i32 [[TMP86]], ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK19-NEXT: [[TMP87:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
+// CHECK19-NEXT: store i32 [[TMP87]], ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4
+// CHECK19-NEXT: [[TMP88:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4
+// CHECK19-NEXT: [[TMP89:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP89]], align 4
+// CHECK19-NEXT: [[TMP90:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP90]], align 4
+// CHECK19-NEXT: [[TMP91:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP91]], align 4
+// CHECK19-NEXT: [[TMP92:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP88]], ptr [[TMP92]], align 4
+// CHECK19-NEXT: [[TMP93:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP88]], ptr [[TMP93]], align 4
+// CHECK19-NEXT: [[TMP94:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 1
+// CHECK19-NEXT: store ptr null, ptr [[TMP94]], align 4
+// CHECK19-NEXT: [[TMP95:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP96:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS28:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP81]], align 4
-// CHECK19-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
-// CHECK19-NEXT: store i32 2, ptr [[TMP82]], align 4
-// CHECK19-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP79]], ptr [[TMP83]], align 4
-// CHECK19-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP80]], ptr [[TMP84]], align 4
-// CHECK19-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
-// CHECK19-NEXT: store ptr @.offload_sizes.36, ptr [[TMP85]], align 4
-// CHECK19-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP86]], align 4
-// CHECK19-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP87]], align 4
-// CHECK19-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP88]], align 4
-// CHECK19-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
-// CHECK19-NEXT: store i64 10, ptr [[TMP89]], align 8
-// CHECK19-NEXT: [[TMP90:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.region_id, ptr [[KERNEL_ARGS28]])
-// CHECK19-NEXT: [[TMP91:%.*]] = icmp ne i32 [[TMP90]], 0
-// CHECK19-NEXT: br i1 [[TMP91]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
+// CHECK19-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP97]], align 4
+// CHECK19-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
+// CHECK19-NEXT: store i32 2, ptr [[TMP98]], align 4
+// CHECK19-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP95]], ptr [[TMP99]], align 4
+// CHECK19-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP96]], ptr [[TMP100]], align 4
+// CHECK19-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
+// CHECK19-NEXT: store ptr @.offload_sizes.36, ptr [[TMP101]], align 4
+// CHECK19-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP102]], align 4
+// CHECK19-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP103]], align 4
+// CHECK19-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP104]], align 4
+// CHECK19-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
+// CHECK19-NEXT: store i64 10, ptr [[TMP105]], align 8
+// CHECK19-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP106]], align 8
+// CHECK19-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP107]], align 4
+// CHECK19-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP108]], align 4
+// CHECK19-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP109]], align 4
+// CHECK19-NEXT: [[TMP110:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.region_id, ptr [[KERNEL_ARGS28]])
+// CHECK19-NEXT: [[TMP111:%.*]] = icmp ne i32 [[TMP110]], 0
+// CHECK19-NEXT: br i1 [[TMP111]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
// CHECK19: omp_offload.failed29:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128(ptr [[A]], i32 [[TMP72]]) #[[ATTR3]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128(ptr [[A]], i32 [[TMP88]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT30]]
// CHECK19: omp_offload.cont30:
// CHECK19-NEXT: ret i32 0
diff --git a/clang/test/OpenMP/target_teams_distribute_private_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_private_codegen.cpp
index 5f96b1a5892d7..9127aec58b9f4 100644
--- a/clang/test/OpenMP/target_teams_distribute_private_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_private_codegen.cpp
@@ -240,7 +240,7 @@ int main() {
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -257,9 +257,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l91.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l91.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l91() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -401,7 +409,7 @@ int main() {
// CHECK1-NEXT: store ptr undef, ptr [[_TMP1]], align 8
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -418,27 +426,35 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK1-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK1: arraydestroy.body:
-// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK1-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK1-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK1: arraydestroy.done2:
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK1-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK1-NEXT: ret i32 [[TMP12]]
+// CHECK1-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK1-NEXT: ret i32 [[TMP16]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN1SIiEC1Ev
@@ -759,7 +775,7 @@ int main() {
// CHECK3-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -776,9 +792,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l91.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK3-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l91.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK3-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l91() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -918,7 +942,7 @@ int main() {
// CHECK3-NEXT: store ptr undef, ptr [[_TMP1]], align 4
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -935,27 +959,35 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK3-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK3-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK3-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK3-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK3: arraydestroy.body:
-// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK3-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK3-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK3: arraydestroy.done2:
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK3-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK3-NEXT: ret i32 [[TMP12]]
+// CHECK3-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK3-NEXT: ret i32 [[TMP16]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN1SIiEC1Ev
diff --git a/clang/test/OpenMP/target_teams_distribute_reduction_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_reduction_codegen.cpp
index 99efdc8e7d6fb..1634e5b6f920b 100644
--- a/clang/test/OpenMP/target_teams_distribute_reduction_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_reduction_codegen.cpp
@@ -97,7 +97,7 @@ int main() {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -114,9 +114,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l60.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l60.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l60(ptr @_ZZ4mainE5sivar) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -266,7 +274,7 @@ int main() {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -283,9 +291,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32(ptr [[T_VAR]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -439,7 +455,7 @@ int main() {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -456,9 +472,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l60.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l60.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l60(ptr @_ZZ4mainE5sivar) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -608,7 +632,7 @@ int main() {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -625,9 +649,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32(ptr [[T_VAR]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/target_teams_distribute_simd_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_simd_codegen.cpp
index 09b8540a6edde..47fd29ebacd75 100644
--- a/clang/test/OpenMP/target_teams_distribute_simd_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_simd_codegen.cpp
@@ -422,7 +422,7 @@ int bar(int n){
// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP52]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP52]], align 4
// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP53]], align 4
// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -439,175 +439,199 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP59]], align 8
// CHECK1-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 10, ptr [[TMP60]], align 8
-// CHECK1-NEXT: [[TMP61:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l111.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP62:%.*]] = icmp ne i32 [[TMP61]], 0
-// CHECK1-NEXT: br i1 [[TMP62]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP61]], align 8
+// CHECK1-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP62]], align 4
+// CHECK1-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP63]], align 4
+// CHECK1-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP64]], align 4
+// CHECK1-NEXT: [[TMP65:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l111.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP66:%.*]] = icmp ne i32 [[TMP65]], 0
+// CHECK1-NEXT: br i1 [[TMP66]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l111(i64 [[TMP46]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
-// CHECK1-NEXT: [[TMP63:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: store i32 [[TMP63]], ptr [[A_CASTED8]], align 4
-// CHECK1-NEXT: [[TMP64:%.*]] = load i64, ptr [[A_CASTED8]], align 8
-// CHECK1-NEXT: [[TMP65:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK1-NEXT: store i16 [[TMP65]], ptr [[AA_CASTED9]], align 2
-// CHECK1-NEXT: [[TMP66:%.*]] = load i64, ptr [[AA_CASTED9]], align 8
-// CHECK1-NEXT: [[TMP67:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP67]], 10
+// CHECK1-NEXT: [[TMP67:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: store i32 [[TMP67]], ptr [[A_CASTED8]], align 4
+// CHECK1-NEXT: [[TMP68:%.*]] = load i64, ptr [[A_CASTED8]], align 8
+// CHECK1-NEXT: [[TMP69:%.*]] = load i16, ptr [[AA]], align 2
+// CHECK1-NEXT: store i16 [[TMP69]], ptr [[AA_CASTED9]], align 2
+// CHECK1-NEXT: [[TMP70:%.*]] = load i64, ptr [[AA_CASTED9]], align 8
+// CHECK1-NEXT: [[TMP71:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP71]], 10
// CHECK1-NEXT: br i1 [[CMP]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK1: omp_if.then:
-// CHECK1-NEXT: [[TMP68:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP64]], ptr [[TMP68]], align 8
-// CHECK1-NEXT: [[TMP69:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP64]], ptr [[TMP69]], align 8
-// CHECK1-NEXT: [[TMP70:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP70]], align 8
-// CHECK1-NEXT: [[TMP71:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP66]], ptr [[TMP71]], align 8
-// CHECK1-NEXT: [[TMP72:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP66]], ptr [[TMP72]], align 8
-// CHECK1-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP73]], align 8
-// CHECK1-NEXT: [[TMP74:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP75:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP72:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP68]], ptr [[TMP72]], align 8
+// CHECK1-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP68]], ptr [[TMP73]], align 8
+// CHECK1-NEXT: [[TMP74:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP74]], align 8
+// CHECK1-NEXT: [[TMP75:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP70]], ptr [[TMP75]], align 8
+// CHECK1-NEXT: [[TMP76:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP70]], ptr [[TMP76]], align 8
+// CHECK1-NEXT: [[TMP77:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP77]], align 8
+// CHECK1-NEXT: [[TMP78:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP79:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS14:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP76]], align 4
-// CHECK1-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
-// CHECK1-NEXT: store i32 2, ptr [[TMP77]], align 4
-// CHECK1-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP74]], ptr [[TMP78]], align 8
-// CHECK1-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP75]], ptr [[TMP79]], align 8
-// CHECK1-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.7, ptr [[TMP80]], align 8
-// CHECK1-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP81]], align 8
-// CHECK1-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP82]], align 8
-// CHECK1-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP83]], align 8
-// CHECK1-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
-// CHECK1-NEXT: store i64 10, ptr [[TMP84]], align 8
-// CHECK1-NEXT: [[TMP85:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118.region_id, ptr [[KERNEL_ARGS14]])
-// CHECK1-NEXT: [[TMP86:%.*]] = icmp ne i32 [[TMP85]], 0
-// CHECK1-NEXT: br i1 [[TMP86]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
+// CHECK1-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP80]], align 4
+// CHECK1-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
+// CHECK1-NEXT: store i32 2, ptr [[TMP81]], align 4
+// CHECK1-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP78]], ptr [[TMP82]], align 8
+// CHECK1-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP79]], ptr [[TMP83]], align 8
+// CHECK1-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.7, ptr [[TMP84]], align 8
+// CHECK1-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP85]], align 8
+// CHECK1-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP86]], align 8
+// CHECK1-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP87]], align 8
+// CHECK1-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
+// CHECK1-NEXT: store i64 10, ptr [[TMP88]], align 8
+// CHECK1-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP89]], align 8
+// CHECK1-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP90]], align 4
+// CHECK1-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP91]], align 4
+// CHECK1-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP92]], align 4
+// CHECK1-NEXT: [[TMP93:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118.region_id, ptr [[KERNEL_ARGS14]])
+// CHECK1-NEXT: [[TMP94:%.*]] = icmp ne i32 [[TMP93]], 0
+// CHECK1-NEXT: br i1 [[TMP94]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
// CHECK1: omp_offload.failed15:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118(i64 [[TMP64]], i64 [[TMP66]]) #[[ATTR4]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118(i64 [[TMP68]], i64 [[TMP70]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT16]]
// CHECK1: omp_offload.cont16:
// CHECK1-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK1: omp_if.else:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118(i64 [[TMP64]], i64 [[TMP66]]) #[[ATTR4]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118(i64 [[TMP68]], i64 [[TMP70]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
-// CHECK1-NEXT: [[TMP87:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: store i32 [[TMP87]], ptr [[A_CASTED17]], align 4
-// CHECK1-NEXT: [[TMP88:%.*]] = load i64, ptr [[A_CASTED17]], align 8
-// CHECK1-NEXT: [[TMP89:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: [[CMP18:%.*]] = icmp sgt i32 [[TMP89]], 20
+// CHECK1-NEXT: [[TMP95:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: store i32 [[TMP95]], ptr [[A_CASTED17]], align 4
+// CHECK1-NEXT: [[TMP96:%.*]] = load i64, ptr [[A_CASTED17]], align 8
+// CHECK1-NEXT: [[TMP97:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: [[CMP18:%.*]] = icmp sgt i32 [[TMP97]], 20
// CHECK1-NEXT: br i1 [[CMP18]], label [[OMP_IF_THEN19:%.*]], label [[OMP_IF_ELSE27:%.*]]
// CHECK1: omp_if.then19:
-// CHECK1-NEXT: [[TMP90:%.*]] = mul nuw i64 [[TMP2]], 4
-// CHECK1-NEXT: [[TMP91:%.*]] = mul nuw i64 5, [[TMP5]]
-// CHECK1-NEXT: [[TMP92:%.*]] = mul nuw i64 [[TMP91]], 8
+// CHECK1-NEXT: [[TMP98:%.*]] = mul nuw i64 [[TMP2]], 4
+// CHECK1-NEXT: [[TMP99:%.*]] = mul nuw i64 5, [[TMP5]]
+// CHECK1-NEXT: [[TMP100:%.*]] = mul nuw i64 [[TMP99]], 8
// CHECK1-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES]], ptr align 8 @.offload_sizes.10, i64 72, i1 false)
-// CHECK1-NEXT: [[TMP93:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP88]], ptr [[TMP93]], align 8
-// CHECK1-NEXT: [[TMP94:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP88]], ptr [[TMP94]], align 8
-// CHECK1-NEXT: [[TMP95:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP95]], align 8
-// CHECK1-NEXT: [[TMP96:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
-// CHECK1-NEXT: store ptr [[B]], ptr [[TMP96]], align 8
-// CHECK1-NEXT: [[TMP97:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
-// CHECK1-NEXT: store ptr [[B]], ptr [[TMP97]], align 8
-// CHECK1-NEXT: [[TMP98:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP98]], align 8
-// CHECK1-NEXT: [[TMP99:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
-// CHECK1-NEXT: store i64 [[TMP2]], ptr [[TMP99]], align 8
-// CHECK1-NEXT: [[TMP100:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
-// CHECK1-NEXT: store i64 [[TMP2]], ptr [[TMP100]], align 8
-// CHECK1-NEXT: [[TMP101:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP101]], align 8
-// CHECK1-NEXT: [[TMP102:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[VLA]], ptr [[TMP102]], align 8
-// CHECK1-NEXT: [[TMP103:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[VLA]], ptr [[TMP103]], align 8
-// CHECK1-NEXT: [[TMP104:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
-// CHECK1-NEXT: store i64 [[TMP90]], ptr [[TMP104]], align 8
-// CHECK1-NEXT: [[TMP105:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP105]], align 8
-// CHECK1-NEXT: [[TMP106:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 4
-// CHECK1-NEXT: store ptr [[C]], ptr [[TMP106]], align 8
-// CHECK1-NEXT: [[TMP107:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 4
-// CHECK1-NEXT: store ptr [[C]], ptr [[TMP107]], align 8
-// CHECK1-NEXT: [[TMP108:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP108]], align 8
-// CHECK1-NEXT: [[TMP109:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 5
-// CHECK1-NEXT: store i64 5, ptr [[TMP109]], align 8
-// CHECK1-NEXT: [[TMP110:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 5
-// CHECK1-NEXT: store i64 5, ptr [[TMP110]], align 8
-// CHECK1-NEXT: [[TMP111:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP111]], align 8
-// CHECK1-NEXT: [[TMP112:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 6
-// CHECK1-NEXT: store i64 [[TMP5]], ptr [[TMP112]], align 8
-// CHECK1-NEXT: [[TMP113:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 6
-// CHECK1-NEXT: store i64 [[TMP5]], ptr [[TMP113]], align 8
-// CHECK1-NEXT: [[TMP114:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP114]], align 8
-// CHECK1-NEXT: [[TMP115:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 7
-// CHECK1-NEXT: store ptr [[VLA1]], ptr [[TMP115]], align 8
-// CHECK1-NEXT: [[TMP116:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 7
-// CHECK1-NEXT: store ptr [[VLA1]], ptr [[TMP116]], align 8
-// CHECK1-NEXT: [[TMP117:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
-// CHECK1-NEXT: store i64 [[TMP92]], ptr [[TMP117]], align 8
-// CHECK1-NEXT: [[TMP118:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP118]], align 8
-// CHECK1-NEXT: [[TMP119:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 8
-// CHECK1-NEXT: store ptr [[D]], ptr [[TMP119]], align 8
-// CHECK1-NEXT: [[TMP120:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 8
-// CHECK1-NEXT: store ptr [[D]], ptr [[TMP120]], align 8
-// CHECK1-NEXT: [[TMP121:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 8
-// CHECK1-NEXT: store ptr null, ptr [[TMP121]], align 8
-// CHECK1-NEXT: [[TMP122:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP123:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP124:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP101:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP96]], ptr [[TMP101]], align 8
+// CHECK1-NEXT: [[TMP102:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP96]], ptr [[TMP102]], align 8
+// CHECK1-NEXT: [[TMP103:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP103]], align 8
+// CHECK1-NEXT: [[TMP104:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
+// CHECK1-NEXT: store ptr [[B]], ptr [[TMP104]], align 8
+// CHECK1-NEXT: [[TMP105:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
+// CHECK1-NEXT: store ptr [[B]], ptr [[TMP105]], align 8
+// CHECK1-NEXT: [[TMP106:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP106]], align 8
+// CHECK1-NEXT: [[TMP107:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
+// CHECK1-NEXT: store i64 [[TMP2]], ptr [[TMP107]], align 8
+// CHECK1-NEXT: [[TMP108:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
+// CHECK1-NEXT: store i64 [[TMP2]], ptr [[TMP108]], align 8
+// CHECK1-NEXT: [[TMP109:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 2
+// CHECK1-NEXT: store ptr null, ptr [[TMP109]], align 8
+// CHECK1-NEXT: [[TMP110:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[VLA]], ptr [[TMP110]], align 8
+// CHECK1-NEXT: [[TMP111:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[VLA]], ptr [[TMP111]], align 8
+// CHECK1-NEXT: [[TMP112:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
+// CHECK1-NEXT: store i64 [[TMP98]], ptr [[TMP112]], align 8
+// CHECK1-NEXT: [[TMP113:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 3
+// CHECK1-NEXT: store ptr null, ptr [[TMP113]], align 8
+// CHECK1-NEXT: [[TMP114:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 4
+// CHECK1-NEXT: store ptr [[C]], ptr [[TMP114]], align 8
+// CHECK1-NEXT: [[TMP115:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 4
+// CHECK1-NEXT: store ptr [[C]], ptr [[TMP115]], align 8
+// CHECK1-NEXT: [[TMP116:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP116]], align 8
+// CHECK1-NEXT: [[TMP117:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 5
+// CHECK1-NEXT: store i64 5, ptr [[TMP117]], align 8
+// CHECK1-NEXT: [[TMP118:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 5
+// CHECK1-NEXT: store i64 5, ptr [[TMP118]], align 8
+// CHECK1-NEXT: [[TMP119:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP119]], align 8
+// CHECK1-NEXT: [[TMP120:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 6
+// CHECK1-NEXT: store i64 [[TMP5]], ptr [[TMP120]], align 8
+// CHECK1-NEXT: [[TMP121:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 6
+// CHECK1-NEXT: store i64 [[TMP5]], ptr [[TMP121]], align 8
+// CHECK1-NEXT: [[TMP122:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP122]], align 8
+// CHECK1-NEXT: [[TMP123:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 7
+// CHECK1-NEXT: store ptr [[VLA1]], ptr [[TMP123]], align 8
+// CHECK1-NEXT: [[TMP124:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 7
+// CHECK1-NEXT: store ptr [[VLA1]], ptr [[TMP124]], align 8
+// CHECK1-NEXT: [[TMP125:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
+// CHECK1-NEXT: store i64 [[TMP100]], ptr [[TMP125]], align 8
+// CHECK1-NEXT: [[TMP126:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP126]], align 8
+// CHECK1-NEXT: [[TMP127:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 8
+// CHECK1-NEXT: store ptr [[D]], ptr [[TMP127]], align 8
+// CHECK1-NEXT: [[TMP128:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 8
+// CHECK1-NEXT: store ptr [[D]], ptr [[TMP128]], align 8
+// CHECK1-NEXT: [[TMP129:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 8
+// CHECK1-NEXT: store ptr null, ptr [[TMP129]], align 8
+// CHECK1-NEXT: [[TMP130:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP131:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP132:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS24:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP125]], align 4
-// CHECK1-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 1
-// CHECK1-NEXT: store i32 9, ptr [[TMP126]], align 4
-// CHECK1-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP122]], ptr [[TMP127]], align 8
-// CHECK1-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP123]], ptr [[TMP128]], align 8
-// CHECK1-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 4
-// CHECK1-NEXT: store ptr [[TMP124]], ptr [[TMP129]], align 8
-// CHECK1-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.11, ptr [[TMP130]], align 8
-// CHECK1-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP131]], align 8
-// CHECK1-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP132]], align 8
-// CHECK1-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 8
-// CHECK1-NEXT: store i64 10, ptr [[TMP133]], align 8
-// CHECK1-NEXT: [[TMP134:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142.region_id, ptr [[KERNEL_ARGS24]])
-// CHECK1-NEXT: [[TMP135:%.*]] = icmp ne i32 [[TMP134]], 0
-// CHECK1-NEXT: br i1 [[TMP135]], label [[OMP_OFFLOAD_FAILED25:%.*]], label [[OMP_OFFLOAD_CONT26:%.*]]
+// CHECK1-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP133]], align 4
+// CHECK1-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 1
+// CHECK1-NEXT: store i32 9, ptr [[TMP134]], align 4
+// CHECK1-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP130]], ptr [[TMP135]], align 8
+// CHECK1-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP131]], ptr [[TMP136]], align 8
+// CHECK1-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 4
+// CHECK1-NEXT: store ptr [[TMP132]], ptr [[TMP137]], align 8
+// CHECK1-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.11, ptr [[TMP138]], align 8
+// CHECK1-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP139]], align 8
+// CHECK1-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP140]], align 8
+// CHECK1-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 8
+// CHECK1-NEXT: store i64 10, ptr [[TMP141]], align 8
+// CHECK1-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP142]], align 8
+// CHECK1-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP143]], align 4
+// CHECK1-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP144]], align 4
+// CHECK1-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP145]], align 4
+// CHECK1-NEXT: [[TMP146:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142.region_id, ptr [[KERNEL_ARGS24]])
+// CHECK1-NEXT: [[TMP147:%.*]] = icmp ne i32 [[TMP146]], 0
+// CHECK1-NEXT: br i1 [[TMP147]], label [[OMP_OFFLOAD_FAILED25:%.*]], label [[OMP_OFFLOAD_CONT26:%.*]]
// CHECK1: omp_offload.failed25:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142(i64 [[TMP88]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]]) #[[ATTR4]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142(i64 [[TMP96]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT26]]
// CHECK1: omp_offload.cont26:
// CHECK1-NEXT: br label [[OMP_IF_END28:%.*]]
// CHECK1: omp_if.else27:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142(i64 [[TMP88]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]]) #[[ATTR4]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142(i64 [[TMP96]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_IF_END28]]
// CHECK1: omp_if.end28:
-// CHECK1-NEXT: [[TMP136:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: [[TMP137:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK1-NEXT: call void @llvm.stackrestore(ptr [[TMP137]])
-// CHECK1-NEXT: ret i32 [[TMP136]]
+// CHECK1-NEXT: [[TMP148:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: [[TMP149:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK1-NEXT: call void @llvm.stackrestore(ptr [[TMP149]])
+// CHECK1-NEXT: ret i32 [[TMP148]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l97
@@ -778,37 +802,46 @@ int bar(int n){
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT_ANON:%.*]], ptr [[TMP9]], i32 0, i32 1
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[TMP9]], i32 0, i32 2
// CHECK1-NEXT: [[TMP18:%.*]] = load i32, ptr [[TMP16]], align 4
-// CHECK1-NEXT: store i32 1, ptr [[KERNEL_ARGS_I]], align 4, !noalias !26
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 1
-// CHECK1-NEXT: store i32 3, ptr [[TMP19]], align 4, !noalias !26
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP13]], ptr [[TMP20]], align 8, !noalias !26
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP14]], ptr [[TMP21]], align 8, !noalias !26
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 4
-// CHECK1-NEXT: store ptr [[TMP15]], ptr [[TMP22]], align 8, !noalias !26
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP23]], align 8, !noalias !26
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP24]], align 8, !noalias !26
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP19:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP18]], 0
+// CHECK1-NEXT: store i32 2, ptr [[KERNEL_ARGS_I]], align 4, !noalias !26
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 1
+// CHECK1-NEXT: store i32 3, ptr [[TMP20]], align 4, !noalias !26
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP13]], ptr [[TMP21]], align 8, !noalias !26
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP14]], ptr [[TMP22]], align 8, !noalias !26
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 4
+// CHECK1-NEXT: store ptr [[TMP15]], ptr [[TMP23]], align 8, !noalias !26
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP24]], align 8, !noalias !26
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 6
// CHECK1-NEXT: store ptr null, ptr [[TMP25]], align 8, !noalias !26
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 8
-// CHECK1-NEXT: store i64 10, ptr [[TMP26]], align 8, !noalias !26
-// CHECK1-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel_nowait(ptr @[[GLOB2]], i64 -1, i32 [[TMP18]], i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l97.region_id, ptr [[KERNEL_ARGS_I]], i32 0, ptr null, i32 0, ptr null)
-// CHECK1-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK1-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__1_EXIT:%.*]]
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP26]], align 8, !noalias !26
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 8
+// CHECK1-NEXT: store i64 10, ptr [[TMP27]], align 8, !noalias !26
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 9
+// CHECK1-NEXT: store i64 1, ptr [[TMP28]], align 8, !noalias !26
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [[TMP19]], ptr [[TMP29]], align 4, !noalias !26
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP30]], align 4, !noalias !26
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP31]], align 4, !noalias !26
+// CHECK1-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 [[TMP18]], i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l97.region_id, ptr [[KERNEL_ARGS_I]])
+// CHECK1-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
+// CHECK1-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__1_EXIT:%.*]]
// CHECK1: omp_offload.failed.i:
-// CHECK1-NEXT: [[TMP29:%.*]] = load i16, ptr [[TMP12]], align 2
-// CHECK1-NEXT: store i16 [[TMP29]], ptr [[AA_CASTED_I]], align 2, !noalias !26
-// CHECK1-NEXT: [[TMP30:%.*]] = load i64, ptr [[AA_CASTED_I]], align 8, !noalias !26
-// CHECK1-NEXT: [[TMP31:%.*]] = load i32, ptr [[TMP16]], align 4
-// CHECK1-NEXT: store i32 [[TMP31]], ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 4, !noalias !26
-// CHECK1-NEXT: [[TMP32:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 8, !noalias !26
-// CHECK1-NEXT: [[TMP33:%.*]] = load i32, ptr [[TMP17]], align 4
-// CHECK1-NEXT: store i32 [[TMP33]], ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 4, !noalias !26
-// CHECK1-NEXT: [[TMP34:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 8, !noalias !26
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l97(i64 [[TMP30]], i64 [[TMP32]], i64 [[TMP34]]) #[[ATTR4]]
+// CHECK1-NEXT: [[TMP34:%.*]] = load i16, ptr [[TMP12]], align 2
+// CHECK1-NEXT: store i16 [[TMP34]], ptr [[AA_CASTED_I]], align 2, !noalias !26
+// CHECK1-NEXT: [[TMP35:%.*]] = load i64, ptr [[AA_CASTED_I]], align 8, !noalias !26
+// CHECK1-NEXT: [[TMP36:%.*]] = load i32, ptr [[TMP16]], align 4
+// CHECK1-NEXT: store i32 [[TMP36]], ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 4, !noalias !26
+// CHECK1-NEXT: [[TMP37:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 8, !noalias !26
+// CHECK1-NEXT: [[TMP38:%.*]] = load i32, ptr [[TMP17]], align 4
+// CHECK1-NEXT: store i32 [[TMP38]], ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 4, !noalias !26
+// CHECK1-NEXT: [[TMP39:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 8, !noalias !26
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l97(i64 [[TMP35]], i64 [[TMP37]], i64 [[TMP39]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[DOTOMP_OUTLINED__1_EXIT]]
// CHECK1: .omp_outlined..1.exit:
// CHECK1-NEXT: ret i32 0
@@ -1359,7 +1392,7 @@ int bar(int n){
// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [5 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP29]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP29]], align 4
// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 5, ptr [[TMP30]], align 4
// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1376,9 +1409,17 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP36]], align 8
// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 10, ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l215.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK1-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP38]], align 8
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP39]], align 4
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP40]], align 4
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP41]], align 4
+// CHECK1-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l215.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
+// CHECK1-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l215(ptr [[THIS1]], i64 [[TMP6]], i64 2, i64 [[TMP2]], ptr [[VLA]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1388,15 +1429,15 @@ int bar(int n){
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l215(ptr [[THIS1]], i64 [[TMP6]], i64 2, i64 [[TMP2]], ptr [[VLA]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
-// CHECK1-NEXT: [[TMP40:%.*]] = mul nsw i64 1, [[TMP2]]
-// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i64 [[TMP40]]
+// CHECK1-NEXT: [[TMP44:%.*]] = mul nsw i64 1, [[TMP2]]
+// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i64 [[TMP44]]
// CHECK1-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i16, ptr [[ARRAYIDX]], i64 1
-// CHECK1-NEXT: [[TMP41:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
-// CHECK1-NEXT: [[CONV:%.*]] = sext i16 [[TMP41]] to i32
-// CHECK1-NEXT: [[TMP42:%.*]] = load i32, ptr [[B]], align 4
-// CHECK1-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP42]]
-// CHECK1-NEXT: [[TMP43:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK1-NEXT: call void @llvm.stackrestore(ptr [[TMP43]])
+// CHECK1-NEXT: [[TMP45:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
+// CHECK1-NEXT: [[CONV:%.*]] = sext i16 [[TMP45]] to i32
+// CHECK1-NEXT: [[TMP46:%.*]] = load i32, ptr [[B]], align 4
+// CHECK1-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP46]]
+// CHECK1-NEXT: [[TMP47:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK1-NEXT: call void @llvm.stackrestore(ptr [[TMP47]])
// CHECK1-NEXT: ret i32 [[ADD3]]
//
//
@@ -1488,7 +1529,7 @@ int bar(int n){
// CHECK1-NEXT: [[TMP31:%.*]] = zext i32 [[ADD5]] to i64
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP32]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP32]], align 4
// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 5, ptr [[TMP33]], align 4
// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1505,9 +1546,17 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP39]], align 8
// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 [[TMP31]], ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l197.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
-// CHECK1-NEXT: br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP42]], align 4
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP43]], align 4
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP44]], align 4
+// CHECK1-NEXT: [[TMP45:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l197.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP46:%.*]] = icmp ne i32 [[TMP45]], 0
+// CHECK1-NEXT: br i1 [[TMP46]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l197(i64 [[TMP1]], i64 [[TMP3]], i64 [[TMP5]], i64 [[TMP7]], ptr [[B]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1517,8 +1566,8 @@ int bar(int n){
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l197(i64 [[TMP1]], i64 [[TMP3]], i64 [[TMP5]], i64 [[TMP7]], ptr [[B]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
-// CHECK1-NEXT: [[TMP43:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: ret i32 [[TMP43]]
+// CHECK1-NEXT: [[TMP47:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: ret i32 [[TMP47]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_Z9ftemplateIiET_i
@@ -1569,7 +1618,7 @@ int bar(int n){
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP16]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP16]], align 4
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 3, ptr [[TMP17]], align 4
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1586,9 +1635,17 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP23]], align 8
// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 10, ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l180.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK1-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP25]], align 8
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP28]], align 4
+// CHECK1-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l180.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK1-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l180(i64 [[TMP1]], i64 [[TMP3]], ptr [[B]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1598,8 +1655,8 @@ int bar(int n){
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l180(i64 [[TMP1]], i64 [[TMP3]], ptr [[B]]) #[[ATTR4]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
-// CHECK1-NEXT: [[TMP27:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: ret i32 [[TMP27]]
+// CHECK1-NEXT: [[TMP31:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: ret i32 [[TMP31]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l215
@@ -2134,7 +2191,7 @@ int bar(int n){
// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP50]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP50]], align 4
// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP51]], align 4
// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2151,177 +2208,201 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP57]], align 4
// CHECK3-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 10, ptr [[TMP58]], align 8
-// CHECK3-NEXT: [[TMP59:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l111.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP60:%.*]] = icmp ne i32 [[TMP59]], 0
-// CHECK3-NEXT: br i1 [[TMP60]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP59]], align 8
+// CHECK3-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP60]], align 4
+// CHECK3-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP61]], align 4
+// CHECK3-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP62]], align 4
+// CHECK3-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l111.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
+// CHECK3-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l111(i32 [[TMP44]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
-// CHECK3-NEXT: [[TMP61:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: store i32 [[TMP61]], ptr [[A_CASTED8]], align 4
-// CHECK3-NEXT: [[TMP62:%.*]] = load i32, ptr [[A_CASTED8]], align 4
-// CHECK3-NEXT: [[TMP63:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK3-NEXT: store i16 [[TMP63]], ptr [[AA_CASTED9]], align 2
-// CHECK3-NEXT: [[TMP64:%.*]] = load i32, ptr [[AA_CASTED9]], align 4
-// CHECK3-NEXT: [[TMP65:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP65]], 10
+// CHECK3-NEXT: [[TMP65:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: store i32 [[TMP65]], ptr [[A_CASTED8]], align 4
+// CHECK3-NEXT: [[TMP66:%.*]] = load i32, ptr [[A_CASTED8]], align 4
+// CHECK3-NEXT: [[TMP67:%.*]] = load i16, ptr [[AA]], align 2
+// CHECK3-NEXT: store i16 [[TMP67]], ptr [[AA_CASTED9]], align 2
+// CHECK3-NEXT: [[TMP68:%.*]] = load i32, ptr [[AA_CASTED9]], align 4
+// CHECK3-NEXT: [[TMP69:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP69]], 10
// CHECK3-NEXT: br i1 [[CMP]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK3: omp_if.then:
-// CHECK3-NEXT: [[TMP66:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP62]], ptr [[TMP66]], align 4
-// CHECK3-NEXT: [[TMP67:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP62]], ptr [[TMP67]], align 4
-// CHECK3-NEXT: [[TMP68:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP68]], align 4
-// CHECK3-NEXT: [[TMP69:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP64]], ptr [[TMP69]], align 4
-// CHECK3-NEXT: [[TMP70:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP64]], ptr [[TMP70]], align 4
-// CHECK3-NEXT: [[TMP71:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i32 0, i32 1
-// CHECK3-NEXT: store ptr null, ptr [[TMP71]], align 4
-// CHECK3-NEXT: [[TMP72:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP70:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP66]], ptr [[TMP70]], align 4
+// CHECK3-NEXT: [[TMP71:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP66]], ptr [[TMP71]], align 4
+// CHECK3-NEXT: [[TMP72:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP72]], align 4
+// CHECK3-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP68]], ptr [[TMP73]], align 4
+// CHECK3-NEXT: [[TMP74:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP68]], ptr [[TMP74]], align 4
+// CHECK3-NEXT: [[TMP75:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i32 0, i32 1
+// CHECK3-NEXT: store ptr null, ptr [[TMP75]], align 4
+// CHECK3-NEXT: [[TMP76:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP77:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS14:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP74]], align 4
-// CHECK3-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
-// CHECK3-NEXT: store i32 2, ptr [[TMP75]], align 4
-// CHECK3-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP72]], ptr [[TMP76]], align 4
-// CHECK3-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP73]], ptr [[TMP77]], align 4
-// CHECK3-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.7, ptr [[TMP78]], align 4
-// CHECK3-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP79]], align 4
-// CHECK3-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP80]], align 4
-// CHECK3-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP81]], align 4
-// CHECK3-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
-// CHECK3-NEXT: store i64 10, ptr [[TMP82]], align 8
-// CHECK3-NEXT: [[TMP83:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118.region_id, ptr [[KERNEL_ARGS14]])
-// CHECK3-NEXT: [[TMP84:%.*]] = icmp ne i32 [[TMP83]], 0
-// CHECK3-NEXT: br i1 [[TMP84]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
+// CHECK3-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP78]], align 4
+// CHECK3-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
+// CHECK3-NEXT: store i32 2, ptr [[TMP79]], align 4
+// CHECK3-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP76]], ptr [[TMP80]], align 4
+// CHECK3-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP77]], ptr [[TMP81]], align 4
+// CHECK3-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.7, ptr [[TMP82]], align 4
+// CHECK3-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP83]], align 4
+// CHECK3-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP84]], align 4
+// CHECK3-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP85]], align 4
+// CHECK3-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
+// CHECK3-NEXT: store i64 10, ptr [[TMP86]], align 8
+// CHECK3-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP87]], align 8
+// CHECK3-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP88]], align 4
+// CHECK3-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP89]], align 4
+// CHECK3-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP90]], align 4
+// CHECK3-NEXT: [[TMP91:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118.region_id, ptr [[KERNEL_ARGS14]])
+// CHECK3-NEXT: [[TMP92:%.*]] = icmp ne i32 [[TMP91]], 0
+// CHECK3-NEXT: br i1 [[TMP92]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
// CHECK3: omp_offload.failed15:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118(i32 [[TMP62]], i32 [[TMP64]]) #[[ATTR4]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118(i32 [[TMP66]], i32 [[TMP68]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT16]]
// CHECK3: omp_offload.cont16:
// CHECK3-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK3: omp_if.else:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118(i32 [[TMP62]], i32 [[TMP64]]) #[[ATTR4]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118(i32 [[TMP66]], i32 [[TMP68]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
-// CHECK3-NEXT: [[TMP85:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: store i32 [[TMP85]], ptr [[A_CASTED17]], align 4
-// CHECK3-NEXT: [[TMP86:%.*]] = load i32, ptr [[A_CASTED17]], align 4
-// CHECK3-NEXT: [[TMP87:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: [[CMP18:%.*]] = icmp sgt i32 [[TMP87]], 20
+// CHECK3-NEXT: [[TMP93:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: store i32 [[TMP93]], ptr [[A_CASTED17]], align 4
+// CHECK3-NEXT: [[TMP94:%.*]] = load i32, ptr [[A_CASTED17]], align 4
+// CHECK3-NEXT: [[TMP95:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: [[CMP18:%.*]] = icmp sgt i32 [[TMP95]], 20
// CHECK3-NEXT: br i1 [[CMP18]], label [[OMP_IF_THEN19:%.*]], label [[OMP_IF_ELSE27:%.*]]
// CHECK3: omp_if.then19:
-// CHECK3-NEXT: [[TMP88:%.*]] = mul nuw i32 [[TMP1]], 4
-// CHECK3-NEXT: [[TMP89:%.*]] = sext i32 [[TMP88]] to i64
-// CHECK3-NEXT: [[TMP90:%.*]] = mul nuw i32 5, [[TMP3]]
-// CHECK3-NEXT: [[TMP91:%.*]] = mul nuw i32 [[TMP90]], 8
-// CHECK3-NEXT: [[TMP92:%.*]] = sext i32 [[TMP91]] to i64
+// CHECK3-NEXT: [[TMP96:%.*]] = mul nuw i32 [[TMP1]], 4
+// CHECK3-NEXT: [[TMP97:%.*]] = sext i32 [[TMP96]] to i64
+// CHECK3-NEXT: [[TMP98:%.*]] = mul nuw i32 5, [[TMP3]]
+// CHECK3-NEXT: [[TMP99:%.*]] = mul nuw i32 [[TMP98]], 8
+// CHECK3-NEXT: [[TMP100:%.*]] = sext i32 [[TMP99]] to i64
// CHECK3-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES]], ptr align 4 @.offload_sizes.10, i32 72, i1 false)
-// CHECK3-NEXT: [[TMP93:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP86]], ptr [[TMP93]], align 4
-// CHECK3-NEXT: [[TMP94:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP86]], ptr [[TMP94]], align 4
-// CHECK3-NEXT: [[TMP95:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP95]], align 4
-// CHECK3-NEXT: [[TMP96:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
-// CHECK3-NEXT: store ptr [[B]], ptr [[TMP96]], align 4
-// CHECK3-NEXT: [[TMP97:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
-// CHECK3-NEXT: store ptr [[B]], ptr [[TMP97]], align 4
-// CHECK3-NEXT: [[TMP98:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 1
-// CHECK3-NEXT: store ptr null, ptr [[TMP98]], align 4
-// CHECK3-NEXT: [[TMP99:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
-// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP99]], align 4
-// CHECK3-NEXT: [[TMP100:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
-// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP100]], align 4
-// CHECK3-NEXT: [[TMP101:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 2
-// CHECK3-NEXT: store ptr null, ptr [[TMP101]], align 4
-// CHECK3-NEXT: [[TMP102:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[VLA]], ptr [[TMP102]], align 4
-// CHECK3-NEXT: [[TMP103:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[VLA]], ptr [[TMP103]], align 4
-// CHECK3-NEXT: [[TMP104:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
-// CHECK3-NEXT: store i64 [[TMP89]], ptr [[TMP104]], align 4
-// CHECK3-NEXT: [[TMP105:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 3
-// CHECK3-NEXT: store ptr null, ptr [[TMP105]], align 4
-// CHECK3-NEXT: [[TMP106:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 4
-// CHECK3-NEXT: store ptr [[C]], ptr [[TMP106]], align 4
-// CHECK3-NEXT: [[TMP107:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 4
-// CHECK3-NEXT: store ptr [[C]], ptr [[TMP107]], align 4
-// CHECK3-NEXT: [[TMP108:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 4
-// CHECK3-NEXT: store ptr null, ptr [[TMP108]], align 4
-// CHECK3-NEXT: [[TMP109:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 5
-// CHECK3-NEXT: store i32 5, ptr [[TMP109]], align 4
-// CHECK3-NEXT: [[TMP110:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 5
-// CHECK3-NEXT: store i32 5, ptr [[TMP110]], align 4
-// CHECK3-NEXT: [[TMP111:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 5
-// CHECK3-NEXT: store ptr null, ptr [[TMP111]], align 4
-// CHECK3-NEXT: [[TMP112:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 6
-// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP112]], align 4
-// CHECK3-NEXT: [[TMP113:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 6
-// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP113]], align 4
-// CHECK3-NEXT: [[TMP114:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP114]], align 4
-// CHECK3-NEXT: [[TMP115:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 7
-// CHECK3-NEXT: store ptr [[VLA1]], ptr [[TMP115]], align 4
-// CHECK3-NEXT: [[TMP116:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 7
-// CHECK3-NEXT: store ptr [[VLA1]], ptr [[TMP116]], align 4
-// CHECK3-NEXT: [[TMP117:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
-// CHECK3-NEXT: store i64 [[TMP92]], ptr [[TMP117]], align 4
-// CHECK3-NEXT: [[TMP118:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP118]], align 4
-// CHECK3-NEXT: [[TMP119:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 8
-// CHECK3-NEXT: store ptr [[D]], ptr [[TMP119]], align 4
-// CHECK3-NEXT: [[TMP120:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 8
-// CHECK3-NEXT: store ptr [[D]], ptr [[TMP120]], align 4
-// CHECK3-NEXT: [[TMP121:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 8
-// CHECK3-NEXT: store ptr null, ptr [[TMP121]], align 4
-// CHECK3-NEXT: [[TMP122:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP123:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP124:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP101:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP94]], ptr [[TMP101]], align 4
+// CHECK3-NEXT: [[TMP102:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP94]], ptr [[TMP102]], align 4
+// CHECK3-NEXT: [[TMP103:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP103]], align 4
+// CHECK3-NEXT: [[TMP104:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
+// CHECK3-NEXT: store ptr [[B]], ptr [[TMP104]], align 4
+// CHECK3-NEXT: [[TMP105:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
+// CHECK3-NEXT: store ptr [[B]], ptr [[TMP105]], align 4
+// CHECK3-NEXT: [[TMP106:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 1
+// CHECK3-NEXT: store ptr null, ptr [[TMP106]], align 4
+// CHECK3-NEXT: [[TMP107:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
+// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP107]], align 4
+// CHECK3-NEXT: [[TMP108:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
+// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP108]], align 4
+// CHECK3-NEXT: [[TMP109:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 2
+// CHECK3-NEXT: store ptr null, ptr [[TMP109]], align 4
+// CHECK3-NEXT: [[TMP110:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[VLA]], ptr [[TMP110]], align 4
+// CHECK3-NEXT: [[TMP111:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[VLA]], ptr [[TMP111]], align 4
+// CHECK3-NEXT: [[TMP112:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
+// CHECK3-NEXT: store i64 [[TMP97]], ptr [[TMP112]], align 4
+// CHECK3-NEXT: [[TMP113:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 3
+// CHECK3-NEXT: store ptr null, ptr [[TMP113]], align 4
+// CHECK3-NEXT: [[TMP114:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 4
+// CHECK3-NEXT: store ptr [[C]], ptr [[TMP114]], align 4
+// CHECK3-NEXT: [[TMP115:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 4
+// CHECK3-NEXT: store ptr [[C]], ptr [[TMP115]], align 4
+// CHECK3-NEXT: [[TMP116:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 4
+// CHECK3-NEXT: store ptr null, ptr [[TMP116]], align 4
+// CHECK3-NEXT: [[TMP117:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 5
+// CHECK3-NEXT: store i32 5, ptr [[TMP117]], align 4
+// CHECK3-NEXT: [[TMP118:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 5
+// CHECK3-NEXT: store i32 5, ptr [[TMP118]], align 4
+// CHECK3-NEXT: [[TMP119:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 5
+// CHECK3-NEXT: store ptr null, ptr [[TMP119]], align 4
+// CHECK3-NEXT: [[TMP120:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 6
+// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP120]], align 4
+// CHECK3-NEXT: [[TMP121:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 6
+// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP121]], align 4
+// CHECK3-NEXT: [[TMP122:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP122]], align 4
+// CHECK3-NEXT: [[TMP123:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 7
+// CHECK3-NEXT: store ptr [[VLA1]], ptr [[TMP123]], align 4
+// CHECK3-NEXT: [[TMP124:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 7
+// CHECK3-NEXT: store ptr [[VLA1]], ptr [[TMP124]], align 4
+// CHECK3-NEXT: [[TMP125:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
+// CHECK3-NEXT: store i64 [[TMP100]], ptr [[TMP125]], align 4
+// CHECK3-NEXT: [[TMP126:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP126]], align 4
+// CHECK3-NEXT: [[TMP127:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 8
+// CHECK3-NEXT: store ptr [[D]], ptr [[TMP127]], align 4
+// CHECK3-NEXT: [[TMP128:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 8
+// CHECK3-NEXT: store ptr [[D]], ptr [[TMP128]], align 4
+// CHECK3-NEXT: [[TMP129:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 8
+// CHECK3-NEXT: store ptr null, ptr [[TMP129]], align 4
+// CHECK3-NEXT: [[TMP130:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP131:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP132:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS24:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP125]], align 4
-// CHECK3-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 1
-// CHECK3-NEXT: store i32 9, ptr [[TMP126]], align 4
-// CHECK3-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP122]], ptr [[TMP127]], align 4
-// CHECK3-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP123]], ptr [[TMP128]], align 4
-// CHECK3-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 4
-// CHECK3-NEXT: store ptr [[TMP124]], ptr [[TMP129]], align 4
-// CHECK3-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.11, ptr [[TMP130]], align 4
-// CHECK3-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP131]], align 4
-// CHECK3-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP132]], align 4
-// CHECK3-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 8
-// CHECK3-NEXT: store i64 10, ptr [[TMP133]], align 8
-// CHECK3-NEXT: [[TMP134:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142.region_id, ptr [[KERNEL_ARGS24]])
-// CHECK3-NEXT: [[TMP135:%.*]] = icmp ne i32 [[TMP134]], 0
-// CHECK3-NEXT: br i1 [[TMP135]], label [[OMP_OFFLOAD_FAILED25:%.*]], label [[OMP_OFFLOAD_CONT26:%.*]]
+// CHECK3-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP133]], align 4
+// CHECK3-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 1
+// CHECK3-NEXT: store i32 9, ptr [[TMP134]], align 4
+// CHECK3-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP130]], ptr [[TMP135]], align 4
+// CHECK3-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP131]], ptr [[TMP136]], align 4
+// CHECK3-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 4
+// CHECK3-NEXT: store ptr [[TMP132]], ptr [[TMP137]], align 4
+// CHECK3-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.11, ptr [[TMP138]], align 4
+// CHECK3-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP139]], align 4
+// CHECK3-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP140]], align 4
+// CHECK3-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 8
+// CHECK3-NEXT: store i64 10, ptr [[TMP141]], align 8
+// CHECK3-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP142]], align 8
+// CHECK3-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP143]], align 4
+// CHECK3-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP144]], align 4
+// CHECK3-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP145]], align 4
+// CHECK3-NEXT: [[TMP146:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142.region_id, ptr [[KERNEL_ARGS24]])
+// CHECK3-NEXT: [[TMP147:%.*]] = icmp ne i32 [[TMP146]], 0
+// CHECK3-NEXT: br i1 [[TMP147]], label [[OMP_OFFLOAD_FAILED25:%.*]], label [[OMP_OFFLOAD_CONT26:%.*]]
// CHECK3: omp_offload.failed25:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142(i32 [[TMP86]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]]) #[[ATTR4]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142(i32 [[TMP94]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT26]]
// CHECK3: omp_offload.cont26:
// CHECK3-NEXT: br label [[OMP_IF_END28:%.*]]
// CHECK3: omp_if.else27:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142(i32 [[TMP86]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]]) #[[ATTR4]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142(i32 [[TMP94]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_IF_END28]]
// CHECK3: omp_if.end28:
-// CHECK3-NEXT: [[TMP136:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: [[TMP137:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK3-NEXT: call void @llvm.stackrestore(ptr [[TMP137]])
-// CHECK3-NEXT: ret i32 [[TMP136]]
+// CHECK3-NEXT: [[TMP148:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: [[TMP149:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK3-NEXT: call void @llvm.stackrestore(ptr [[TMP149]])
+// CHECK3-NEXT: ret i32 [[TMP148]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l97
@@ -2492,37 +2573,46 @@ int bar(int n){
// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT_ANON:%.*]], ptr [[TMP9]], i32 0, i32 1
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[TMP9]], i32 0, i32 2
// CHECK3-NEXT: [[TMP18:%.*]] = load i32, ptr [[TMP16]], align 4
-// CHECK3-NEXT: store i32 1, ptr [[KERNEL_ARGS_I]], align 4, !noalias !27
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 1
-// CHECK3-NEXT: store i32 3, ptr [[TMP19]], align 4, !noalias !27
-// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP13]], ptr [[TMP20]], align 4, !noalias !27
-// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP14]], ptr [[TMP21]], align 4, !noalias !27
-// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 4
-// CHECK3-NEXT: store ptr [[TMP15]], ptr [[TMP22]], align 4, !noalias !27
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP23]], align 4, !noalias !27
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP24]], align 4, !noalias !27
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 7
+// CHECK3-NEXT: [[TMP19:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP18]], 0
+// CHECK3-NEXT: store i32 2, ptr [[KERNEL_ARGS_I]], align 4, !noalias !27
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 1
+// CHECK3-NEXT: store i32 3, ptr [[TMP20]], align 4, !noalias !27
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP13]], ptr [[TMP21]], align 4, !noalias !27
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP14]], ptr [[TMP22]], align 4, !noalias !27
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 4
+// CHECK3-NEXT: store ptr [[TMP15]], ptr [[TMP23]], align 4, !noalias !27
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP24]], align 4, !noalias !27
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 6
// CHECK3-NEXT: store ptr null, ptr [[TMP25]], align 4, !noalias !27
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 8
-// CHECK3-NEXT: store i64 10, ptr [[TMP26]], align 8, !noalias !27
-// CHECK3-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel_nowait(ptr @[[GLOB2]], i64 -1, i32 [[TMP18]], i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l97.region_id, ptr [[KERNEL_ARGS_I]], i32 0, ptr null, i32 0, ptr null)
-// CHECK3-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK3-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__1_EXIT:%.*]]
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP26]], align 4, !noalias !27
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 8
+// CHECK3-NEXT: store i64 10, ptr [[TMP27]], align 8, !noalias !27
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 9
+// CHECK3-NEXT: store i64 1, ptr [[TMP28]], align 8, !noalias !27
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [[TMP19]], ptr [[TMP29]], align 4, !noalias !27
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP30]], align 4, !noalias !27
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP31]], align 4, !noalias !27
+// CHECK3-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 [[TMP18]], i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l97.region_id, ptr [[KERNEL_ARGS_I]])
+// CHECK3-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
+// CHECK3-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__1_EXIT:%.*]]
// CHECK3: omp_offload.failed.i:
-// CHECK3-NEXT: [[TMP29:%.*]] = load i16, ptr [[TMP12]], align 2
-// CHECK3-NEXT: store i16 [[TMP29]], ptr [[AA_CASTED_I]], align 2, !noalias !27
-// CHECK3-NEXT: [[TMP30:%.*]] = load i32, ptr [[AA_CASTED_I]], align 4, !noalias !27
-// CHECK3-NEXT: [[TMP31:%.*]] = load i32, ptr [[TMP16]], align 4
-// CHECK3-NEXT: store i32 [[TMP31]], ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 4, !noalias !27
-// CHECK3-NEXT: [[TMP32:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 4, !noalias !27
-// CHECK3-NEXT: [[TMP33:%.*]] = load i32, ptr [[TMP17]], align 4
-// CHECK3-NEXT: store i32 [[TMP33]], ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 4, !noalias !27
-// CHECK3-NEXT: [[TMP34:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 4, !noalias !27
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l97(i32 [[TMP30]], i32 [[TMP32]], i32 [[TMP34]]) #[[ATTR4]]
+// CHECK3-NEXT: [[TMP34:%.*]] = load i16, ptr [[TMP12]], align 2
+// CHECK3-NEXT: store i16 [[TMP34]], ptr [[AA_CASTED_I]], align 2, !noalias !27
+// CHECK3-NEXT: [[TMP35:%.*]] = load i32, ptr [[AA_CASTED_I]], align 4, !noalias !27
+// CHECK3-NEXT: [[TMP36:%.*]] = load i32, ptr [[TMP16]], align 4
+// CHECK3-NEXT: store i32 [[TMP36]], ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 4, !noalias !27
+// CHECK3-NEXT: [[TMP37:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 4, !noalias !27
+// CHECK3-NEXT: [[TMP38:%.*]] = load i32, ptr [[TMP17]], align 4
+// CHECK3-NEXT: store i32 [[TMP38]], ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 4, !noalias !27
+// CHECK3-NEXT: [[TMP39:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 4, !noalias !27
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l97(i32 [[TMP35]], i32 [[TMP37]], i32 [[TMP39]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[DOTOMP_OUTLINED__1_EXIT]]
// CHECK3: .omp_outlined..1.exit:
// CHECK3-NEXT: ret i32 0
@@ -3073,7 +3163,7 @@ int bar(int n){
// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [5 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP29]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP29]], align 4
// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 5, ptr [[TMP30]], align 4
// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3090,9 +3180,17 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP36]], align 4
// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 10, ptr [[TMP37]], align 8
-// CHECK3-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l215.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK3-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP38]], align 8
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP39]], align 4
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l215.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
+// CHECK3-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l215(ptr [[THIS1]], i32 [[TMP5]], i32 2, i32 [[TMP1]], ptr [[VLA]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -3102,15 +3200,15 @@ int bar(int n){
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l215(ptr [[THIS1]], i32 [[TMP5]], i32 2, i32 [[TMP1]], ptr [[VLA]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
-// CHECK3-NEXT: [[TMP40:%.*]] = mul nsw i32 1, [[TMP1]]
-// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i32 [[TMP40]]
+// CHECK3-NEXT: [[TMP44:%.*]] = mul nsw i32 1, [[TMP1]]
+// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i32 [[TMP44]]
// CHECK3-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i16, ptr [[ARRAYIDX]], i32 1
-// CHECK3-NEXT: [[TMP41:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
-// CHECK3-NEXT: [[CONV:%.*]] = sext i16 [[TMP41]] to i32
-// CHECK3-NEXT: [[TMP42:%.*]] = load i32, ptr [[B]], align 4
-// CHECK3-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP42]]
-// CHECK3-NEXT: [[TMP43:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK3-NEXT: call void @llvm.stackrestore(ptr [[TMP43]])
+// CHECK3-NEXT: [[TMP45:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
+// CHECK3-NEXT: [[CONV:%.*]] = sext i16 [[TMP45]] to i32
+// CHECK3-NEXT: [[TMP46:%.*]] = load i32, ptr [[B]], align 4
+// CHECK3-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP46]]
+// CHECK3-NEXT: [[TMP47:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK3-NEXT: call void @llvm.stackrestore(ptr [[TMP47]])
// CHECK3-NEXT: ret i32 [[ADD3]]
//
//
@@ -3202,7 +3300,7 @@ int bar(int n){
// CHECK3-NEXT: [[TMP31:%.*]] = zext i32 [[ADD5]] to i64
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP32]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP32]], align 4
// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 5, ptr [[TMP33]], align 4
// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3219,9 +3317,17 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP39]], align 4
// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 [[TMP31]], ptr [[TMP40]], align 8
-// CHECK3-NEXT: [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l197.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
-// CHECK3-NEXT: br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP41]], align 8
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP42]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP43]], align 4
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP44]], align 4
+// CHECK3-NEXT: [[TMP45:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l197.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP46:%.*]] = icmp ne i32 [[TMP45]], 0
+// CHECK3-NEXT: br i1 [[TMP46]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l197(i32 [[TMP1]], i32 [[TMP3]], i32 [[TMP5]], i32 [[TMP7]], ptr [[B]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -3231,8 +3337,8 @@ int bar(int n){
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l197(i32 [[TMP1]], i32 [[TMP3]], i32 [[TMP5]], i32 [[TMP7]], ptr [[B]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
-// CHECK3-NEXT: [[TMP43:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: ret i32 [[TMP43]]
+// CHECK3-NEXT: [[TMP47:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: ret i32 [[TMP47]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_Z9ftemplateIiET_i
@@ -3283,7 +3389,7 @@ int bar(int n){
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP16]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP16]], align 4
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 3, ptr [[TMP17]], align 4
// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3300,9 +3406,17 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP23]], align 4
// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 10, ptr [[TMP24]], align 8
-// CHECK3-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l180.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK3-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP25]], align 8
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l180.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK3-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l180(i32 [[TMP1]], i32 [[TMP3]], ptr [[B]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -3312,8 +3426,8 @@ int bar(int n){
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l180(i32 [[TMP1]], i32 [[TMP3]], ptr [[B]]) #[[ATTR4]]
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
-// CHECK3-NEXT: [[TMP27:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: ret i32 [[TMP27]]
+// CHECK3-NEXT: [[TMP31:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: ret i32 [[TMP31]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l215
@@ -3850,7 +3964,7 @@ int bar(int n){
// CHECK5-NEXT: [[TMP51:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP52]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP52]], align 4
// CHECK5-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 1, ptr [[TMP53]], align 4
// CHECK5-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3867,175 +3981,199 @@ int bar(int n){
// CHECK5-NEXT: store ptr null, ptr [[TMP59]], align 8
// CHECK5-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 10, ptr [[TMP60]], align 8
-// CHECK5-NEXT: [[TMP61:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l111.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP62:%.*]] = icmp ne i32 [[TMP61]], 0
-// CHECK5-NEXT: br i1 [[TMP62]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP61]], align 8
+// CHECK5-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP62]], align 4
+// CHECK5-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP63]], align 4
+// CHECK5-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP64]], align 4
+// CHECK5-NEXT: [[TMP65:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l111.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP66:%.*]] = icmp ne i32 [[TMP65]], 0
+// CHECK5-NEXT: br i1 [[TMP66]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l111(i64 [[TMP46]]) #[[ATTR4]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK5: omp_offload.cont:
-// CHECK5-NEXT: [[TMP63:%.*]] = load i32, ptr [[A]], align 4
-// CHECK5-NEXT: store i32 [[TMP63]], ptr [[A_CASTED8]], align 4
-// CHECK5-NEXT: [[TMP64:%.*]] = load i64, ptr [[A_CASTED8]], align 8
-// CHECK5-NEXT: [[TMP65:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK5-NEXT: store i16 [[TMP65]], ptr [[AA_CASTED9]], align 2
-// CHECK5-NEXT: [[TMP66:%.*]] = load i64, ptr [[AA_CASTED9]], align 8
-// CHECK5-NEXT: [[TMP67:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK5-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP67]], 10
+// CHECK5-NEXT: [[TMP67:%.*]] = load i32, ptr [[A]], align 4
+// CHECK5-NEXT: store i32 [[TMP67]], ptr [[A_CASTED8]], align 4
+// CHECK5-NEXT: [[TMP68:%.*]] = load i64, ptr [[A_CASTED8]], align 8
+// CHECK5-NEXT: [[TMP69:%.*]] = load i16, ptr [[AA]], align 2
+// CHECK5-NEXT: store i16 [[TMP69]], ptr [[AA_CASTED9]], align 2
+// CHECK5-NEXT: [[TMP70:%.*]] = load i64, ptr [[AA_CASTED9]], align 8
+// CHECK5-NEXT: [[TMP71:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK5-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP71]], 10
// CHECK5-NEXT: br i1 [[CMP]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK5: omp_if.then:
-// CHECK5-NEXT: [[TMP68:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
-// CHECK5-NEXT: store i64 [[TMP64]], ptr [[TMP68]], align 8
-// CHECK5-NEXT: [[TMP69:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
-// CHECK5-NEXT: store i64 [[TMP64]], ptr [[TMP69]], align 8
-// CHECK5-NEXT: [[TMP70:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i64 0, i64 0
-// CHECK5-NEXT: store ptr null, ptr [[TMP70]], align 8
-// CHECK5-NEXT: [[TMP71:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 1
-// CHECK5-NEXT: store i64 [[TMP66]], ptr [[TMP71]], align 8
-// CHECK5-NEXT: [[TMP72:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 1
-// CHECK5-NEXT: store i64 [[TMP66]], ptr [[TMP72]], align 8
-// CHECK5-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i64 0, i64 1
-// CHECK5-NEXT: store ptr null, ptr [[TMP73]], align 8
-// CHECK5-NEXT: [[TMP74:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP75:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP72:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
+// CHECK5-NEXT: store i64 [[TMP68]], ptr [[TMP72]], align 8
+// CHECK5-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
+// CHECK5-NEXT: store i64 [[TMP68]], ptr [[TMP73]], align 8
+// CHECK5-NEXT: [[TMP74:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i64 0, i64 0
+// CHECK5-NEXT: store ptr null, ptr [[TMP74]], align 8
+// CHECK5-NEXT: [[TMP75:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 1
+// CHECK5-NEXT: store i64 [[TMP70]], ptr [[TMP75]], align 8
+// CHECK5-NEXT: [[TMP76:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 1
+// CHECK5-NEXT: store i64 [[TMP70]], ptr [[TMP76]], align 8
+// CHECK5-NEXT: [[TMP77:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i64 0, i64 1
+// CHECK5-NEXT: store ptr null, ptr [[TMP77]], align 8
+// CHECK5-NEXT: [[TMP78:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP79:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS14:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP76]], align 4
-// CHECK5-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
-// CHECK5-NEXT: store i32 2, ptr [[TMP77]], align 4
-// CHECK5-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
-// CHECK5-NEXT: store ptr [[TMP74]], ptr [[TMP78]], align 8
-// CHECK5-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[TMP75]], ptr [[TMP79]], align 8
-// CHECK5-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
-// CHECK5-NEXT: store ptr @.offload_sizes.7, ptr [[TMP80]], align 8
-// CHECK5-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
-// CHECK5-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP81]], align 8
-// CHECK5-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP82]], align 8
-// CHECK5-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
-// CHECK5-NEXT: store ptr null, ptr [[TMP83]], align 8
-// CHECK5-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
-// CHECK5-NEXT: store i64 10, ptr [[TMP84]], align 8
-// CHECK5-NEXT: [[TMP85:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118.region_id, ptr [[KERNEL_ARGS14]])
-// CHECK5-NEXT: [[TMP86:%.*]] = icmp ne i32 [[TMP85]], 0
-// CHECK5-NEXT: br i1 [[TMP86]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
+// CHECK5-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP80]], align 4
+// CHECK5-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
+// CHECK5-NEXT: store i32 2, ptr [[TMP81]], align 4
+// CHECK5-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
+// CHECK5-NEXT: store ptr [[TMP78]], ptr [[TMP82]], align 8
+// CHECK5-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[TMP79]], ptr [[TMP83]], align 8
+// CHECK5-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
+// CHECK5-NEXT: store ptr @.offload_sizes.7, ptr [[TMP84]], align 8
+// CHECK5-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
+// CHECK5-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP85]], align 8
+// CHECK5-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP86]], align 8
+// CHECK5-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP87]], align 8
+// CHECK5-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
+// CHECK5-NEXT: store i64 10, ptr [[TMP88]], align 8
+// CHECK5-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP89]], align 8
+// CHECK5-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP90]], align 4
+// CHECK5-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP91]], align 4
+// CHECK5-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP92]], align 4
+// CHECK5-NEXT: [[TMP93:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118.region_id, ptr [[KERNEL_ARGS14]])
+// CHECK5-NEXT: [[TMP94:%.*]] = icmp ne i32 [[TMP93]], 0
+// CHECK5-NEXT: br i1 [[TMP94]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
// CHECK5: omp_offload.failed15:
-// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118(i64 [[TMP64]], i64 [[TMP66]]) #[[ATTR4]]
+// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118(i64 [[TMP68]], i64 [[TMP70]]) #[[ATTR4]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT16]]
// CHECK5: omp_offload.cont16:
// CHECK5-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK5: omp_if.else:
-// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118(i64 [[TMP64]], i64 [[TMP66]]) #[[ATTR4]]
+// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118(i64 [[TMP68]], i64 [[TMP70]]) #[[ATTR4]]
// CHECK5-NEXT: br label [[OMP_IF_END]]
// CHECK5: omp_if.end:
-// CHECK5-NEXT: [[TMP87:%.*]] = load i32, ptr [[A]], align 4
-// CHECK5-NEXT: store i32 [[TMP87]], ptr [[A_CASTED17]], align 4
-// CHECK5-NEXT: [[TMP88:%.*]] = load i64, ptr [[A_CASTED17]], align 8
-// CHECK5-NEXT: [[TMP89:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK5-NEXT: [[CMP18:%.*]] = icmp sgt i32 [[TMP89]], 20
+// CHECK5-NEXT: [[TMP95:%.*]] = load i32, ptr [[A]], align 4
+// CHECK5-NEXT: store i32 [[TMP95]], ptr [[A_CASTED17]], align 4
+// CHECK5-NEXT: [[TMP96:%.*]] = load i64, ptr [[A_CASTED17]], align 8
+// CHECK5-NEXT: [[TMP97:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK5-NEXT: [[CMP18:%.*]] = icmp sgt i32 [[TMP97]], 20
// CHECK5-NEXT: br i1 [[CMP18]], label [[OMP_IF_THEN19:%.*]], label [[OMP_IF_ELSE27:%.*]]
// CHECK5: omp_if.then19:
-// CHECK5-NEXT: [[TMP90:%.*]] = mul nuw i64 [[TMP2]], 4
-// CHECK5-NEXT: [[TMP91:%.*]] = mul nuw i64 5, [[TMP5]]
-// CHECK5-NEXT: [[TMP92:%.*]] = mul nuw i64 [[TMP91]], 8
+// CHECK5-NEXT: [[TMP98:%.*]] = mul nuw i64 [[TMP2]], 4
+// CHECK5-NEXT: [[TMP99:%.*]] = mul nuw i64 5, [[TMP5]]
+// CHECK5-NEXT: [[TMP100:%.*]] = mul nuw i64 [[TMP99]], 8
// CHECK5-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES]], ptr align 8 @.offload_sizes.10, i64 72, i1 false)
-// CHECK5-NEXT: [[TMP93:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK5-NEXT: store i64 [[TMP88]], ptr [[TMP93]], align 8
-// CHECK5-NEXT: [[TMP94:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK5-NEXT: store i64 [[TMP88]], ptr [[TMP94]], align 8
-// CHECK5-NEXT: [[TMP95:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 0
-// CHECK5-NEXT: store ptr null, ptr [[TMP95]], align 8
-// CHECK5-NEXT: [[TMP96:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
-// CHECK5-NEXT: store ptr [[B]], ptr [[TMP96]], align 8
-// CHECK5-NEXT: [[TMP97:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
-// CHECK5-NEXT: store ptr [[B]], ptr [[TMP97]], align 8
-// CHECK5-NEXT: [[TMP98:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 1
-// CHECK5-NEXT: store ptr null, ptr [[TMP98]], align 8
-// CHECK5-NEXT: [[TMP99:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
-// CHECK5-NEXT: store i64 [[TMP2]], ptr [[TMP99]], align 8
-// CHECK5-NEXT: [[TMP100:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
-// CHECK5-NEXT: store i64 [[TMP2]], ptr [[TMP100]], align 8
-// CHECK5-NEXT: [[TMP101:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 2
-// CHECK5-NEXT: store ptr null, ptr [[TMP101]], align 8
-// CHECK5-NEXT: [[TMP102:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[VLA]], ptr [[TMP102]], align 8
-// CHECK5-NEXT: [[TMP103:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[VLA]], ptr [[TMP103]], align 8
-// CHECK5-NEXT: [[TMP104:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
-// CHECK5-NEXT: store i64 [[TMP90]], ptr [[TMP104]], align 8
-// CHECK5-NEXT: [[TMP105:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 3
-// CHECK5-NEXT: store ptr null, ptr [[TMP105]], align 8
-// CHECK5-NEXT: [[TMP106:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 4
-// CHECK5-NEXT: store ptr [[C]], ptr [[TMP106]], align 8
-// CHECK5-NEXT: [[TMP107:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 4
-// CHECK5-NEXT: store ptr [[C]], ptr [[TMP107]], align 8
-// CHECK5-NEXT: [[TMP108:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 4
-// CHECK5-NEXT: store ptr null, ptr [[TMP108]], align 8
-// CHECK5-NEXT: [[TMP109:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 5
-// CHECK5-NEXT: store i64 5, ptr [[TMP109]], align 8
-// CHECK5-NEXT: [[TMP110:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 5
-// CHECK5-NEXT: store i64 5, ptr [[TMP110]], align 8
-// CHECK5-NEXT: [[TMP111:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 5
-// CHECK5-NEXT: store ptr null, ptr [[TMP111]], align 8
-// CHECK5-NEXT: [[TMP112:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 6
-// CHECK5-NEXT: store i64 [[TMP5]], ptr [[TMP112]], align 8
-// CHECK5-NEXT: [[TMP113:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 6
-// CHECK5-NEXT: store i64 [[TMP5]], ptr [[TMP113]], align 8
-// CHECK5-NEXT: [[TMP114:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP114]], align 8
-// CHECK5-NEXT: [[TMP115:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 7
-// CHECK5-NEXT: store ptr [[VLA1]], ptr [[TMP115]], align 8
-// CHECK5-NEXT: [[TMP116:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 7
-// CHECK5-NEXT: store ptr [[VLA1]], ptr [[TMP116]], align 8
-// CHECK5-NEXT: [[TMP117:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
-// CHECK5-NEXT: store i64 [[TMP92]], ptr [[TMP117]], align 8
-// CHECK5-NEXT: [[TMP118:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 7
-// CHECK5-NEXT: store ptr null, ptr [[TMP118]], align 8
-// CHECK5-NEXT: [[TMP119:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 8
-// CHECK5-NEXT: store ptr [[D]], ptr [[TMP119]], align 8
-// CHECK5-NEXT: [[TMP120:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 8
-// CHECK5-NEXT: store ptr [[D]], ptr [[TMP120]], align 8
-// CHECK5-NEXT: [[TMP121:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 8
-// CHECK5-NEXT: store ptr null, ptr [[TMP121]], align 8
-// CHECK5-NEXT: [[TMP122:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP123:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP124:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP101:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK5-NEXT: store i64 [[TMP96]], ptr [[TMP101]], align 8
+// CHECK5-NEXT: [[TMP102:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK5-NEXT: store i64 [[TMP96]], ptr [[TMP102]], align 8
+// CHECK5-NEXT: [[TMP103:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 0
+// CHECK5-NEXT: store ptr null, ptr [[TMP103]], align 8
+// CHECK5-NEXT: [[TMP104:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
+// CHECK5-NEXT: store ptr [[B]], ptr [[TMP104]], align 8
+// CHECK5-NEXT: [[TMP105:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
+// CHECK5-NEXT: store ptr [[B]], ptr [[TMP105]], align 8
+// CHECK5-NEXT: [[TMP106:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 1
+// CHECK5-NEXT: store ptr null, ptr [[TMP106]], align 8
+// CHECK5-NEXT: [[TMP107:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
+// CHECK5-NEXT: store i64 [[TMP2]], ptr [[TMP107]], align 8
+// CHECK5-NEXT: [[TMP108:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
+// CHECK5-NEXT: store i64 [[TMP2]], ptr [[TMP108]], align 8
+// CHECK5-NEXT: [[TMP109:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 2
+// CHECK5-NEXT: store ptr null, ptr [[TMP109]], align 8
+// CHECK5-NEXT: [[TMP110:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[VLA]], ptr [[TMP110]], align 8
+// CHECK5-NEXT: [[TMP111:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[VLA]], ptr [[TMP111]], align 8
+// CHECK5-NEXT: [[TMP112:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
+// CHECK5-NEXT: store i64 [[TMP98]], ptr [[TMP112]], align 8
+// CHECK5-NEXT: [[TMP113:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 3
+// CHECK5-NEXT: store ptr null, ptr [[TMP113]], align 8
+// CHECK5-NEXT: [[TMP114:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 4
+// CHECK5-NEXT: store ptr [[C]], ptr [[TMP114]], align 8
+// CHECK5-NEXT: [[TMP115:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 4
+// CHECK5-NEXT: store ptr [[C]], ptr [[TMP115]], align 8
+// CHECK5-NEXT: [[TMP116:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 4
+// CHECK5-NEXT: store ptr null, ptr [[TMP116]], align 8
+// CHECK5-NEXT: [[TMP117:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 5
+// CHECK5-NEXT: store i64 5, ptr [[TMP117]], align 8
+// CHECK5-NEXT: [[TMP118:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 5
+// CHECK5-NEXT: store i64 5, ptr [[TMP118]], align 8
+// CHECK5-NEXT: [[TMP119:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 5
+// CHECK5-NEXT: store ptr null, ptr [[TMP119]], align 8
+// CHECK5-NEXT: [[TMP120:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 6
+// CHECK5-NEXT: store i64 [[TMP5]], ptr [[TMP120]], align 8
+// CHECK5-NEXT: [[TMP121:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 6
+// CHECK5-NEXT: store i64 [[TMP5]], ptr [[TMP121]], align 8
+// CHECK5-NEXT: [[TMP122:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP122]], align 8
+// CHECK5-NEXT: [[TMP123:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 7
+// CHECK5-NEXT: store ptr [[VLA1]], ptr [[TMP123]], align 8
+// CHECK5-NEXT: [[TMP124:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 7
+// CHECK5-NEXT: store ptr [[VLA1]], ptr [[TMP124]], align 8
+// CHECK5-NEXT: [[TMP125:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
+// CHECK5-NEXT: store i64 [[TMP100]], ptr [[TMP125]], align 8
+// CHECK5-NEXT: [[TMP126:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP126]], align 8
+// CHECK5-NEXT: [[TMP127:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 8
+// CHECK5-NEXT: store ptr [[D]], ptr [[TMP127]], align 8
+// CHECK5-NEXT: [[TMP128:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 8
+// CHECK5-NEXT: store ptr [[D]], ptr [[TMP128]], align 8
+// CHECK5-NEXT: [[TMP129:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 8
+// CHECK5-NEXT: store ptr null, ptr [[TMP129]], align 8
+// CHECK5-NEXT: [[TMP130:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP131:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP132:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS24:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP125]], align 4
-// CHECK5-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 1
-// CHECK5-NEXT: store i32 9, ptr [[TMP126]], align 4
-// CHECK5-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 2
-// CHECK5-NEXT: store ptr [[TMP122]], ptr [[TMP127]], align 8
-// CHECK5-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[TMP123]], ptr [[TMP128]], align 8
-// CHECK5-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 4
-// CHECK5-NEXT: store ptr [[TMP124]], ptr [[TMP129]], align 8
-// CHECK5-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 5
-// CHECK5-NEXT: store ptr @.offload_maptypes.11, ptr [[TMP130]], align 8
-// CHECK5-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP131]], align 8
-// CHECK5-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 7
-// CHECK5-NEXT: store ptr null, ptr [[TMP132]], align 8
-// CHECK5-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 8
-// CHECK5-NEXT: store i64 10, ptr [[TMP133]], align 8
-// CHECK5-NEXT: [[TMP134:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142.region_id, ptr [[KERNEL_ARGS24]])
-// CHECK5-NEXT: [[TMP135:%.*]] = icmp ne i32 [[TMP134]], 0
-// CHECK5-NEXT: br i1 [[TMP135]], label [[OMP_OFFLOAD_FAILED25:%.*]], label [[OMP_OFFLOAD_CONT26:%.*]]
+// CHECK5-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP133]], align 4
+// CHECK5-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 1
+// CHECK5-NEXT: store i32 9, ptr [[TMP134]], align 4
+// CHECK5-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 2
+// CHECK5-NEXT: store ptr [[TMP130]], ptr [[TMP135]], align 8
+// CHECK5-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[TMP131]], ptr [[TMP136]], align 8
+// CHECK5-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 4
+// CHECK5-NEXT: store ptr [[TMP132]], ptr [[TMP137]], align 8
+// CHECK5-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 5
+// CHECK5-NEXT: store ptr @.offload_maptypes.11, ptr [[TMP138]], align 8
+// CHECK5-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP139]], align 8
+// CHECK5-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP140]], align 8
+// CHECK5-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 8
+// CHECK5-NEXT: store i64 10, ptr [[TMP141]], align 8
+// CHECK5-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP142]], align 8
+// CHECK5-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP143]], align 4
+// CHECK5-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP144]], align 4
+// CHECK5-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP145]], align 4
+// CHECK5-NEXT: [[TMP146:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142.region_id, ptr [[KERNEL_ARGS24]])
+// CHECK5-NEXT: [[TMP147:%.*]] = icmp ne i32 [[TMP146]], 0
+// CHECK5-NEXT: br i1 [[TMP147]], label [[OMP_OFFLOAD_FAILED25:%.*]], label [[OMP_OFFLOAD_CONT26:%.*]]
// CHECK5: omp_offload.failed25:
-// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142(i64 [[TMP88]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]]) #[[ATTR4]]
+// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142(i64 [[TMP96]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]]) #[[ATTR4]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT26]]
// CHECK5: omp_offload.cont26:
// CHECK5-NEXT: br label [[OMP_IF_END28:%.*]]
// CHECK5: omp_if.else27:
-// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142(i64 [[TMP88]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]]) #[[ATTR4]]
+// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142(i64 [[TMP96]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]]) #[[ATTR4]]
// CHECK5-NEXT: br label [[OMP_IF_END28]]
// CHECK5: omp_if.end28:
-// CHECK5-NEXT: [[TMP136:%.*]] = load i32, ptr [[A]], align 4
-// CHECK5-NEXT: [[TMP137:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK5-NEXT: call void @llvm.stackrestore(ptr [[TMP137]])
-// CHECK5-NEXT: ret i32 [[TMP136]]
+// CHECK5-NEXT: [[TMP148:%.*]] = load i32, ptr [[A]], align 4
+// CHECK5-NEXT: [[TMP149:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK5-NEXT: call void @llvm.stackrestore(ptr [[TMP149]])
+// CHECK5-NEXT: ret i32 [[TMP148]]
//
//
// CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l97
@@ -4206,37 +4344,46 @@ int bar(int n){
// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT_ANON:%.*]], ptr [[TMP9]], i32 0, i32 1
// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[TMP9]], i32 0, i32 2
// CHECK5-NEXT: [[TMP18:%.*]] = load i32, ptr [[TMP16]], align 4
-// CHECK5-NEXT: store i32 1, ptr [[KERNEL_ARGS_I]], align 4, !noalias !26
-// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 1
-// CHECK5-NEXT: store i32 3, ptr [[TMP19]], align 4, !noalias !26
-// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 2
-// CHECK5-NEXT: store ptr [[TMP13]], ptr [[TMP20]], align 8, !noalias !26
-// CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[TMP14]], ptr [[TMP21]], align 8, !noalias !26
-// CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 4
-// CHECK5-NEXT: store ptr [[TMP15]], ptr [[TMP22]], align 8, !noalias !26
-// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 5
-// CHECK5-NEXT: store ptr @.offload_maptypes, ptr [[TMP23]], align 8, !noalias !26
-// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP24]], align 8, !noalias !26
-// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 7
+// CHECK5-NEXT: [[TMP19:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP18]], 0
+// CHECK5-NEXT: store i32 2, ptr [[KERNEL_ARGS_I]], align 4, !noalias !26
+// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 1
+// CHECK5-NEXT: store i32 3, ptr [[TMP20]], align 4, !noalias !26
+// CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 2
+// CHECK5-NEXT: store ptr [[TMP13]], ptr [[TMP21]], align 8, !noalias !26
+// CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[TMP14]], ptr [[TMP22]], align 8, !noalias !26
+// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 4
+// CHECK5-NEXT: store ptr [[TMP15]], ptr [[TMP23]], align 8, !noalias !26
+// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 5
+// CHECK5-NEXT: store ptr @.offload_maptypes, ptr [[TMP24]], align 8, !noalias !26
+// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 6
// CHECK5-NEXT: store ptr null, ptr [[TMP25]], align 8, !noalias !26
-// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 8
-// CHECK5-NEXT: store i64 10, ptr [[TMP26]], align 8, !noalias !26
-// CHECK5-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel_nowait(ptr @[[GLOB2]], i64 -1, i32 [[TMP18]], i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l97.region_id, ptr [[KERNEL_ARGS_I]], i32 0, ptr null, i32 0, ptr null)
-// CHECK5-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK5-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__1_EXIT:%.*]]
+// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP26]], align 8, !noalias !26
+// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 8
+// CHECK5-NEXT: store i64 10, ptr [[TMP27]], align 8, !noalias !26
+// CHECK5-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 9
+// CHECK5-NEXT: store i64 1, ptr [[TMP28]], align 8, !noalias !26
+// CHECK5-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] [[TMP19]], ptr [[TMP29]], align 4, !noalias !26
+// CHECK5-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP30]], align 4, !noalias !26
+// CHECK5-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP31]], align 4, !noalias !26
+// CHECK5-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 [[TMP18]], i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l97.region_id, ptr [[KERNEL_ARGS_I]])
+// CHECK5-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
+// CHECK5-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__1_EXIT:%.*]]
// CHECK5: omp_offload.failed.i:
-// CHECK5-NEXT: [[TMP29:%.*]] = load i16, ptr [[TMP12]], align 2
-// CHECK5-NEXT: store i16 [[TMP29]], ptr [[AA_CASTED_I]], align 2, !noalias !26
-// CHECK5-NEXT: [[TMP30:%.*]] = load i64, ptr [[AA_CASTED_I]], align 8, !noalias !26
-// CHECK5-NEXT: [[TMP31:%.*]] = load i32, ptr [[TMP16]], align 4
-// CHECK5-NEXT: store i32 [[TMP31]], ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 4, !noalias !26
-// CHECK5-NEXT: [[TMP32:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 8, !noalias !26
-// CHECK5-NEXT: [[TMP33:%.*]] = load i32, ptr [[TMP17]], align 4
-// CHECK5-NEXT: store i32 [[TMP33]], ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 4, !noalias !26
-// CHECK5-NEXT: [[TMP34:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 8, !noalias !26
-// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l97(i64 [[TMP30]], i64 [[TMP32]], i64 [[TMP34]]) #[[ATTR4]]
+// CHECK5-NEXT: [[TMP34:%.*]] = load i16, ptr [[TMP12]], align 2
+// CHECK5-NEXT: store i16 [[TMP34]], ptr [[AA_CASTED_I]], align 2, !noalias !26
+// CHECK5-NEXT: [[TMP35:%.*]] = load i64, ptr [[AA_CASTED_I]], align 8, !noalias !26
+// CHECK5-NEXT: [[TMP36:%.*]] = load i32, ptr [[TMP16]], align 4
+// CHECK5-NEXT: store i32 [[TMP36]], ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 4, !noalias !26
+// CHECK5-NEXT: [[TMP37:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 8, !noalias !26
+// CHECK5-NEXT: [[TMP38:%.*]] = load i32, ptr [[TMP17]], align 4
+// CHECK5-NEXT: store i32 [[TMP38]], ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 4, !noalias !26
+// CHECK5-NEXT: [[TMP39:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 8, !noalias !26
+// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l97(i64 [[TMP35]], i64 [[TMP37]], i64 [[TMP39]]) #[[ATTR4]]
// CHECK5-NEXT: br label [[DOTOMP_OUTLINED__1_EXIT]]
// CHECK5: .omp_outlined..1.exit:
// CHECK5-NEXT: ret i32 0
@@ -4804,7 +4951,7 @@ int bar(int n){
// CHECK5-NEXT: [[TMP34:%.*]] = getelementptr inbounds [6 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP35]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP35]], align 4
// CHECK5-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 6, ptr [[TMP36]], align 4
// CHECK5-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -4821,9 +4968,17 @@ int bar(int n){
// CHECK5-NEXT: store ptr null, ptr [[TMP42]], align 8
// CHECK5-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 10, ptr [[TMP43]], align 8
-// CHECK5-NEXT: [[TMP44:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l215.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP45:%.*]] = icmp ne i32 [[TMP44]], 0
-// CHECK5-NEXT: br i1 [[TMP45]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP44]], align 8
+// CHECK5-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP45]], align 4
+// CHECK5-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP46]], align 4
+// CHECK5-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP47]], align 4
+// CHECK5-NEXT: [[TMP48:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l215.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP49:%.*]] = icmp ne i32 [[TMP48]], 0
+// CHECK5-NEXT: br i1 [[TMP49]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l215(ptr [[THIS1]], i64 [[TMP7]], i64 2, i64 [[TMP2]], ptr [[VLA]], i64 [[TMP9]]) #[[ATTR4]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -4833,15 +4988,15 @@ int bar(int n){
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l215(ptr [[THIS1]], i64 [[TMP7]], i64 2, i64 [[TMP2]], ptr [[VLA]], i64 [[TMP9]]) #[[ATTR4]]
// CHECK5-NEXT: br label [[OMP_IF_END]]
// CHECK5: omp_if.end:
-// CHECK5-NEXT: [[TMP46:%.*]] = mul nsw i64 1, [[TMP2]]
-// CHECK5-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i64 [[TMP46]]
+// CHECK5-NEXT: [[TMP50:%.*]] = mul nsw i64 1, [[TMP2]]
+// CHECK5-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i64 [[TMP50]]
// CHECK5-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds i16, ptr [[ARRAYIDX]], i64 1
-// CHECK5-NEXT: [[TMP47:%.*]] = load i16, ptr [[ARRAYIDX4]], align 2
-// CHECK5-NEXT: [[CONV:%.*]] = sext i16 [[TMP47]] to i32
-// CHECK5-NEXT: [[TMP48:%.*]] = load i32, ptr [[B]], align 4
-// CHECK5-NEXT: [[ADD5:%.*]] = add nsw i32 [[CONV]], [[TMP48]]
-// CHECK5-NEXT: [[TMP49:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK5-NEXT: call void @llvm.stackrestore(ptr [[TMP49]])
+// CHECK5-NEXT: [[TMP51:%.*]] = load i16, ptr [[ARRAYIDX4]], align 2
+// CHECK5-NEXT: [[CONV:%.*]] = sext i16 [[TMP51]] to i32
+// CHECK5-NEXT: [[TMP52:%.*]] = load i32, ptr [[B]], align 4
+// CHECK5-NEXT: [[ADD5:%.*]] = add nsw i32 [[CONV]], [[TMP52]]
+// CHECK5-NEXT: [[TMP53:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK5-NEXT: call void @llvm.stackrestore(ptr [[TMP53]])
// CHECK5-NEXT: ret i32 [[ADD5]]
//
//
@@ -4933,7 +5088,7 @@ int bar(int n){
// CHECK5-NEXT: [[TMP31:%.*]] = zext i32 [[ADD5]] to i64
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP32]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP32]], align 4
// CHECK5-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 5, ptr [[TMP33]], align 4
// CHECK5-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -4950,9 +5105,17 @@ int bar(int n){
// CHECK5-NEXT: store ptr null, ptr [[TMP39]], align 8
// CHECK5-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 [[TMP31]], ptr [[TMP40]], align 8
-// CHECK5-NEXT: [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l197.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
-// CHECK5-NEXT: br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP41]], align 8
+// CHECK5-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP42]], align 4
+// CHECK5-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP43]], align 4
+// CHECK5-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP44]], align 4
+// CHECK5-NEXT: [[TMP45:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l197.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP46:%.*]] = icmp ne i32 [[TMP45]], 0
+// CHECK5-NEXT: br i1 [[TMP46]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l197(i64 [[TMP1]], i64 [[TMP3]], i64 [[TMP5]], i64 [[TMP7]], ptr [[B]]) #[[ATTR4]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -4962,8 +5125,8 @@ int bar(int n){
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l197(i64 [[TMP1]], i64 [[TMP3]], i64 [[TMP5]], i64 [[TMP7]], ptr [[B]]) #[[ATTR4]]
// CHECK5-NEXT: br label [[OMP_IF_END]]
// CHECK5: omp_if.end:
-// CHECK5-NEXT: [[TMP43:%.*]] = load i32, ptr [[A]], align 4
-// CHECK5-NEXT: ret i32 [[TMP43]]
+// CHECK5-NEXT: [[TMP47:%.*]] = load i32, ptr [[A]], align 4
+// CHECK5-NEXT: ret i32 [[TMP47]]
//
//
// CHECK5-LABEL: define {{[^@]+}}@_Z9ftemplateIiET_i
@@ -5014,7 +5177,7 @@ int bar(int n){
// CHECK5-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP16]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP16]], align 4
// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 3, ptr [[TMP17]], align 4
// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -5031,9 +5194,17 @@ int bar(int n){
// CHECK5-NEXT: store ptr null, ptr [[TMP23]], align 8
// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 10, ptr [[TMP24]], align 8
-// CHECK5-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l180.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK5-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP25]], align 8
+// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP27]], align 4
+// CHECK5-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP28]], align 4
+// CHECK5-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l180.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK5-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l180(i64 [[TMP1]], i64 [[TMP3]], ptr [[B]]) #[[ATTR4]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -5043,8 +5214,8 @@ int bar(int n){
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l180(i64 [[TMP1]], i64 [[TMP3]], ptr [[B]]) #[[ATTR4]]
// CHECK5-NEXT: br label [[OMP_IF_END]]
// CHECK5: omp_if.end:
-// CHECK5-NEXT: [[TMP27:%.*]] = load i32, ptr [[A]], align 4
-// CHECK5-NEXT: ret i32 [[TMP27]]
+// CHECK5-NEXT: [[TMP31:%.*]] = load i32, ptr [[A]], align 4
+// CHECK5-NEXT: ret i32 [[TMP31]]
//
//
// CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l215
@@ -5631,7 +5802,7 @@ int bar(int n){
// CHECK7-NEXT: [[TMP49:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK7-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP50]], align 4
+// CHECK7-NEXT: store i32 2, ptr [[TMP50]], align 4
// CHECK7-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK7-NEXT: store i32 1, ptr [[TMP51]], align 4
// CHECK7-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -5648,177 +5819,201 @@ int bar(int n){
// CHECK7-NEXT: store ptr null, ptr [[TMP57]], align 4
// CHECK7-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK7-NEXT: store i64 10, ptr [[TMP58]], align 8
-// CHECK7-NEXT: [[TMP59:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l111.region_id, ptr [[KERNEL_ARGS]])
-// CHECK7-NEXT: [[TMP60:%.*]] = icmp ne i32 [[TMP59]], 0
-// CHECK7-NEXT: br i1 [[TMP60]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK7-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP59]], align 8
+// CHECK7-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP60]], align 4
+// CHECK7-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP61]], align 4
+// CHECK7-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP62]], align 4
+// CHECK7-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l111.region_id, ptr [[KERNEL_ARGS]])
+// CHECK7-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
+// CHECK7-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK7: omp_offload.failed:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l111(i32 [[TMP44]]) #[[ATTR4]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK7: omp_offload.cont:
-// CHECK7-NEXT: [[TMP61:%.*]] = load i32, ptr [[A]], align 4
-// CHECK7-NEXT: store i32 [[TMP61]], ptr [[A_CASTED8]], align 4
-// CHECK7-NEXT: [[TMP62:%.*]] = load i32, ptr [[A_CASTED8]], align 4
-// CHECK7-NEXT: [[TMP63:%.*]] = load i16, ptr [[AA]], align 2
-// CHECK7-NEXT: store i16 [[TMP63]], ptr [[AA_CASTED9]], align 2
-// CHECK7-NEXT: [[TMP64:%.*]] = load i32, ptr [[AA_CASTED9]], align 4
-// CHECK7-NEXT: [[TMP65:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK7-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP65]], 10
+// CHECK7-NEXT: [[TMP65:%.*]] = load i32, ptr [[A]], align 4
+// CHECK7-NEXT: store i32 [[TMP65]], ptr [[A_CASTED8]], align 4
+// CHECK7-NEXT: [[TMP66:%.*]] = load i32, ptr [[A_CASTED8]], align 4
+// CHECK7-NEXT: [[TMP67:%.*]] = load i16, ptr [[AA]], align 2
+// CHECK7-NEXT: store i16 [[TMP67]], ptr [[AA_CASTED9]], align 2
+// CHECK7-NEXT: [[TMP68:%.*]] = load i32, ptr [[AA_CASTED9]], align 4
+// CHECK7-NEXT: [[TMP69:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK7-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP69]], 10
// CHECK7-NEXT: br i1 [[CMP]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK7: omp_if.then:
-// CHECK7-NEXT: [[TMP66:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
-// CHECK7-NEXT: store i32 [[TMP62]], ptr [[TMP66]], align 4
-// CHECK7-NEXT: [[TMP67:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
-// CHECK7-NEXT: store i32 [[TMP62]], ptr [[TMP67]], align 4
-// CHECK7-NEXT: [[TMP68:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i32 0, i32 0
-// CHECK7-NEXT: store ptr null, ptr [[TMP68]], align 4
-// CHECK7-NEXT: [[TMP69:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 1
-// CHECK7-NEXT: store i32 [[TMP64]], ptr [[TMP69]], align 4
-// CHECK7-NEXT: [[TMP70:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 1
-// CHECK7-NEXT: store i32 [[TMP64]], ptr [[TMP70]], align 4
-// CHECK7-NEXT: [[TMP71:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i32 0, i32 1
-// CHECK7-NEXT: store ptr null, ptr [[TMP71]], align 4
-// CHECK7-NEXT: [[TMP72:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP70:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
+// CHECK7-NEXT: store i32 [[TMP66]], ptr [[TMP70]], align 4
+// CHECK7-NEXT: [[TMP71:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
+// CHECK7-NEXT: store i32 [[TMP66]], ptr [[TMP71]], align 4
+// CHECK7-NEXT: [[TMP72:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i32 0, i32 0
+// CHECK7-NEXT: store ptr null, ptr [[TMP72]], align 4
+// CHECK7-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 1
+// CHECK7-NEXT: store i32 [[TMP68]], ptr [[TMP73]], align 4
+// CHECK7-NEXT: [[TMP74:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 1
+// CHECK7-NEXT: store i32 [[TMP68]], ptr [[TMP74]], align 4
+// CHECK7-NEXT: [[TMP75:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS12]], i32 0, i32 1
+// CHECK7-NEXT: store ptr null, ptr [[TMP75]], align 4
+// CHECK7-NEXT: [[TMP76:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS10]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP77:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS11]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS14:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK7-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP74]], align 4
-// CHECK7-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
-// CHECK7-NEXT: store i32 2, ptr [[TMP75]], align 4
-// CHECK7-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
-// CHECK7-NEXT: store ptr [[TMP72]], ptr [[TMP76]], align 4
-// CHECK7-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
-// CHECK7-NEXT: store ptr [[TMP73]], ptr [[TMP77]], align 4
-// CHECK7-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
-// CHECK7-NEXT: store ptr @.offload_sizes.7, ptr [[TMP78]], align 4
-// CHECK7-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
-// CHECK7-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP79]], align 4
-// CHECK7-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
-// CHECK7-NEXT: store ptr null, ptr [[TMP80]], align 4
-// CHECK7-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
-// CHECK7-NEXT: store ptr null, ptr [[TMP81]], align 4
-// CHECK7-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
-// CHECK7-NEXT: store i64 10, ptr [[TMP82]], align 8
-// CHECK7-NEXT: [[TMP83:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118.region_id, ptr [[KERNEL_ARGS14]])
-// CHECK7-NEXT: [[TMP84:%.*]] = icmp ne i32 [[TMP83]], 0
-// CHECK7-NEXT: br i1 [[TMP84]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
+// CHECK7-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
+// CHECK7-NEXT: store i32 2, ptr [[TMP78]], align 4
+// CHECK7-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
+// CHECK7-NEXT: store i32 2, ptr [[TMP79]], align 4
+// CHECK7-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
+// CHECK7-NEXT: store ptr [[TMP76]], ptr [[TMP80]], align 4
+// CHECK7-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
+// CHECK7-NEXT: store ptr [[TMP77]], ptr [[TMP81]], align 4
+// CHECK7-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
+// CHECK7-NEXT: store ptr @.offload_sizes.7, ptr [[TMP82]], align 4
+// CHECK7-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
+// CHECK7-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP83]], align 4
+// CHECK7-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
+// CHECK7-NEXT: store ptr null, ptr [[TMP84]], align 4
+// CHECK7-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
+// CHECK7-NEXT: store ptr null, ptr [[TMP85]], align 4
+// CHECK7-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
+// CHECK7-NEXT: store i64 10, ptr [[TMP86]], align 8
+// CHECK7-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP87]], align 8
+// CHECK7-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP88]], align 4
+// CHECK7-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP89]], align 4
+// CHECK7-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP90]], align 4
+// CHECK7-NEXT: [[TMP91:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118.region_id, ptr [[KERNEL_ARGS14]])
+// CHECK7-NEXT: [[TMP92:%.*]] = icmp ne i32 [[TMP91]], 0
+// CHECK7-NEXT: br i1 [[TMP92]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
// CHECK7: omp_offload.failed15:
-// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118(i32 [[TMP62]], i32 [[TMP64]]) #[[ATTR4]]
+// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118(i32 [[TMP66]], i32 [[TMP68]]) #[[ATTR4]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT16]]
// CHECK7: omp_offload.cont16:
// CHECK7-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK7: omp_if.else:
-// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118(i32 [[TMP62]], i32 [[TMP64]]) #[[ATTR4]]
+// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l118(i32 [[TMP66]], i32 [[TMP68]]) #[[ATTR4]]
// CHECK7-NEXT: br label [[OMP_IF_END]]
// CHECK7: omp_if.end:
-// CHECK7-NEXT: [[TMP85:%.*]] = load i32, ptr [[A]], align 4
-// CHECK7-NEXT: store i32 [[TMP85]], ptr [[A_CASTED17]], align 4
-// CHECK7-NEXT: [[TMP86:%.*]] = load i32, ptr [[A_CASTED17]], align 4
-// CHECK7-NEXT: [[TMP87:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK7-NEXT: [[CMP18:%.*]] = icmp sgt i32 [[TMP87]], 20
+// CHECK7-NEXT: [[TMP93:%.*]] = load i32, ptr [[A]], align 4
+// CHECK7-NEXT: store i32 [[TMP93]], ptr [[A_CASTED17]], align 4
+// CHECK7-NEXT: [[TMP94:%.*]] = load i32, ptr [[A_CASTED17]], align 4
+// CHECK7-NEXT: [[TMP95:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK7-NEXT: [[CMP18:%.*]] = icmp sgt i32 [[TMP95]], 20
// CHECK7-NEXT: br i1 [[CMP18]], label [[OMP_IF_THEN19:%.*]], label [[OMP_IF_ELSE27:%.*]]
// CHECK7: omp_if.then19:
-// CHECK7-NEXT: [[TMP88:%.*]] = mul nuw i32 [[TMP1]], 4
-// CHECK7-NEXT: [[TMP89:%.*]] = sext i32 [[TMP88]] to i64
-// CHECK7-NEXT: [[TMP90:%.*]] = mul nuw i32 5, [[TMP3]]
-// CHECK7-NEXT: [[TMP91:%.*]] = mul nuw i32 [[TMP90]], 8
-// CHECK7-NEXT: [[TMP92:%.*]] = sext i32 [[TMP91]] to i64
+// CHECK7-NEXT: [[TMP96:%.*]] = mul nuw i32 [[TMP1]], 4
+// CHECK7-NEXT: [[TMP97:%.*]] = sext i32 [[TMP96]] to i64
+// CHECK7-NEXT: [[TMP98:%.*]] = mul nuw i32 5, [[TMP3]]
+// CHECK7-NEXT: [[TMP99:%.*]] = mul nuw i32 [[TMP98]], 8
+// CHECK7-NEXT: [[TMP100:%.*]] = sext i32 [[TMP99]] to i64
// CHECK7-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES]], ptr align 4 @.offload_sizes.10, i32 72, i1 false)
-// CHECK7-NEXT: [[TMP93:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK7-NEXT: store i32 [[TMP86]], ptr [[TMP93]], align 4
-// CHECK7-NEXT: [[TMP94:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK7-NEXT: store i32 [[TMP86]], ptr [[TMP94]], align 4
-// CHECK7-NEXT: [[TMP95:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 0
-// CHECK7-NEXT: store ptr null, ptr [[TMP95]], align 4
-// CHECK7-NEXT: [[TMP96:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
-// CHECK7-NEXT: store ptr [[B]], ptr [[TMP96]], align 4
-// CHECK7-NEXT: [[TMP97:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
-// CHECK7-NEXT: store ptr [[B]], ptr [[TMP97]], align 4
-// CHECK7-NEXT: [[TMP98:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 1
-// CHECK7-NEXT: store ptr null, ptr [[TMP98]], align 4
-// CHECK7-NEXT: [[TMP99:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
-// CHECK7-NEXT: store i32 [[TMP1]], ptr [[TMP99]], align 4
-// CHECK7-NEXT: [[TMP100:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
-// CHECK7-NEXT: store i32 [[TMP1]], ptr [[TMP100]], align 4
-// CHECK7-NEXT: [[TMP101:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 2
-// CHECK7-NEXT: store ptr null, ptr [[TMP101]], align 4
-// CHECK7-NEXT: [[TMP102:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
-// CHECK7-NEXT: store ptr [[VLA]], ptr [[TMP102]], align 4
-// CHECK7-NEXT: [[TMP103:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
-// CHECK7-NEXT: store ptr [[VLA]], ptr [[TMP103]], align 4
-// CHECK7-NEXT: [[TMP104:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
-// CHECK7-NEXT: store i64 [[TMP89]], ptr [[TMP104]], align 4
-// CHECK7-NEXT: [[TMP105:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 3
-// CHECK7-NEXT: store ptr null, ptr [[TMP105]], align 4
-// CHECK7-NEXT: [[TMP106:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 4
-// CHECK7-NEXT: store ptr [[C]], ptr [[TMP106]], align 4
-// CHECK7-NEXT: [[TMP107:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 4
-// CHECK7-NEXT: store ptr [[C]], ptr [[TMP107]], align 4
-// CHECK7-NEXT: [[TMP108:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 4
-// CHECK7-NEXT: store ptr null, ptr [[TMP108]], align 4
-// CHECK7-NEXT: [[TMP109:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 5
-// CHECK7-NEXT: store i32 5, ptr [[TMP109]], align 4
-// CHECK7-NEXT: [[TMP110:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 5
-// CHECK7-NEXT: store i32 5, ptr [[TMP110]], align 4
-// CHECK7-NEXT: [[TMP111:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 5
-// CHECK7-NEXT: store ptr null, ptr [[TMP111]], align 4
-// CHECK7-NEXT: [[TMP112:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 6
-// CHECK7-NEXT: store i32 [[TMP3]], ptr [[TMP112]], align 4
-// CHECK7-NEXT: [[TMP113:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 6
-// CHECK7-NEXT: store i32 [[TMP3]], ptr [[TMP113]], align 4
-// CHECK7-NEXT: [[TMP114:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 6
-// CHECK7-NEXT: store ptr null, ptr [[TMP114]], align 4
-// CHECK7-NEXT: [[TMP115:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 7
-// CHECK7-NEXT: store ptr [[VLA1]], ptr [[TMP115]], align 4
-// CHECK7-NEXT: [[TMP116:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 7
-// CHECK7-NEXT: store ptr [[VLA1]], ptr [[TMP116]], align 4
-// CHECK7-NEXT: [[TMP117:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
-// CHECK7-NEXT: store i64 [[TMP92]], ptr [[TMP117]], align 4
-// CHECK7-NEXT: [[TMP118:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 7
-// CHECK7-NEXT: store ptr null, ptr [[TMP118]], align 4
-// CHECK7-NEXT: [[TMP119:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 8
-// CHECK7-NEXT: store ptr [[D]], ptr [[TMP119]], align 4
-// CHECK7-NEXT: [[TMP120:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 8
-// CHECK7-NEXT: store ptr [[D]], ptr [[TMP120]], align 4
-// CHECK7-NEXT: [[TMP121:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 8
-// CHECK7-NEXT: store ptr null, ptr [[TMP121]], align 4
-// CHECK7-NEXT: [[TMP122:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP123:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP124:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP101:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK7-NEXT: store i32 [[TMP94]], ptr [[TMP101]], align 4
+// CHECK7-NEXT: [[TMP102:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK7-NEXT: store i32 [[TMP94]], ptr [[TMP102]], align 4
+// CHECK7-NEXT: [[TMP103:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 0
+// CHECK7-NEXT: store ptr null, ptr [[TMP103]], align 4
+// CHECK7-NEXT: [[TMP104:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
+// CHECK7-NEXT: store ptr [[B]], ptr [[TMP104]], align 4
+// CHECK7-NEXT: [[TMP105:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
+// CHECK7-NEXT: store ptr [[B]], ptr [[TMP105]], align 4
+// CHECK7-NEXT: [[TMP106:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 1
+// CHECK7-NEXT: store ptr null, ptr [[TMP106]], align 4
+// CHECK7-NEXT: [[TMP107:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
+// CHECK7-NEXT: store i32 [[TMP1]], ptr [[TMP107]], align 4
+// CHECK7-NEXT: [[TMP108:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
+// CHECK7-NEXT: store i32 [[TMP1]], ptr [[TMP108]], align 4
+// CHECK7-NEXT: [[TMP109:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 2
+// CHECK7-NEXT: store ptr null, ptr [[TMP109]], align 4
+// CHECK7-NEXT: [[TMP110:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
+// CHECK7-NEXT: store ptr [[VLA]], ptr [[TMP110]], align 4
+// CHECK7-NEXT: [[TMP111:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
+// CHECK7-NEXT: store ptr [[VLA]], ptr [[TMP111]], align 4
+// CHECK7-NEXT: [[TMP112:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
+// CHECK7-NEXT: store i64 [[TMP97]], ptr [[TMP112]], align 4
+// CHECK7-NEXT: [[TMP113:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 3
+// CHECK7-NEXT: store ptr null, ptr [[TMP113]], align 4
+// CHECK7-NEXT: [[TMP114:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 4
+// CHECK7-NEXT: store ptr [[C]], ptr [[TMP114]], align 4
+// CHECK7-NEXT: [[TMP115:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 4
+// CHECK7-NEXT: store ptr [[C]], ptr [[TMP115]], align 4
+// CHECK7-NEXT: [[TMP116:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 4
+// CHECK7-NEXT: store ptr null, ptr [[TMP116]], align 4
+// CHECK7-NEXT: [[TMP117:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 5
+// CHECK7-NEXT: store i32 5, ptr [[TMP117]], align 4
+// CHECK7-NEXT: [[TMP118:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 5
+// CHECK7-NEXT: store i32 5, ptr [[TMP118]], align 4
+// CHECK7-NEXT: [[TMP119:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 5
+// CHECK7-NEXT: store ptr null, ptr [[TMP119]], align 4
+// CHECK7-NEXT: [[TMP120:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 6
+// CHECK7-NEXT: store i32 [[TMP3]], ptr [[TMP120]], align 4
+// CHECK7-NEXT: [[TMP121:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 6
+// CHECK7-NEXT: store i32 [[TMP3]], ptr [[TMP121]], align 4
+// CHECK7-NEXT: [[TMP122:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 6
+// CHECK7-NEXT: store ptr null, ptr [[TMP122]], align 4
+// CHECK7-NEXT: [[TMP123:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 7
+// CHECK7-NEXT: store ptr [[VLA1]], ptr [[TMP123]], align 4
+// CHECK7-NEXT: [[TMP124:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 7
+// CHECK7-NEXT: store ptr [[VLA1]], ptr [[TMP124]], align 4
+// CHECK7-NEXT: [[TMP125:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
+// CHECK7-NEXT: store i64 [[TMP100]], ptr [[TMP125]], align 4
+// CHECK7-NEXT: [[TMP126:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 7
+// CHECK7-NEXT: store ptr null, ptr [[TMP126]], align 4
+// CHECK7-NEXT: [[TMP127:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 8
+// CHECK7-NEXT: store ptr [[D]], ptr [[TMP127]], align 4
+// CHECK7-NEXT: [[TMP128:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 8
+// CHECK7-NEXT: store ptr [[D]], ptr [[TMP128]], align 4
+// CHECK7-NEXT: [[TMP129:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 8
+// CHECK7-NEXT: store ptr null, ptr [[TMP129]], align 4
+// CHECK7-NEXT: [[TMP130:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP131:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP132:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS24:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK7-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP125]], align 4
-// CHECK7-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 1
-// CHECK7-NEXT: store i32 9, ptr [[TMP126]], align 4
-// CHECK7-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 2
-// CHECK7-NEXT: store ptr [[TMP122]], ptr [[TMP127]], align 4
-// CHECK7-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 3
-// CHECK7-NEXT: store ptr [[TMP123]], ptr [[TMP128]], align 4
-// CHECK7-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 4
-// CHECK7-NEXT: store ptr [[TMP124]], ptr [[TMP129]], align 4
-// CHECK7-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 5
-// CHECK7-NEXT: store ptr @.offload_maptypes.11, ptr [[TMP130]], align 4
-// CHECK7-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 6
-// CHECK7-NEXT: store ptr null, ptr [[TMP131]], align 4
-// CHECK7-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 7
-// CHECK7-NEXT: store ptr null, ptr [[TMP132]], align 4
-// CHECK7-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 8
-// CHECK7-NEXT: store i64 10, ptr [[TMP133]], align 8
-// CHECK7-NEXT: [[TMP134:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142.region_id, ptr [[KERNEL_ARGS24]])
-// CHECK7-NEXT: [[TMP135:%.*]] = icmp ne i32 [[TMP134]], 0
-// CHECK7-NEXT: br i1 [[TMP135]], label [[OMP_OFFLOAD_FAILED25:%.*]], label [[OMP_OFFLOAD_CONT26:%.*]]
+// CHECK7-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 0
+// CHECK7-NEXT: store i32 2, ptr [[TMP133]], align 4
+// CHECK7-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 1
+// CHECK7-NEXT: store i32 9, ptr [[TMP134]], align 4
+// CHECK7-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 2
+// CHECK7-NEXT: store ptr [[TMP130]], ptr [[TMP135]], align 4
+// CHECK7-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 3
+// CHECK7-NEXT: store ptr [[TMP131]], ptr [[TMP136]], align 4
+// CHECK7-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 4
+// CHECK7-NEXT: store ptr [[TMP132]], ptr [[TMP137]], align 4
+// CHECK7-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 5
+// CHECK7-NEXT: store ptr @.offload_maptypes.11, ptr [[TMP138]], align 4
+// CHECK7-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 6
+// CHECK7-NEXT: store ptr null, ptr [[TMP139]], align 4
+// CHECK7-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 7
+// CHECK7-NEXT: store ptr null, ptr [[TMP140]], align 4
+// CHECK7-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 8
+// CHECK7-NEXT: store i64 10, ptr [[TMP141]], align 8
+// CHECK7-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP142]], align 8
+// CHECK7-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP143]], align 4
+// CHECK7-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP144]], align 4
+// CHECK7-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS24]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP145]], align 4
+// CHECK7-NEXT: [[TMP146:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142.region_id, ptr [[KERNEL_ARGS24]])
+// CHECK7-NEXT: [[TMP147:%.*]] = icmp ne i32 [[TMP146]], 0
+// CHECK7-NEXT: br i1 [[TMP147]], label [[OMP_OFFLOAD_FAILED25:%.*]], label [[OMP_OFFLOAD_CONT26:%.*]]
// CHECK7: omp_offload.failed25:
-// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142(i32 [[TMP86]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]]) #[[ATTR4]]
+// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142(i32 [[TMP94]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]]) #[[ATTR4]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT26]]
// CHECK7: omp_offload.cont26:
// CHECK7-NEXT: br label [[OMP_IF_END28:%.*]]
// CHECK7: omp_if.else27:
-// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142(i32 [[TMP86]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]]) #[[ATTR4]]
+// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l142(i32 [[TMP94]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]]) #[[ATTR4]]
// CHECK7-NEXT: br label [[OMP_IF_END28]]
// CHECK7: omp_if.end28:
-// CHECK7-NEXT: [[TMP136:%.*]] = load i32, ptr [[A]], align 4
-// CHECK7-NEXT: [[TMP137:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK7-NEXT: call void @llvm.stackrestore(ptr [[TMP137]])
-// CHECK7-NEXT: ret i32 [[TMP136]]
+// CHECK7-NEXT: [[TMP148:%.*]] = load i32, ptr [[A]], align 4
+// CHECK7-NEXT: [[TMP149:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK7-NEXT: call void @llvm.stackrestore(ptr [[TMP149]])
+// CHECK7-NEXT: ret i32 [[TMP148]]
//
//
// CHECK7-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l97
@@ -5989,37 +6184,46 @@ int bar(int n){
// CHECK7-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT_ANON:%.*]], ptr [[TMP9]], i32 0, i32 1
// CHECK7-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[TMP9]], i32 0, i32 2
// CHECK7-NEXT: [[TMP18:%.*]] = load i32, ptr [[TMP16]], align 4
-// CHECK7-NEXT: store i32 1, ptr [[KERNEL_ARGS_I]], align 4, !noalias !27
-// CHECK7-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 1
-// CHECK7-NEXT: store i32 3, ptr [[TMP19]], align 4, !noalias !27
-// CHECK7-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 2
-// CHECK7-NEXT: store ptr [[TMP13]], ptr [[TMP20]], align 4, !noalias !27
-// CHECK7-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 3
-// CHECK7-NEXT: store ptr [[TMP14]], ptr [[TMP21]], align 4, !noalias !27
-// CHECK7-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 4
-// CHECK7-NEXT: store ptr [[TMP15]], ptr [[TMP22]], align 4, !noalias !27
-// CHECK7-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 5
-// CHECK7-NEXT: store ptr @.offload_maptypes, ptr [[TMP23]], align 4, !noalias !27
-// CHECK7-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 6
-// CHECK7-NEXT: store ptr null, ptr [[TMP24]], align 4, !noalias !27
-// CHECK7-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 7
+// CHECK7-NEXT: [[TMP19:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP18]], 0
+// CHECK7-NEXT: store i32 2, ptr [[KERNEL_ARGS_I]], align 4, !noalias !27
+// CHECK7-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 1
+// CHECK7-NEXT: store i32 3, ptr [[TMP20]], align 4, !noalias !27
+// CHECK7-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 2
+// CHECK7-NEXT: store ptr [[TMP13]], ptr [[TMP21]], align 4, !noalias !27
+// CHECK7-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 3
+// CHECK7-NEXT: store ptr [[TMP14]], ptr [[TMP22]], align 4, !noalias !27
+// CHECK7-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 4
+// CHECK7-NEXT: store ptr [[TMP15]], ptr [[TMP23]], align 4, !noalias !27
+// CHECK7-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 5
+// CHECK7-NEXT: store ptr @.offload_maptypes, ptr [[TMP24]], align 4, !noalias !27
+// CHECK7-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 6
// CHECK7-NEXT: store ptr null, ptr [[TMP25]], align 4, !noalias !27
-// CHECK7-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 8
-// CHECK7-NEXT: store i64 10, ptr [[TMP26]], align 8, !noalias !27
-// CHECK7-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel_nowait(ptr @[[GLOB2]], i64 -1, i32 [[TMP18]], i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l97.region_id, ptr [[KERNEL_ARGS_I]], i32 0, ptr null, i32 0, ptr null)
-// CHECK7-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK7-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__1_EXIT:%.*]]
+// CHECK7-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 7
+// CHECK7-NEXT: store ptr null, ptr [[TMP26]], align 4, !noalias !27
+// CHECK7-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 8
+// CHECK7-NEXT: store i64 10, ptr [[TMP27]], align 8, !noalias !27
+// CHECK7-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 9
+// CHECK7-NEXT: store i64 1, ptr [[TMP28]], align 8, !noalias !27
+// CHECK7-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] [[TMP19]], ptr [[TMP29]], align 4, !noalias !27
+// CHECK7-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP30]], align 4, !noalias !27
+// CHECK7-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP31]], align 4, !noalias !27
+// CHECK7-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 [[TMP18]], i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l97.region_id, ptr [[KERNEL_ARGS_I]])
+// CHECK7-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
+// CHECK7-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__1_EXIT:%.*]]
// CHECK7: omp_offload.failed.i:
-// CHECK7-NEXT: [[TMP29:%.*]] = load i16, ptr [[TMP12]], align 2
-// CHECK7-NEXT: store i16 [[TMP29]], ptr [[AA_CASTED_I]], align 2, !noalias !27
-// CHECK7-NEXT: [[TMP30:%.*]] = load i32, ptr [[AA_CASTED_I]], align 4, !noalias !27
-// CHECK7-NEXT: [[TMP31:%.*]] = load i32, ptr [[TMP16]], align 4
-// CHECK7-NEXT: store i32 [[TMP31]], ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 4, !noalias !27
-// CHECK7-NEXT: [[TMP32:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 4, !noalias !27
-// CHECK7-NEXT: [[TMP33:%.*]] = load i32, ptr [[TMP17]], align 4
-// CHECK7-NEXT: store i32 [[TMP33]], ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 4, !noalias !27
-// CHECK7-NEXT: [[TMP34:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 4, !noalias !27
-// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l97(i32 [[TMP30]], i32 [[TMP32]], i32 [[TMP34]]) #[[ATTR4]]
+// CHECK7-NEXT: [[TMP34:%.*]] = load i16, ptr [[TMP12]], align 2
+// CHECK7-NEXT: store i16 [[TMP34]], ptr [[AA_CASTED_I]], align 2, !noalias !27
+// CHECK7-NEXT: [[TMP35:%.*]] = load i32, ptr [[AA_CASTED_I]], align 4, !noalias !27
+// CHECK7-NEXT: [[TMP36:%.*]] = load i32, ptr [[TMP16]], align 4
+// CHECK7-NEXT: store i32 [[TMP36]], ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 4, !noalias !27
+// CHECK7-NEXT: [[TMP37:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED_I]], align 4, !noalias !27
+// CHECK7-NEXT: [[TMP38:%.*]] = load i32, ptr [[TMP17]], align 4
+// CHECK7-NEXT: store i32 [[TMP38]], ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 4, !noalias !27
+// CHECK7-NEXT: [[TMP39:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED4_I]], align 4, !noalias !27
+// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l97(i32 [[TMP35]], i32 [[TMP37]], i32 [[TMP39]]) #[[ATTR4]]
// CHECK7-NEXT: br label [[DOTOMP_OUTLINED__1_EXIT]]
// CHECK7: .omp_outlined..1.exit:
// CHECK7-NEXT: ret i32 0
@@ -6587,7 +6791,7 @@ int bar(int n){
// CHECK7-NEXT: [[TMP34:%.*]] = getelementptr inbounds [6 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK7-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP35]], align 4
+// CHECK7-NEXT: store i32 2, ptr [[TMP35]], align 4
// CHECK7-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK7-NEXT: store i32 6, ptr [[TMP36]], align 4
// CHECK7-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -6604,9 +6808,17 @@ int bar(int n){
// CHECK7-NEXT: store ptr null, ptr [[TMP42]], align 4
// CHECK7-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK7-NEXT: store i64 10, ptr [[TMP43]], align 8
-// CHECK7-NEXT: [[TMP44:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l215.region_id, ptr [[KERNEL_ARGS]])
-// CHECK7-NEXT: [[TMP45:%.*]] = icmp ne i32 [[TMP44]], 0
-// CHECK7-NEXT: br i1 [[TMP45]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK7-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP44]], align 8
+// CHECK7-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP45]], align 4
+// CHECK7-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP46]], align 4
+// CHECK7-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP47]], align 4
+// CHECK7-NEXT: [[TMP48:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l215.region_id, ptr [[KERNEL_ARGS]])
+// CHECK7-NEXT: [[TMP49:%.*]] = icmp ne i32 [[TMP48]], 0
+// CHECK7-NEXT: br i1 [[TMP49]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK7: omp_offload.failed:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l215(ptr [[THIS1]], i32 [[TMP6]], i32 2, i32 [[TMP1]], ptr [[VLA]], i32 [[TMP8]]) #[[ATTR4]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -6616,15 +6828,15 @@ int bar(int n){
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l215(ptr [[THIS1]], i32 [[TMP6]], i32 2, i32 [[TMP1]], ptr [[VLA]], i32 [[TMP8]]) #[[ATTR4]]
// CHECK7-NEXT: br label [[OMP_IF_END]]
// CHECK7: omp_if.end:
-// CHECK7-NEXT: [[TMP46:%.*]] = mul nsw i32 1, [[TMP1]]
-// CHECK7-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i32 [[TMP46]]
+// CHECK7-NEXT: [[TMP50:%.*]] = mul nsw i32 1, [[TMP1]]
+// CHECK7-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i32 [[TMP50]]
// CHECK7-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds i16, ptr [[ARRAYIDX]], i32 1
-// CHECK7-NEXT: [[TMP47:%.*]] = load i16, ptr [[ARRAYIDX4]], align 2
-// CHECK7-NEXT: [[CONV:%.*]] = sext i16 [[TMP47]] to i32
-// CHECK7-NEXT: [[TMP48:%.*]] = load i32, ptr [[B]], align 4
-// CHECK7-NEXT: [[ADD5:%.*]] = add nsw i32 [[CONV]], [[TMP48]]
-// CHECK7-NEXT: [[TMP49:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK7-NEXT: call void @llvm.stackrestore(ptr [[TMP49]])
+// CHECK7-NEXT: [[TMP51:%.*]] = load i16, ptr [[ARRAYIDX4]], align 2
+// CHECK7-NEXT: [[CONV:%.*]] = sext i16 [[TMP51]] to i32
+// CHECK7-NEXT: [[TMP52:%.*]] = load i32, ptr [[B]], align 4
+// CHECK7-NEXT: [[ADD5:%.*]] = add nsw i32 [[CONV]], [[TMP52]]
+// CHECK7-NEXT: [[TMP53:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK7-NEXT: call void @llvm.stackrestore(ptr [[TMP53]])
// CHECK7-NEXT: ret i32 [[ADD5]]
//
//
@@ -6716,7 +6928,7 @@ int bar(int n){
// CHECK7-NEXT: [[TMP31:%.*]] = zext i32 [[ADD5]] to i64
// CHECK7-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK7-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP32]], align 4
+// CHECK7-NEXT: store i32 2, ptr [[TMP32]], align 4
// CHECK7-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK7-NEXT: store i32 5, ptr [[TMP33]], align 4
// CHECK7-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -6733,9 +6945,17 @@ int bar(int n){
// CHECK7-NEXT: store ptr null, ptr [[TMP39]], align 4
// CHECK7-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK7-NEXT: store i64 [[TMP31]], ptr [[TMP40]], align 8
-// CHECK7-NEXT: [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l197.region_id, ptr [[KERNEL_ARGS]])
-// CHECK7-NEXT: [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
-// CHECK7-NEXT: br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK7-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP41]], align 8
+// CHECK7-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP42]], align 4
+// CHECK7-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP43]], align 4
+// CHECK7-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP44]], align 4
+// CHECK7-NEXT: [[TMP45:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l197.region_id, ptr [[KERNEL_ARGS]])
+// CHECK7-NEXT: [[TMP46:%.*]] = icmp ne i32 [[TMP45]], 0
+// CHECK7-NEXT: br i1 [[TMP46]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK7: omp_offload.failed:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l197(i32 [[TMP1]], i32 [[TMP3]], i32 [[TMP5]], i32 [[TMP7]], ptr [[B]]) #[[ATTR4]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -6745,8 +6965,8 @@ int bar(int n){
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l197(i32 [[TMP1]], i32 [[TMP3]], i32 [[TMP5]], i32 [[TMP7]], ptr [[B]]) #[[ATTR4]]
// CHECK7-NEXT: br label [[OMP_IF_END]]
// CHECK7: omp_if.end:
-// CHECK7-NEXT: [[TMP43:%.*]] = load i32, ptr [[A]], align 4
-// CHECK7-NEXT: ret i32 [[TMP43]]
+// CHECK7-NEXT: [[TMP47:%.*]] = load i32, ptr [[A]], align 4
+// CHECK7-NEXT: ret i32 [[TMP47]]
//
//
// CHECK7-LABEL: define {{[^@]+}}@_Z9ftemplateIiET_i
@@ -6797,7 +7017,7 @@ int bar(int n){
// CHECK7-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK7-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP16]], align 4
+// CHECK7-NEXT: store i32 2, ptr [[TMP16]], align 4
// CHECK7-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK7-NEXT: store i32 3, ptr [[TMP17]], align 4
// CHECK7-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -6814,9 +7034,17 @@ int bar(int n){
// CHECK7-NEXT: store ptr null, ptr [[TMP23]], align 4
// CHECK7-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK7-NEXT: store i64 10, ptr [[TMP24]], align 8
-// CHECK7-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l180.region_id, ptr [[KERNEL_ARGS]])
-// CHECK7-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK7-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK7-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP25]], align 8
+// CHECK7-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK7-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP27]], align 4
+// CHECK7-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP28]], align 4
+// CHECK7-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l180.region_id, ptr [[KERNEL_ARGS]])
+// CHECK7-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK7-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK7: omp_offload.failed:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l180(i32 [[TMP1]], i32 [[TMP3]], ptr [[B]]) #[[ATTR4]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -6826,8 +7054,8 @@ int bar(int n){
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l180(i32 [[TMP1]], i32 [[TMP3]], ptr [[B]]) #[[ATTR4]]
// CHECK7-NEXT: br label [[OMP_IF_END]]
// CHECK7: omp_if.end:
-// CHECK7-NEXT: [[TMP27:%.*]] = load i32, ptr [[A]], align 4
-// CHECK7-NEXT: ret i32 [[TMP27]]
+// CHECK7-NEXT: [[TMP31:%.*]] = load i32, ptr [[A]], align 4
+// CHECK7-NEXT: ret i32 [[TMP31]]
//
//
// CHECK7-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l215
diff --git a/clang/test/OpenMP/target_teams_distribute_simd_collapse_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_simd_collapse_codegen.cpp
index fde0114f2428a..4aa8008841c04 100644
--- a/clang/test/OpenMP/target_teams_distribute_simd_collapse_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_simd_collapse_codegen.cpp
@@ -124,7 +124,7 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -141,9 +141,17 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 56088, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR3:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -151,8 +159,8 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[A3:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x [456 x i32]], ptr [[A3]], i64 0, i64 0
// CHECK1-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds [456 x i32], ptr [[ARRAYIDX]], i64 0, i64 0
-// CHECK1-NEXT: [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
-// CHECK1-NEXT: ret i32 [[TMP16]]
+// CHECK1-NEXT: [[TMP20:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
+// CHECK1-NEXT: ret i32 [[TMP20]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -292,7 +300,7 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -309,9 +317,17 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 56088, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR3:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -319,8 +335,8 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[A3:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x [456 x i32]], ptr [[A3]], i32 0, i32 0
// CHECK3-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds [456 x i32], ptr [[ARRAYIDX]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
-// CHECK3-NEXT: ret i32 [[TMP16]]
+// CHECK3-NEXT: [[TMP20:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
+// CHECK3-NEXT: ret i32 [[TMP20]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -668,7 +684,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[ADD:%.*]] = add nsw i64 [[TMP35]], 1
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP36]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP36]], align 4
// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 5, ptr [[TMP37]], align 4
// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -685,20 +701,28 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP43]], align 8
// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 [[ADD]], ptr [[TMP44]], align 8
-// CHECK9-NEXT: [[TMP45:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l80.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP46:%.*]] = icmp ne i32 [[TMP45]], 0
-// CHECK9-NEXT: br i1 [[TMP46]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP46]], align 4
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP47]], align 4
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP48]], align 4
+// CHECK9-NEXT: [[TMP49:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l80.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP50:%.*]] = icmp ne i32 [[TMP49]], 0
+// CHECK9-NEXT: br i1 [[TMP50]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l80(i64 [[TMP7]], i64 [[TMP9]], i64 [[TMP1]], i64 [[TMP3]], ptr [[VLA]]) #[[ATTR4:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP47:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef signext [[TMP47]])
+// CHECK9-NEXT: [[TMP51:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef signext [[TMP51]])
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK9-NEXT: [[TMP48:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP48]])
-// CHECK9-NEXT: [[TMP49:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP49]]
+// CHECK9-NEXT: [[TMP52:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP52]])
+// CHECK9-NEXT: [[TMP53:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP53]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l80
@@ -914,7 +938,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK9-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK9-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -931,9 +955,17 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 20, ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l67.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK9-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK9-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l67.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK9-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l67(ptr [[A]]) #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1146,7 +1178,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[ADD:%.*]] = add nsw i64 [[TMP34]], 1
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP35]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP35]], align 4
// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 5, ptr [[TMP36]], align 4
// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1163,20 +1195,28 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP42]], align 4
// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 [[ADD]], ptr [[TMP43]], align 8
-// CHECK11-NEXT: [[TMP44:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l80.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP45:%.*]] = icmp ne i32 [[TMP44]], 0
-// CHECK11-NEXT: br i1 [[TMP45]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP44]], align 8
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP47]], align 4
+// CHECK11-NEXT: [[TMP48:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l80.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP49:%.*]] = icmp ne i32 [[TMP48]], 0
+// CHECK11-NEXT: br i1 [[TMP49]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l80(i32 [[TMP5]], i32 [[TMP7]], i32 [[TMP0]], i32 [[TMP1]], ptr [[VLA]]) #[[ATTR4:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP46:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef [[TMP46]])
+// CHECK11-NEXT: [[TMP50:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef [[TMP50]])
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK11-NEXT: [[TMP47:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP47]])
-// CHECK11-NEXT: [[TMP48:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP48]]
+// CHECK11-NEXT: [[TMP51:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP51]])
+// CHECK11-NEXT: [[TMP52:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP52]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l80
@@ -1390,7 +1430,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK11-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK11-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1407,9 +1447,17 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK11-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 20, ptr [[TMP13]], align 8
-// CHECK11-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l67.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK11-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP16]], align 4
+// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK11-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l67.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK11-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l67(ptr [[A]]) #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/target_teams_distribute_simd_depend_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_simd_depend_codegen.cpp
index 0076db7750e20..1bcf4ebfa2fd1 100644
--- a/clang/test/OpenMP/target_teams_distribute_simd_depend_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_simd_depend_codegen.cpp
@@ -242,8 +242,8 @@ int foo(int n) {
// OMP50-32: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 3
// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]],
// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64
-// OMP45: [[RET:%.+]] = call i32 @__tgt_target_kernel_nowait(%struct.ident_t* @{{.+}}, i64 [[DEVICE]], i32 0, i32 1, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* %{{.+}})
-// OMP50: [[RET:%.+]] = call i32 @__tgt_target_kernel_nowait(%struct.ident_t* @{{.+}}, i64 [[DEVICE]], i32 0, i32 1, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* %{{.+}})
+// OMP45: [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 [[DEVICE]], i32 0, i32 1, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* %{{.+}})
+// OMP50: [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 [[DEVICE]], i32 0, i32 1, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* %{{.+}})
// CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]]
diff --git a/clang/test/OpenMP/target_teams_distribute_simd_dist_schedule_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_simd_dist_schedule_codegen.cpp
index e86c331ae5efd..5a14f0ab1939e 100644
--- a/clang/test/OpenMP/target_teams_distribute_simd_dist_schedule_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_simd_dist_schedule_codegen.cpp
@@ -159,7 +159,7 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -176,87 +176,111 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR3:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.2, ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l32.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.2, ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP36]], align 4
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l32.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK1-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK1-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK1: omp_offload.failed8:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l32(ptr [[THIS1]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK1: omp_offload.cont9:
// CHECK1-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A10]], ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A10]], ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.5, ptr [[TMP41]], align 8
-// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP42]], align 8
-// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP43]], align 8
-// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP44]], align 8
-// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK1-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK1-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK1-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 8
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 8
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.5, ptr [[TMP49]], align 8
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP50]], align 8
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP51]], align 8
+// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP52]], align 8
+// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP56]], align 4
+// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK1-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK1-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK1-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK1: omp_offload.failed16:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36(ptr [[THIS1]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK1: omp_offload.cont17:
// CHECK1-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A18]], i64 0, i64 0
-// CHECK1-NEXT: [[TMP48:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK1-NEXT: ret i32 [[TMP48]]
+// CHECK1-NEXT: [[TMP60:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK1-NEXT: ret i32 [[TMP60]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -575,7 +599,7 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -592,87 +616,111 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR3:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 4
-// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A2]], ptr [[TMP17]], align 4
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 4
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A2]], ptr [[TMP21]], align 4
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.2, ptr [[TMP25]], align 4
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP26]], align 4
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK3-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l32.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK3-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK3-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.2, ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK3-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP36]], align 4
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l32.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK3-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK3-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK3: omp_offload.failed8:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l32(ptr [[THIS1]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK3: omp_offload.cont9:
// CHECK3-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 4
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A10]], ptr [[TMP33]], align 4
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP34]], align 4
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A10]], ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP42]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 4
-// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 4
-// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.5, ptr [[TMP41]], align 4
-// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP42]], align 4
-// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP44]], align 4
-// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK3-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK3-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK3-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK3-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 4
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 4
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.5, ptr [[TMP49]], align 4
+// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP50]], align 4
+// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP52]], align 4
+// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK3-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP56]], align 4
+// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK3-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK3-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK3-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK3: omp_offload.failed16:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36(ptr [[THIS1]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK3: omp_offload.cont17:
// CHECK3-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A18]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP48:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK3-NEXT: ret i32 [[TMP48]]
+// CHECK3-NEXT: [[TMP60:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK3-NEXT: ret i32 [[TMP60]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -1283,7 +1331,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1300,159 +1348,183 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK9-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK9-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP34]], align 4
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK9-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94(i64 [[TMP4]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR4:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK9-NEXT: [[TMP35:%.*]] = load i64, ptr [[N_CASTED3]], align 8
-// CHECK9-NEXT: [[TMP36:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK9-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK9-NEXT: [[TMP39:%.*]] = load i64, ptr [[N_CASTED3]], align 8
+// CHECK9-NEXT: [[TMP40:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK9-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES7]], ptr align 8 @.offload_sizes.2, i64 24, i1 false)
-// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP35]], ptr [[TMP37]], align 8
-// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP35]], ptr [[TMP38]], align 8
-// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP40]], align 8
-// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP41]], align 8
-// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP42]], align 8
-// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP43]], align 8
-// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 8
-// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK9-NEXT: store i64 [[TMP36]], ptr [[TMP45]], align 8
-// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP39]], ptr [[TMP41]], align 8
+// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP39]], ptr [[TMP42]], align 8
+// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP44]], align 8
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
// CHECK9-NEXT: store ptr null, ptr [[TMP46]], align 8
-// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP50:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP50]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK9-NEXT: [[TMP51:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK9-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP51]], 0
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP47]], align 8
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 8
+// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK9-NEXT: store i64 [[TMP40]], ptr [[TMP49]], align 8
+// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP50]], align 8
+// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP54:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP54]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK9-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK9-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP55]], 0
// CHECK9-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK9-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK9-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK9-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK9-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP52]], 1
-// CHECK9-NEXT: [[TMP53:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK9-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK9-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP56]], 1
+// CHECK9-NEXT: [[TMP57:%.*]] = zext i32 [[ADD14]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK9-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK9-NEXT: store i32 3, ptr [[TMP55]], align 4
-// CHECK9-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP47]], ptr [[TMP56]], align 8
-// CHECK9-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 8
-// CHECK9-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 8
-// CHECK9-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP59]], align 8
-// CHECK9-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP60]], align 8
-// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP61]], align 8
-// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP53]], ptr [[TMP62]], align 8
-// CHECK9-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l98.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK9-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
-// CHECK9-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK9-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP58]], align 4
+// CHECK9-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK9-NEXT: store i32 3, ptr [[TMP59]], align 4
+// CHECK9-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP51]], ptr [[TMP60]], align 8
+// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 8
+// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 8
+// CHECK9-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP63]], align 8
+// CHECK9-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP64]], align 8
+// CHECK9-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP65]], align 8
+// CHECK9-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP57]], ptr [[TMP66]], align 8
+// CHECK9-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP67]], align 8
+// CHECK9-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP68]], align 4
+// CHECK9-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP69]], align 4
+// CHECK9-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP70]], align 4
+// CHECK9-NEXT: [[TMP71:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l98.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK9-NEXT: [[TMP72:%.*]] = icmp ne i32 [[TMP71]], 0
+// CHECK9-NEXT: br i1 [[TMP72]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK9: omp_offload.failed16:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l98(i64 [[TMP35]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR4]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l98(i64 [[TMP39]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK9: omp_offload.cont17:
-// CHECK9-NEXT: [[TMP65:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP65]], ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK9-NEXT: [[TMP66:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP66]], ptr [[N_CASTED19]], align 4
-// CHECK9-NEXT: [[TMP67:%.*]] = load i64, ptr [[N_CASTED19]], align 8
-// CHECK9-NEXT: [[TMP68:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK9-NEXT: store i32 [[TMP68]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK9-NEXT: [[TMP69:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK9-NEXT: [[TMP70:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK9-NEXT: [[TMP73:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP73]], ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK9-NEXT: [[TMP74:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP74]], ptr [[N_CASTED19]], align 4
+// CHECK9-NEXT: [[TMP75:%.*]] = load i64, ptr [[N_CASTED19]], align 8
+// CHECK9-NEXT: [[TMP76:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK9-NEXT: store i32 [[TMP76]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK9-NEXT: [[TMP77:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK9-NEXT: [[TMP78:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK9-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES23]], ptr align 8 @.offload_sizes.5, i64 32, i1 false)
-// CHECK9-NEXT: [[TMP71:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP67]], ptr [[TMP71]], align 8
-// CHECK9-NEXT: [[TMP72:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP67]], ptr [[TMP72]], align 8
-// CHECK9-NEXT: [[TMP73:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP73]], align 8
-// CHECK9-NEXT: [[TMP74:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP74]], align 8
-// CHECK9-NEXT: [[TMP75:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP75]], align 8
-// CHECK9-NEXT: [[TMP76:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP76]], align 8
-// CHECK9-NEXT: [[TMP77:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP77]], align 8
-// CHECK9-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP78]], align 8
-// CHECK9-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
-// CHECK9-NEXT: store i64 [[TMP70]], ptr [[TMP79]], align 8
-// CHECK9-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP80]], align 8
-// CHECK9-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
-// CHECK9-NEXT: store i64 [[TMP69]], ptr [[TMP81]], align 8
-// CHECK9-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
-// CHECK9-NEXT: store i64 [[TMP69]], ptr [[TMP82]], align 8
-// CHECK9-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP83]], align 8
-// CHECK9-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP87:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP87]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK9-NEXT: [[TMP88:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK9-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP88]], 0
+// CHECK9-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP75]], ptr [[TMP79]], align 8
+// CHECK9-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP75]], ptr [[TMP80]], align 8
+// CHECK9-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP81]], align 8
+// CHECK9-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP82]], align 8
+// CHECK9-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP83]], align 8
+// CHECK9-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP84]], align 8
+// CHECK9-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP85]], align 8
+// CHECK9-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP86]], align 8
+// CHECK9-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
+// CHECK9-NEXT: store i64 [[TMP78]], ptr [[TMP87]], align 8
+// CHECK9-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP88]], align 8
+// CHECK9-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
+// CHECK9-NEXT: store i64 [[TMP77]], ptr [[TMP89]], align 8
+// CHECK9-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
+// CHECK9-NEXT: store i64 [[TMP77]], ptr [[TMP90]], align 8
+// CHECK9-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP91]], align 8
+// CHECK9-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP94:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP95:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP95]], ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK9-NEXT: [[TMP96:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK9-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP96]], 0
// CHECK9-NEXT: [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1
// CHECK9-NEXT: [[SUB29:%.*]] = sub nsw i32 [[DIV28]], 1
// CHECK9-NEXT: store i32 [[SUB29]], ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK9-NEXT: [[TMP89:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK9-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP89]], 1
-// CHECK9-NEXT: [[TMP90:%.*]] = zext i32 [[ADD30]] to i64
+// CHECK9-NEXT: [[TMP97:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
+// CHECK9-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP97]], 1
+// CHECK9-NEXT: [[TMP98:%.*]] = zext i32 [[ADD30]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP91]], align 4
-// CHECK9-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK9-NEXT: store i32 4, ptr [[TMP92]], align 4
-// CHECK9-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP84]], ptr [[TMP93]], align 8
-// CHECK9-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP85]], ptr [[TMP94]], align 8
-// CHECK9-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP86]], ptr [[TMP95]], align 8
-// CHECK9-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP96]], align 8
-// CHECK9-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP97]], align 8
-// CHECK9-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP98]], align 8
-// CHECK9-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP90]], ptr [[TMP99]], align 8
-// CHECK9-NEXT: [[TMP100:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK9-NEXT: [[TMP101:%.*]] = icmp ne i32 [[TMP100]], 0
-// CHECK9-NEXT: br i1 [[TMP101]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK9-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP99]], align 4
+// CHECK9-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK9-NEXT: store i32 4, ptr [[TMP100]], align 4
+// CHECK9-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP92]], ptr [[TMP101]], align 8
+// CHECK9-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP93]], ptr [[TMP102]], align 8
+// CHECK9-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP94]], ptr [[TMP103]], align 8
+// CHECK9-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP104]], align 8
+// CHECK9-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP105]], align 8
+// CHECK9-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP106]], align 8
+// CHECK9-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP98]], ptr [[TMP107]], align 8
+// CHECK9-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP108]], align 8
+// CHECK9-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP109]], align 4
+// CHECK9-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP110]], align 4
+// CHECK9-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP111]], align 4
+// CHECK9-NEXT: [[TMP112:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK9-NEXT: [[TMP113:%.*]] = icmp ne i32 [[TMP112]], 0
+// CHECK9-NEXT: br i1 [[TMP113]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK9: omp_offload.failed32:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102(i64 [[TMP67]], i64 [[TMP1]], ptr [[VLA]], i64 [[TMP69]]) #[[ATTR4]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102(i64 [[TMP75]], i64 [[TMP1]], ptr [[VLA]], i64 [[TMP77]]) #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK9: omp_offload.cont33:
-// CHECK9-NEXT: [[TMP102:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP102]])
+// CHECK9-NEXT: [[TMP114:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP114]])
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK9-NEXT: [[TMP103:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP103]])
-// CHECK9-NEXT: [[TMP104:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP104]]
+// CHECK9-NEXT: [[TMP115:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP115]])
+// CHECK9-NEXT: [[TMP116:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP116]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94
@@ -1884,7 +1956,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK9-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK9-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1901,77 +1973,101 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l76.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK9-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK9-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l76.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK9-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l76(ptr [[A]]) #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP16]], align 8
-// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP17]], align 8
-// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP20]], align 8
+// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP21]], align 8
+// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK9-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.11, ptr [[TMP25]], align 8
-// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP26]], align 8
-// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK9-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK9-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l80.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK9-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK9-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK9-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.11, ptr [[TMP29]], align 8
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP30]], align 8
+// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK9-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP36]], align 4
+// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK9-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l80.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK9-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK9-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK9: omp_offload.failed6:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l80(ptr [[A]]) #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK9: omp_offload.cont7:
-// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP32]], align 8
-// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP33]], align 8
-// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP34]], align 8
-// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP40]], align 8
+// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP41]], align 8
+// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP42]], align 8
+// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK9-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
-// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
-// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.14, ptr [[TMP41]], align 8
-// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.15, ptr [[TMP42]], align 8
-// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP43]], align 8
-// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP44]], align 8
-// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK9-NEXT: store i64 10, ptr [[TMP45]], align 8
-// CHECK9-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK9-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK9-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK9-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 8
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 8
+// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.14, ptr [[TMP49]], align 8
+// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.15, ptr [[TMP50]], align 8
+// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP51]], align 8
+// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP52]], align 8
+// CHECK9-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK9-NEXT: store i64 10, ptr [[TMP53]], align 8
+// CHECK9-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK9-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK9-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP56]], align 4
+// CHECK9-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK9-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK9-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK9-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK9: omp_offload.failed13:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84(ptr [[A]]) #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT14]]
@@ -2340,7 +2436,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2357,161 +2453,185 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK11-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK11-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP34]], align 4
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK11-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94(i32 [[TMP3]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR4:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK11-NEXT: [[TMP35:%.*]] = load i32, ptr [[N_CASTED3]], align 4
-// CHECK11-NEXT: [[TMP36:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK11-NEXT: [[TMP37:%.*]] = sext i32 [[TMP36]] to i64
+// CHECK11-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK11-NEXT: [[TMP39:%.*]] = load i32, ptr [[N_CASTED3]], align 4
+// CHECK11-NEXT: [[TMP40:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK11-NEXT: [[TMP41:%.*]] = sext i32 [[TMP40]] to i64
// CHECK11-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES7]], ptr align 4 @.offload_sizes.2, i32 24, i1 false)
-// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP35]], ptr [[TMP38]], align 4
-// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP35]], ptr [[TMP39]], align 4
-// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP40]], align 4
-// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP41]], align 4
-// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP42]], align 4
-// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 4
-// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP45]], align 4
-// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK11-NEXT: store i64 [[TMP37]], ptr [[TMP46]], align 4
-// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP39]], ptr [[TMP42]], align 4
+// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP39]], ptr [[TMP43]], align 4
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP44]], align 4
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
// CHECK11-NEXT: store ptr null, ptr [[TMP47]], align 4
-// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP51:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP51]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK11-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK11-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP52]], 0
+// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 4
+// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP49]], align 4
+// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK11-NEXT: store i64 [[TMP41]], ptr [[TMP50]], align 4
+// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP54:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP55:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP55]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK11-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK11-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP56]], 0
// CHECK11-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK11-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK11-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK11-NEXT: [[TMP53:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK11-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP53]], 1
-// CHECK11-NEXT: [[TMP54:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK11-NEXT: [[TMP57:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK11-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP57]], 1
+// CHECK11-NEXT: [[TMP58:%.*]] = zext i32 [[ADD14]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP55]], align 4
-// CHECK11-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK11-NEXT: store i32 3, ptr [[TMP56]], align 4
-// CHECK11-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 4
-// CHECK11-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 4
-// CHECK11-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP50]], ptr [[TMP59]], align 4
-// CHECK11-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP60]], align 4
-// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP61]], align 4
-// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP62]], align 4
-// CHECK11-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP54]], ptr [[TMP63]], align 8
-// CHECK11-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l98.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK11-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
-// CHECK11-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK11-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP59]], align 4
+// CHECK11-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK11-NEXT: store i32 3, ptr [[TMP60]], align 4
+// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 4
+// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 4
+// CHECK11-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP54]], ptr [[TMP63]], align 4
+// CHECK11-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP64]], align 4
+// CHECK11-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP65]], align 4
+// CHECK11-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP66]], align 4
+// CHECK11-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP58]], ptr [[TMP67]], align 8
+// CHECK11-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP68]], align 8
+// CHECK11-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK11-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP70]], align 4
+// CHECK11-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP71]], align 4
+// CHECK11-NEXT: [[TMP72:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l98.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK11-NEXT: [[TMP73:%.*]] = icmp ne i32 [[TMP72]], 0
+// CHECK11-NEXT: br i1 [[TMP73]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK11: omp_offload.failed16:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l98(i32 [[TMP35]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR4]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l98(i32 [[TMP39]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK11: omp_offload.cont17:
-// CHECK11-NEXT: [[TMP66:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP66]], ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK11-NEXT: [[TMP67:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP67]], ptr [[N_CASTED19]], align 4
-// CHECK11-NEXT: [[TMP68:%.*]] = load i32, ptr [[N_CASTED19]], align 4
-// CHECK11-NEXT: [[TMP69:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
-// CHECK11-NEXT: store i32 [[TMP69]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK11-NEXT: [[TMP70:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK11-NEXT: [[TMP71:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK11-NEXT: [[TMP72:%.*]] = sext i32 [[TMP71]] to i64
+// CHECK11-NEXT: [[TMP74:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP74]], ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK11-NEXT: [[TMP75:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP75]], ptr [[N_CASTED19]], align 4
+// CHECK11-NEXT: [[TMP76:%.*]] = load i32, ptr [[N_CASTED19]], align 4
+// CHECK11-NEXT: [[TMP77:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4
+// CHECK11-NEXT: store i32 [[TMP77]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK11-NEXT: [[TMP78:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK11-NEXT: [[TMP79:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK11-NEXT: [[TMP80:%.*]] = sext i32 [[TMP79]] to i64
// CHECK11-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES23]], ptr align 4 @.offload_sizes.5, i32 32, i1 false)
-// CHECK11-NEXT: [[TMP73:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP68]], ptr [[TMP73]], align 4
-// CHECK11-NEXT: [[TMP74:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP68]], ptr [[TMP74]], align 4
-// CHECK11-NEXT: [[TMP75:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP75]], align 4
-// CHECK11-NEXT: [[TMP76:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP76]], align 4
-// CHECK11-NEXT: [[TMP77:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP77]], align 4
-// CHECK11-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP78]], align 4
-// CHECK11-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP79]], align 4
-// CHECK11-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP80]], align 4
-// CHECK11-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
-// CHECK11-NEXT: store i64 [[TMP72]], ptr [[TMP81]], align 4
-// CHECK11-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP82]], align 4
-// CHECK11-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
-// CHECK11-NEXT: store i32 [[TMP70]], ptr [[TMP83]], align 4
-// CHECK11-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
-// CHECK11-NEXT: store i32 [[TMP70]], ptr [[TMP84]], align 4
-// CHECK11-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP85]], align 4
-// CHECK11-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP89:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP89]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK11-NEXT: [[TMP90:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK11-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP90]], 0
+// CHECK11-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP76]], ptr [[TMP81]], align 4
+// CHECK11-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP76]], ptr [[TMP82]], align 4
+// CHECK11-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP83]], align 4
+// CHECK11-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP84]], align 4
+// CHECK11-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP85]], align 4
+// CHECK11-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP86]], align 4
+// CHECK11-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP87]], align 4
+// CHECK11-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP88]], align 4
+// CHECK11-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2
+// CHECK11-NEXT: store i64 [[TMP80]], ptr [[TMP89]], align 4
+// CHECK11-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP90]], align 4
+// CHECK11-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3
+// CHECK11-NEXT: store i32 [[TMP78]], ptr [[TMP91]], align 4
+// CHECK11-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3
+// CHECK11-NEXT: store i32 [[TMP78]], ptr [[TMP92]], align 4
+// CHECK11-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP93]], align 4
+// CHECK11-NEXT: [[TMP94:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP95:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP96:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP97:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP97]], ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK11-NEXT: [[TMP98:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK11-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP98]], 0
// CHECK11-NEXT: [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1
// CHECK11-NEXT: [[SUB29:%.*]] = sub nsw i32 [[DIV28]], 1
// CHECK11-NEXT: store i32 [[SUB29]], ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK11-NEXT: [[TMP91:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
-// CHECK11-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP91]], 1
-// CHECK11-NEXT: [[TMP92:%.*]] = zext i32 [[ADD30]] to i64
+// CHECK11-NEXT: [[TMP99:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4
+// CHECK11-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP99]], 1
+// CHECK11-NEXT: [[TMP100:%.*]] = zext i32 [[ADD30]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP93]], align 4
-// CHECK11-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK11-NEXT: store i32 4, ptr [[TMP94]], align 4
-// CHECK11-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP86]], ptr [[TMP95]], align 4
-// CHECK11-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP87]], ptr [[TMP96]], align 4
-// CHECK11-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP88]], ptr [[TMP97]], align 4
-// CHECK11-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP98]], align 4
-// CHECK11-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP99]], align 4
-// CHECK11-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP100]], align 4
-// CHECK11-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP92]], ptr [[TMP101]], align 8
-// CHECK11-NEXT: [[TMP102:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK11-NEXT: [[TMP103:%.*]] = icmp ne i32 [[TMP102]], 0
-// CHECK11-NEXT: br i1 [[TMP103]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK11-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP101]], align 4
+// CHECK11-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK11-NEXT: store i32 4, ptr [[TMP102]], align 4
+// CHECK11-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP94]], ptr [[TMP103]], align 4
+// CHECK11-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP95]], ptr [[TMP104]], align 4
+// CHECK11-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP96]], ptr [[TMP105]], align 4
+// CHECK11-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP106]], align 4
+// CHECK11-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP107]], align 4
+// CHECK11-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP108]], align 4
+// CHECK11-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP100]], ptr [[TMP109]], align 8
+// CHECK11-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP110]], align 8
+// CHECK11-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP111]], align 4
+// CHECK11-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP112]], align 4
+// CHECK11-NEXT: [[TMP113:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP113]], align 4
+// CHECK11-NEXT: [[TMP114:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK11-NEXT: [[TMP115:%.*]] = icmp ne i32 [[TMP114]], 0
+// CHECK11-NEXT: br i1 [[TMP115]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK11: omp_offload.failed32:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102(i32 [[TMP68]], i32 [[TMP0]], ptr [[VLA]], i32 [[TMP70]]) #[[ATTR4]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l102(i32 [[TMP76]], i32 [[TMP0]], ptr [[VLA]], i32 [[TMP78]]) #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK11: omp_offload.cont33:
-// CHECK11-NEXT: [[TMP104:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP104]])
+// CHECK11-NEXT: [[TMP116:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP116]])
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK11-NEXT: [[TMP105:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP105]])
-// CHECK11-NEXT: [[TMP106:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP106]]
+// CHECK11-NEXT: [[TMP117:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP117]])
+// CHECK11-NEXT: [[TMP118:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP118]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94
@@ -2940,7 +3060,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK11-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK11-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2957,77 +3077,101 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK11-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK11-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l76.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK11-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP16]], align 4
+// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK11-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l76.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK11-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l76(ptr [[A]]) #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP16]], align 4
-// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP17]], align 4
-// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP20]], align 4
+// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP21]], align 4
+// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK11-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.11, ptr [[TMP25]], align 4
-// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP26]], align 4
-// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK11-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK11-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l80.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK11-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK11-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK11-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.11, ptr [[TMP29]], align 4
+// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP30]], align 4
+// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK11-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP36]], align 4
+// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK11-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l80.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK11-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK11-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK11: omp_offload.failed6:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l80(ptr [[A]]) #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK11: omp_offload.cont7:
-// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP32]], align 4
-// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP33]], align 4
-// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP34]], align 4
-// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP40]], align 4
+// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP41]], align 4
+// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP42]], align 4
+// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK11-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 4
-// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 4
-// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.14, ptr [[TMP41]], align 4
-// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.15, ptr [[TMP42]], align 4
-// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP44]], align 4
-// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK11-NEXT: store i64 10, ptr [[TMP45]], align 8
-// CHECK11-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK11-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK11-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK11-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 4
+// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 4
+// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.14, ptr [[TMP49]], align 4
+// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.15, ptr [[TMP50]], align 4
+// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP52]], align 4
+// CHECK11-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK11-NEXT: store i64 10, ptr [[TMP53]], align 8
+// CHECK11-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK11-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK11-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP56]], align 4
+// CHECK11-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK11-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK11-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK11-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK11: omp_offload.failed13:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84(ptr [[A]]) #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT14]]
diff --git a/clang/test/OpenMP/target_teams_distribute_simd_firstprivate_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_simd_firstprivate_codegen.cpp
index 0d85f1b765a56..3fc376bbfc9c3 100644
--- a/clang/test/OpenMP/target_teams_distribute_simd_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_simd_firstprivate_codegen.cpp
@@ -303,7 +303,7 @@ int main() {
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 5, ptr [[TMP22]], align 4
// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -320,9 +320,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP32]], align 4
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK1-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK1-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92(ptr @vec, i64 [[TMP1]], ptr @s_arr, ptr @var, i64 [[TMP3]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -581,7 +589,7 @@ int main() {
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -598,27 +606,35 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP25]], align 8
// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK1-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP29]], align 4
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK1-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56(ptr [[VEC]], i64 [[TMP2]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK1-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK1: arraydestroy.body:
-// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP29]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP33]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK1-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK1-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK1: arraydestroy.done2:
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK1-NEXT: [[TMP30:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK1-NEXT: ret i32 [[TMP30]]
+// CHECK1-NEXT: [[TMP34:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK1-NEXT: ret i32 [[TMP34]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN2StC2Ev
@@ -1102,7 +1118,7 @@ int main() {
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 5, ptr [[TMP22]], align 4
// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1119,9 +1135,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4
// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP29]], align 8
-// CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK3-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK3-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK3-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92(ptr @vec, i32 [[TMP1]], ptr @s_arr, ptr @var, i32 [[TMP3]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1378,7 +1402,7 @@ int main() {
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1395,27 +1419,35 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP25]], align 4
// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP26]], align 8
-// CHECK3-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK3-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK3-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56(ptr [[VEC]], i32 [[TMP2]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK3-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK3-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK3: arraydestroy.body:
-// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP29]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP33]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK3-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK3-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK3: arraydestroy.done2:
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK3-NEXT: [[TMP30:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK3-NEXT: ret i32 [[TMP30]]
+// CHECK3-NEXT: [[TMP34:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK3-NEXT: ret i32 [[TMP34]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN2StC2Ev
diff --git a/clang/test/OpenMP/target_teams_distribute_simd_lastprivate_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_simd_lastprivate_codegen.cpp
index 6b28f798b2ed6..f0786333ad125 100644
--- a/clang/test/OpenMP/target_teams_distribute_simd_lastprivate_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_simd_lastprivate_codegen.cpp
@@ -586,7 +586,7 @@ int main() {
// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 5, ptr [[TMP24]], align 4
// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -603,9 +603,17 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP31]], align 8
-// CHECK9-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK9-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP34]], align 4
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK9-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94(ptr [[VEC]], i64 [[TMP2]], ptr [[S_ARR]], ptr [[TMP3]], i64 [[TMP5]]) #[[ATTR5:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -613,18 +621,18 @@ int main() {
// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiET_v()
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP34]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP38]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR5]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR5]]
-// CHECK9-NEXT: [[TMP35:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP35]]
+// CHECK9-NEXT: [[TMP39:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP39]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -900,7 +908,7 @@ int main() {
// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -917,27 +925,35 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP25]], align 8
// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP26]], align 8
-// CHECK9-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK9-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP29]], align 4
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK9-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK9-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49(ptr [[VEC]], i64 [[TMP2]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR5]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP29]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP33]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR5]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR5]]
-// CHECK9-NEXT: [[TMP30:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP30]]
+// CHECK9-NEXT: [[TMP34:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP34]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
@@ -1292,7 +1308,7 @@ int main() {
// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 5, ptr [[TMP24]], align 4
// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1309,9 +1325,17 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP31]], align 8
-// CHECK11-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK11-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP34]], align 4
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK11-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94(ptr [[VEC]], i32 [[TMP2]], ptr [[S_ARR]], ptr [[TMP3]], i32 [[TMP5]]) #[[ATTR5:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1319,18 +1343,18 @@ int main() {
// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiET_v()
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP34]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP38]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR5]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR5]]
-// CHECK11-NEXT: [[TMP35:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP35]]
+// CHECK11-NEXT: [[TMP39:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP39]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -1604,7 +1628,7 @@ int main() {
// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 4, ptr [[TMP19]], align 4
// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1621,27 +1645,35 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP25]], align 4
// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP26]], align 8
-// CHECK11-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK11-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP29]], align 4
+// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK11-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK11-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49(ptr [[VEC]], i32 [[TMP2]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR5]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP29]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP33]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR5]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR5]]
-// CHECK11-NEXT: [[TMP30:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP30]]
+// CHECK11-NEXT: [[TMP34:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP34]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
diff --git a/clang/test/OpenMP/target_teams_distribute_simd_private_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_simd_private_codegen.cpp
index 17fa78da1b5ee..e4f62469a8b26 100644
--- a/clang/test/OpenMP/target_teams_distribute_simd_private_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_simd_private_codegen.cpp
@@ -240,7 +240,7 @@ int main() {
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -257,9 +257,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l91.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l91.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l91() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -408,7 +416,7 @@ int main() {
// CHECK1-NEXT: store ptr undef, ptr [[_TMP1]], align 8
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -425,27 +433,35 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK1-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK1: arraydestroy.body:
-// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK1-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK1-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK1: arraydestroy.done2:
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK1-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK1-NEXT: ret i32 [[TMP12]]
+// CHECK1-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK1-NEXT: ret i32 [[TMP16]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN1SIiEC1Ev
@@ -773,7 +789,7 @@ int main() {
// CHECK3-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -790,9 +806,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l91.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK3-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l91.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK3-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l91() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -939,7 +963,7 @@ int main() {
// CHECK3-NEXT: store ptr undef, ptr [[_TMP1]], align 4
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -956,27 +980,35 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK3-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK3-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK3-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK3-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK3: arraydestroy.body:
-// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK3-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK3-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK3: arraydestroy.done2:
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK3-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK3-NEXT: ret i32 [[TMP12]]
+// CHECK3-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK3-NEXT: ret i32 [[TMP16]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN1SIiEC1Ev
diff --git a/clang/test/OpenMP/target_teams_distribute_simd_reduction_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_simd_reduction_codegen.cpp
index 338c177446f4e..38adc315e8e82 100644
--- a/clang/test/OpenMP/target_teams_distribute_simd_reduction_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_simd_reduction_codegen.cpp
@@ -97,7 +97,7 @@ int main() {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -114,9 +114,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l60.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l60.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l60(ptr @_ZZ4mainE5sivar) #[[ATTR3:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -273,7 +281,7 @@ int main() {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -290,9 +298,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32(ptr [[T_VAR]]) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -453,7 +469,7 @@ int main() {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -470,9 +486,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l60.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l60.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l60(ptr @_ZZ4mainE5sivar) #[[ATTR3:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -629,7 +653,7 @@ int main() {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -646,9 +670,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32(ptr [[T_VAR]]) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/target_teams_map_codegen.cpp b/clang/test/OpenMP/target_teams_map_codegen.cpp
index c81730a7e2630..69aa5002ca591 100644
--- a/clang/test/OpenMP/target_teams_map_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_map_codegen.cpp
@@ -100,7 +100,7 @@ void mapInt128() {
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP8]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP8]], align 4
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 2, ptr [[TMP9]], align 4
// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -117,9 +117,17 @@ void mapInt128() {
// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z14mapWithPrivatev_l27.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP18:%.*]] = icmp ne i32 [[TMP17]], 0
-// CHECK1-NEXT: br i1 [[TMP18]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP17]], align 8
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP19]], align 4
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP20]], align 4
+// CHECK1-NEXT: [[TMP21:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z14mapWithPrivatev_l27.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP22:%.*]] = icmp ne i32 [[TMP21]], 0
+// CHECK1-NEXT: br i1 [[TMP22]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z14mapWithPrivatev_l27() #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -170,7 +178,7 @@ void mapInt128() {
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP8]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP8]], align 4
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 2, ptr [[TMP9]], align 4
// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -187,9 +195,17 @@ void mapInt128() {
// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z19mapWithFirstprivatev_l33.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP18:%.*]] = icmp ne i32 [[TMP17]], 0
-// CHECK1-NEXT: br i1 [[TMP18]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP17]], align 8
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP19]], align 4
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP20]], align 4
+// CHECK1-NEXT: [[TMP21:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z19mapWithFirstprivatev_l33.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP22:%.*]] = icmp ne i32 [[TMP21]], 0
+// CHECK1-NEXT: br i1 [[TMP22]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z19mapWithFirstprivatev_l33(ptr [[X]], ptr [[Y]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -256,7 +272,7 @@ void mapInt128() {
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP8]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP8]], align 4
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 2, ptr [[TMP9]], align 4
// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -273,9 +289,17 @@ void mapInt128() {
// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16mapWithReductionv_l39.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP18:%.*]] = icmp ne i32 [[TMP17]], 0
-// CHECK1-NEXT: br i1 [[TMP18]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP17]], align 8
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP19]], align 4
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP20]], align 4
+// CHECK1-NEXT: [[TMP21:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16mapWithReductionv_l39.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP22:%.*]] = icmp ne i32 [[TMP21]], 0
+// CHECK1-NEXT: br i1 [[TMP22]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16mapWithReductionv_l39(ptr [[X]], ptr [[Y]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -392,7 +416,7 @@ void mapInt128() {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -409,9 +433,17 @@ void mapInt128() {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z7mapFromv_l45.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z7mapFromv_l45.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z7mapFromv_l45(ptr [[X]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -462,7 +494,7 @@ void mapInt128() {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -479,9 +511,17 @@ void mapInt128() {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5mapTov_l51.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5mapTov_l51.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5mapTov_l51(ptr [[X]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -532,7 +572,7 @@ void mapInt128() {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -549,9 +589,17 @@ void mapInt128() {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z8mapAllocv_l57.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z8mapAllocv_l57.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z8mapAllocv_l57(ptr [[X]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -619,7 +667,7 @@ void mapInt128() {
// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP11]], align 4
// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 3, ptr [[TMP12]], align 4
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -636,55 +684,71 @@ void mapInt128() {
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z8mapArrayv_l63.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP21]], align 4
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP22]], align 4
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP23]], align 4
+// CHECK1-NEXT: [[TMP24:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z8mapArrayv_l63.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP25:%.*]] = icmp ne i32 [[TMP24]], 0
+// CHECK1-NEXT: br i1 [[TMP25]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z8mapArrayv_l63(ptr [[Y]], ptr [[Z]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[Y]], ptr [[TMP22]], align 8
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[Y]], ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 1
-// CHECK1-NEXT: store ptr [[Z]], ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 1
-// CHECK1-NEXT: store ptr [[Z]], ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[X]], ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[X]], ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP30]], align 8
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[Y]], ptr [[TMP26]], align 8
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[Y]], ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 1
+// CHECK1-NEXT: store ptr [[Z]], ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 1
+// CHECK1-NEXT: store ptr [[Z]], ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[X]], ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[X]], ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 2
+// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS4:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP33]], align 4
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 1
-// CHECK1-NEXT: store i32 3, ptr [[TMP34]], align 4
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP31]], ptr [[TMP35]], align 8
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP32]], ptr [[TMP36]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.22, ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.23, ptr [[TMP38]], align 8
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP41]], align 8
-// CHECK1-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z8mapArrayv_l65.region_id, ptr [[KERNEL_ARGS4]])
-// CHECK1-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
-// CHECK1-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED5:%.*]], label [[OMP_OFFLOAD_CONT6:%.*]]
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 1
+// CHECK1-NEXT: store i32 3, ptr [[TMP38]], align 4
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.22, ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.23, ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP44]], align 8
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP45]], align 8
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP46]], align 8
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP47]], align 4
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP48]], align 4
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP49]], align 4
+// CHECK1-NEXT: [[TMP50:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z8mapArrayv_l65.region_id, ptr [[KERNEL_ARGS4]])
+// CHECK1-NEXT: [[TMP51:%.*]] = icmp ne i32 [[TMP50]], 0
+// CHECK1-NEXT: br i1 [[TMP51]], label [[OMP_OFFLOAD_FAILED5:%.*]], label [[OMP_OFFLOAD_CONT6:%.*]]
// CHECK1: omp_offload.failed5:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z8mapArrayv_l65(ptr [[Y]], ptr [[Z]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT6]]
@@ -966,7 +1030,7 @@ void mapInt128() {
// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP11]], align 4
// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 3, ptr [[TMP12]], align 4
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -983,55 +1047,71 @@ void mapInt128() {
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9mapInt128v_l72.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP21]], align 4
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP22]], align 4
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP23]], align 4
+// CHECK1-NEXT: [[TMP24:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9mapInt128v_l72.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP25:%.*]] = icmp ne i32 [[TMP24]], 0
+// CHECK1-NEXT: br i1 [[TMP25]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9mapInt128v_l72(ptr [[Y]], ptr [[Z]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[Y]], ptr [[TMP22]], align 8
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[Y]], ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 1
-// CHECK1-NEXT: store ptr [[Z]], ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 1
-// CHECK1-NEXT: store ptr [[Z]], ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[X]], ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[X]], ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP30]], align 8
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[Y]], ptr [[TMP26]], align 8
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[Y]], ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 1
+// CHECK1-NEXT: store ptr [[Z]], ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 1
+// CHECK1-NEXT: store ptr [[Z]], ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[X]], ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[X]], ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 2
+// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS4:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP33]], align 4
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 1
-// CHECK1-NEXT: store i32 3, ptr [[TMP34]], align 4
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP31]], ptr [[TMP35]], align 8
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP32]], ptr [[TMP36]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.30, ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.31, ptr [[TMP38]], align 8
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP41]], align 8
-// CHECK1-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9mapInt128v_l74.region_id, ptr [[KERNEL_ARGS4]])
-// CHECK1-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
-// CHECK1-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED5:%.*]], label [[OMP_OFFLOAD_CONT6:%.*]]
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 1
+// CHECK1-NEXT: store i32 3, ptr [[TMP38]], align 4
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.30, ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.31, ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP44]], align 8
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP45]], align 8
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP46]], align 8
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP47]], align 4
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP48]], align 4
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP49]], align 4
+// CHECK1-NEXT: [[TMP50:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9mapInt128v_l74.region_id, ptr [[KERNEL_ARGS4]])
+// CHECK1-NEXT: [[TMP51:%.*]] = icmp ne i32 [[TMP50]], 0
+// CHECK1-NEXT: br i1 [[TMP51]], label [[OMP_OFFLOAD_FAILED5:%.*]], label [[OMP_OFFLOAD_CONT6:%.*]]
// CHECK1: omp_offload.failed5:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9mapInt128v_l74(ptr [[Y]], ptr [[Z]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT6]]
@@ -1254,7 +1334,7 @@ void mapInt128() {
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP8]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP8]], align 4
// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 2, ptr [[TMP9]], align 4
// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1271,9 +1351,17 @@ void mapInt128() {
// CHECK3-NEXT: store ptr null, ptr [[TMP15]], align 4
// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP16]], align 8
-// CHECK3-NEXT: [[TMP17:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z14mapWithPrivatev_l27.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP18:%.*]] = icmp ne i32 [[TMP17]], 0
-// CHECK3-NEXT: br i1 [[TMP18]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP17]], align 8
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP19]], align 4
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP20]], align 4
+// CHECK3-NEXT: [[TMP21:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z14mapWithPrivatev_l27.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP22:%.*]] = icmp ne i32 [[TMP21]], 0
+// CHECK3-NEXT: br i1 [[TMP22]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z14mapWithPrivatev_l27() #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1324,7 +1412,7 @@ void mapInt128() {
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP8]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP8]], align 4
// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 2, ptr [[TMP9]], align 4
// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1341,9 +1429,17 @@ void mapInt128() {
// CHECK3-NEXT: store ptr null, ptr [[TMP15]], align 4
// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP16]], align 8
-// CHECK3-NEXT: [[TMP17:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z19mapWithFirstprivatev_l33.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP18:%.*]] = icmp ne i32 [[TMP17]], 0
-// CHECK3-NEXT: br i1 [[TMP18]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP17]], align 8
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP19]], align 4
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP20]], align 4
+// CHECK3-NEXT: [[TMP21:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z19mapWithFirstprivatev_l33.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP22:%.*]] = icmp ne i32 [[TMP21]], 0
+// CHECK3-NEXT: br i1 [[TMP22]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z19mapWithFirstprivatev_l33(ptr [[X]], ptr [[Y]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1410,7 +1506,7 @@ void mapInt128() {
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP8]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP8]], align 4
// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 2, ptr [[TMP9]], align 4
// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1427,9 +1523,17 @@ void mapInt128() {
// CHECK3-NEXT: store ptr null, ptr [[TMP15]], align 4
// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP16]], align 8
-// CHECK3-NEXT: [[TMP17:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16mapWithReductionv_l39.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP18:%.*]] = icmp ne i32 [[TMP17]], 0
-// CHECK3-NEXT: br i1 [[TMP18]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP17]], align 8
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP19]], align 4
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP20]], align 4
+// CHECK3-NEXT: [[TMP21:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16mapWithReductionv_l39.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP22:%.*]] = icmp ne i32 [[TMP21]], 0
+// CHECK3-NEXT: br i1 [[TMP22]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16mapWithReductionv_l39(ptr [[X]], ptr [[Y]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1546,7 +1650,7 @@ void mapInt128() {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1563,9 +1667,17 @@ void mapInt128() {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z7mapFromv_l45.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z7mapFromv_l45.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z7mapFromv_l45(ptr [[X]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1616,7 +1728,7 @@ void mapInt128() {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1633,9 +1745,17 @@ void mapInt128() {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5mapTov_l51.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5mapTov_l51.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5mapTov_l51(ptr [[X]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1686,7 +1806,7 @@ void mapInt128() {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1703,9 +1823,17 @@ void mapInt128() {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z8mapAllocv_l57.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z8mapAllocv_l57.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z8mapAllocv_l57(ptr [[X]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1773,7 +1901,7 @@ void mapInt128() {
// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP11]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP11]], align 4
// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 3, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1790,55 +1918,71 @@ void mapInt128() {
// CHECK3-NEXT: store ptr null, ptr [[TMP18]], align 4
// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP19]], align 8
-// CHECK3-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z8mapArrayv_l63.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK3-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP20]], align 8
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP21]], align 4
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP22]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP23]], align 4
+// CHECK3-NEXT: [[TMP24:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z8mapArrayv_l63.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP25:%.*]] = icmp ne i32 [[TMP24]], 0
+// CHECK3-NEXT: br i1 [[TMP25]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z8mapArrayv_l63(ptr [[Y]], ptr [[Z]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
-// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[Y]], ptr [[TMP22]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[Y]], ptr [[TMP23]], align 4
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP24]], align 4
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 1
-// CHECK3-NEXT: store ptr [[Z]], ptr [[TMP25]], align 4
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 1
-// CHECK3-NEXT: store ptr [[Z]], ptr [[TMP26]], align 4
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 1
-// CHECK3-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[X]], ptr [[TMP28]], align 4
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[X]], ptr [[TMP29]], align 4
-// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 2
-// CHECK3-NEXT: store ptr null, ptr [[TMP30]], align 4
-// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[Y]], ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[Y]], ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 1
+// CHECK3-NEXT: store ptr [[Z]], ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 1
+// CHECK3-NEXT: store ptr [[Z]], ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 1
+// CHECK3-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[X]], ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[X]], ptr [[TMP33]], align 4
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 2
+// CHECK3-NEXT: store ptr null, ptr [[TMP34]], align 4
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS4:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP33]], align 4
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 1
-// CHECK3-NEXT: store i32 3, ptr [[TMP34]], align 4
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP31]], ptr [[TMP35]], align 4
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP32]], ptr [[TMP36]], align 4
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.22, ptr [[TMP37]], align 4
-// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.23, ptr [[TMP38]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP39]], align 4
-// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP40]], align 4
-// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP41]], align 8
-// CHECK3-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z8mapArrayv_l65.region_id, ptr [[KERNEL_ARGS4]])
-// CHECK3-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
-// CHECK3-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED5:%.*]], label [[OMP_OFFLOAD_CONT6:%.*]]
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 1
+// CHECK3-NEXT: store i32 3, ptr [[TMP38]], align 4
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 4
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.22, ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.23, ptr [[TMP42]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP43]], align 4
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP44]], align 4
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP45]], align 8
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP46]], align 8
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP47]], align 4
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP48]], align 4
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP49]], align 4
+// CHECK3-NEXT: [[TMP50:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z8mapArrayv_l65.region_id, ptr [[KERNEL_ARGS4]])
+// CHECK3-NEXT: [[TMP51:%.*]] = icmp ne i32 [[TMP50]], 0
+// CHECK3-NEXT: br i1 [[TMP51]], label [[OMP_OFFLOAD_FAILED5:%.*]], label [[OMP_OFFLOAD_CONT6:%.*]]
// CHECK3: omp_offload.failed5:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z8mapArrayv_l65(ptr [[Y]], ptr [[Z]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT6]]
diff --git a/clang/test/OpenMP/target_teams_num_teams_codegen.cpp b/clang/test/OpenMP/target_teams_num_teams_codegen.cpp
index 6fcacfe3a30aa..b25d10dac42c2 100644
--- a/clang/test/OpenMP/target_teams_num_teams_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_num_teams_codegen.cpp
@@ -252,70 +252,87 @@ int bar(int n){
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP17]], 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP18]], align 4
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK1-NEXT: store i32 3, ptr [[TMP19]], align 4
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP15]], ptr [[TMP20]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP16]], ptr [[TMP21]], align 8
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP22]], align 8
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP19]], align 4
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK1-NEXT: store i32 3, ptr [[TMP20]], align 4
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP15]], ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP16]], ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK1-NEXT: store ptr null, ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 [[TMP17]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK1-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP26]], align 8
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [[TMP18]], ptr [[TMP29]], align 4
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP31]], align 4
+// CHECK1-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 [[TMP17]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
+// CHECK1-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121(ptr [[THIS1]], i64 [[TMP3]], i64 [[TMP5]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_S1]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP30]], align 8
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP31]], align 8
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP35]], align 8
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP36]], align 8
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP34]], align 4
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP35]], align 4
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP32]], ptr [[TMP36]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP33]], ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.2, ptr [[TMP38]], align 8
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP41]], align 8
-// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP42]], align 8
-// CHECK1-NEXT: [[TMP43:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1024, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126.region_id, ptr [[KERNEL_ARGS6]])
-// CHECK1-NEXT: [[TMP44:%.*]] = icmp ne i32 [[TMP43]], 0
-// CHECK1-NEXT: br i1 [[TMP44]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP39]], align 4
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP40]], align 4
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP37]], ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP38]], ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.2, ptr [[TMP43]], align 8
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP44]], align 8
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP45]], align 8
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP46]], align 8
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP47]], align 8
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP48]], align 8
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 1024, i32 0, i32 0], ptr [[TMP49]], align 4
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP50]], align 4
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP51]], align 4
+// CHECK1-NEXT: [[TMP52:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1024, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126.region_id, ptr [[KERNEL_ARGS6]])
+// CHECK1-NEXT: [[TMP53:%.*]] = icmp ne i32 [[TMP52]], 0
+// CHECK1-NEXT: br i1 [[TMP53]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK1: omp_offload.failed7:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK1: omp_offload.cont8:
// CHECK1-NEXT: [[A9:%.*]] = getelementptr inbounds [[STRUCT_S1]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP45:%.*]] = load double, ptr [[A9]], align 8
-// CHECK1-NEXT: [[CONV:%.*]] = fptosi double [[TMP45]] to i32
+// CHECK1-NEXT: [[TMP54:%.*]] = load double, ptr [[A9]], align 8
+// CHECK1-NEXT: [[CONV:%.*]] = fptosi double [[TMP54]] to i32
// CHECK1-NEXT: ret i32 [[CONV]]
//
//
@@ -348,75 +365,93 @@ int bar(int n){
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK1-NEXT: [[TMP9:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP8]], 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP9]], align 4
-// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP10]], align 4
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP6]], ptr [[TMP11]], align 8
-// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP7]], ptr [[TMP12]], align 8
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.5, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP6]], ptr [[TMP12]], align 8
+// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP7]], ptr [[TMP13]], align 8
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.5, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP15]], align 8
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP8]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l104.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
-// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP18]], align 8
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [[TMP9]], ptr [[TMP20]], align 4
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP21]], align 4
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP22]], align 4
+// CHECK1-NEXT: [[TMP23:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP8]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l104.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP24:%.*]] = icmp ne i32 [[TMP23]], 0
+// CHECK1-NEXT: br i1 [[TMP24]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l104(i64 [[TMP2]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
-// CHECK1-NEXT: [[TMP20:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 32, [[TMP20]]
+// CHECK1-NEXT: [[TMP25:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 32, [[TMP25]]
// CHECK1-NEXT: store i32 [[ADD]], ptr [[DOTCAPTURE_EXPR_1]], align 4
-// CHECK1-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
-// CHECK1-NEXT: store i32 [[TMP21]], ptr [[DOTCAPTURE_EXPR__CASTED2]], align 4
-// CHECK1-NEXT: [[TMP22:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED2]], align 8
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP22]], ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP22]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP28:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
+// CHECK1-NEXT: store i32 [[TMP26]], ptr [[DOTCAPTURE_EXPR__CASTED2]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED2]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP27]], ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP27]], ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP33:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
+// CHECK1-NEXT: [[TMP34:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP33]], 0
// CHECK1-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP29]], align 4
-// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP30]], align 4
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP26]], ptr [[TMP31]], align 8
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP27]], ptr [[TMP32]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.8, ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP35]], align 8
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP36]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP28]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108.region_id, ptr [[KERNEL_ARGS6]])
-// CHECK1-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK1-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP35]], align 4
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP36]], align 4
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP31]], ptr [[TMP37]], align 8
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP32]], ptr [[TMP38]], align 8
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.8, ptr [[TMP39]], align 8
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP43]], align 8
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP44]], align 8
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [[TMP34]], ptr [[TMP45]], align 4
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP46]], align 4
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP47]], align 4
+// CHECK1-NEXT: [[TMP48:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP33]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108.region_id, ptr [[KERNEL_ARGS6]])
+// CHECK1-NEXT: [[TMP49:%.*]] = icmp ne i32 [[TMP48]], 0
+// CHECK1-NEXT: br i1 [[TMP49]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK1: omp_offload.failed7:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108(i64 [[TMP22]]) #[[ATTR2]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108(i64 [[TMP27]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK1: omp_offload.cont8:
-// CHECK1-NEXT: [[TMP40:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: [[ADD9:%.*]] = add nsw i32 [[TMP40]], 1
+// CHECK1-NEXT: [[TMP50:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: [[ADD9:%.*]] = add nsw i32 [[TMP50]], 1
// CHECK1-NEXT: ret i32 [[ADD9]]
//
//
@@ -437,7 +472,7 @@ int bar(int n){
// CHECK1-NEXT: store i32 0, ptr [[A]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -454,75 +489,92 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 20, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l88.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [i32 20, i32 0, i32 0], ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 20, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l88.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l88() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: store i16 1, ptr [[B]], align 2
-// CHECK1-NEXT: [[TMP11:%.*]] = load i16, ptr [[B]], align 2
-// CHECK1-NEXT: store i16 [[TMP11]], ptr [[DOTCAPTURE_EXPR_]], align 2
-// CHECK1-NEXT: [[TMP12:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: store i32 [[TMP12]], ptr [[A_CASTED]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = load i64, ptr [[A_CASTED]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = load i16, ptr [[B]], align 2
-// CHECK1-NEXT: store i16 [[TMP14]], ptr [[B_CASTED]], align 2
-// CHECK1-NEXT: [[TMP15:%.*]] = load i64, ptr [[B_CASTED]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = load i16, ptr [[DOTCAPTURE_EXPR_]], align 2
-// CHECK1-NEXT: store i16 [[TMP16]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 2
-// CHECK1-NEXT: [[TMP17:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP13]], ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP13]], ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP15]], ptr [[TMP21]], align 8
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP15]], ptr [[TMP22]], align 8
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
-// CHECK1-NEXT: store i64 [[TMP17]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
-// CHECK1-NEXT: store i64 [[TMP17]], ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP29:%.*]] = load i16, ptr [[DOTCAPTURE_EXPR_]], align 2
-// CHECK1-NEXT: [[TMP30:%.*]] = sext i16 [[TMP29]] to i32
+// CHECK1-NEXT: [[TMP15:%.*]] = load i16, ptr [[B]], align 2
+// CHECK1-NEXT: store i16 [[TMP15]], ptr [[DOTCAPTURE_EXPR_]], align 2
+// CHECK1-NEXT: [[TMP16:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: store i32 [[TMP16]], ptr [[A_CASTED]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = load i64, ptr [[A_CASTED]], align 8
+// CHECK1-NEXT: [[TMP18:%.*]] = load i16, ptr [[B]], align 2
+// CHECK1-NEXT: store i16 [[TMP18]], ptr [[B_CASTED]], align 2
+// CHECK1-NEXT: [[TMP19:%.*]] = load i64, ptr [[B_CASTED]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = load i16, ptr [[DOTCAPTURE_EXPR_]], align 2
+// CHECK1-NEXT: store i16 [[TMP20]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 2
+// CHECK1-NEXT: [[TMP21:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP17]], ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP17]], ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP19]], ptr [[TMP25]], align 8
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP19]], ptr [[TMP26]], align 8
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
+// CHECK1-NEXT: store i64 [[TMP21]], ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
+// CHECK1-NEXT: store i64 [[TMP21]], ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 2
+// CHECK1-NEXT: store ptr null, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP33:%.*]] = load i16, ptr [[DOTCAPTURE_EXPR_]], align 2
+// CHECK1-NEXT: [[TMP34:%.*]] = sext i16 [[TMP33]] to i32
+// CHECK1-NEXT: [[TMP35:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP34]], 0
// CHECK1-NEXT: [[KERNEL_ARGS1:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP31]], align 4
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 1
-// CHECK1-NEXT: store i32 3, ptr [[TMP32]], align 4
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP27]], ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP28]], ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.12, ptr [[TMP35]], align 8
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP36]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP38]], align 8
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP30]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93.region_id, ptr [[KERNEL_ARGS1]])
-// CHECK1-NEXT: [[TMP41:%.*]] = icmp ne i32 [[TMP40]], 0
-// CHECK1-NEXT: br i1 [[TMP41]], label [[OMP_OFFLOAD_FAILED2:%.*]], label [[OMP_OFFLOAD_CONT3:%.*]]
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP36]], align 4
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 1
+// CHECK1-NEXT: store i32 3, ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP31]], ptr [[TMP38]], align 8
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP32]], ptr [[TMP39]], align 8
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.12, ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP44]], align 8
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP45]], align 8
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [[TMP35]], ptr [[TMP46]], align 4
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP47]], align 4
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP48]], align 4
+// CHECK1-NEXT: [[TMP49:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP34]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93.region_id, ptr [[KERNEL_ARGS1]])
+// CHECK1-NEXT: [[TMP50:%.*]] = icmp ne i32 [[TMP49]], 0
+// CHECK1-NEXT: br i1 [[TMP50]], label [[OMP_OFFLOAD_FAILED2:%.*]], label [[OMP_OFFLOAD_CONT3:%.*]]
// CHECK1: omp_offload.failed2:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93(i64 [[TMP13]], i64 [[TMP15]], i64 [[TMP17]]) #[[ATTR2]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93(i64 [[TMP17]], i64 [[TMP19]], i64 [[TMP21]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT3]]
// CHECK1: omp_offload.cont3:
-// CHECK1-NEXT: [[TMP42:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: ret i32 [[TMP42]]
+// CHECK1-NEXT: [[TMP51:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: ret i32 [[TMP51]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121
@@ -785,70 +837,87 @@ int bar(int n){
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP17]], 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP18]], align 4
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK3-NEXT: store i32 3, ptr [[TMP19]], align 4
-// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP15]], ptr [[TMP20]], align 4
-// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP16]], ptr [[TMP21]], align 4
-// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes, ptr [[TMP22]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP23]], align 4
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP24]], align 4
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP19]], align 4
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK3-NEXT: store i32 3, ptr [[TMP20]], align 4
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP15]], ptr [[TMP21]], align 4
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP16]], ptr [[TMP22]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes, ptr [[TMP23]], align 4
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP24]], align 4
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK3-NEXT: store ptr null, ptr [[TMP25]], align 4
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP26]], align 8
-// CHECK3-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 [[TMP17]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK3-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP28]], align 8
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [[TMP18]], ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 [[TMP17]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
+// CHECK3-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121(ptr [[THIS1]], i32 [[TMP3]], i32 [[TMP5]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_S1]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP29]], align 4
-// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A2]], ptr [[TMP30]], align 4
-// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP31]], align 4
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP34]], align 4
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A2]], ptr [[TMP35]], align 4
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP36]], align 4
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP34]], align 4
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP35]], align 4
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP32]], ptr [[TMP36]], align 4
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP33]], ptr [[TMP37]], align 4
-// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.2, ptr [[TMP38]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP39]], align 4
-// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP40]], align 4
-// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP41]], align 4
-// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP42]], align 8
-// CHECK3-NEXT: [[TMP43:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1024, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126.region_id, ptr [[KERNEL_ARGS6]])
-// CHECK3-NEXT: [[TMP44:%.*]] = icmp ne i32 [[TMP43]], 0
-// CHECK3-NEXT: br i1 [[TMP44]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP39]], align 4
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP37]], ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP38]], ptr [[TMP42]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.2, ptr [[TMP43]], align 4
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP44]], align 4
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP45]], align 4
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP46]], align 4
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP47]], align 8
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP48]], align 8
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 1024, i32 0, i32 0], ptr [[TMP49]], align 4
+// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP50]], align 4
+// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP51]], align 4
+// CHECK3-NEXT: [[TMP52:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1024, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126.region_id, ptr [[KERNEL_ARGS6]])
+// CHECK3-NEXT: [[TMP53:%.*]] = icmp ne i32 [[TMP52]], 0
+// CHECK3-NEXT: br i1 [[TMP53]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK3: omp_offload.failed7:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126(ptr [[THIS1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK3: omp_offload.cont8:
// CHECK3-NEXT: [[A9:%.*]] = getelementptr inbounds [[STRUCT_S1]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP45:%.*]] = load double, ptr [[A9]], align 4
-// CHECK3-NEXT: [[CONV:%.*]] = fptosi double [[TMP45]] to i32
+// CHECK3-NEXT: [[TMP54:%.*]] = load double, ptr [[A9]], align 4
+// CHECK3-NEXT: [[CONV:%.*]] = fptosi double [[TMP54]] to i32
// CHECK3-NEXT: ret i32 [[CONV]]
//
//
@@ -881,75 +950,93 @@ int bar(int n){
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK3-NEXT: [[TMP9:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP8]], 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP9]], align 4
-// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP10]], align 4
-// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP6]], ptr [[TMP11]], align 4
-// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP7]], ptr [[TMP12]], align 4
-// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.5, ptr [[TMP13]], align 4
-// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP14]], align 4
-// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP15]], align 4
-// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP6]], ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP7]], ptr [[TMP13]], align 4
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.5, ptr [[TMP14]], align 4
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK3-NEXT: store ptr null, ptr [[TMP16]], align 4
-// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP17]], align 8
-// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP8]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l104.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
-// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP18]], align 8
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP19]], align 8
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [[TMP9]], ptr [[TMP20]], align 4
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP21]], align 4
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP22]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP8]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l104.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP24:%.*]] = icmp ne i32 [[TMP23]], 0
+// CHECK3-NEXT: br i1 [[TMP24]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l104(i32 [[TMP2]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
-// CHECK3-NEXT: [[TMP20:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 32, [[TMP20]]
+// CHECK3-NEXT: [[TMP25:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 32, [[TMP25]]
// CHECK3-NEXT: store i32 [[ADD]], ptr [[DOTCAPTURE_EXPR_1]], align 4
-// CHECK3-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
-// CHECK3-NEXT: store i32 [[TMP21]], ptr [[DOTCAPTURE_EXPR__CASTED2]], align 4
-// CHECK3-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED2]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP22]], ptr [[TMP23]], align 4
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP22]], ptr [[TMP24]], align 4
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP25]], align 4
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP28:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
+// CHECK3-NEXT: [[TMP26:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
+// CHECK3-NEXT: store i32 [[TMP26]], ptr [[DOTCAPTURE_EXPR__CASTED2]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED2]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP27]], ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP27]], ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP33:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
+// CHECK3-NEXT: [[TMP34:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP33]], 0
// CHECK3-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP29]], align 4
-// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP30]], align 4
-// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP26]], ptr [[TMP31]], align 4
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP27]], ptr [[TMP32]], align 4
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.8, ptr [[TMP33]], align 4
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP34]], align 4
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP35]], align 4
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP36]], align 4
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP37]], align 8
-// CHECK3-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP28]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108.region_id, ptr [[KERNEL_ARGS6]])
-// CHECK3-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
-// CHECK3-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP35]], align 4
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP36]], align 4
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP31]], ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP32]], ptr [[TMP38]], align 4
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.8, ptr [[TMP39]], align 4
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP42]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP43]], align 8
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP44]], align 8
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [[TMP34]], ptr [[TMP45]], align 4
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP46]], align 4
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP47]], align 4
+// CHECK3-NEXT: [[TMP48:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP33]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108.region_id, ptr [[KERNEL_ARGS6]])
+// CHECK3-NEXT: [[TMP49:%.*]] = icmp ne i32 [[TMP48]], 0
+// CHECK3-NEXT: br i1 [[TMP49]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK3: omp_offload.failed7:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108(i32 [[TMP22]]) #[[ATTR2]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108(i32 [[TMP27]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK3: omp_offload.cont8:
-// CHECK3-NEXT: [[TMP40:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: [[ADD9:%.*]] = add nsw i32 [[TMP40]], 1
+// CHECK3-NEXT: [[TMP50:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: [[ADD9:%.*]] = add nsw i32 [[TMP50]], 1
// CHECK3-NEXT: ret i32 [[ADD9]]
//
//
@@ -970,7 +1057,7 @@ int bar(int n){
// CHECK3-NEXT: store i32 0, ptr [[A]], align 4
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -987,75 +1074,92 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP8]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 20, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l88.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK3-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [i32 20, i32 0, i32 0], ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 20, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l88.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK3-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l88() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: store i16 1, ptr [[B]], align 2
-// CHECK3-NEXT: [[TMP11:%.*]] = load i16, ptr [[B]], align 2
-// CHECK3-NEXT: store i16 [[TMP11]], ptr [[DOTCAPTURE_EXPR_]], align 2
-// CHECK3-NEXT: [[TMP12:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: store i32 [[TMP12]], ptr [[A_CASTED]], align 4
-// CHECK3-NEXT: [[TMP13:%.*]] = load i32, ptr [[A_CASTED]], align 4
-// CHECK3-NEXT: [[TMP14:%.*]] = load i16, ptr [[B]], align 2
-// CHECK3-NEXT: store i16 [[TMP14]], ptr [[B_CASTED]], align 2
-// CHECK3-NEXT: [[TMP15:%.*]] = load i32, ptr [[B_CASTED]], align 4
-// CHECK3-NEXT: [[TMP16:%.*]] = load i16, ptr [[DOTCAPTURE_EXPR_]], align 2
-// CHECK3-NEXT: store i16 [[TMP16]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 2
-// CHECK3-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP13]], ptr [[TMP18]], align 4
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP13]], ptr [[TMP19]], align 4
-// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP20]], align 4
-// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP15]], ptr [[TMP21]], align 4
-// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP15]], ptr [[TMP22]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 1
-// CHECK3-NEXT: store ptr null, ptr [[TMP23]], align 4
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
-// CHECK3-NEXT: store i32 [[TMP17]], ptr [[TMP24]], align 4
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
-// CHECK3-NEXT: store i32 [[TMP17]], ptr [[TMP25]], align 4
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 2
-// CHECK3-NEXT: store ptr null, ptr [[TMP26]], align 4
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP29:%.*]] = load i16, ptr [[DOTCAPTURE_EXPR_]], align 2
-// CHECK3-NEXT: [[TMP30:%.*]] = sext i16 [[TMP29]] to i32
+// CHECK3-NEXT: [[TMP15:%.*]] = load i16, ptr [[B]], align 2
+// CHECK3-NEXT: store i16 [[TMP15]], ptr [[DOTCAPTURE_EXPR_]], align 2
+// CHECK3-NEXT: [[TMP16:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: store i32 [[TMP16]], ptr [[A_CASTED]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = load i32, ptr [[A_CASTED]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = load i16, ptr [[B]], align 2
+// CHECK3-NEXT: store i16 [[TMP18]], ptr [[B_CASTED]], align 2
+// CHECK3-NEXT: [[TMP19:%.*]] = load i32, ptr [[B_CASTED]], align 4
+// CHECK3-NEXT: [[TMP20:%.*]] = load i16, ptr [[DOTCAPTURE_EXPR_]], align 2
+// CHECK3-NEXT: store i16 [[TMP20]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 2
+// CHECK3-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP17]], ptr [[TMP22]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP17]], ptr [[TMP23]], align 4
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP24]], align 4
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP19]], ptr [[TMP25]], align 4
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP19]], ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 1
+// CHECK3-NEXT: store ptr null, ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
+// CHECK3-NEXT: store i32 [[TMP21]], ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
+// CHECK3-NEXT: store i32 [[TMP21]], ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 2
+// CHECK3-NEXT: store ptr null, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP33:%.*]] = load i16, ptr [[DOTCAPTURE_EXPR_]], align 2
+// CHECK3-NEXT: [[TMP34:%.*]] = sext i16 [[TMP33]] to i32
+// CHECK3-NEXT: [[TMP35:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP34]], 0
// CHECK3-NEXT: [[KERNEL_ARGS1:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP31]], align 4
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 1
-// CHECK3-NEXT: store i32 3, ptr [[TMP32]], align 4
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP27]], ptr [[TMP33]], align 4
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP28]], ptr [[TMP34]], align 4
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.12, ptr [[TMP35]], align 4
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP36]], align 4
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP37]], align 4
-// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP38]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP39]], align 8
-// CHECK3-NEXT: [[TMP40:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP30]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93.region_id, ptr [[KERNEL_ARGS1]])
-// CHECK3-NEXT: [[TMP41:%.*]] = icmp ne i32 [[TMP40]], 0
-// CHECK3-NEXT: br i1 [[TMP41]], label [[OMP_OFFLOAD_FAILED2:%.*]], label [[OMP_OFFLOAD_CONT3:%.*]]
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP36]], align 4
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 1
+// CHECK3-NEXT: store i32 3, ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP31]], ptr [[TMP38]], align 4
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP32]], ptr [[TMP39]], align 4
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.12, ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP42]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP43]], align 4
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP44]], align 8
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP45]], align 8
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [[TMP35]], ptr [[TMP46]], align 4
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP47]], align 4
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP48]], align 4
+// CHECK3-NEXT: [[TMP49:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP34]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93.region_id, ptr [[KERNEL_ARGS1]])
+// CHECK3-NEXT: [[TMP50:%.*]] = icmp ne i32 [[TMP49]], 0
+// CHECK3-NEXT: br i1 [[TMP50]], label [[OMP_OFFLOAD_FAILED2:%.*]], label [[OMP_OFFLOAD_CONT3:%.*]]
// CHECK3: omp_offload.failed2:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93(i32 [[TMP13]], i32 [[TMP15]], i32 [[TMP17]]) #[[ATTR2]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93(i32 [[TMP17]], i32 [[TMP19]], i32 [[TMP21]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT3]]
// CHECK3: omp_offload.cont3:
-// CHECK3-NEXT: [[TMP42:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: ret i32 [[TMP42]]
+// CHECK3-NEXT: [[TMP51:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: ret i32 [[TMP51]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121
diff --git a/clang/test/OpenMP/target_teams_thread_limit_codegen.cpp b/clang/test/OpenMP/target_teams_thread_limit_codegen.cpp
index 2b0d2edae7e36..6a1dfdfa5fca0 100644
--- a/clang/test/OpenMP/target_teams_thread_limit_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_thread_limit_codegen.cpp
@@ -252,70 +252,87 @@ int bar(int n){
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP17]], 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP18]], align 4
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK1-NEXT: store i32 3, ptr [[TMP19]], align 4
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP15]], ptr [[TMP20]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP16]], ptr [[TMP21]], align 8
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP22]], align 8
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP19]], align 4
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK1-NEXT: store i32 3, ptr [[TMP20]], align 4
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP15]], ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP16]], ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK1-NEXT: store ptr null, ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 0, i32 [[TMP17]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK1-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP26]], align 8
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP29]], align 4
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP18]], ptr [[TMP30]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP31]], align 4
+// CHECK1-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 0, i32 [[TMP17]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
+// CHECK1-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121(ptr [[THIS1]], i64 [[TMP3]], i64 [[TMP5]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_S1]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP30]], align 8
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP31]], align 8
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP35]], align 8
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP36]], align 8
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP34]], align 4
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP35]], align 4
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP32]], ptr [[TMP36]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP33]], ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.2, ptr [[TMP38]], align 8
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP41]], align 8
-// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP42]], align 8
-// CHECK1-NEXT: [[TMP43:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 1024, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126.region_id, ptr [[KERNEL_ARGS6]])
-// CHECK1-NEXT: [[TMP44:%.*]] = icmp ne i32 [[TMP43]], 0
-// CHECK1-NEXT: br i1 [[TMP44]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP39]], align 4
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP40]], align 4
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP37]], ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP38]], ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.2, ptr [[TMP43]], align 8
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP44]], align 8
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP45]], align 8
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP46]], align 8
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP47]], align 8
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP48]], align 8
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP49]], align 4
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1024, i32 0, i32 0], ptr [[TMP50]], align 4
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP51]], align 4
+// CHECK1-NEXT: [[TMP52:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 1024, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126.region_id, ptr [[KERNEL_ARGS6]])
+// CHECK1-NEXT: [[TMP53:%.*]] = icmp ne i32 [[TMP52]], 0
+// CHECK1-NEXT: br i1 [[TMP53]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK1: omp_offload.failed7:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK1: omp_offload.cont8:
// CHECK1-NEXT: [[A9:%.*]] = getelementptr inbounds [[STRUCT_S1]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP45:%.*]] = load double, ptr [[A9]], align 8
-// CHECK1-NEXT: [[CONV:%.*]] = fptosi double [[TMP45]] to i32
+// CHECK1-NEXT: [[TMP54:%.*]] = load double, ptr [[A9]], align 8
+// CHECK1-NEXT: [[CONV:%.*]] = fptosi double [[TMP54]] to i32
// CHECK1-NEXT: ret i32 [[CONV]]
//
//
@@ -363,75 +380,94 @@ int bar(int n){
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP14]], 0
+// CHECK1-NEXT: [[TMP17:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP15]], 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP16]], align 4
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK1-NEXT: store i32 2, ptr [[TMP17]], align 4
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP12]], ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP13]], ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.5, ptr [[TMP20]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP21]], align 8
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP14]], i32 [[TMP15]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l104.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK1-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP18]], align 4
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK1-NEXT: store i32 2, ptr [[TMP19]], align 4
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP12]], ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP13]], ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.5, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP25]], align 8
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP26]], align 8
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [[TMP16]], ptr [[TMP28]], align 4
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP17]], ptr [[TMP29]], align 4
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP14]], i32 [[TMP15]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l104.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK1-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l104(i64 [[TMP3]], i64 [[TMP5]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
-// CHECK1-NEXT: [[TMP27:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 32, [[TMP27]]
+// CHECK1-NEXT: [[TMP33:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 32, [[TMP33]]
// CHECK1-NEXT: store i32 [[ADD]], ptr [[DOTCAPTURE_EXPR_3]], align 4
-// CHECK1-NEXT: [[TMP28:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_3]], align 4
-// CHECK1-NEXT: store i32 [[TMP28]], ptr [[DOTCAPTURE_EXPR__CASTED4]], align 4
-// CHECK1-NEXT: [[TMP29:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED4]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP29]], ptr [[TMP30]], align 8
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP29]], ptr [[TMP31]], align 8
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP32]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP35:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_3]], align 4
+// CHECK1-NEXT: [[TMP34:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_3]], align 4
+// CHECK1-NEXT: store i32 [[TMP34]], ptr [[DOTCAPTURE_EXPR__CASTED4]], align 4
+// CHECK1-NEXT: [[TMP35:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED4]], align 8
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP35]], ptr [[TMP36]], align 8
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP35]], ptr [[TMP37]], align 8
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP38]], align 8
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP41:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_3]], align 4
+// CHECK1-NEXT: [[TMP42:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP41]], 0
// CHECK1-NEXT: [[KERNEL_ARGS8:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP36]], align 4
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP33]], ptr [[TMP38]], align 8
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP34]], ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.8, ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP41]], align 8
-// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP42]], align 8
-// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP43]], align 8
-// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP44]], align 8
-// CHECK1-NEXT: [[TMP45:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 [[TMP35]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108.region_id, ptr [[KERNEL_ARGS8]])
-// CHECK1-NEXT: [[TMP46:%.*]] = icmp ne i32 [[TMP45]], 0
-// CHECK1-NEXT: br i1 [[TMP46]], label [[OMP_OFFLOAD_FAILED9:%.*]], label [[OMP_OFFLOAD_CONT10:%.*]]
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP43]], align 4
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP44]], align 4
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP39]], ptr [[TMP45]], align 8
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP40]], ptr [[TMP46]], align 8
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.8, ptr [[TMP47]], align 8
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP48]], align 8
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP49]], align 8
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP50]], align 8
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP51]], align 8
+// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP52]], align 8
+// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP53]], align 4
+// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP42]], ptr [[TMP54]], align 4
+// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP55]], align 4
+// CHECK1-NEXT: [[TMP56:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 [[TMP41]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108.region_id, ptr [[KERNEL_ARGS8]])
+// CHECK1-NEXT: [[TMP57:%.*]] = icmp ne i32 [[TMP56]], 0
+// CHECK1-NEXT: br i1 [[TMP57]], label [[OMP_OFFLOAD_FAILED9:%.*]], label [[OMP_OFFLOAD_CONT10:%.*]]
// CHECK1: omp_offload.failed9:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108(i64 [[TMP29]]) #[[ATTR2]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108(i64 [[TMP35]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT10]]
// CHECK1: omp_offload.cont10:
-// CHECK1-NEXT: [[TMP47:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: [[ADD11:%.*]] = add nsw i32 [[TMP47]], 1
+// CHECK1-NEXT: [[TMP58:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: [[ADD11:%.*]] = add nsw i32 [[TMP58]], 1
// CHECK1-NEXT: ret i32 [[ADD11]]
//
//
@@ -452,7 +488,7 @@ int bar(int n){
// CHECK1-NEXT: store i32 0, ptr [[A]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -469,75 +505,92 @@ int bar(int n){
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 20, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l88.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 20, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 20, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l88.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l88() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: store i16 1, ptr [[B]], align 2
-// CHECK1-NEXT: [[TMP11:%.*]] = load i16, ptr [[B]], align 2
-// CHECK1-NEXT: store i16 [[TMP11]], ptr [[DOTCAPTURE_EXPR_]], align 2
-// CHECK1-NEXT: [[TMP12:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: store i32 [[TMP12]], ptr [[A_CASTED]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = load i64, ptr [[A_CASTED]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = load i16, ptr [[B]], align 2
-// CHECK1-NEXT: store i16 [[TMP14]], ptr [[B_CASTED]], align 2
-// CHECK1-NEXT: [[TMP15:%.*]] = load i64, ptr [[B_CASTED]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = load i16, ptr [[DOTCAPTURE_EXPR_]], align 2
-// CHECK1-NEXT: store i16 [[TMP16]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 2
-// CHECK1-NEXT: [[TMP17:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP13]], ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP13]], ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP15]], ptr [[TMP21]], align 8
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP15]], ptr [[TMP22]], align 8
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
-// CHECK1-NEXT: store i64 [[TMP17]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
-// CHECK1-NEXT: store i64 [[TMP17]], ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP29:%.*]] = load i16, ptr [[DOTCAPTURE_EXPR_]], align 2
-// CHECK1-NEXT: [[TMP30:%.*]] = sext i16 [[TMP29]] to i32
+// CHECK1-NEXT: [[TMP15:%.*]] = load i16, ptr [[B]], align 2
+// CHECK1-NEXT: store i16 [[TMP15]], ptr [[DOTCAPTURE_EXPR_]], align 2
+// CHECK1-NEXT: [[TMP16:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: store i32 [[TMP16]], ptr [[A_CASTED]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = load i64, ptr [[A_CASTED]], align 8
+// CHECK1-NEXT: [[TMP18:%.*]] = load i16, ptr [[B]], align 2
+// CHECK1-NEXT: store i16 [[TMP18]], ptr [[B_CASTED]], align 2
+// CHECK1-NEXT: [[TMP19:%.*]] = load i64, ptr [[B_CASTED]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = load i16, ptr [[DOTCAPTURE_EXPR_]], align 2
+// CHECK1-NEXT: store i16 [[TMP20]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 2
+// CHECK1-NEXT: [[TMP21:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP17]], ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP17]], ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP19]], ptr [[TMP25]], align 8
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP19]], ptr [[TMP26]], align 8
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
+// CHECK1-NEXT: store i64 [[TMP21]], ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
+// CHECK1-NEXT: store i64 [[TMP21]], ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 2
+// CHECK1-NEXT: store ptr null, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP33:%.*]] = load i16, ptr [[DOTCAPTURE_EXPR_]], align 2
+// CHECK1-NEXT: [[TMP34:%.*]] = sext i16 [[TMP33]] to i32
+// CHECK1-NEXT: [[TMP35:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP34]], 0
// CHECK1-NEXT: [[KERNEL_ARGS1:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP31]], align 4
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 1
-// CHECK1-NEXT: store i32 3, ptr [[TMP32]], align 4
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP27]], ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP28]], ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.12, ptr [[TMP35]], align 8
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP36]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP38]], align 8
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP30]], i32 1024, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93.region_id, ptr [[KERNEL_ARGS1]])
-// CHECK1-NEXT: [[TMP41:%.*]] = icmp ne i32 [[TMP40]], 0
-// CHECK1-NEXT: br i1 [[TMP41]], label [[OMP_OFFLOAD_FAILED2:%.*]], label [[OMP_OFFLOAD_CONT3:%.*]]
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP36]], align 4
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 1
+// CHECK1-NEXT: store i32 3, ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP31]], ptr [[TMP38]], align 8
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP32]], ptr [[TMP39]], align 8
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.12, ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP44]], align 8
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP45]], align 8
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [[TMP35]], ptr [[TMP46]], align 4
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1024, i32 0, i32 0], ptr [[TMP47]], align 4
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP48]], align 4
+// CHECK1-NEXT: [[TMP49:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP34]], i32 1024, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93.region_id, ptr [[KERNEL_ARGS1]])
+// CHECK1-NEXT: [[TMP50:%.*]] = icmp ne i32 [[TMP49]], 0
+// CHECK1-NEXT: br i1 [[TMP50]], label [[OMP_OFFLOAD_FAILED2:%.*]], label [[OMP_OFFLOAD_CONT3:%.*]]
// CHECK1: omp_offload.failed2:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93(i64 [[TMP13]], i64 [[TMP15]], i64 [[TMP17]]) #[[ATTR2]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93(i64 [[TMP17]], i64 [[TMP19]], i64 [[TMP21]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT3]]
// CHECK1: omp_offload.cont3:
-// CHECK1-NEXT: [[TMP42:%.*]] = load i32, ptr [[A]], align 4
-// CHECK1-NEXT: ret i32 [[TMP42]]
+// CHECK1-NEXT: [[TMP51:%.*]] = load i32, ptr [[A]], align 4
+// CHECK1-NEXT: ret i32 [[TMP51]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121
@@ -803,70 +856,87 @@ int bar(int n){
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP17]], 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP18]], align 4
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK3-NEXT: store i32 3, ptr [[TMP19]], align 4
-// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP15]], ptr [[TMP20]], align 4
-// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP16]], ptr [[TMP21]], align 4
-// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes, ptr [[TMP22]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP23]], align 4
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP24]], align 4
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP19]], align 4
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK3-NEXT: store i32 3, ptr [[TMP20]], align 4
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP15]], ptr [[TMP21]], align 4
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP16]], ptr [[TMP22]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes, ptr [[TMP23]], align 4
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP24]], align 4
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK3-NEXT: store ptr null, ptr [[TMP25]], align 4
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP26]], align 8
-// CHECK3-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 0, i32 [[TMP17]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK3-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP28]], align 8
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [[TMP18]], ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 0, i32 [[TMP17]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
+// CHECK3-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121(ptr [[THIS1]], i32 [[TMP3]], i32 [[TMP5]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_S1]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP29]], align 4
-// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A2]], ptr [[TMP30]], align 4
-// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP31]], align 4
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP34]], align 4
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A2]], ptr [[TMP35]], align 4
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP36]], align 4
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP34]], align 4
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP35]], align 4
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP32]], ptr [[TMP36]], align 4
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP33]], ptr [[TMP37]], align 4
-// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.2, ptr [[TMP38]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP39]], align 4
-// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP40]], align 4
-// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP41]], align 4
-// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP42]], align 8
-// CHECK3-NEXT: [[TMP43:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 1024, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126.region_id, ptr [[KERNEL_ARGS6]])
-// CHECK3-NEXT: [[TMP44:%.*]] = icmp ne i32 [[TMP43]], 0
-// CHECK3-NEXT: br i1 [[TMP44]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP39]], align 4
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP37]], ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP38]], ptr [[TMP42]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.2, ptr [[TMP43]], align 4
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP44]], align 4
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP45]], align 4
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP46]], align 4
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP47]], align 8
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP48]], align 8
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP49]], align 4
+// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1024, i32 0, i32 0], ptr [[TMP50]], align 4
+// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP51]], align 4
+// CHECK3-NEXT: [[TMP52:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 1024, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126.region_id, ptr [[KERNEL_ARGS6]])
+// CHECK3-NEXT: [[TMP53:%.*]] = icmp ne i32 [[TMP52]], 0
+// CHECK3-NEXT: br i1 [[TMP53]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK3: omp_offload.failed7:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l126(ptr [[THIS1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT8]]
// CHECK3: omp_offload.cont8:
// CHECK3-NEXT: [[A9:%.*]] = getelementptr inbounds [[STRUCT_S1]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP45:%.*]] = load double, ptr [[A9]], align 4
-// CHECK3-NEXT: [[CONV:%.*]] = fptosi double [[TMP45]] to i32
+// CHECK3-NEXT: [[TMP54:%.*]] = load double, ptr [[A9]], align 4
+// CHECK3-NEXT: [[CONV:%.*]] = fptosi double [[TMP54]] to i32
// CHECK3-NEXT: ret i32 [[CONV]]
//
//
@@ -914,75 +984,94 @@ int bar(int n){
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP14]], 0
+// CHECK3-NEXT: [[TMP17:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP15]], 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP16]], align 4
-// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK3-NEXT: store i32 2, ptr [[TMP17]], align 4
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP12]], ptr [[TMP18]], align 4
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP13]], ptr [[TMP19]], align 4
-// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.5, ptr [[TMP20]], align 4
-// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP21]], align 4
-// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP22]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP23]], align 4
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP24]], align 8
-// CHECK3-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP14]], i32 [[TMP15]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l104.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK3-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP18]], align 4
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK3-NEXT: store i32 2, ptr [[TMP19]], align 4
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP12]], ptr [[TMP20]], align 4
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP13]], ptr [[TMP21]], align 4
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.5, ptr [[TMP22]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP23]], align 4
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP24]], align 4
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP25]], align 4
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP26]], align 8
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [[TMP16]], ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [[TMP17]], ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP14]], i32 [[TMP15]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l104.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK3-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l104(i32 [[TMP3]], i32 [[TMP5]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
-// CHECK3-NEXT: [[TMP27:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 32, [[TMP27]]
+// CHECK3-NEXT: [[TMP33:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 32, [[TMP33]]
// CHECK3-NEXT: store i32 [[ADD]], ptr [[DOTCAPTURE_EXPR_3]], align 4
-// CHECK3-NEXT: [[TMP28:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_3]], align 4
-// CHECK3-NEXT: store i32 [[TMP28]], ptr [[DOTCAPTURE_EXPR__CASTED4]], align 4
-// CHECK3-NEXT: [[TMP29:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED4]], align 4
-// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP29]], ptr [[TMP30]], align 4
-// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP29]], ptr [[TMP31]], align 4
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP32]], align 4
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP35:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_3]], align 4
+// CHECK3-NEXT: [[TMP34:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_3]], align 4
+// CHECK3-NEXT: store i32 [[TMP34]], ptr [[DOTCAPTURE_EXPR__CASTED4]], align 4
+// CHECK3-NEXT: [[TMP35:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED4]], align 4
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP35]], ptr [[TMP36]], align 4
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP35]], ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP38]], align 4
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP41:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_3]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP41]], 0
// CHECK3-NEXT: [[KERNEL_ARGS8:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP36]], align 4
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP33]], ptr [[TMP38]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP34]], ptr [[TMP39]], align 4
-// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.8, ptr [[TMP40]], align 4
-// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP41]], align 4
-// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP42]], align 4
-// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP44]], align 8
-// CHECK3-NEXT: [[TMP45:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 [[TMP35]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108.region_id, ptr [[KERNEL_ARGS8]])
-// CHECK3-NEXT: [[TMP46:%.*]] = icmp ne i32 [[TMP45]], 0
-// CHECK3-NEXT: br i1 [[TMP46]], label [[OMP_OFFLOAD_FAILED9:%.*]], label [[OMP_OFFLOAD_CONT10:%.*]]
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP43]], align 4
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP44]], align 4
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP39]], ptr [[TMP45]], align 4
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP40]], ptr [[TMP46]], align 4
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.8, ptr [[TMP47]], align 4
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP48]], align 4
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP49]], align 4
+// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP50]], align 4
+// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP51]], align 8
+// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP52]], align 8
+// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP53]], align 4
+// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [[TMP42]], ptr [[TMP54]], align 4
+// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP55]], align 4
+// CHECK3-NEXT: [[TMP56:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 [[TMP41]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108.region_id, ptr [[KERNEL_ARGS8]])
+// CHECK3-NEXT: [[TMP57:%.*]] = icmp ne i32 [[TMP56]], 0
+// CHECK3-NEXT: br i1 [[TMP57]], label [[OMP_OFFLOAD_FAILED9:%.*]], label [[OMP_OFFLOAD_CONT10:%.*]]
// CHECK3: omp_offload.failed9:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108(i32 [[TMP29]]) #[[ATTR2]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l108(i32 [[TMP35]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT10]]
// CHECK3: omp_offload.cont10:
-// CHECK3-NEXT: [[TMP47:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: [[ADD11:%.*]] = add nsw i32 [[TMP47]], 1
+// CHECK3-NEXT: [[TMP58:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: [[ADD11:%.*]] = add nsw i32 [[TMP58]], 1
// CHECK3-NEXT: ret i32 [[ADD11]]
//
//
@@ -1003,7 +1092,7 @@ int bar(int n){
// CHECK3-NEXT: store i32 0, ptr [[A]], align 4
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1020,75 +1109,92 @@ int bar(int n){
// CHECK3-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP8]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 20, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l88.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK3-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 20, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 20, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l88.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK3-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l88() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: store i16 1, ptr [[B]], align 2
-// CHECK3-NEXT: [[TMP11:%.*]] = load i16, ptr [[B]], align 2
-// CHECK3-NEXT: store i16 [[TMP11]], ptr [[DOTCAPTURE_EXPR_]], align 2
-// CHECK3-NEXT: [[TMP12:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: store i32 [[TMP12]], ptr [[A_CASTED]], align 4
-// CHECK3-NEXT: [[TMP13:%.*]] = load i32, ptr [[A_CASTED]], align 4
-// CHECK3-NEXT: [[TMP14:%.*]] = load i16, ptr [[B]], align 2
-// CHECK3-NEXT: store i16 [[TMP14]], ptr [[B_CASTED]], align 2
-// CHECK3-NEXT: [[TMP15:%.*]] = load i32, ptr [[B_CASTED]], align 4
-// CHECK3-NEXT: [[TMP16:%.*]] = load i16, ptr [[DOTCAPTURE_EXPR_]], align 2
-// CHECK3-NEXT: store i16 [[TMP16]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 2
-// CHECK3-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP13]], ptr [[TMP18]], align 4
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP13]], ptr [[TMP19]], align 4
-// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP20]], align 4
-// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP15]], ptr [[TMP21]], align 4
-// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP15]], ptr [[TMP22]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 1
-// CHECK3-NEXT: store ptr null, ptr [[TMP23]], align 4
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
-// CHECK3-NEXT: store i32 [[TMP17]], ptr [[TMP24]], align 4
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
-// CHECK3-NEXT: store i32 [[TMP17]], ptr [[TMP25]], align 4
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 2
-// CHECK3-NEXT: store ptr null, ptr [[TMP26]], align 4
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP29:%.*]] = load i16, ptr [[DOTCAPTURE_EXPR_]], align 2
-// CHECK3-NEXT: [[TMP30:%.*]] = sext i16 [[TMP29]] to i32
+// CHECK3-NEXT: [[TMP15:%.*]] = load i16, ptr [[B]], align 2
+// CHECK3-NEXT: store i16 [[TMP15]], ptr [[DOTCAPTURE_EXPR_]], align 2
+// CHECK3-NEXT: [[TMP16:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: store i32 [[TMP16]], ptr [[A_CASTED]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = load i32, ptr [[A_CASTED]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = load i16, ptr [[B]], align 2
+// CHECK3-NEXT: store i16 [[TMP18]], ptr [[B_CASTED]], align 2
+// CHECK3-NEXT: [[TMP19:%.*]] = load i32, ptr [[B_CASTED]], align 4
+// CHECK3-NEXT: [[TMP20:%.*]] = load i16, ptr [[DOTCAPTURE_EXPR_]], align 2
+// CHECK3-NEXT: store i16 [[TMP20]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 2
+// CHECK3-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP17]], ptr [[TMP22]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP17]], ptr [[TMP23]], align 4
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP24]], align 4
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP19]], ptr [[TMP25]], align 4
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP19]], ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 1
+// CHECK3-NEXT: store ptr null, ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
+// CHECK3-NEXT: store i32 [[TMP21]], ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
+// CHECK3-NEXT: store i32 [[TMP21]], ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 2
+// CHECK3-NEXT: store ptr null, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP33:%.*]] = load i16, ptr [[DOTCAPTURE_EXPR_]], align 2
+// CHECK3-NEXT: [[TMP34:%.*]] = sext i16 [[TMP33]] to i32
+// CHECK3-NEXT: [[TMP35:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP34]], 0
// CHECK3-NEXT: [[KERNEL_ARGS1:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP31]], align 4
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 1
-// CHECK3-NEXT: store i32 3, ptr [[TMP32]], align 4
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP27]], ptr [[TMP33]], align 4
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP28]], ptr [[TMP34]], align 4
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.12, ptr [[TMP35]], align 4
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP36]], align 4
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP37]], align 4
-// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP38]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP39]], align 8
-// CHECK3-NEXT: [[TMP40:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP30]], i32 1024, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93.region_id, ptr [[KERNEL_ARGS1]])
-// CHECK3-NEXT: [[TMP41:%.*]] = icmp ne i32 [[TMP40]], 0
-// CHECK3-NEXT: br i1 [[TMP41]], label [[OMP_OFFLOAD_FAILED2:%.*]], label [[OMP_OFFLOAD_CONT3:%.*]]
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP36]], align 4
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 1
+// CHECK3-NEXT: store i32 3, ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP31]], ptr [[TMP38]], align 4
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP32]], ptr [[TMP39]], align 4
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.12, ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP42]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP43]], align 4
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP44]], align 8
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP45]], align 8
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [[TMP35]], ptr [[TMP46]], align 4
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1024, i32 0, i32 0], ptr [[TMP47]], align 4
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS1]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP48]], align 4
+// CHECK3-NEXT: [[TMP49:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP34]], i32 1024, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93.region_id, ptr [[KERNEL_ARGS1]])
+// CHECK3-NEXT: [[TMP50:%.*]] = icmp ne i32 [[TMP49]], 0
+// CHECK3-NEXT: br i1 [[TMP50]], label [[OMP_OFFLOAD_FAILED2:%.*]], label [[OMP_OFFLOAD_CONT3:%.*]]
// CHECK3: omp_offload.failed2:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93(i32 [[TMP13]], i32 [[TMP15]], i32 [[TMP17]]) #[[ATTR2]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l93(i32 [[TMP17]], i32 [[TMP19]], i32 [[TMP21]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT3]]
// CHECK3: omp_offload.cont3:
-// CHECK3-NEXT: [[TMP42:%.*]] = load i32, ptr [[A]], align 4
-// CHECK3-NEXT: ret i32 [[TMP42]]
+// CHECK3-NEXT: [[TMP51:%.*]] = load i32, ptr [[A]], align 4
+// CHECK3-NEXT: ret i32 [[TMP51]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l121
diff --git a/clang/test/OpenMP/teams_codegen.cpp b/clang/test/OpenMP/teams_codegen.cpp
index d8fc00919af37..9d6da1ec7fd4e 100644
--- a/clang/test/OpenMP/teams_codegen.cpp
+++ b/clang/test/OpenMP/teams_codegen.cpp
@@ -356,7 +356,7 @@ void foo() {
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -373,283 +373,337 @@ void foo() {
// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l31.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK1-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l31.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l31(i64 [[TMP1]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
-// CHECK1-NEXT: [[TMP18:%.*]] = load i32, ptr [[COMP]], align 4
-// CHECK1-NEXT: store i32 [[TMP18]], ptr [[COMP_CASTED1]], align 4
-// CHECK1-NEXT: [[TMP19:%.*]] = load i64, ptr [[COMP_CASTED1]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP19]], ptr [[TMP20]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP19]], ptr [[TMP21]], align 8
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS4]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP22:%.*]] = load i32, ptr [[COMP]], align 4
+// CHECK1-NEXT: store i32 [[TMP22]], ptr [[COMP_CASTED1]], align 4
+// CHECK1-NEXT: [[TMP23:%.*]] = load i64, ptr [[COMP_CASTED1]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP23]], ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP23]], ptr [[TMP25]], align 8
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS4]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP26]], align 8
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP25]], align 4
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP26]], align 4
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.2, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP30]], align 8
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP31]], align 8
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP32]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l37.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK1-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
-// CHECK1-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP29]], align 4
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP30]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP27]], ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP28]], ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.2, ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP35]], align 8
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP36]], align 8
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP37]], align 8
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP38]], align 8
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP39]], align 4
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP40]], align 4
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP41]], align 4
+// CHECK1-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l37.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK1-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
+// CHECK1-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK1: omp_offload.failed6:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l37(i64 [[TMP19]]) #[[ATTR2]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l37(i64 [[TMP23]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK1: omp_offload.cont7:
-// CHECK1-NEXT: [[TMP36:%.*]] = load i32, ptr [[LA]], align 4
-// CHECK1-NEXT: store i32 [[TMP36]], ptr [[LA_CASTED]], align 4
-// CHECK1-NEXT: [[TMP37:%.*]] = load i64, ptr [[LA_CASTED]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = load i32, ptr [[COMP]], align 4
-// CHECK1-NEXT: store i32 [[TMP38]], ptr [[COMP_CASTED8]], align 4
-// CHECK1-NEXT: [[TMP39:%.*]] = load i64, ptr [[COMP_CASTED8]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS9]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP37]], ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS10]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP37]], ptr [[TMP41]], align 8
-// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS11]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP42]], align 8
-// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS9]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP39]], ptr [[TMP43]], align 8
-// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS10]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP39]], ptr [[TMP44]], align 8
-// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS11]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP45]], align 8
-// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS9]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS10]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP48:%.*]] = load i32, ptr [[LA]], align 4
+// CHECK1-NEXT: [[TMP44:%.*]] = load i32, ptr [[LA]], align 4
+// CHECK1-NEXT: store i32 [[TMP44]], ptr [[LA_CASTED]], align 4
+// CHECK1-NEXT: [[TMP45:%.*]] = load i64, ptr [[LA_CASTED]], align 8
+// CHECK1-NEXT: [[TMP46:%.*]] = load i32, ptr [[COMP]], align 4
+// CHECK1-NEXT: store i32 [[TMP46]], ptr [[COMP_CASTED8]], align 4
+// CHECK1-NEXT: [[TMP47:%.*]] = load i64, ptr [[COMP_CASTED8]], align 8
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS9]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP45]], ptr [[TMP48]], align 8
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS10]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP45]], ptr [[TMP49]], align 8
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS11]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP50]], align 8
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS9]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP47]], ptr [[TMP51]], align 8
+// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS10]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP47]], ptr [[TMP52]], align 8
+// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS11]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP53]], align 8
+// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS9]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS10]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP56:%.*]] = load i32, ptr [[LA]], align 4
+// CHECK1-NEXT: [[TMP57:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP56]], 0
// CHECK1-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP49]], align 4
-// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK1-NEXT: store i32 2, ptr [[TMP50]], align 4
-// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP46]], ptr [[TMP51]], align 8
-// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP47]], ptr [[TMP52]], align 8
-// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.5, ptr [[TMP53]], align 8
-// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP54]], align 8
-// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP55]], align 8
-// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP56]], align 8
-// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP57]], align 8
-// CHECK1-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP48]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l46.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK1-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
-// CHECK1-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK1-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP58]], align 4
+// CHECK1-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK1-NEXT: store i32 2, ptr [[TMP59]], align 4
+// CHECK1-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP54]], ptr [[TMP60]], align 8
+// CHECK1-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP55]], ptr [[TMP61]], align 8
+// CHECK1-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.5, ptr [[TMP62]], align 8
+// CHECK1-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP63]], align 8
+// CHECK1-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP64]], align 8
+// CHECK1-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP65]], align 8
+// CHECK1-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP66]], align 8
+// CHECK1-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP67]], align 8
+// CHECK1-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [[TMP57]], ptr [[TMP68]], align 4
+// CHECK1-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK1-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP70]], align 4
+// CHECK1-NEXT: [[TMP71:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP56]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l46.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK1-NEXT: [[TMP72:%.*]] = icmp ne i32 [[TMP71]], 0
+// CHECK1-NEXT: br i1 [[TMP72]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK1: omp_offload.failed13:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l46(i64 [[TMP37]], i64 [[TMP39]]) #[[ATTR2]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l46(i64 [[TMP45]], i64 [[TMP47]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT14]]
// CHECK1: omp_offload.cont14:
-// CHECK1-NEXT: [[TMP60:%.*]] = load i32, ptr [[LA]], align 4
-// CHECK1-NEXT: store i32 [[TMP60]], ptr [[LA_CASTED15]], align 4
-// CHECK1-NEXT: [[TMP61:%.*]] = load i64, ptr [[LA_CASTED15]], align 8
-// CHECK1-NEXT: [[TMP62:%.*]] = load i32, ptr [[COMP]], align 4
-// CHECK1-NEXT: store i32 [[TMP62]], ptr [[COMP_CASTED16]], align 4
-// CHECK1-NEXT: [[TMP63:%.*]] = load i64, ptr [[COMP_CASTED16]], align 8
-// CHECK1-NEXT: [[TMP64:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS17]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP61]], ptr [[TMP64]], align 8
-// CHECK1-NEXT: [[TMP65:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS18]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP61]], ptr [[TMP65]], align 8
-// CHECK1-NEXT: [[TMP66:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS19]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP66]], align 8
-// CHECK1-NEXT: [[TMP67:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS17]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP63]], ptr [[TMP67]], align 8
-// CHECK1-NEXT: [[TMP68:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS18]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP63]], ptr [[TMP68]], align 8
-// CHECK1-NEXT: [[TMP69:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS19]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP69]], align 8
-// CHECK1-NEXT: [[TMP70:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS17]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP71:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS18]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP72:%.*]] = load i32, ptr [[LA]], align 4
-// CHECK1-NEXT: [[KERNEL_ARGS20:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP73]], align 4
-// CHECK1-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 1
-// CHECK1-NEXT: store i32 2, ptr [[TMP74]], align 4
-// CHECK1-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP70]], ptr [[TMP75]], align 8
-// CHECK1-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP71]], ptr [[TMP76]], align 8
-// CHECK1-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.8, ptr [[TMP77]], align 8
-// CHECK1-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP78]], align 8
-// CHECK1-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 6
+// CHECK1-NEXT: [[TMP73:%.*]] = load i32, ptr [[LA]], align 4
+// CHECK1-NEXT: store i32 [[TMP73]], ptr [[LA_CASTED15]], align 4
+// CHECK1-NEXT: [[TMP74:%.*]] = load i64, ptr [[LA_CASTED15]], align 8
+// CHECK1-NEXT: [[TMP75:%.*]] = load i32, ptr [[COMP]], align 4
+// CHECK1-NEXT: store i32 [[TMP75]], ptr [[COMP_CASTED16]], align 4
+// CHECK1-NEXT: [[TMP76:%.*]] = load i64, ptr [[COMP_CASTED16]], align 8
+// CHECK1-NEXT: [[TMP77:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS17]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP74]], ptr [[TMP77]], align 8
+// CHECK1-NEXT: [[TMP78:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS18]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP74]], ptr [[TMP78]], align 8
+// CHECK1-NEXT: [[TMP79:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS19]], i64 0, i64 0
// CHECK1-NEXT: store ptr null, ptr [[TMP79]], align 8
-// CHECK1-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP80]], align 8
-// CHECK1-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP81]], align 8
-// CHECK1-NEXT: [[TMP82:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 [[TMP72]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l53.region_id, ptr [[KERNEL_ARGS20]])
-// CHECK1-NEXT: [[TMP83:%.*]] = icmp ne i32 [[TMP82]], 0
-// CHECK1-NEXT: br i1 [[TMP83]], label [[OMP_OFFLOAD_FAILED21:%.*]], label [[OMP_OFFLOAD_CONT22:%.*]]
+// CHECK1-NEXT: [[TMP80:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS17]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP76]], ptr [[TMP80]], align 8
+// CHECK1-NEXT: [[TMP81:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS18]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP76]], ptr [[TMP81]], align 8
+// CHECK1-NEXT: [[TMP82:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS19]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP82]], align 8
+// CHECK1-NEXT: [[TMP83:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS17]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP84:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS18]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP85:%.*]] = load i32, ptr [[LA]], align 4
+// CHECK1-NEXT: [[TMP86:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP85]], 0
+// CHECK1-NEXT: [[KERNEL_ARGS20:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
+// CHECK1-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP87]], align 4
+// CHECK1-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 1
+// CHECK1-NEXT: store i32 2, ptr [[TMP88]], align 4
+// CHECK1-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP83]], ptr [[TMP89]], align 8
+// CHECK1-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP84]], ptr [[TMP90]], align 8
+// CHECK1-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.8, ptr [[TMP91]], align 8
+// CHECK1-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP92]], align 8
+// CHECK1-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP93]], align 8
+// CHECK1-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP94]], align 8
+// CHECK1-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP95]], align 8
+// CHECK1-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP96]], align 8
+// CHECK1-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP97]], align 4
+// CHECK1-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP86]], ptr [[TMP98]], align 4
+// CHECK1-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP99]], align 4
+// CHECK1-NEXT: [[TMP100:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 [[TMP85]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l53.region_id, ptr [[KERNEL_ARGS20]])
+// CHECK1-NEXT: [[TMP101:%.*]] = icmp ne i32 [[TMP100]], 0
+// CHECK1-NEXT: br i1 [[TMP101]], label [[OMP_OFFLOAD_FAILED21:%.*]], label [[OMP_OFFLOAD_CONT22:%.*]]
// CHECK1: omp_offload.failed21:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l53(i64 [[TMP61]], i64 [[TMP63]]) #[[ATTR2]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l53(i64 [[TMP74]], i64 [[TMP76]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT22]]
// CHECK1: omp_offload.cont22:
-// CHECK1-NEXT: [[TMP84:%.*]] = load i32, ptr @Gbla, align 4
-// CHECK1-NEXT: store i32 [[TMP84]], ptr [[GBLA_CASTED]], align 4
-// CHECK1-NEXT: [[TMP85:%.*]] = load i64, ptr [[GBLA_CASTED]], align 8
-// CHECK1-NEXT: [[TMP86:%.*]] = load i32, ptr [[A_ADDR]], align 4
-// CHECK1-NEXT: store i32 [[TMP86]], ptr [[A_CASTED]], align 4
-// CHECK1-NEXT: [[TMP87:%.*]] = load i64, ptr [[A_CASTED]], align 8
-// CHECK1-NEXT: [[TMP88:%.*]] = load i64, ptr @Gblb, align 8
-// CHECK1-NEXT: store i64 [[TMP88]], ptr [[GBLB_CASTED]], align 8
-// CHECK1-NEXT: [[TMP89:%.*]] = load i64, ptr [[GBLB_CASTED]], align 8
-// CHECK1-NEXT: [[TMP90:%.*]] = load float, ptr [[LC]], align 4
-// CHECK1-NEXT: store float [[TMP90]], ptr [[LC_CASTED]], align 4
-// CHECK1-NEXT: [[TMP91:%.*]] = load i64, ptr [[LC_CASTED]], align 8
-// CHECK1-NEXT: [[TMP92:%.*]] = load i32, ptr [[COMP]], align 4
-// CHECK1-NEXT: store i32 [[TMP92]], ptr [[COMP_CASTED23]], align 4
-// CHECK1-NEXT: [[TMP93:%.*]] = load i64, ptr [[COMP_CASTED23]], align 8
-// CHECK1-NEXT: [[TMP94:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP85]], ptr [[TMP94]], align 8
-// CHECK1-NEXT: [[TMP95:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP85]], ptr [[TMP95]], align 8
-// CHECK1-NEXT: [[TMP96:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP96]], align 8
-// CHECK1-NEXT: [[TMP97:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP87]], ptr [[TMP97]], align 8
-// CHECK1-NEXT: [[TMP98:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP87]], ptr [[TMP98]], align 8
-// CHECK1-NEXT: [[TMP99:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP99]], align 8
-// CHECK1-NEXT: [[TMP100:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 2
-// CHECK1-NEXT: store i64 [[TMP89]], ptr [[TMP100]], align 8
-// CHECK1-NEXT: [[TMP101:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 2
-// CHECK1-NEXT: store i64 [[TMP89]], ptr [[TMP101]], align 8
-// CHECK1-NEXT: [[TMP102:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP102]], align 8
-// CHECK1-NEXT: [[TMP103:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 3
-// CHECK1-NEXT: store i64 [[TMP91]], ptr [[TMP103]], align 8
-// CHECK1-NEXT: [[TMP104:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 3
-// CHECK1-NEXT: store i64 [[TMP91]], ptr [[TMP104]], align 8
-// CHECK1-NEXT: [[TMP105:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP105]], align 8
-// CHECK1-NEXT: [[TMP106:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 4
-// CHECK1-NEXT: store i64 [[TMP93]], ptr [[TMP106]], align 8
-// CHECK1-NEXT: [[TMP107:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 4
-// CHECK1-NEXT: store i64 [[TMP93]], ptr [[TMP107]], align 8
-// CHECK1-NEXT: [[TMP108:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP108]], align 8
-// CHECK1-NEXT: [[TMP109:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP110:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP111:%.*]] = load i32, ptr @Gbla, align 4
-// CHECK1-NEXT: [[TMP112:%.*]] = load i32, ptr [[A_ADDR]], align 4
-// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP111]], [[TMP112]]
-// CHECK1-NEXT: [[TMP113:%.*]] = load i64, ptr @Gblb, align 8
-// CHECK1-NEXT: [[TMP114:%.*]] = load float, ptr [[LC]], align 4
-// CHECK1-NEXT: [[CONV:%.*]] = fptosi float [[TMP114]] to i64
-// CHECK1-NEXT: [[ADD27:%.*]] = add nsw i64 [[TMP113]], [[CONV]]
-// CHECK1-NEXT: [[TMP115:%.*]] = trunc i64 [[ADD27]] to i32
-// CHECK1-NEXT: [[KERNEL_ARGS28:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP116:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP116]], align 4
-// CHECK1-NEXT: [[TMP117:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
-// CHECK1-NEXT: store i32 5, ptr [[TMP117]], align 4
-// CHECK1-NEXT: [[TMP118:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP109]], ptr [[TMP118]], align 8
-// CHECK1-NEXT: [[TMP119:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP110]], ptr [[TMP119]], align 8
-// CHECK1-NEXT: [[TMP120:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.11, ptr [[TMP120]], align 8
-// CHECK1-NEXT: [[TMP121:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP121]], align 8
-// CHECK1-NEXT: [[TMP122:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP122]], align 8
-// CHECK1-NEXT: [[TMP123:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP102:%.*]] = load i32, ptr @Gbla, align 4
+// CHECK1-NEXT: store i32 [[TMP102]], ptr [[GBLA_CASTED]], align 4
+// CHECK1-NEXT: [[TMP103:%.*]] = load i64, ptr [[GBLA_CASTED]], align 8
+// CHECK1-NEXT: [[TMP104:%.*]] = load i32, ptr [[A_ADDR]], align 4
+// CHECK1-NEXT: store i32 [[TMP104]], ptr [[A_CASTED]], align 4
+// CHECK1-NEXT: [[TMP105:%.*]] = load i64, ptr [[A_CASTED]], align 8
+// CHECK1-NEXT: [[TMP106:%.*]] = load i64, ptr @Gblb, align 8
+// CHECK1-NEXT: store i64 [[TMP106]], ptr [[GBLB_CASTED]], align 8
+// CHECK1-NEXT: [[TMP107:%.*]] = load i64, ptr [[GBLB_CASTED]], align 8
+// CHECK1-NEXT: [[TMP108:%.*]] = load float, ptr [[LC]], align 4
+// CHECK1-NEXT: store float [[TMP108]], ptr [[LC_CASTED]], align 4
+// CHECK1-NEXT: [[TMP109:%.*]] = load i64, ptr [[LC_CASTED]], align 8
+// CHECK1-NEXT: [[TMP110:%.*]] = load i32, ptr [[COMP]], align 4
+// CHECK1-NEXT: store i32 [[TMP110]], ptr [[COMP_CASTED23]], align 4
+// CHECK1-NEXT: [[TMP111:%.*]] = load i64, ptr [[COMP_CASTED23]], align 8
+// CHECK1-NEXT: [[TMP112:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP103]], ptr [[TMP112]], align 8
+// CHECK1-NEXT: [[TMP113:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP103]], ptr [[TMP113]], align 8
+// CHECK1-NEXT: [[TMP114:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP114]], align 8
+// CHECK1-NEXT: [[TMP115:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP105]], ptr [[TMP115]], align 8
+// CHECK1-NEXT: [[TMP116:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP105]], ptr [[TMP116]], align 8
+// CHECK1-NEXT: [[TMP117:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP117]], align 8
+// CHECK1-NEXT: [[TMP118:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 2
+// CHECK1-NEXT: store i64 [[TMP107]], ptr [[TMP118]], align 8
+// CHECK1-NEXT: [[TMP119:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 2
+// CHECK1-NEXT: store i64 [[TMP107]], ptr [[TMP119]], align 8
+// CHECK1-NEXT: [[TMP120:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 2
+// CHECK1-NEXT: store ptr null, ptr [[TMP120]], align 8
+// CHECK1-NEXT: [[TMP121:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 3
+// CHECK1-NEXT: store i64 [[TMP109]], ptr [[TMP121]], align 8
+// CHECK1-NEXT: [[TMP122:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 3
+// CHECK1-NEXT: store i64 [[TMP109]], ptr [[TMP122]], align 8
+// CHECK1-NEXT: [[TMP123:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 3
// CHECK1-NEXT: store ptr null, ptr [[TMP123]], align 8
-// CHECK1-NEXT: [[TMP124:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP124]], align 8
-// CHECK1-NEXT: [[TMP125:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[ADD]], i32 [[TMP115]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l62.region_id, ptr [[KERNEL_ARGS28]])
-// CHECK1-NEXT: [[TMP126:%.*]] = icmp ne i32 [[TMP125]], 0
-// CHECK1-NEXT: br i1 [[TMP126]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
+// CHECK1-NEXT: [[TMP124:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 4
+// CHECK1-NEXT: store i64 [[TMP111]], ptr [[TMP124]], align 8
+// CHECK1-NEXT: [[TMP125:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 4
+// CHECK1-NEXT: store i64 [[TMP111]], ptr [[TMP125]], align 8
+// CHECK1-NEXT: [[TMP126:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP126]], align 8
+// CHECK1-NEXT: [[TMP127:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP128:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP129:%.*]] = load i32, ptr @Gbla, align 4
+// CHECK1-NEXT: [[TMP130:%.*]] = load i32, ptr [[A_ADDR]], align 4
+// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP129]], [[TMP130]]
+// CHECK1-NEXT: [[TMP131:%.*]] = load i64, ptr @Gblb, align 8
+// CHECK1-NEXT: [[TMP132:%.*]] = load float, ptr [[LC]], align 4
+// CHECK1-NEXT: [[CONV:%.*]] = fptosi float [[TMP132]] to i64
+// CHECK1-NEXT: [[ADD27:%.*]] = add nsw i64 [[TMP131]], [[CONV]]
+// CHECK1-NEXT: [[TMP133:%.*]] = trunc i64 [[ADD27]] to i32
+// CHECK1-NEXT: [[TMP134:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[ADD]], 0
+// CHECK1-NEXT: [[TMP135:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP133]], 0
+// CHECK1-NEXT: [[KERNEL_ARGS28:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
+// CHECK1-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP136]], align 4
+// CHECK1-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
+// CHECK1-NEXT: store i32 5, ptr [[TMP137]], align 4
+// CHECK1-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP127]], ptr [[TMP138]], align 8
+// CHECK1-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP128]], ptr [[TMP139]], align 8
+// CHECK1-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.11, ptr [[TMP140]], align 8
+// CHECK1-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP141]], align 8
+// CHECK1-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP142]], align 8
+// CHECK1-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP143]], align 8
+// CHECK1-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP144]], align 8
+// CHECK1-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP145]], align 8
+// CHECK1-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [[TMP134]], ptr [[TMP146]], align 4
+// CHECK1-NEXT: [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP135]], ptr [[TMP147]], align 4
+// CHECK1-NEXT: [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP148]], align 4
+// CHECK1-NEXT: [[TMP149:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[ADD]], i32 [[TMP133]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l62.region_id, ptr [[KERNEL_ARGS28]])
+// CHECK1-NEXT: [[TMP150:%.*]] = icmp ne i32 [[TMP149]], 0
+// CHECK1-NEXT: br i1 [[TMP150]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
// CHECK1: omp_offload.failed29:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l62(i64 [[TMP85]], i64 [[TMP87]], i64 [[TMP89]], i64 [[TMP91]], i64 [[TMP93]]) #[[ATTR2]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l62(i64 [[TMP103]], i64 [[TMP105]], i64 [[TMP107]], i64 [[TMP109]], i64 [[TMP111]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT30]]
// CHECK1: omp_offload.cont30:
-// CHECK1-NEXT: [[TMP127:%.*]] = load ptr, ptr @Gblc, align 8
-// CHECK1-NEXT: store ptr [[TMP127]], ptr [[TMP]], align 8
-// CHECK1-NEXT: [[TMP128:%.*]] = load i32, ptr @Gbla, align 4
-// CHECK1-NEXT: store i32 [[TMP128]], ptr [[GBLC_CASTED]], align 4
-// CHECK1-NEXT: [[TMP129:%.*]] = load i64, ptr [[GBLC_CASTED]], align 8
-// CHECK1-NEXT: [[TMP130:%.*]] = load i32, ptr [[COMP]], align 4
-// CHECK1-NEXT: store i32 [[TMP130]], ptr [[COMP_CASTED31]], align 4
-// CHECK1-NEXT: [[TMP131:%.*]] = load i64, ptr [[COMP_CASTED31]], align 8
-// CHECK1-NEXT: [[TMP132:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP129]], ptr [[TMP132]], align 8
-// CHECK1-NEXT: [[TMP133:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP129]], ptr [[TMP133]], align 8
-// CHECK1-NEXT: [[TMP134:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP134]], align 8
-// CHECK1-NEXT: [[TMP135:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP131]], ptr [[TMP135]], align 8
-// CHECK1-NEXT: [[TMP136:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP131]], ptr [[TMP136]], align 8
-// CHECK1-NEXT: [[TMP137:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP137]], align 8
-// CHECK1-NEXT: [[TMP138:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP139:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP140:%.*]] = load ptr, ptr @Gblc, align 8
-// CHECK1-NEXT: store ptr [[TMP140]], ptr [[_TMP35]], align 8
-// CHECK1-NEXT: [[TMP141:%.*]] = load i32, ptr @Gbla, align 4
-// CHECK1-NEXT: [[ADD36:%.*]] = add nsw i32 [[TMP141]], 1
-// CHECK1-NEXT: [[TMP142:%.*]] = load ptr, ptr @Gblc, align 8
-// CHECK1-NEXT: store ptr [[TMP142]], ptr [[_TMP37]], align 8
-// CHECK1-NEXT: [[TMP143:%.*]] = load i32, ptr @Gbla, align 4
-// CHECK1-NEXT: [[ADD38:%.*]] = add nsw i32 [[TMP143]], 2
+// CHECK1-NEXT: [[TMP151:%.*]] = load ptr, ptr @Gblc, align 8
+// CHECK1-NEXT: store ptr [[TMP151]], ptr [[TMP]], align 8
+// CHECK1-NEXT: [[TMP152:%.*]] = load i32, ptr @Gbla, align 4
+// CHECK1-NEXT: store i32 [[TMP152]], ptr [[GBLC_CASTED]], align 4
+// CHECK1-NEXT: [[TMP153:%.*]] = load i64, ptr [[GBLC_CASTED]], align 8
+// CHECK1-NEXT: [[TMP154:%.*]] = load i32, ptr [[COMP]], align 4
+// CHECK1-NEXT: store i32 [[TMP154]], ptr [[COMP_CASTED31]], align 4
+// CHECK1-NEXT: [[TMP155:%.*]] = load i64, ptr [[COMP_CASTED31]], align 8
+// CHECK1-NEXT: [[TMP156:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP153]], ptr [[TMP156]], align 8
+// CHECK1-NEXT: [[TMP157:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP153]], ptr [[TMP157]], align 8
+// CHECK1-NEXT: [[TMP158:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP158]], align 8
+// CHECK1-NEXT: [[TMP159:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP155]], ptr [[TMP159]], align 8
+// CHECK1-NEXT: [[TMP160:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP155]], ptr [[TMP160]], align 8
+// CHECK1-NEXT: [[TMP161:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP161]], align 8
+// CHECK1-NEXT: [[TMP162:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP163:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP164:%.*]] = load ptr, ptr @Gblc, align 8
+// CHECK1-NEXT: store ptr [[TMP164]], ptr [[_TMP35]], align 8
+// CHECK1-NEXT: [[TMP165:%.*]] = load i32, ptr @Gbla, align 4
+// CHECK1-NEXT: [[ADD36:%.*]] = add nsw i32 [[TMP165]], 1
+// CHECK1-NEXT: [[TMP166:%.*]] = load ptr, ptr @Gblc, align 8
+// CHECK1-NEXT: store ptr [[TMP166]], ptr [[_TMP37]], align 8
+// CHECK1-NEXT: [[TMP167:%.*]] = load i32, ptr @Gbla, align 4
+// CHECK1-NEXT: [[ADD38:%.*]] = add nsw i32 [[TMP167]], 2
+// CHECK1-NEXT: [[TMP168:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[ADD36]], 0
+// CHECK1-NEXT: [[TMP169:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[ADD38]], 0
// CHECK1-NEXT: [[KERNEL_ARGS39:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP144]], align 4
-// CHECK1-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 1
-// CHECK1-NEXT: store i32 2, ptr [[TMP145]], align 4
-// CHECK1-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP138]], ptr [[TMP146]], align 8
-// CHECK1-NEXT: [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP139]], ptr [[TMP147]], align 8
-// CHECK1-NEXT: [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.14, ptr [[TMP148]], align 8
-// CHECK1-NEXT: [[TMP149:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.15, ptr [[TMP149]], align 8
-// CHECK1-NEXT: [[TMP150:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP150]], align 8
-// CHECK1-NEXT: [[TMP151:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP151]], align 8
-// CHECK1-NEXT: [[TMP152:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 8
-// CHECK1-NEXT: store i64 0, ptr [[TMP152]], align 8
-// CHECK1-NEXT: [[TMP153:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[ADD36]], i32 [[ADD38]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l71.region_id, ptr [[KERNEL_ARGS39]])
-// CHECK1-NEXT: [[TMP154:%.*]] = icmp ne i32 [[TMP153]], 0
-// CHECK1-NEXT: br i1 [[TMP154]], label [[OMP_OFFLOAD_FAILED40:%.*]], label [[OMP_OFFLOAD_CONT41:%.*]]
+// CHECK1-NEXT: [[TMP170:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP170]], align 4
+// CHECK1-NEXT: [[TMP171:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 1
+// CHECK1-NEXT: store i32 2, ptr [[TMP171]], align 4
+// CHECK1-NEXT: [[TMP172:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP162]], ptr [[TMP172]], align 8
+// CHECK1-NEXT: [[TMP173:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP163]], ptr [[TMP173]], align 8
+// CHECK1-NEXT: [[TMP174:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.14, ptr [[TMP174]], align 8
+// CHECK1-NEXT: [[TMP175:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.15, ptr [[TMP175]], align 8
+// CHECK1-NEXT: [[TMP176:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP176]], align 8
+// CHECK1-NEXT: [[TMP177:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP177]], align 8
+// CHECK1-NEXT: [[TMP178:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 8
+// CHECK1-NEXT: store i64 0, ptr [[TMP178]], align 8
+// CHECK1-NEXT: [[TMP179:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP179]], align 8
+// CHECK1-NEXT: [[TMP180:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [[TMP168]], ptr [[TMP180]], align 4
+// CHECK1-NEXT: [[TMP181:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP169]], ptr [[TMP181]], align 4
+// CHECK1-NEXT: [[TMP182:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP182]], align 4
+// CHECK1-NEXT: [[TMP183:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[ADD36]], i32 [[ADD38]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l71.region_id, ptr [[KERNEL_ARGS39]])
+// CHECK1-NEXT: [[TMP184:%.*]] = icmp ne i32 [[TMP183]], 0
+// CHECK1-NEXT: br i1 [[TMP184]], label [[OMP_OFFLOAD_FAILED40:%.*]], label [[OMP_OFFLOAD_CONT41:%.*]]
// CHECK1: omp_offload.failed40:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l71(i64 [[TMP129]], i64 [[TMP131]]) #[[ATTR2]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l71(i64 [[TMP153]], i64 [[TMP155]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT41]]
// CHECK1: omp_offload.cont41:
-// CHECK1-NEXT: [[TMP155:%.*]] = load i32, ptr [[COMP]], align 4
-// CHECK1-NEXT: ret i32 [[TMP155]]
+// CHECK1-NEXT: [[TMP185:%.*]] = load i32, ptr [[COMP]], align 4
+// CHECK1-NEXT: ret i32 [[TMP185]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l31
@@ -911,7 +965,7 @@ void foo() {
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -928,280 +982,334 @@ void foo() {
// CHECK3-NEXT: store ptr null, ptr [[TMP14]], align 4
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP15]], align 8
-// CHECK3-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l31.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK3-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK3-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l31.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK3-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l31(i32 [[TMP1]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
-// CHECK3-NEXT: [[TMP18:%.*]] = load i32, ptr [[COMP]], align 4
-// CHECK3-NEXT: store i32 [[TMP18]], ptr [[COMP_CASTED1]], align 4
-// CHECK3-NEXT: [[TMP19:%.*]] = load i32, ptr [[COMP_CASTED1]], align 4
-// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP19]], ptr [[TMP20]], align 4
-// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP19]], ptr [[TMP21]], align 4
-// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS4]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP22]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP22:%.*]] = load i32, ptr [[COMP]], align 4
+// CHECK3-NEXT: store i32 [[TMP22]], ptr [[COMP_CASTED1]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = load i32, ptr [[COMP_CASTED1]], align 4
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP23]], ptr [[TMP24]], align 4
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP23]], ptr [[TMP25]], align 4
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS4]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP25]], align 4
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP26]], align 4
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
-// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.2, ptr [[TMP29]], align 4
-// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP30]], align 4
-// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP31]], align 4
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP32]], align 4
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP33]], align 8
-// CHECK3-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l37.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK3-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
-// CHECK3-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP27]], ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP28]], ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.2, ptr [[TMP33]], align 4
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP34]], align 4
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP35]], align 4
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP36]], align 4
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP37]], align 8
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP38]], align 8
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP39]], align 4
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l37.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK3-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
+// CHECK3-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK3: omp_offload.failed6:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l37(i32 [[TMP19]]) #[[ATTR2]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l37(i32 [[TMP23]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK3: omp_offload.cont7:
-// CHECK3-NEXT: [[TMP36:%.*]] = load i32, ptr [[LA]], align 4
-// CHECK3-NEXT: store i32 [[TMP36]], ptr [[LA_CASTED]], align 4
-// CHECK3-NEXT: [[TMP37:%.*]] = load i32, ptr [[LA_CASTED]], align 4
-// CHECK3-NEXT: [[TMP38:%.*]] = load i32, ptr [[COMP]], align 4
-// CHECK3-NEXT: store i32 [[TMP38]], ptr [[COMP_CASTED8]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = load i32, ptr [[COMP_CASTED8]], align 4
-// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS9]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP37]], ptr [[TMP40]], align 4
-// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS10]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP37]], ptr [[TMP41]], align 4
-// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS11]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP42]], align 4
-// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS9]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP39]], ptr [[TMP43]], align 4
-// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS10]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP39]], ptr [[TMP44]], align 4
-// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS11]], i32 0, i32 1
-// CHECK3-NEXT: store ptr null, ptr [[TMP45]], align 4
-// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS9]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS10]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP48:%.*]] = load i32, ptr [[LA]], align 4
+// CHECK3-NEXT: [[TMP44:%.*]] = load i32, ptr [[LA]], align 4
+// CHECK3-NEXT: store i32 [[TMP44]], ptr [[LA_CASTED]], align 4
+// CHECK3-NEXT: [[TMP45:%.*]] = load i32, ptr [[LA_CASTED]], align 4
+// CHECK3-NEXT: [[TMP46:%.*]] = load i32, ptr [[COMP]], align 4
+// CHECK3-NEXT: store i32 [[TMP46]], ptr [[COMP_CASTED8]], align 4
+// CHECK3-NEXT: [[TMP47:%.*]] = load i32, ptr [[COMP_CASTED8]], align 4
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS9]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP45]], ptr [[TMP48]], align 4
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS10]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP45]], ptr [[TMP49]], align 4
+// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS11]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP50]], align 4
+// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS9]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP47]], ptr [[TMP51]], align 4
+// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS10]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP47]], ptr [[TMP52]], align 4
+// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS11]], i32 0, i32 1
+// CHECK3-NEXT: store ptr null, ptr [[TMP53]], align 4
+// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS9]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS10]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP56:%.*]] = load i32, ptr [[LA]], align 4
+// CHECK3-NEXT: [[TMP57:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP56]], 0
// CHECK3-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP49]], align 4
-// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK3-NEXT: store i32 2, ptr [[TMP50]], align 4
-// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP46]], ptr [[TMP51]], align 4
-// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP47]], ptr [[TMP52]], align 4
-// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.5, ptr [[TMP53]], align 4
-// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP54]], align 4
-// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP55]], align 4
-// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP56]], align 4
-// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP57]], align 8
-// CHECK3-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP48]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l46.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK3-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
-// CHECK3-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK3-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP58]], align 4
+// CHECK3-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK3-NEXT: store i32 2, ptr [[TMP59]], align 4
+// CHECK3-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP54]], ptr [[TMP60]], align 4
+// CHECK3-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP55]], ptr [[TMP61]], align 4
+// CHECK3-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.5, ptr [[TMP62]], align 4
+// CHECK3-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP63]], align 4
+// CHECK3-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP64]], align 4
+// CHECK3-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP65]], align 4
+// CHECK3-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP66]], align 8
+// CHECK3-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP67]], align 8
+// CHECK3-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [[TMP57]], ptr [[TMP68]], align 4
+// CHECK3-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK3-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP70]], align 4
+// CHECK3-NEXT: [[TMP71:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP56]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l46.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK3-NEXT: [[TMP72:%.*]] = icmp ne i32 [[TMP71]], 0
+// CHECK3-NEXT: br i1 [[TMP72]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK3: omp_offload.failed13:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l46(i32 [[TMP37]], i32 [[TMP39]]) #[[ATTR2]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l46(i32 [[TMP45]], i32 [[TMP47]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT14]]
// CHECK3: omp_offload.cont14:
-// CHECK3-NEXT: [[TMP60:%.*]] = load i32, ptr [[LA]], align 4
-// CHECK3-NEXT: store i32 [[TMP60]], ptr [[LA_CASTED15]], align 4
-// CHECK3-NEXT: [[TMP61:%.*]] = load i32, ptr [[LA_CASTED15]], align 4
-// CHECK3-NEXT: [[TMP62:%.*]] = load i32, ptr [[COMP]], align 4
-// CHECK3-NEXT: store i32 [[TMP62]], ptr [[COMP_CASTED16]], align 4
-// CHECK3-NEXT: [[TMP63:%.*]] = load i32, ptr [[COMP_CASTED16]], align 4
-// CHECK3-NEXT: [[TMP64:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS17]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP61]], ptr [[TMP64]], align 4
-// CHECK3-NEXT: [[TMP65:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS18]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP61]], ptr [[TMP65]], align 4
-// CHECK3-NEXT: [[TMP66:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS19]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP66]], align 4
-// CHECK3-NEXT: [[TMP67:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS17]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP63]], ptr [[TMP67]], align 4
-// CHECK3-NEXT: [[TMP68:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS18]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP63]], ptr [[TMP68]], align 4
-// CHECK3-NEXT: [[TMP69:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS19]], i32 0, i32 1
-// CHECK3-NEXT: store ptr null, ptr [[TMP69]], align 4
-// CHECK3-NEXT: [[TMP70:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS17]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP71:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS18]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP72:%.*]] = load i32, ptr [[LA]], align 4
-// CHECK3-NEXT: [[KERNEL_ARGS20:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP73]], align 4
-// CHECK3-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 1
-// CHECK3-NEXT: store i32 2, ptr [[TMP74]], align 4
-// CHECK3-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP70]], ptr [[TMP75]], align 4
-// CHECK3-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP71]], ptr [[TMP76]], align 4
-// CHECK3-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.8, ptr [[TMP77]], align 4
-// CHECK3-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP78]], align 4
-// CHECK3-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 6
+// CHECK3-NEXT: [[TMP73:%.*]] = load i32, ptr [[LA]], align 4
+// CHECK3-NEXT: store i32 [[TMP73]], ptr [[LA_CASTED15]], align 4
+// CHECK3-NEXT: [[TMP74:%.*]] = load i32, ptr [[LA_CASTED15]], align 4
+// CHECK3-NEXT: [[TMP75:%.*]] = load i32, ptr [[COMP]], align 4
+// CHECK3-NEXT: store i32 [[TMP75]], ptr [[COMP_CASTED16]], align 4
+// CHECK3-NEXT: [[TMP76:%.*]] = load i32, ptr [[COMP_CASTED16]], align 4
+// CHECK3-NEXT: [[TMP77:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS17]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP74]], ptr [[TMP77]], align 4
+// CHECK3-NEXT: [[TMP78:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS18]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP74]], ptr [[TMP78]], align 4
+// CHECK3-NEXT: [[TMP79:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS19]], i32 0, i32 0
// CHECK3-NEXT: store ptr null, ptr [[TMP79]], align 4
-// CHECK3-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP80]], align 4
-// CHECK3-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP81]], align 8
-// CHECK3-NEXT: [[TMP82:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 [[TMP72]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l53.region_id, ptr [[KERNEL_ARGS20]])
-// CHECK3-NEXT: [[TMP83:%.*]] = icmp ne i32 [[TMP82]], 0
-// CHECK3-NEXT: br i1 [[TMP83]], label [[OMP_OFFLOAD_FAILED21:%.*]], label [[OMP_OFFLOAD_CONT22:%.*]]
+// CHECK3-NEXT: [[TMP80:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS17]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP76]], ptr [[TMP80]], align 4
+// CHECK3-NEXT: [[TMP81:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS18]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP76]], ptr [[TMP81]], align 4
+// CHECK3-NEXT: [[TMP82:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS19]], i32 0, i32 1
+// CHECK3-NEXT: store ptr null, ptr [[TMP82]], align 4
+// CHECK3-NEXT: [[TMP83:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS17]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP84:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS18]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP85:%.*]] = load i32, ptr [[LA]], align 4
+// CHECK3-NEXT: [[TMP86:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP85]], 0
+// CHECK3-NEXT: [[KERNEL_ARGS20:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
+// CHECK3-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP87]], align 4
+// CHECK3-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 1
+// CHECK3-NEXT: store i32 2, ptr [[TMP88]], align 4
+// CHECK3-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP83]], ptr [[TMP89]], align 4
+// CHECK3-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP84]], ptr [[TMP90]], align 4
+// CHECK3-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.8, ptr [[TMP91]], align 4
+// CHECK3-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP92]], align 4
+// CHECK3-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP93]], align 4
+// CHECK3-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP94]], align 4
+// CHECK3-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP95]], align 8
+// CHECK3-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP96]], align 8
+// CHECK3-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP97]], align 4
+// CHECK3-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [[TMP86]], ptr [[TMP98]], align 4
+// CHECK3-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS20]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP99]], align 4
+// CHECK3-NEXT: [[TMP100:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 [[TMP85]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l53.region_id, ptr [[KERNEL_ARGS20]])
+// CHECK3-NEXT: [[TMP101:%.*]] = icmp ne i32 [[TMP100]], 0
+// CHECK3-NEXT: br i1 [[TMP101]], label [[OMP_OFFLOAD_FAILED21:%.*]], label [[OMP_OFFLOAD_CONT22:%.*]]
// CHECK3: omp_offload.failed21:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l53(i32 [[TMP61]], i32 [[TMP63]]) #[[ATTR2]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l53(i32 [[TMP74]], i32 [[TMP76]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT22]]
// CHECK3: omp_offload.cont22:
-// CHECK3-NEXT: [[TMP84:%.*]] = load i32, ptr @Gbla, align 4
-// CHECK3-NEXT: store i32 [[TMP84]], ptr [[GBLA_CASTED]], align 4
-// CHECK3-NEXT: [[TMP85:%.*]] = load i32, ptr [[GBLA_CASTED]], align 4
-// CHECK3-NEXT: [[TMP86:%.*]] = load i32, ptr [[A_ADDR]], align 4
-// CHECK3-NEXT: store i32 [[TMP86]], ptr [[A_CASTED]], align 4
-// CHECK3-NEXT: [[TMP87:%.*]] = load i32, ptr [[A_CASTED]], align 4
-// CHECK3-NEXT: [[TMP88:%.*]] = load float, ptr [[LC]], align 4
-// CHECK3-NEXT: store float [[TMP88]], ptr [[LC_CASTED]], align 4
-// CHECK3-NEXT: [[TMP89:%.*]] = load i32, ptr [[LC_CASTED]], align 4
-// CHECK3-NEXT: [[TMP90:%.*]] = load i32, ptr [[COMP]], align 4
-// CHECK3-NEXT: store i32 [[TMP90]], ptr [[COMP_CASTED23]], align 4
-// CHECK3-NEXT: [[TMP91:%.*]] = load i32, ptr [[COMP_CASTED23]], align 4
-// CHECK3-NEXT: [[TMP92:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP85]], ptr [[TMP92]], align 4
-// CHECK3-NEXT: [[TMP93:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP85]], ptr [[TMP93]], align 4
-// CHECK3-NEXT: [[TMP94:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP94]], align 4
-// CHECK3-NEXT: [[TMP95:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP87]], ptr [[TMP95]], align 4
-// CHECK3-NEXT: [[TMP96:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP87]], ptr [[TMP96]], align 4
-// CHECK3-NEXT: [[TMP97:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 1
-// CHECK3-NEXT: store ptr null, ptr [[TMP97]], align 4
-// CHECK3-NEXT: [[TMP98:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 2
-// CHECK3-NEXT: store ptr @Gblb, ptr [[TMP98]], align 4
-// CHECK3-NEXT: [[TMP99:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 2
-// CHECK3-NEXT: store ptr @Gblb, ptr [[TMP99]], align 4
-// CHECK3-NEXT: [[TMP100:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 2
-// CHECK3-NEXT: store ptr null, ptr [[TMP100]], align 4
-// CHECK3-NEXT: [[TMP101:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 3
-// CHECK3-NEXT: store i32 [[TMP89]], ptr [[TMP101]], align 4
-// CHECK3-NEXT: [[TMP102:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 3
-// CHECK3-NEXT: store i32 [[TMP89]], ptr [[TMP102]], align 4
-// CHECK3-NEXT: [[TMP103:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 3
-// CHECK3-NEXT: store ptr null, ptr [[TMP103]], align 4
-// CHECK3-NEXT: [[TMP104:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 4
-// CHECK3-NEXT: store i32 [[TMP91]], ptr [[TMP104]], align 4
-// CHECK3-NEXT: [[TMP105:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 4
-// CHECK3-NEXT: store i32 [[TMP91]], ptr [[TMP105]], align 4
-// CHECK3-NEXT: [[TMP106:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 4
-// CHECK3-NEXT: store ptr null, ptr [[TMP106]], align 4
-// CHECK3-NEXT: [[TMP107:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP108:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP109:%.*]] = load i32, ptr @Gbla, align 4
-// CHECK3-NEXT: [[TMP110:%.*]] = load i32, ptr [[A_ADDR]], align 4
-// CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP109]], [[TMP110]]
-// CHECK3-NEXT: [[TMP111:%.*]] = load i64, ptr @Gblb, align 8
-// CHECK3-NEXT: [[TMP112:%.*]] = load float, ptr [[LC]], align 4
-// CHECK3-NEXT: [[CONV:%.*]] = fptosi float [[TMP112]] to i64
-// CHECK3-NEXT: [[ADD27:%.*]] = add nsw i64 [[TMP111]], [[CONV]]
-// CHECK3-NEXT: [[TMP113:%.*]] = trunc i64 [[ADD27]] to i32
-// CHECK3-NEXT: [[KERNEL_ARGS28:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP114:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP114]], align 4
-// CHECK3-NEXT: [[TMP115:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
-// CHECK3-NEXT: store i32 5, ptr [[TMP115]], align 4
-// CHECK3-NEXT: [[TMP116:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP107]], ptr [[TMP116]], align 4
-// CHECK3-NEXT: [[TMP117:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP108]], ptr [[TMP117]], align 4
-// CHECK3-NEXT: [[TMP118:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.11, ptr [[TMP118]], align 4
-// CHECK3-NEXT: [[TMP119:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP119]], align 4
-// CHECK3-NEXT: [[TMP120:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP120]], align 4
-// CHECK3-NEXT: [[TMP121:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
+// CHECK3-NEXT: [[TMP102:%.*]] = load i32, ptr @Gbla, align 4
+// CHECK3-NEXT: store i32 [[TMP102]], ptr [[GBLA_CASTED]], align 4
+// CHECK3-NEXT: [[TMP103:%.*]] = load i32, ptr [[GBLA_CASTED]], align 4
+// CHECK3-NEXT: [[TMP104:%.*]] = load i32, ptr [[A_ADDR]], align 4
+// CHECK3-NEXT: store i32 [[TMP104]], ptr [[A_CASTED]], align 4
+// CHECK3-NEXT: [[TMP105:%.*]] = load i32, ptr [[A_CASTED]], align 4
+// CHECK3-NEXT: [[TMP106:%.*]] = load float, ptr [[LC]], align 4
+// CHECK3-NEXT: store float [[TMP106]], ptr [[LC_CASTED]], align 4
+// CHECK3-NEXT: [[TMP107:%.*]] = load i32, ptr [[LC_CASTED]], align 4
+// CHECK3-NEXT: [[TMP108:%.*]] = load i32, ptr [[COMP]], align 4
+// CHECK3-NEXT: store i32 [[TMP108]], ptr [[COMP_CASTED23]], align 4
+// CHECK3-NEXT: [[TMP109:%.*]] = load i32, ptr [[COMP_CASTED23]], align 4
+// CHECK3-NEXT: [[TMP110:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP103]], ptr [[TMP110]], align 4
+// CHECK3-NEXT: [[TMP111:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP103]], ptr [[TMP111]], align 4
+// CHECK3-NEXT: [[TMP112:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP112]], align 4
+// CHECK3-NEXT: [[TMP113:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP105]], ptr [[TMP113]], align 4
+// CHECK3-NEXT: [[TMP114:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP105]], ptr [[TMP114]], align 4
+// CHECK3-NEXT: [[TMP115:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 1
+// CHECK3-NEXT: store ptr null, ptr [[TMP115]], align 4
+// CHECK3-NEXT: [[TMP116:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 2
+// CHECK3-NEXT: store ptr @Gblb, ptr [[TMP116]], align 4
+// CHECK3-NEXT: [[TMP117:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 2
+// CHECK3-NEXT: store ptr @Gblb, ptr [[TMP117]], align 4
+// CHECK3-NEXT: [[TMP118:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 2
+// CHECK3-NEXT: store ptr null, ptr [[TMP118]], align 4
+// CHECK3-NEXT: [[TMP119:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 3
+// CHECK3-NEXT: store i32 [[TMP107]], ptr [[TMP119]], align 4
+// CHECK3-NEXT: [[TMP120:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 3
+// CHECK3-NEXT: store i32 [[TMP107]], ptr [[TMP120]], align 4
+// CHECK3-NEXT: [[TMP121:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 3
// CHECK3-NEXT: store ptr null, ptr [[TMP121]], align 4
-// CHECK3-NEXT: [[TMP122:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP122]], align 8
-// CHECK3-NEXT: [[TMP123:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[ADD]], i32 [[TMP113]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l62.region_id, ptr [[KERNEL_ARGS28]])
-// CHECK3-NEXT: [[TMP124:%.*]] = icmp ne i32 [[TMP123]], 0
-// CHECK3-NEXT: br i1 [[TMP124]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
+// CHECK3-NEXT: [[TMP122:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 4
+// CHECK3-NEXT: store i32 [[TMP109]], ptr [[TMP122]], align 4
+// CHECK3-NEXT: [[TMP123:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 4
+// CHECK3-NEXT: store i32 [[TMP109]], ptr [[TMP123]], align 4
+// CHECK3-NEXT: [[TMP124:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 4
+// CHECK3-NEXT: store ptr null, ptr [[TMP124]], align 4
+// CHECK3-NEXT: [[TMP125:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP126:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP127:%.*]] = load i32, ptr @Gbla, align 4
+// CHECK3-NEXT: [[TMP128:%.*]] = load i32, ptr [[A_ADDR]], align 4
+// CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP127]], [[TMP128]]
+// CHECK3-NEXT: [[TMP129:%.*]] = load i64, ptr @Gblb, align 8
+// CHECK3-NEXT: [[TMP130:%.*]] = load float, ptr [[LC]], align 4
+// CHECK3-NEXT: [[CONV:%.*]] = fptosi float [[TMP130]] to i64
+// CHECK3-NEXT: [[ADD27:%.*]] = add nsw i64 [[TMP129]], [[CONV]]
+// CHECK3-NEXT: [[TMP131:%.*]] = trunc i64 [[ADD27]] to i32
+// CHECK3-NEXT: [[TMP132:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[ADD]], 0
+// CHECK3-NEXT: [[TMP133:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP131]], 0
+// CHECK3-NEXT: [[KERNEL_ARGS28:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
+// CHECK3-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP134]], align 4
+// CHECK3-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
+// CHECK3-NEXT: store i32 5, ptr [[TMP135]], align 4
+// CHECK3-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP125]], ptr [[TMP136]], align 4
+// CHECK3-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP126]], ptr [[TMP137]], align 4
+// CHECK3-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.11, ptr [[TMP138]], align 4
+// CHECK3-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP139]], align 4
+// CHECK3-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP140]], align 4
+// CHECK3-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP141]], align 4
+// CHECK3-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP142]], align 8
+// CHECK3-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP143]], align 8
+// CHECK3-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [[TMP132]], ptr [[TMP144]], align 4
+// CHECK3-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [[TMP133]], ptr [[TMP145]], align 4
+// CHECK3-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP146]], align 4
+// CHECK3-NEXT: [[TMP147:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[ADD]], i32 [[TMP131]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l62.region_id, ptr [[KERNEL_ARGS28]])
+// CHECK3-NEXT: [[TMP148:%.*]] = icmp ne i32 [[TMP147]], 0
+// CHECK3-NEXT: br i1 [[TMP148]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
// CHECK3: omp_offload.failed29:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l62(i32 [[TMP85]], i32 [[TMP87]], ptr @Gblb, i32 [[TMP89]], i32 [[TMP91]]) #[[ATTR2]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l62(i32 [[TMP103]], i32 [[TMP105]], ptr @Gblb, i32 [[TMP107]], i32 [[TMP109]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT30]]
// CHECK3: omp_offload.cont30:
-// CHECK3-NEXT: [[TMP125:%.*]] = load ptr, ptr @Gblc, align 4
-// CHECK3-NEXT: store ptr [[TMP125]], ptr [[TMP]], align 4
-// CHECK3-NEXT: [[TMP126:%.*]] = load i32, ptr @Gbla, align 4
-// CHECK3-NEXT: store i32 [[TMP126]], ptr [[GBLC_CASTED]], align 4
-// CHECK3-NEXT: [[TMP127:%.*]] = load i32, ptr [[GBLC_CASTED]], align 4
-// CHECK3-NEXT: [[TMP128:%.*]] = load i32, ptr [[COMP]], align 4
-// CHECK3-NEXT: store i32 [[TMP128]], ptr [[COMP_CASTED31]], align 4
-// CHECK3-NEXT: [[TMP129:%.*]] = load i32, ptr [[COMP_CASTED31]], align 4
-// CHECK3-NEXT: [[TMP130:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP127]], ptr [[TMP130]], align 4
-// CHECK3-NEXT: [[TMP131:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP127]], ptr [[TMP131]], align 4
-// CHECK3-NEXT: [[TMP132:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP132]], align 4
-// CHECK3-NEXT: [[TMP133:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP129]], ptr [[TMP133]], align 4
-// CHECK3-NEXT: [[TMP134:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP129]], ptr [[TMP134]], align 4
-// CHECK3-NEXT: [[TMP135:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 1
-// CHECK3-NEXT: store ptr null, ptr [[TMP135]], align 4
-// CHECK3-NEXT: [[TMP136:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP137:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP138:%.*]] = load ptr, ptr @Gblc, align 4
-// CHECK3-NEXT: store ptr [[TMP138]], ptr [[_TMP35]], align 4
-// CHECK3-NEXT: [[TMP139:%.*]] = load i32, ptr @Gbla, align 4
-// CHECK3-NEXT: [[ADD36:%.*]] = add nsw i32 [[TMP139]], 1
-// CHECK3-NEXT: [[TMP140:%.*]] = load ptr, ptr @Gblc, align 4
-// CHECK3-NEXT: store ptr [[TMP140]], ptr [[_TMP37]], align 4
-// CHECK3-NEXT: [[TMP141:%.*]] = load i32, ptr @Gbla, align 4
-// CHECK3-NEXT: [[ADD38:%.*]] = add nsw i32 [[TMP141]], 2
+// CHECK3-NEXT: [[TMP149:%.*]] = load ptr, ptr @Gblc, align 4
+// CHECK3-NEXT: store ptr [[TMP149]], ptr [[TMP]], align 4
+// CHECK3-NEXT: [[TMP150:%.*]] = load i32, ptr @Gbla, align 4
+// CHECK3-NEXT: store i32 [[TMP150]], ptr [[GBLC_CASTED]], align 4
+// CHECK3-NEXT: [[TMP151:%.*]] = load i32, ptr [[GBLC_CASTED]], align 4
+// CHECK3-NEXT: [[TMP152:%.*]] = load i32, ptr [[COMP]], align 4
+// CHECK3-NEXT: store i32 [[TMP152]], ptr [[COMP_CASTED31]], align 4
+// CHECK3-NEXT: [[TMP153:%.*]] = load i32, ptr [[COMP_CASTED31]], align 4
+// CHECK3-NEXT: [[TMP154:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP151]], ptr [[TMP154]], align 4
+// CHECK3-NEXT: [[TMP155:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP151]], ptr [[TMP155]], align 4
+// CHECK3-NEXT: [[TMP156:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP156]], align 4
+// CHECK3-NEXT: [[TMP157:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP153]], ptr [[TMP157]], align 4
+// CHECK3-NEXT: [[TMP158:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP153]], ptr [[TMP158]], align 4
+// CHECK3-NEXT: [[TMP159:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 1
+// CHECK3-NEXT: store ptr null, ptr [[TMP159]], align 4
+// CHECK3-NEXT: [[TMP160:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP161:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP162:%.*]] = load ptr, ptr @Gblc, align 4
+// CHECK3-NEXT: store ptr [[TMP162]], ptr [[_TMP35]], align 4
+// CHECK3-NEXT: [[TMP163:%.*]] = load i32, ptr @Gbla, align 4
+// CHECK3-NEXT: [[ADD36:%.*]] = add nsw i32 [[TMP163]], 1
+// CHECK3-NEXT: [[TMP164:%.*]] = load ptr, ptr @Gblc, align 4
+// CHECK3-NEXT: store ptr [[TMP164]], ptr [[_TMP37]], align 4
+// CHECK3-NEXT: [[TMP165:%.*]] = load i32, ptr @Gbla, align 4
+// CHECK3-NEXT: [[ADD38:%.*]] = add nsw i32 [[TMP165]], 2
+// CHECK3-NEXT: [[TMP166:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[ADD36]], 0
+// CHECK3-NEXT: [[TMP167:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[ADD38]], 0
// CHECK3-NEXT: [[KERNEL_ARGS39:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP142]], align 4
-// CHECK3-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 1
-// CHECK3-NEXT: store i32 2, ptr [[TMP143]], align 4
-// CHECK3-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP136]], ptr [[TMP144]], align 4
-// CHECK3-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP137]], ptr [[TMP145]], align 4
-// CHECK3-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.14, ptr [[TMP146]], align 4
-// CHECK3-NEXT: [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.15, ptr [[TMP147]], align 4
-// CHECK3-NEXT: [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP148]], align 4
-// CHECK3-NEXT: [[TMP149:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP149]], align 4
-// CHECK3-NEXT: [[TMP150:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 8
-// CHECK3-NEXT: store i64 0, ptr [[TMP150]], align 8
-// CHECK3-NEXT: [[TMP151:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[ADD36]], i32 [[ADD38]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l71.region_id, ptr [[KERNEL_ARGS39]])
-// CHECK3-NEXT: [[TMP152:%.*]] = icmp ne i32 [[TMP151]], 0
-// CHECK3-NEXT: br i1 [[TMP152]], label [[OMP_OFFLOAD_FAILED40:%.*]], label [[OMP_OFFLOAD_CONT41:%.*]]
+// CHECK3-NEXT: [[TMP168:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP168]], align 4
+// CHECK3-NEXT: [[TMP169:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 1
+// CHECK3-NEXT: store i32 2, ptr [[TMP169]], align 4
+// CHECK3-NEXT: [[TMP170:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP160]], ptr [[TMP170]], align 4
+// CHECK3-NEXT: [[TMP171:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP161]], ptr [[TMP171]], align 4
+// CHECK3-NEXT: [[TMP172:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.14, ptr [[TMP172]], align 4
+// CHECK3-NEXT: [[TMP173:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.15, ptr [[TMP173]], align 4
+// CHECK3-NEXT: [[TMP174:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP174]], align 4
+// CHECK3-NEXT: [[TMP175:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP175]], align 4
+// CHECK3-NEXT: [[TMP176:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 8
+// CHECK3-NEXT: store i64 0, ptr [[TMP176]], align 8
+// CHECK3-NEXT: [[TMP177:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP177]], align 8
+// CHECK3-NEXT: [[TMP178:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [[TMP166]], ptr [[TMP178]], align 4
+// CHECK3-NEXT: [[TMP179:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [[TMP167]], ptr [[TMP179]], align 4
+// CHECK3-NEXT: [[TMP180:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS39]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP180]], align 4
+// CHECK3-NEXT: [[TMP181:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[ADD36]], i32 [[ADD38]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l71.region_id, ptr [[KERNEL_ARGS39]])
+// CHECK3-NEXT: [[TMP182:%.*]] = icmp ne i32 [[TMP181]], 0
+// CHECK3-NEXT: br i1 [[TMP182]], label [[OMP_OFFLOAD_FAILED40:%.*]], label [[OMP_OFFLOAD_CONT41:%.*]]
// CHECK3: omp_offload.failed40:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l71(i32 [[TMP127]], i32 [[TMP129]]) #[[ATTR2]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l71(i32 [[TMP151]], i32 [[TMP153]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT41]]
// CHECK3: omp_offload.cont41:
-// CHECK3-NEXT: [[TMP153:%.*]] = load i32, ptr [[COMP]], align 4
-// CHECK3-NEXT: ret i32 [[TMP153]]
+// CHECK3-NEXT: [[TMP183:%.*]] = load i32, ptr [[COMP]], align 4
+// CHECK3-NEXT: ret i32 [[TMP183]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z27teams_argument_global_locali_l31
@@ -1453,89 +1561,109 @@ void foo() {
// CHECK9-NEXT: [[TMP14:%.*]] = load float, ptr [[B]], align 4
// CHECK9-NEXT: [[CONV:%.*]] = fptosi float [[TMP14]] to i64
// CHECK9-NEXT: [[TMP15:%.*]] = trunc i64 [[CONV]] to i32
+// CHECK9-NEXT: [[TMP16:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP13]], 0
+// CHECK9-NEXT: [[TMP17:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP15]], 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP16]], align 4
-// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK9-NEXT: store i32 3, ptr [[TMP17]], align 4
-// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP11]], ptr [[TMP18]], align 8
-// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP12]], ptr [[TMP19]], align 8
-// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes, ptr [[TMP20]], align 8
-// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes, ptr [[TMP21]], align 8
-// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP22]], align 8
-// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP23]], align 8
-// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK9-NEXT: store i64 0, ptr [[TMP24]], align 8
-// CHECK9-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 [[TMP13]], i32 [[TMP15]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18teams_template_argv_l116.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK9-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP18]], align 4
+// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK9-NEXT: store i32 3, ptr [[TMP19]], align 4
+// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP11]], ptr [[TMP20]], align 8
+// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP12]], ptr [[TMP21]], align 8
+// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes, ptr [[TMP22]], align 8
+// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes, ptr [[TMP23]], align 8
+// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP24]], align 8
+// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP25]], align 8
+// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK9-NEXT: store i64 0, ptr [[TMP26]], align 8
+// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] [[TMP16]], ptr [[TMP28]], align 4
+// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [[TMP17]], ptr [[TMP29]], align 4
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK9-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 [[TMP13]], i32 [[TMP15]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18teams_template_argv_l116.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK9-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18teams_template_argv_l116(ptr @Gbla, ptr [[LA]], i64 [[TMP1]]) #[[ATTR2:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP27:%.*]] = load i32, ptr [[COMP]], align 4
-// CHECK9-NEXT: store i32 [[TMP27]], ptr [[COMP_CASTED1]], align 4
-// CHECK9-NEXT: [[TMP28:%.*]] = load i64, ptr [[COMP_CASTED1]], align 8
-// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[LB]], ptr [[TMP29]], align 8
-// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[LB]], ptr [[TMP30]], align 8
-// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS4]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP31]], align 8
-// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 1
-// CHECK9-NEXT: store ptr @Gblb, ptr [[TMP32]], align 8
-// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 1
-// CHECK9-NEXT: store ptr @Gblb, ptr [[TMP33]], align 8
-// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS4]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP34]], align 8
-// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 2
-// CHECK9-NEXT: store i64 [[TMP28]], ptr [[TMP35]], align 8
-// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 2
-// CHECK9-NEXT: store i64 [[TMP28]], ptr [[TMP36]], align 8
-// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS4]], i64 0, i64 2
+// CHECK9-NEXT: [[TMP33:%.*]] = load i32, ptr [[COMP]], align 4
+// CHECK9-NEXT: store i32 [[TMP33]], ptr [[COMP_CASTED1]], align 4
+// CHECK9-NEXT: [[TMP34:%.*]] = load i64, ptr [[COMP_CASTED1]], align 8
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[LB]], ptr [[TMP35]], align 8
+// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[LB]], ptr [[TMP36]], align 8
+// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS4]], i64 0, i64 0
// CHECK9-NEXT: store ptr null, ptr [[TMP37]], align 8
-// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 1
+// CHECK9-NEXT: store ptr @Gblb, ptr [[TMP38]], align 8
+// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 1
+// CHECK9-NEXT: store ptr @Gblb, ptr [[TMP39]], align 8
+// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS4]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP40]], align 8
+// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 2
+// CHECK9-NEXT: store i64 [[TMP34]], ptr [[TMP41]], align 8
+// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 2
+// CHECK9-NEXT: store i64 [[TMP34]], ptr [[TMP42]], align 8
+// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS4]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
// CHECK9-NEXT: [[B5:%.*]] = getelementptr inbounds [[STRUCT_SS_0]], ptr [[LB]], i32 0, i32 1
-// CHECK9-NEXT: [[TMP40:%.*]] = load float, ptr [[B5]], align 8
-// CHECK9-NEXT: [[CONV6:%.*]] = fptosi float [[TMP40]] to i64
-// CHECK9-NEXT: [[TMP41:%.*]] = trunc i64 [[CONV6]] to i32
-// CHECK9-NEXT: [[TMP42:%.*]] = load i64, ptr @Gblb, align 8
-// CHECK9-NEXT: [[TMP43:%.*]] = trunc i64 [[TMP42]] to i32
+// CHECK9-NEXT: [[TMP46:%.*]] = load float, ptr [[B5]], align 8
+// CHECK9-NEXT: [[CONV6:%.*]] = fptosi float [[TMP46]] to i64
+// CHECK9-NEXT: [[TMP47:%.*]] = trunc i64 [[CONV6]] to i32
+// CHECK9-NEXT: [[TMP48:%.*]] = load i64, ptr @Gblb, align 8
+// CHECK9-NEXT: [[TMP49:%.*]] = trunc i64 [[TMP48]] to i32
+// CHECK9-NEXT: [[TMP50:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP47]], 0
+// CHECK9-NEXT: [[TMP51:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP49]], 0
// CHECK9-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP44]], align 4
-// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK9-NEXT: store i32 3, ptr [[TMP45]], align 4
-// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP38]], ptr [[TMP46]], align 8
-// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP39]], ptr [[TMP47]], align 8
-// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.2, ptr [[TMP48]], align 8
-// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP49]], align 8
-// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP50]], align 8
-// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP51]], align 8
-// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK9-NEXT: store i64 0, ptr [[TMP52]], align 8
-// CHECK9-NEXT: [[TMP53:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP41]], i32 [[TMP43]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18teams_template_argv_l125.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK9-NEXT: [[TMP54:%.*]] = icmp ne i32 [[TMP53]], 0
-// CHECK9-NEXT: br i1 [[TMP54]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP52]], align 4
+// CHECK9-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK9-NEXT: store i32 3, ptr [[TMP53]], align 4
+// CHECK9-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP44]], ptr [[TMP54]], align 8
+// CHECK9-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP45]], ptr [[TMP55]], align 8
+// CHECK9-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.2, ptr [[TMP56]], align 8
+// CHECK9-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP57]], align 8
+// CHECK9-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP58]], align 8
+// CHECK9-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP59]], align 8
+// CHECK9-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK9-NEXT: store i64 0, ptr [[TMP60]], align 8
+// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP61]], align 8
+// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] [[TMP50]], ptr [[TMP62]], align 4
+// CHECK9-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [[TMP51]], ptr [[TMP63]], align 4
+// CHECK9-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP64]], align 4
+// CHECK9-NEXT: [[TMP65:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP47]], i32 [[TMP49]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18teams_template_argv_l125.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK9-NEXT: [[TMP66:%.*]] = icmp ne i32 [[TMP65]], 0
+// CHECK9-NEXT: br i1 [[TMP66]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK9: omp_offload.failed8:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18teams_template_argv_l125(ptr [[LB]], ptr @Gblb, i64 [[TMP28]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18teams_template_argv_l125(ptr [[LB]], ptr @Gblb, i64 [[TMP34]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK9: omp_offload.cont9:
-// CHECK9-NEXT: [[TMP55:%.*]] = load i32, ptr [[COMP]], align 4
-// CHECK9-NEXT: ret i32 [[TMP55]]
+// CHECK9-NEXT: [[TMP67:%.*]] = load i32, ptr [[COMP]], align 4
+// CHECK9-NEXT: ret i32 [[TMP67]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18teams_template_argv_l116
@@ -1667,89 +1795,109 @@ void foo() {
// CHECK11-NEXT: [[TMP14:%.*]] = load float, ptr [[B]], align 4
// CHECK11-NEXT: [[CONV:%.*]] = fptosi float [[TMP14]] to i64
// CHECK11-NEXT: [[TMP15:%.*]] = trunc i64 [[CONV]] to i32
+// CHECK11-NEXT: [[TMP16:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP13]], 0
+// CHECK11-NEXT: [[TMP17:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP15]], 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP16]], align 4
-// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK11-NEXT: store i32 3, ptr [[TMP17]], align 4
-// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP11]], ptr [[TMP18]], align 4
-// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP12]], ptr [[TMP19]], align 4
-// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes, ptr [[TMP20]], align 4
-// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes, ptr [[TMP21]], align 4
-// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP22]], align 4
-// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP23]], align 4
-// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK11-NEXT: store i64 0, ptr [[TMP24]], align 8
-// CHECK11-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 [[TMP13]], i32 [[TMP15]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18teams_template_argv_l116.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK11-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP18]], align 4
+// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK11-NEXT: store i32 3, ptr [[TMP19]], align 4
+// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP11]], ptr [[TMP20]], align 4
+// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP12]], ptr [[TMP21]], align 4
+// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes, ptr [[TMP22]], align 4
+// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes, ptr [[TMP23]], align 4
+// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP24]], align 4
+// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP25]], align 4
+// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK11-NEXT: store i64 0, ptr [[TMP26]], align 8
+// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] [[TMP16]], ptr [[TMP28]], align 4
+// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [[TMP17]], ptr [[TMP29]], align 4
+// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK11-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 [[TMP13]], i32 [[TMP15]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18teams_template_argv_l116.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK11-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18teams_template_argv_l116(ptr @Gbla, ptr [[LA]], i32 [[TMP1]]) #[[ATTR2:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP27:%.*]] = load i32, ptr [[COMP]], align 4
-// CHECK11-NEXT: store i32 [[TMP27]], ptr [[COMP_CASTED1]], align 4
-// CHECK11-NEXT: [[TMP28:%.*]] = load i32, ptr [[COMP_CASTED1]], align 4
-// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[LB]], ptr [[TMP29]], align 4
-// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[LB]], ptr [[TMP30]], align 4
-// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS4]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP31]], align 4
-// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 1
-// CHECK11-NEXT: store ptr @Gblb, ptr [[TMP32]], align 4
-// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 1
-// CHECK11-NEXT: store ptr @Gblb, ptr [[TMP33]], align 4
-// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS4]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP34]], align 4
-// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 2
-// CHECK11-NEXT: store i32 [[TMP28]], ptr [[TMP35]], align 4
-// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 2
-// CHECK11-NEXT: store i32 [[TMP28]], ptr [[TMP36]], align 4
-// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS4]], i32 0, i32 2
+// CHECK11-NEXT: [[TMP33:%.*]] = load i32, ptr [[COMP]], align 4
+// CHECK11-NEXT: store i32 [[TMP33]], ptr [[COMP_CASTED1]], align 4
+// CHECK11-NEXT: [[TMP34:%.*]] = load i32, ptr [[COMP_CASTED1]], align 4
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[LB]], ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[LB]], ptr [[TMP36]], align 4
+// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS4]], i32 0, i32 0
// CHECK11-NEXT: store ptr null, ptr [[TMP37]], align 4
-// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 1
+// CHECK11-NEXT: store ptr @Gblb, ptr [[TMP38]], align 4
+// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 1
+// CHECK11-NEXT: store ptr @Gblb, ptr [[TMP39]], align 4
+// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS4]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP40]], align 4
+// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 2
+// CHECK11-NEXT: store i32 [[TMP34]], ptr [[TMP41]], align 4
+// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 2
+// CHECK11-NEXT: store i32 [[TMP34]], ptr [[TMP42]], align 4
+// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS4]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP43]], align 4
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
// CHECK11-NEXT: [[B5:%.*]] = getelementptr inbounds [[STRUCT_SS_0]], ptr [[LB]], i32 0, i32 1
-// CHECK11-NEXT: [[TMP40:%.*]] = load float, ptr [[B5]], align 4
-// CHECK11-NEXT: [[CONV6:%.*]] = fptosi float [[TMP40]] to i64
-// CHECK11-NEXT: [[TMP41:%.*]] = trunc i64 [[CONV6]] to i32
-// CHECK11-NEXT: [[TMP42:%.*]] = load i64, ptr @Gblb, align 4
-// CHECK11-NEXT: [[TMP43:%.*]] = trunc i64 [[TMP42]] to i32
+// CHECK11-NEXT: [[TMP46:%.*]] = load float, ptr [[B5]], align 4
+// CHECK11-NEXT: [[CONV6:%.*]] = fptosi float [[TMP46]] to i64
+// CHECK11-NEXT: [[TMP47:%.*]] = trunc i64 [[CONV6]] to i32
+// CHECK11-NEXT: [[TMP48:%.*]] = load i64, ptr @Gblb, align 4
+// CHECK11-NEXT: [[TMP49:%.*]] = trunc i64 [[TMP48]] to i32
+// CHECK11-NEXT: [[TMP50:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP47]], 0
+// CHECK11-NEXT: [[TMP51:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP49]], 0
// CHECK11-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP44]], align 4
-// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK11-NEXT: store i32 3, ptr [[TMP45]], align 4
-// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP38]], ptr [[TMP46]], align 4
-// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP39]], ptr [[TMP47]], align 4
-// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.2, ptr [[TMP48]], align 4
-// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP49]], align 4
-// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP50]], align 4
-// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP51]], align 4
-// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK11-NEXT: store i64 0, ptr [[TMP52]], align 8
-// CHECK11-NEXT: [[TMP53:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP41]], i32 [[TMP43]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18teams_template_argv_l125.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK11-NEXT: [[TMP54:%.*]] = icmp ne i32 [[TMP53]], 0
-// CHECK11-NEXT: br i1 [[TMP54]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP52]], align 4
+// CHECK11-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK11-NEXT: store i32 3, ptr [[TMP53]], align 4
+// CHECK11-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP44]], ptr [[TMP54]], align 4
+// CHECK11-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP45]], ptr [[TMP55]], align 4
+// CHECK11-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.2, ptr [[TMP56]], align 4
+// CHECK11-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP57]], align 4
+// CHECK11-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP58]], align 4
+// CHECK11-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP59]], align 4
+// CHECK11-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK11-NEXT: store i64 0, ptr [[TMP60]], align 8
+// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP61]], align 8
+// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] [[TMP50]], ptr [[TMP62]], align 4
+// CHECK11-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [[TMP51]], ptr [[TMP63]], align 4
+// CHECK11-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP64]], align 4
+// CHECK11-NEXT: [[TMP65:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 [[TMP47]], i32 [[TMP49]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18teams_template_argv_l125.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK11-NEXT: [[TMP66:%.*]] = icmp ne i32 [[TMP65]], 0
+// CHECK11-NEXT: br i1 [[TMP66]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK11: omp_offload.failed8:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18teams_template_argv_l125(ptr [[LB]], ptr @Gblb, i32 [[TMP28]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18teams_template_argv_l125(ptr [[LB]], ptr @Gblb, i32 [[TMP34]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK11: omp_offload.cont9:
-// CHECK11-NEXT: [[TMP55:%.*]] = load i32, ptr [[COMP]], align 4
-// CHECK11-NEXT: ret i32 [[TMP55]]
+// CHECK11-NEXT: [[TMP67:%.*]] = load i32, ptr [[COMP]], align 4
+// CHECK11-NEXT: ret i32 [[TMP67]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18teams_template_argv_l116
@@ -1882,82 +2030,100 @@ void foo() {
// CHECK17-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK17-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK17-NEXT: [[TMP10:%.*]] = load i32, ptr [[A2]], align 4
+// CHECK17-NEXT: [[TMP11:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP10]], 0
// CHECK17-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK17-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK17-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK17-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
// CHECK17-NEXT: store i32 2, ptr [[TMP12]], align 4
-// CHECK17-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP8]], ptr [[TMP13]], align 8
-// CHECK17-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP9]], ptr [[TMP14]], align 8
-// CHECK17-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK17-NEXT: store ptr @.offload_sizes, ptr [[TMP15]], align 8
-// CHECK17-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes, ptr [[TMP16]], align 8
-// CHECK17-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK17-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK17-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK17-NEXT: store i32 2, ptr [[TMP13]], align 4
+// CHECK17-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP8]], ptr [[TMP14]], align 8
+// CHECK17-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP9]], ptr [[TMP15]], align 8
+// CHECK17-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK17-NEXT: store ptr @.offload_sizes, ptr [[TMP16]], align 8
+// CHECK17-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes, ptr [[TMP17]], align 8
+// CHECK17-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK17-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK17-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK17-NEXT: store i64 0, ptr [[TMP19]], align 8
-// CHECK17-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 [[TMP10]], i32 123, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l161.region_id, ptr [[KERNEL_ARGS]])
-// CHECK17-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK17-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK17-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK17-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK17-NEXT: store i64 0, ptr [[TMP20]], align 8
+// CHECK17-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP21]], align 8
+// CHECK17-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] [[TMP11]], ptr [[TMP22]], align 4
+// CHECK17-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] [i32 123, i32 0, i32 0], ptr [[TMP23]], align 4
+// CHECK17-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP24]], align 4
+// CHECK17-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 [[TMP10]], i32 123, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l161.region_id, ptr [[KERNEL_ARGS]])
+// CHECK17-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
+// CHECK17-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK17: omp_offload.failed:
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l161(ptr [[THIS1]], i64 [[TMP1]]) #[[ATTR2:[0-9]+]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK17: omp_offload.cont:
-// CHECK17-NEXT: [[TMP22:%.*]] = load i32, ptr [[COMP]], align 4
-// CHECK17-NEXT: store i32 [[TMP22]], ptr [[COMP_CASTED3]], align 4
-// CHECK17-NEXT: [[TMP23:%.*]] = load i64, ptr [[COMP_CASTED3]], align 8
+// CHECK17-NEXT: [[TMP27:%.*]] = load i32, ptr [[COMP]], align 4
+// CHECK17-NEXT: store i32 [[TMP27]], ptr [[COMP_CASTED3]], align 4
+// CHECK17-NEXT: [[TMP28:%.*]] = load i64, ptr [[COMP_CASTED3]], align 8
// CHECK17-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 1
-// CHECK17-NEXT: [[TMP24:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK17-NEXT: store ptr [[THIS1]], ptr [[TMP24]], align 8
-// CHECK17-NEXT: [[TMP25:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK17-NEXT: store ptr [[B]], ptr [[TMP25]], align 8
-// CHECK17-NEXT: [[TMP26:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP26]], align 8
-// CHECK17-NEXT: [[TMP27:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP23]], ptr [[TMP27]], align 8
-// CHECK17-NEXT: [[TMP28:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP23]], ptr [[TMP28]], align 8
-// CHECK17-NEXT: [[TMP29:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
-// CHECK17-NEXT: store ptr null, ptr [[TMP29]], align 8
-// CHECK17-NEXT: [[TMP30:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP31:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP29:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK17-NEXT: store ptr [[THIS1]], ptr [[TMP29]], align 8
+// CHECK17-NEXT: [[TMP30:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK17-NEXT: store ptr [[B]], ptr [[TMP30]], align 8
+// CHECK17-NEXT: [[TMP31:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK17-NEXT: [[TMP32:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP28]], ptr [[TMP32]], align 8
+// CHECK17-NEXT: [[TMP33:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP28]], ptr [[TMP33]], align 8
+// CHECK17-NEXT: [[TMP34:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
+// CHECK17-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK17-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
// CHECK17-NEXT: [[B7:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 1
-// CHECK17-NEXT: [[TMP32:%.*]] = load float, ptr [[B7]], align 4
-// CHECK17-NEXT: [[CONV:%.*]] = fptosi float [[TMP32]] to i32
+// CHECK17-NEXT: [[TMP37:%.*]] = load float, ptr [[B7]], align 4
+// CHECK17-NEXT: [[CONV:%.*]] = fptosi float [[TMP37]] to i32
// CHECK17-NEXT: [[ADD:%.*]] = add nsw i32 [[CONV]], 123
+// CHECK17-NEXT: [[TMP38:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[ADD]], 0
// CHECK17-NEXT: [[KERNEL_ARGS8:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP33]], align 4
-// CHECK17-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 1
-// CHECK17-NEXT: store i32 2, ptr [[TMP34]], align 4
-// CHECK17-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP30]], ptr [[TMP35]], align 8
-// CHECK17-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP31]], ptr [[TMP36]], align 8
-// CHECK17-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 4
-// CHECK17-NEXT: store ptr @.offload_sizes.2, ptr [[TMP37]], align 8
-// CHECK17-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP38]], align 8
-// CHECK17-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK17-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP40]], align 8
-// CHECK17-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 8
-// CHECK17-NEXT: store i64 0, ptr [[TMP41]], align 8
-// CHECK17-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 456, i32 [[ADD]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l169.region_id, ptr [[KERNEL_ARGS8]])
-// CHECK17-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
-// CHECK17-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED9:%.*]], label [[OMP_OFFLOAD_CONT10:%.*]]
+// CHECK17-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP39]], align 4
+// CHECK17-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 1
+// CHECK17-NEXT: store i32 2, ptr [[TMP40]], align 4
+// CHECK17-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP35]], ptr [[TMP41]], align 8
+// CHECK17-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP36]], ptr [[TMP42]], align 8
+// CHECK17-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 4
+// CHECK17-NEXT: store ptr @.offload_sizes.2, ptr [[TMP43]], align 8
+// CHECK17-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP44]], align 8
+// CHECK17-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP45]], align 8
+// CHECK17-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP46]], align 8
+// CHECK17-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 8
+// CHECK17-NEXT: store i64 0, ptr [[TMP47]], align 8
+// CHECK17-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP48]], align 8
+// CHECK17-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] [i32 456, i32 0, i32 0], ptr [[TMP49]], align 4
+// CHECK17-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] [[TMP38]], ptr [[TMP50]], align 4
+// CHECK17-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP51]], align 4
+// CHECK17-NEXT: [[TMP52:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 456, i32 [[ADD]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l169.region_id, ptr [[KERNEL_ARGS8]])
+// CHECK17-NEXT: [[TMP53:%.*]] = icmp ne i32 [[TMP52]], 0
+// CHECK17-NEXT: br i1 [[TMP53]], label [[OMP_OFFLOAD_FAILED9:%.*]], label [[OMP_OFFLOAD_CONT10:%.*]]
// CHECK17: omp_offload.failed9:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l169(ptr [[THIS1]], i64 [[TMP23]]) #[[ATTR2]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l169(ptr [[THIS1]], i64 [[TMP28]]) #[[ATTR2]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT10]]
// CHECK17: omp_offload.cont10:
-// CHECK17-NEXT: [[TMP44:%.*]] = load i32, ptr [[COMP]], align 4
-// CHECK17-NEXT: ret i32 [[TMP44]]
+// CHECK17-NEXT: [[TMP54:%.*]] = load i32, ptr [[COMP]], align 4
+// CHECK17-NEXT: ret i32 [[TMP54]]
//
//
// CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l161
@@ -2077,82 +2243,100 @@ void foo() {
// CHECK19-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK19-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK19-NEXT: [[TMP10:%.*]] = load i32, ptr [[A2]], align 4
+// CHECK19-NEXT: [[TMP11:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP10]], 0
// CHECK19-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK19-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK19-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK19-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
// CHECK19-NEXT: store i32 2, ptr [[TMP12]], align 4
-// CHECK19-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP8]], ptr [[TMP13]], align 4
-// CHECK19-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP9]], ptr [[TMP14]], align 4
-// CHECK19-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK19-NEXT: store ptr @.offload_sizes, ptr [[TMP15]], align 4
-// CHECK19-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes, ptr [[TMP16]], align 4
-// CHECK19-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP17]], align 4
-// CHECK19-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK19-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK19-NEXT: store i32 2, ptr [[TMP13]], align 4
+// CHECK19-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP8]], ptr [[TMP14]], align 4
+// CHECK19-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP9]], ptr [[TMP15]], align 4
+// CHECK19-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK19-NEXT: store ptr @.offload_sizes, ptr [[TMP16]], align 4
+// CHECK19-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes, ptr [[TMP17]], align 4
+// CHECK19-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK19-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK19-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK19-NEXT: store i64 0, ptr [[TMP19]], align 8
-// CHECK19-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 [[TMP10]], i32 123, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l161.region_id, ptr [[KERNEL_ARGS]])
-// CHECK19-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK19-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK19-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP19]], align 4
+// CHECK19-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK19-NEXT: store i64 0, ptr [[TMP20]], align 8
+// CHECK19-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP21]], align 8
+// CHECK19-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] [[TMP11]], ptr [[TMP22]], align 4
+// CHECK19-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] [i32 123, i32 0, i32 0], ptr [[TMP23]], align 4
+// CHECK19-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP24]], align 4
+// CHECK19-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 [[TMP10]], i32 123, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l161.region_id, ptr [[KERNEL_ARGS]])
+// CHECK19-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
+// CHECK19-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK19: omp_offload.failed:
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l161(ptr [[THIS1]], i32 [[TMP1]]) #[[ATTR2:[0-9]+]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK19: omp_offload.cont:
-// CHECK19-NEXT: [[TMP22:%.*]] = load i32, ptr [[COMP]], align 4
-// CHECK19-NEXT: store i32 [[TMP22]], ptr [[COMP_CASTED3]], align 4
-// CHECK19-NEXT: [[TMP23:%.*]] = load i32, ptr [[COMP_CASTED3]], align 4
+// CHECK19-NEXT: [[TMP27:%.*]] = load i32, ptr [[COMP]], align 4
+// CHECK19-NEXT: store i32 [[TMP27]], ptr [[COMP_CASTED3]], align 4
+// CHECK19-NEXT: [[TMP28:%.*]] = load i32, ptr [[COMP_CASTED3]], align 4
// CHECK19-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 1
-// CHECK19-NEXT: [[TMP24:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK19-NEXT: store ptr [[THIS1]], ptr [[TMP24]], align 4
-// CHECK19-NEXT: [[TMP25:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK19-NEXT: store ptr [[B]], ptr [[TMP25]], align 4
-// CHECK19-NEXT: [[TMP26:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP26]], align 4
-// CHECK19-NEXT: [[TMP27:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP23]], ptr [[TMP27]], align 4
-// CHECK19-NEXT: [[TMP28:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP23]], ptr [[TMP28]], align 4
-// CHECK19-NEXT: [[TMP29:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
-// CHECK19-NEXT: store ptr null, ptr [[TMP29]], align 4
-// CHECK19-NEXT: [[TMP30:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP31:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP29:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK19-NEXT: store ptr [[THIS1]], ptr [[TMP29]], align 4
+// CHECK19-NEXT: [[TMP30:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK19-NEXT: store ptr [[B]], ptr [[TMP30]], align 4
+// CHECK19-NEXT: [[TMP31:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK19-NEXT: [[TMP32:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP28]], ptr [[TMP32]], align 4
+// CHECK19-NEXT: [[TMP33:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP28]], ptr [[TMP33]], align 4
+// CHECK19-NEXT: [[TMP34:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
+// CHECK19-NEXT: store ptr null, ptr [[TMP34]], align 4
+// CHECK19-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
// CHECK19-NEXT: [[B7:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 1
-// CHECK19-NEXT: [[TMP32:%.*]] = load float, ptr [[B7]], align 4
-// CHECK19-NEXT: [[CONV:%.*]] = fptosi float [[TMP32]] to i32
+// CHECK19-NEXT: [[TMP37:%.*]] = load float, ptr [[B7]], align 4
+// CHECK19-NEXT: [[CONV:%.*]] = fptosi float [[TMP37]] to i32
// CHECK19-NEXT: [[ADD:%.*]] = add nsw i32 [[CONV]], 123
+// CHECK19-NEXT: [[TMP38:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[ADD]], 0
// CHECK19-NEXT: [[KERNEL_ARGS8:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP33]], align 4
-// CHECK19-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 1
-// CHECK19-NEXT: store i32 2, ptr [[TMP34]], align 4
-// CHECK19-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP30]], ptr [[TMP35]], align 4
-// CHECK19-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP31]], ptr [[TMP36]], align 4
-// CHECK19-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 4
-// CHECK19-NEXT: store ptr @.offload_sizes.2, ptr [[TMP37]], align 4
-// CHECK19-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP38]], align 4
-// CHECK19-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP39]], align 4
-// CHECK19-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP40]], align 4
-// CHECK19-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 8
-// CHECK19-NEXT: store i64 0, ptr [[TMP41]], align 8
-// CHECK19-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 456, i32 [[ADD]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l169.region_id, ptr [[KERNEL_ARGS8]])
-// CHECK19-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
-// CHECK19-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED9:%.*]], label [[OMP_OFFLOAD_CONT10:%.*]]
+// CHECK19-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP39]], align 4
+// CHECK19-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 1
+// CHECK19-NEXT: store i32 2, ptr [[TMP40]], align 4
+// CHECK19-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP35]], ptr [[TMP41]], align 4
+// CHECK19-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP36]], ptr [[TMP42]], align 4
+// CHECK19-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 4
+// CHECK19-NEXT: store ptr @.offload_sizes.2, ptr [[TMP43]], align 4
+// CHECK19-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP44]], align 4
+// CHECK19-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP45]], align 4
+// CHECK19-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP46]], align 4
+// CHECK19-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 8
+// CHECK19-NEXT: store i64 0, ptr [[TMP47]], align 8
+// CHECK19-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP48]], align 8
+// CHECK19-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] [i32 456, i32 0, i32 0], ptr [[TMP49]], align 4
+// CHECK19-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] [[TMP38]], ptr [[TMP50]], align 4
+// CHECK19-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS8]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP51]], align 4
+// CHECK19-NEXT: [[TMP52:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 456, i32 [[ADD]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l169.region_id, ptr [[KERNEL_ARGS8]])
+// CHECK19-NEXT: [[TMP53:%.*]] = icmp ne i32 [[TMP52]], 0
+// CHECK19-NEXT: br i1 [[TMP53]], label [[OMP_OFFLOAD_FAILED9:%.*]], label [[OMP_OFFLOAD_CONT10:%.*]]
// CHECK19: omp_offload.failed9:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l169(ptr [[THIS1]], i32 [[TMP23]]) #[[ATTR2]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l169(ptr [[THIS1]], i32 [[TMP28]]) #[[ATTR2]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT10]]
// CHECK19: omp_offload.cont10:
-// CHECK19-NEXT: [[TMP44:%.*]] = load i32, ptr [[COMP]], align 4
-// CHECK19-NEXT: ret i32 [[TMP44]]
+// CHECK19-NEXT: [[TMP54:%.*]] = load i32, ptr [[COMP]], align 4
+// CHECK19-NEXT: ret i32 [[TMP54]]
//
//
// CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l161
diff --git a/clang/test/OpenMP/teams_distribute_codegen.cpp b/clang/test/OpenMP/teams_distribute_codegen.cpp
index 56c6b15f5d73d..a11bae68a7e5a 100644
--- a/clang/test/OpenMP/teams_distribute_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_codegen.cpp
@@ -249,87 +249,105 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[TMP25:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP25]], 1
// CHECK1-NEXT: [[TMP26:%.*]] = zext i32 [[ADD]] to i64
+// CHECK1-NEXT: [[TMP27:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP21]], 0
+// CHECK1-NEXT: [[TMP28:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP22]], 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP27]], align 4
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK1-NEXT: store i32 4, ptr [[TMP28]], align 4
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP19]], ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP20]], ptr [[TMP30]], align 8
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP31]], align 8
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP32]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK1-NEXT: store i64 [[TMP26]], ptr [[TMP35]], align 8
-// CHECK1-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 [[TMP21]], i32 [[TMP22]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l29.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
-// CHECK1-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP29]], align 4
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK1-NEXT: store i32 4, ptr [[TMP30]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP19]], ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP20]], ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP35]], align 8
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP36]], align 8
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK1-NEXT: store i64 [[TMP26]], ptr [[TMP37]], align 8
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP38]], align 8
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [[TMP27]], ptr [[TMP39]], align 4
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP28]], ptr [[TMP40]], align 4
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP41]], align 4
+// CHECK1-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 [[TMP21]], i32 [[TMP22]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l29.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
+// CHECK1-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l29(i64 [[TMP2]], i64 [[TMP4]], i64 [[TMP6]], ptr @a) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
-// CHECK1-NEXT: [[TMP38:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: store i32 [[TMP38]], ptr [[N_CASTED4]], align 4
-// CHECK1-NEXT: [[TMP39:%.*]] = load i64, ptr [[N_CASTED4]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP39]], ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP39]], ptr [[TMP41]], align 8
-// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP42]], align 8
-// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
-// CHECK1-NEXT: store ptr @a, ptr [[TMP43]], align 8
-// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
-// CHECK1-NEXT: store ptr @a, ptr [[TMP44]], align 8
-// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP45]], align 8
+// CHECK1-NEXT: [[TMP44:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: store i32 [[TMP44]], ptr [[N_CASTED4]], align 4
+// CHECK1-NEXT: [[TMP45:%.*]] = load i64, ptr [[N_CASTED4]], align 8
// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP45]], ptr [[TMP46]], align 8
// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP48:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: store i32 [[TMP48]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK1-NEXT: [[TMP49:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK1-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP49]], 0
+// CHECK1-NEXT: store i64 [[TMP45]], ptr [[TMP47]], align 8
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP48]], align 8
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
+// CHECK1-NEXT: store ptr @a, ptr [[TMP49]], align 8
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
+// CHECK1-NEXT: store ptr @a, ptr [[TMP50]], align 8
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP51]], align 8
+// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP54:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: store i32 [[TMP54]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK1-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK1-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP55]], 0
// CHECK1-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK1-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK1-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK1-NEXT: [[TMP50:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK1-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP50]], 1
-// CHECK1-NEXT: [[TMP51:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK1-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK1-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP56]], 1
+// CHECK1-NEXT: [[TMP57:%.*]] = zext i32 [[ADD14]] to i64
// CHECK1-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP52]], align 4
-// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK1-NEXT: store i32 2, ptr [[TMP53]], align 4
-// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP46]], ptr [[TMP54]], align 8
-// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP47]], ptr [[TMP55]], align 8
-// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.2, ptr [[TMP56]], align 8
-// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP57]], align 8
-// CHECK1-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP58]], align 8
-// CHECK1-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP59]], align 8
-// CHECK1-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK1-NEXT: store i64 [[TMP51]], ptr [[TMP60]], align 8
-// CHECK1-NEXT: [[TMP61:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l35.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK1-NEXT: [[TMP62:%.*]] = icmp ne i32 [[TMP61]], 0
-// CHECK1-NEXT: br i1 [[TMP62]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK1-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP58]], align 4
+// CHECK1-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK1-NEXT: store i32 2, ptr [[TMP59]], align 4
+// CHECK1-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP52]], ptr [[TMP60]], align 8
+// CHECK1-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP53]], ptr [[TMP61]], align 8
+// CHECK1-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.2, ptr [[TMP62]], align 8
+// CHECK1-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP63]], align 8
+// CHECK1-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP64]], align 8
+// CHECK1-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP65]], align 8
+// CHECK1-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK1-NEXT: store i64 [[TMP57]], ptr [[TMP66]], align 8
+// CHECK1-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP67]], align 8
+// CHECK1-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP68]], align 4
+// CHECK1-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK1-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP70]], align 4
+// CHECK1-NEXT: [[TMP71:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l35.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK1-NEXT: [[TMP72:%.*]] = icmp ne i32 [[TMP71]], 0
+// CHECK1-NEXT: br i1 [[TMP72]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK1: omp_offload.failed16:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l35(i64 [[TMP39]], ptr @a) #[[ATTR2]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l35(i64 [[TMP45]], ptr @a) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK1: omp_offload.cont17:
-// CHECK1-NEXT: [[TMP63:%.*]] = load i32, ptr @a, align 4
-// CHECK1-NEXT: ret i32 [[TMP63]]
+// CHECK1-NEXT: [[TMP73:%.*]] = load i32, ptr @a, align 4
+// CHECK1-NEXT: ret i32 [[TMP73]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l29
@@ -629,87 +647,105 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[TMP25:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP25]], 1
// CHECK3-NEXT: [[TMP26:%.*]] = zext i32 [[ADD]] to i64
+// CHECK3-NEXT: [[TMP27:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP21]], 0
+// CHECK3-NEXT: [[TMP28:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP22]], 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP27]], align 4
-// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK3-NEXT: store i32 4, ptr [[TMP28]], align 4
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP19]], ptr [[TMP29]], align 4
-// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP20]], ptr [[TMP30]], align 4
-// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes, ptr [[TMP31]], align 4
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP32]], align 4
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP33]], align 4
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP34]], align 4
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK3-NEXT: store i64 [[TMP26]], ptr [[TMP35]], align 8
-// CHECK3-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 [[TMP21]], i32 [[TMP22]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l29.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
-// CHECK3-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK3-NEXT: store i32 4, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP19]], ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP20]], ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes, ptr [[TMP33]], align 4
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP34]], align 4
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP35]], align 4
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP36]], align 4
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK3-NEXT: store i64 [[TMP26]], ptr [[TMP37]], align 8
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP38]], align 8
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [[TMP27]], ptr [[TMP39]], align 4
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [[TMP28]], ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 [[TMP21]], i32 [[TMP22]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l29.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
+// CHECK3-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l29(i32 [[TMP2]], i32 [[TMP4]], i32 [[TMP6]], ptr @a) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
-// CHECK3-NEXT: [[TMP38:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: store i32 [[TMP38]], ptr [[N_CASTED4]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = load i32, ptr [[N_CASTED4]], align 4
-// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP39]], ptr [[TMP40]], align 4
-// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP39]], ptr [[TMP41]], align 4
-// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP42]], align 4
-// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
-// CHECK3-NEXT: store ptr @a, ptr [[TMP43]], align 4
-// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
-// CHECK3-NEXT: store ptr @a, ptr [[TMP44]], align 4
-// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 1
-// CHECK3-NEXT: store ptr null, ptr [[TMP45]], align 4
+// CHECK3-NEXT: [[TMP44:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: store i32 [[TMP44]], ptr [[N_CASTED4]], align 4
+// CHECK3-NEXT: [[TMP45:%.*]] = load i32, ptr [[N_CASTED4]], align 4
// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP45]], ptr [[TMP46]], align 4
// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP48:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: store i32 [[TMP48]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK3-NEXT: [[TMP49:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK3-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP49]], 0
+// CHECK3-NEXT: store i32 [[TMP45]], ptr [[TMP47]], align 4
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP48]], align 4
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
+// CHECK3-NEXT: store ptr @a, ptr [[TMP49]], align 4
+// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
+// CHECK3-NEXT: store ptr @a, ptr [[TMP50]], align 4
+// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 1
+// CHECK3-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP54:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: store i32 [[TMP54]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK3-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK3-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP55]], 0
// CHECK3-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK3-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK3-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK3-NEXT: [[TMP50:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK3-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP50]], 1
-// CHECK3-NEXT: [[TMP51:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK3-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK3-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP56]], 1
+// CHECK3-NEXT: [[TMP57:%.*]] = zext i32 [[ADD14]] to i64
// CHECK3-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP52]], align 4
-// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK3-NEXT: store i32 2, ptr [[TMP53]], align 4
-// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP46]], ptr [[TMP54]], align 4
-// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP47]], ptr [[TMP55]], align 4
-// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.2, ptr [[TMP56]], align 4
-// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP57]], align 4
-// CHECK3-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP58]], align 4
-// CHECK3-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP59]], align 4
-// CHECK3-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK3-NEXT: store i64 [[TMP51]], ptr [[TMP60]], align 8
-// CHECK3-NEXT: [[TMP61:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l35.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK3-NEXT: [[TMP62:%.*]] = icmp ne i32 [[TMP61]], 0
-// CHECK3-NEXT: br i1 [[TMP62]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK3-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP58]], align 4
+// CHECK3-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK3-NEXT: store i32 2, ptr [[TMP59]], align 4
+// CHECK3-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP52]], ptr [[TMP60]], align 4
+// CHECK3-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP53]], ptr [[TMP61]], align 4
+// CHECK3-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.2, ptr [[TMP62]], align 4
+// CHECK3-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP63]], align 4
+// CHECK3-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP64]], align 4
+// CHECK3-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP65]], align 4
+// CHECK3-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK3-NEXT: store i64 [[TMP57]], ptr [[TMP66]], align 8
+// CHECK3-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP67]], align 8
+// CHECK3-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP68]], align 4
+// CHECK3-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK3-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP70]], align 4
+// CHECK3-NEXT: [[TMP71:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l35.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK3-NEXT: [[TMP72:%.*]] = icmp ne i32 [[TMP71]], 0
+// CHECK3-NEXT: br i1 [[TMP72]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK3: omp_offload.failed16:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l35(i32 [[TMP39]], ptr @a) #[[ATTR2]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l35(i32 [[TMP45]], ptr @a) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK3: omp_offload.cont17:
-// CHECK3-NEXT: [[TMP63:%.*]] = load i32, ptr @a, align 4
-// CHECK3-NEXT: ret i32 [[TMP63]]
+// CHECK3-NEXT: [[TMP73:%.*]] = load i32, ptr @a, align 4
+// CHECK3-NEXT: ret i32 [[TMP73]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l29
@@ -994,7 +1030,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1011,18 +1047,26 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK9-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l73.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK9-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l73.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK9-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l73(i64 [[TMP4]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[VLA]], i64 0
-// CHECK9-NEXT: [[TMP34:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK9-NEXT: [[TMP35:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP35]])
-// CHECK9-NEXT: ret i32 [[TMP34]]
+// CHECK9-NEXT: [[TMP38:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK9-NEXT: [[TMP39:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP39]])
+// CHECK9-NEXT: ret i32 [[TMP38]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l73
@@ -1203,7 +1247,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1220,18 +1264,26 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK11-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l73.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK11-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l73.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK11-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l73(i32 [[TMP3]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[VLA]], i32 0
-// CHECK11-NEXT: [[TMP34:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK11-NEXT: [[TMP35:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP35]])
-// CHECK11-NEXT: ret i32 [[TMP34]]
+// CHECK11-NEXT: [[TMP38:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK11-NEXT: [[TMP39:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP39]])
+// CHECK11-NEXT: ret i32 [[TMP38]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l73
@@ -1379,7 +1431,7 @@ int main (int argc, char **argv) {
// CHECK17-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK17-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK17-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK17-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK17-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK17-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1396,17 +1448,25 @@ int main (int argc, char **argv) {
// CHECK17-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK17-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK17-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK17-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l109.region_id, ptr [[KERNEL_ARGS]])
-// CHECK17-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK17-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK17-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK17-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK17-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK17-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK17-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l109.region_id, ptr [[KERNEL_ARGS]])
+// CHECK17-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK17-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK17: omp_offload.failed:
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l109(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK17: omp_offload.cont:
// CHECK17-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK17-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A2]], i64 0, i64 0
-// CHECK17-NEXT: [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK17-NEXT: ret i32 [[TMP16]]
+// CHECK17-NEXT: [[TMP20:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK17-NEXT: ret i32 [[TMP20]]
//
//
// CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l109
@@ -1523,7 +1583,7 @@ int main (int argc, char **argv) {
// CHECK19-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK19-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK19-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK19-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK19-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK19-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1540,17 +1600,25 @@ int main (int argc, char **argv) {
// CHECK19-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK19-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK19-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK19-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l109.region_id, ptr [[KERNEL_ARGS]])
-// CHECK19-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK19-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK19-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK19-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK19-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK19-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK19-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l109.region_id, ptr [[KERNEL_ARGS]])
+// CHECK19-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK19-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK19: omp_offload.failed:
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l109(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK19: omp_offload.cont:
// CHECK19-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK19-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A2]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK19-NEXT: ret i32 [[TMP16]]
+// CHECK19-NEXT: [[TMP20:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK19-NEXT: ret i32 [[TMP20]]
//
//
// CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l109
@@ -1704,7 +1772,7 @@ int main (int argc, char **argv) {
// CHECK25-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK25-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK25-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK25-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK25-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK25-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK25-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK25-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1721,20 +1789,28 @@ int main (int argc, char **argv) {
// CHECK25-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK25-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK25-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK25-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l162.region_id, ptr [[KERNEL_ARGS]])
-// CHECK25-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK25-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK25-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK25-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK25-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK25-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK25-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK25-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK25-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK25-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK25-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l162.region_id, ptr [[KERNEL_ARGS]])
+// CHECK25-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK25-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK25: omp_offload.failed:
// CHECK25-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l162(i64 [[TMP4]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK25-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK25: omp_offload.cont:
-// CHECK25-NEXT: [[TMP34:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK25-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP34]])
+// CHECK25-NEXT: [[TMP38:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK25-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP38]])
// CHECK25-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK25-NEXT: [[TMP35:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK25-NEXT: call void @llvm.stackrestore(ptr [[TMP35]])
-// CHECK25-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK25-NEXT: ret i32 [[TMP36]]
+// CHECK25-NEXT: [[TMP39:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK25-NEXT: call void @llvm.stackrestore(ptr [[TMP39]])
+// CHECK25-NEXT: [[TMP40:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK25-NEXT: ret i32 [[TMP40]]
//
//
// CHECK25-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l162
@@ -1891,28 +1967,38 @@ int main (int argc, char **argv) {
// CHECK25-NEXT: [[TMP14:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK25-NEXT: [[TMP15:%.*]] = load i32, ptr [[TE]], align 4
// CHECK25-NEXT: [[TMP16:%.*]] = load i32, ptr [[TH]], align 4
+// CHECK25-NEXT: [[TMP17:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP15]], 0
+// CHECK25-NEXT: [[TMP18:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP16]], 0
// CHECK25-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK25-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK25-NEXT: store i32 1, ptr [[TMP17]], align 4
-// CHECK25-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK25-NEXT: store i32 3, ptr [[TMP18]], align 4
-// CHECK25-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK25-NEXT: store ptr [[TMP13]], ptr [[TMP19]], align 8
-// CHECK25-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK25-NEXT: store ptr [[TMP14]], ptr [[TMP20]], align 8
-// CHECK25-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK25-NEXT: store ptr @.offload_sizes.2, ptr [[TMP21]], align 8
-// CHECK25-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK25-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP22]], align 8
-// CHECK25-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK25-NEXT: store ptr null, ptr [[TMP23]], align 8
-// CHECK25-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
-// CHECK25-NEXT: store ptr null, ptr [[TMP24]], align 8
-// CHECK25-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK25-NEXT: store i64 10, ptr [[TMP25]], align 8
-// CHECK25-NEXT: [[TMP26:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 [[TMP15]], i32 [[TMP16]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l151.region_id, ptr [[KERNEL_ARGS]])
-// CHECK25-NEXT: [[TMP27:%.*]] = icmp ne i32 [[TMP26]], 0
-// CHECK25-NEXT: br i1 [[TMP27]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK25-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK25-NEXT: store i32 2, ptr [[TMP19]], align 4
+// CHECK25-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK25-NEXT: store i32 3, ptr [[TMP20]], align 4
+// CHECK25-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK25-NEXT: store ptr [[TMP13]], ptr [[TMP21]], align 8
+// CHECK25-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK25-NEXT: store ptr [[TMP14]], ptr [[TMP22]], align 8
+// CHECK25-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK25-NEXT: store ptr @.offload_sizes.2, ptr [[TMP23]], align 8
+// CHECK25-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK25-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP24]], align 8
+// CHECK25-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
+// CHECK25-NEXT: store ptr null, ptr [[TMP25]], align 8
+// CHECK25-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK25-NEXT: store ptr null, ptr [[TMP26]], align 8
+// CHECK25-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK25-NEXT: store i64 10, ptr [[TMP27]], align 8
+// CHECK25-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK25-NEXT: store i64 0, ptr [[TMP28]], align 8
+// CHECK25-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK25-NEXT: store [3 x i32] [[TMP17]], ptr [[TMP29]], align 4
+// CHECK25-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK25-NEXT: store [3 x i32] [[TMP18]], ptr [[TMP30]], align 4
+// CHECK25-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK25-NEXT: store i32 0, ptr [[TMP31]], align 4
+// CHECK25-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 [[TMP15]], i32 [[TMP16]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l151.region_id, ptr [[KERNEL_ARGS]])
+// CHECK25-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
+// CHECK25-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK25: omp_offload.failed:
// CHECK25-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l151(i64 [[TMP1]], i64 [[TMP3]], ptr [[A]]) #[[ATTR3]]
// CHECK25-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -2079,7 +2165,7 @@ int main (int argc, char **argv) {
// CHECK27-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK27-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK27-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK27-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK27-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK27-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK27-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK27-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2096,20 +2182,28 @@ int main (int argc, char **argv) {
// CHECK27-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK27-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK27-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK27-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l162.region_id, ptr [[KERNEL_ARGS]])
-// CHECK27-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK27-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK27-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK27-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK27-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK27-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK27-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK27-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK27-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK27-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK27-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l162.region_id, ptr [[KERNEL_ARGS]])
+// CHECK27-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK27-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK27: omp_offload.failed:
// CHECK27-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l162(i32 [[TMP3]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK27-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK27: omp_offload.cont:
-// CHECK27-NEXT: [[TMP34:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK27-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP34]])
+// CHECK27-NEXT: [[TMP38:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK27-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP38]])
// CHECK27-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK27-NEXT: [[TMP35:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK27-NEXT: call void @llvm.stackrestore(ptr [[TMP35]])
-// CHECK27-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK27-NEXT: ret i32 [[TMP36]]
+// CHECK27-NEXT: [[TMP39:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK27-NEXT: call void @llvm.stackrestore(ptr [[TMP39]])
+// CHECK27-NEXT: [[TMP40:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK27-NEXT: ret i32 [[TMP40]]
//
//
// CHECK27-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l162
@@ -2265,28 +2359,38 @@ int main (int argc, char **argv) {
// CHECK27-NEXT: [[TMP14:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK27-NEXT: [[TMP15:%.*]] = load i32, ptr [[TE]], align 4
// CHECK27-NEXT: [[TMP16:%.*]] = load i32, ptr [[TH]], align 4
+// CHECK27-NEXT: [[TMP17:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP15]], 0
+// CHECK27-NEXT: [[TMP18:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP16]], 0
// CHECK27-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK27-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK27-NEXT: store i32 1, ptr [[TMP17]], align 4
-// CHECK27-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK27-NEXT: store i32 3, ptr [[TMP18]], align 4
-// CHECK27-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK27-NEXT: store ptr [[TMP13]], ptr [[TMP19]], align 4
-// CHECK27-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK27-NEXT: store ptr [[TMP14]], ptr [[TMP20]], align 4
-// CHECK27-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK27-NEXT: store ptr @.offload_sizes.2, ptr [[TMP21]], align 4
-// CHECK27-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK27-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP22]], align 4
-// CHECK27-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK27-NEXT: store ptr null, ptr [[TMP23]], align 4
-// CHECK27-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
-// CHECK27-NEXT: store ptr null, ptr [[TMP24]], align 4
-// CHECK27-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK27-NEXT: store i64 10, ptr [[TMP25]], align 8
-// CHECK27-NEXT: [[TMP26:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 [[TMP15]], i32 [[TMP16]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l151.region_id, ptr [[KERNEL_ARGS]])
-// CHECK27-NEXT: [[TMP27:%.*]] = icmp ne i32 [[TMP26]], 0
-// CHECK27-NEXT: br i1 [[TMP27]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK27-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK27-NEXT: store i32 2, ptr [[TMP19]], align 4
+// CHECK27-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK27-NEXT: store i32 3, ptr [[TMP20]], align 4
+// CHECK27-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK27-NEXT: store ptr [[TMP13]], ptr [[TMP21]], align 4
+// CHECK27-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK27-NEXT: store ptr [[TMP14]], ptr [[TMP22]], align 4
+// CHECK27-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK27-NEXT: store ptr @.offload_sizes.2, ptr [[TMP23]], align 4
+// CHECK27-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK27-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP24]], align 4
+// CHECK27-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
+// CHECK27-NEXT: store ptr null, ptr [[TMP25]], align 4
+// CHECK27-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK27-NEXT: store ptr null, ptr [[TMP26]], align 4
+// CHECK27-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK27-NEXT: store i64 10, ptr [[TMP27]], align 8
+// CHECK27-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK27-NEXT: store i64 0, ptr [[TMP28]], align 8
+// CHECK27-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK27-NEXT: store [3 x i32] [[TMP17]], ptr [[TMP29]], align 4
+// CHECK27-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK27-NEXT: store [3 x i32] [[TMP18]], ptr [[TMP30]], align 4
+// CHECK27-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK27-NEXT: store i32 0, ptr [[TMP31]], align 4
+// CHECK27-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 [[TMP15]], i32 [[TMP16]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l151.region_id, ptr [[KERNEL_ARGS]])
+// CHECK27-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
+// CHECK27-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK27: omp_offload.failed:
// CHECK27-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l151(i32 [[TMP1]], i32 [[TMP3]], ptr [[A]]) #[[ATTR3]]
// CHECK27-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/teams_distribute_collapse_codegen.cpp b/clang/test/OpenMP/teams_distribute_collapse_codegen.cpp
index b55086de61648..256c823223aa3 100644
--- a/clang/test/OpenMP/teams_distribute_collapse_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_collapse_codegen.cpp
@@ -127,7 +127,7 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -144,9 +144,17 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 56088, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -154,8 +162,8 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[A3:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x [456 x i32]], ptr [[A3]], i64 0, i64 0
// CHECK1-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds [456 x i32], ptr [[ARRAYIDX]], i64 0, i64 0
-// CHECK1-NEXT: [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
-// CHECK1-NEXT: ret i32 [[TMP16]]
+// CHECK1-NEXT: [[TMP20:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
+// CHECK1-NEXT: ret i32 [[TMP20]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -287,7 +295,7 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -304,9 +312,17 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 56088, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -314,8 +330,8 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[A3:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x [456 x i32]], ptr [[A3]], i32 0, i32 0
// CHECK3-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds [456 x i32], ptr [[ARRAYIDX]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
-// CHECK3-NEXT: ret i32 [[TMP16]]
+// CHECK3-NEXT: [[TMP20:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
+// CHECK3-NEXT: ret i32 [[TMP20]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -515,7 +531,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[ADD:%.*]] = add nsw i64 [[TMP35]], 1
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP36]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP36]], align 4
// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 5, ptr [[TMP37]], align 4
// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -532,20 +548,28 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP43]], align 8
// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 [[ADD]], ptr [[TMP44]], align 8
-// CHECK9-NEXT: [[TMP45:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l82.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP46:%.*]] = icmp ne i32 [[TMP45]], 0
-// CHECK9-NEXT: br i1 [[TMP46]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP46]], align 4
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP47]], align 4
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP48]], align 4
+// CHECK9-NEXT: [[TMP49:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l82.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP50:%.*]] = icmp ne i32 [[TMP49]], 0
+// CHECK9-NEXT: br i1 [[TMP50]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l82(i64 [[TMP7]], i64 [[TMP9]], i64 [[TMP1]], i64 [[TMP3]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP47:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef signext [[TMP47]])
+// CHECK9-NEXT: [[TMP51:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef signext [[TMP51]])
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK9-NEXT: [[TMP48:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP48]])
-// CHECK9-NEXT: [[TMP49:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP49]]
+// CHECK9-NEXT: [[TMP52:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP52]])
+// CHECK9-NEXT: [[TMP53:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP53]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l82
@@ -737,7 +761,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK9-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK9-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -754,9 +778,17 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 20, ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l68.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK9-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK9-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l68.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK9-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l68(ptr [[A]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -961,7 +993,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[ADD:%.*]] = add nsw i64 [[TMP34]], 1
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP35]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP35]], align 4
// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 5, ptr [[TMP36]], align 4
// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -978,20 +1010,28 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP42]], align 4
// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 [[ADD]], ptr [[TMP43]], align 8
-// CHECK11-NEXT: [[TMP44:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l82.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP45:%.*]] = icmp ne i32 [[TMP44]], 0
-// CHECK11-NEXT: br i1 [[TMP45]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP44]], align 8
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP47]], align 4
+// CHECK11-NEXT: [[TMP48:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l82.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP49:%.*]] = icmp ne i32 [[TMP48]], 0
+// CHECK11-NEXT: br i1 [[TMP49]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l82(i32 [[TMP5]], i32 [[TMP7]], i32 [[TMP0]], i32 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP46:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef [[TMP46]])
+// CHECK11-NEXT: [[TMP50:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef [[TMP50]])
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK11-NEXT: [[TMP47:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP47]])
-// CHECK11-NEXT: [[TMP48:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP48]]
+// CHECK11-NEXT: [[TMP51:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP51]])
+// CHECK11-NEXT: [[TMP52:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP52]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l82
@@ -1181,7 +1221,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK11-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK11-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1198,9 +1238,17 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK11-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 20, ptr [[TMP13]], align 8
-// CHECK11-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l68.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK11-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK11-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l68.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK11-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l68(ptr [[A]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/teams_distribute_dist_schedule_codegen.cpp b/clang/test/OpenMP/teams_distribute_dist_schedule_codegen.cpp
index 4d4d2762ef550..94e47786fd9f7 100644
--- a/clang/test/OpenMP/teams_distribute_dist_schedule_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_dist_schedule_codegen.cpp
@@ -168,7 +168,7 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -185,87 +185,111 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.2, ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l33.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.2, ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l33.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK1-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK1-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK1: omp_offload.failed8:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l33(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK1: omp_offload.cont9:
// CHECK1-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A10]], ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A10]], ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.5, ptr [[TMP41]], align 8
-// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP42]], align 8
-// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP43]], align 8
-// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP44]], align 8
-// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK1-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l38.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK1-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK1-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 8
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 8
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.5, ptr [[TMP49]], align 8
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP50]], align 8
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP51]], align 8
+// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP52]], align 8
+// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK1-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l38.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK1-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK1-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK1: omp_offload.failed16:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l38(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK1: omp_offload.cont17:
// CHECK1-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A18]], i64 0, i64 0
-// CHECK1-NEXT: [[TMP48:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK1-NEXT: ret i32 [[TMP48]]
+// CHECK1-NEXT: [[TMP60:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK1-NEXT: ret i32 [[TMP60]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -563,7 +587,7 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -580,87 +604,111 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 4
-// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A2]], ptr [[TMP17]], align 4
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 4
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A2]], ptr [[TMP21]], align 4
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.2, ptr [[TMP25]], align 4
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP26]], align 4
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK3-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l33.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK3-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK3-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.2, ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK3-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l33.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK3-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK3-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK3: omp_offload.failed8:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l33(ptr [[THIS1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK3: omp_offload.cont9:
// CHECK3-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 4
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A10]], ptr [[TMP33]], align 4
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP34]], align 4
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A10]], ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP42]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 4
-// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 4
-// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.5, ptr [[TMP41]], align 4
-// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP42]], align 4
-// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP44]], align 4
-// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK3-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK3-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l38.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK3-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK3-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 4
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 4
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.5, ptr [[TMP49]], align 4
+// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP50]], align 4
+// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP52]], align 4
+// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK3-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK3-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l38.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK3-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK3-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK3: omp_offload.failed16:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l38(ptr [[THIS1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK3: omp_offload.cont17:
// CHECK3-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A18]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP48:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK3-NEXT: ret i32 [[TMP48]]
+// CHECK3-NEXT: [[TMP60:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK3-NEXT: ret i32 [[TMP60]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -1001,7 +1049,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1018,148 +1066,172 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK9-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK9-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK9-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100(i64 [[TMP4]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK9-NEXT: [[TMP35:%.*]] = load i64, ptr [[N_CASTED3]], align 8
-// CHECK9-NEXT: [[TMP36:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK9-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK9-NEXT: [[TMP39:%.*]] = load i64, ptr [[N_CASTED3]], align 8
+// CHECK9-NEXT: [[TMP40:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK9-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES7]], ptr align 8 @.offload_sizes.2, i64 24, i1 false)
-// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP35]], ptr [[TMP37]], align 8
-// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP35]], ptr [[TMP38]], align 8
-// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP40]], align 8
-// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP41]], align 8
-// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP42]], align 8
-// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP43]], align 8
-// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 8
-// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK9-NEXT: store i64 [[TMP36]], ptr [[TMP45]], align 8
-// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP39]], ptr [[TMP41]], align 8
+// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP39]], ptr [[TMP42]], align 8
+// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP44]], align 8
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
// CHECK9-NEXT: store ptr null, ptr [[TMP46]], align 8
-// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP50:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP50]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK9-NEXT: [[TMP51:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK9-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP51]], 0
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP47]], align 8
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 8
+// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK9-NEXT: store i64 [[TMP40]], ptr [[TMP49]], align 8
+// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP50]], align 8
+// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP54:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP54]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK9-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK9-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP55]], 0
// CHECK9-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK9-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK9-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK9-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK9-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP52]], 1
-// CHECK9-NEXT: [[TMP53:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK9-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK9-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP56]], 1
+// CHECK9-NEXT: [[TMP57:%.*]] = zext i32 [[ADD14]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK9-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK9-NEXT: store i32 3, ptr [[TMP55]], align 4
-// CHECK9-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP47]], ptr [[TMP56]], align 8
-// CHECK9-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 8
-// CHECK9-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 8
-// CHECK9-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP59]], align 8
-// CHECK9-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP60]], align 8
-// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP61]], align 8
-// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP53]], ptr [[TMP62]], align 8
-// CHECK9-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l105.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK9-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
-// CHECK9-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK9-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP58]], align 4
+// CHECK9-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK9-NEXT: store i32 3, ptr [[TMP59]], align 4
+// CHECK9-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP51]], ptr [[TMP60]], align 8
+// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 8
+// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 8
+// CHECK9-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP63]], align 8
+// CHECK9-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP64]], align 8
+// CHECK9-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP65]], align 8
+// CHECK9-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP57]], ptr [[TMP66]], align 8
+// CHECK9-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP67]], align 8
+// CHECK9-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP68]], align 4
+// CHECK9-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK9-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP70]], align 4
+// CHECK9-NEXT: [[TMP71:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l105.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK9-NEXT: [[TMP72:%.*]] = icmp ne i32 [[TMP71]], 0
+// CHECK9-NEXT: br i1 [[TMP72]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK9: omp_offload.failed16:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l105(i64 [[TMP35]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l105(i64 [[TMP39]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK9: omp_offload.cont17:
-// CHECK9-NEXT: [[TMP65:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP65]], ptr [[N_CASTED18]], align 4
-// CHECK9-NEXT: [[TMP66:%.*]] = load i64, ptr [[N_CASTED18]], align 8
-// CHECK9-NEXT: [[TMP67:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK9-NEXT: [[TMP73:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP73]], ptr [[N_CASTED18]], align 4
+// CHECK9-NEXT: [[TMP74:%.*]] = load i64, ptr [[N_CASTED18]], align 8
+// CHECK9-NEXT: [[TMP75:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK9-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES22]], ptr align 8 @.offload_sizes.5, i64 24, i1 false)
-// CHECK9-NEXT: [[TMP68:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP66]], ptr [[TMP68]], align 8
-// CHECK9-NEXT: [[TMP69:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP66]], ptr [[TMP69]], align 8
-// CHECK9-NEXT: [[TMP70:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP70]], align 8
-// CHECK9-NEXT: [[TMP71:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP71]], align 8
-// CHECK9-NEXT: [[TMP72:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP72]], align 8
-// CHECK9-NEXT: [[TMP73:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP73]], align 8
-// CHECK9-NEXT: [[TMP74:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP74]], align 8
-// CHECK9-NEXT: [[TMP75:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP75]], align 8
-// CHECK9-NEXT: [[TMP76:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 2
-// CHECK9-NEXT: store i64 [[TMP67]], ptr [[TMP76]], align 8
-// CHECK9-NEXT: [[TMP77:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP77]], align 8
-// CHECK9-NEXT: [[TMP78:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP79:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP80:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP81:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP81]], ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK9-NEXT: [[TMP82:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK9-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP82]], 0
+// CHECK9-NEXT: [[TMP76:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP74]], ptr [[TMP76]], align 8
+// CHECK9-NEXT: [[TMP77:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP74]], ptr [[TMP77]], align 8
+// CHECK9-NEXT: [[TMP78:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP78]], align 8
+// CHECK9-NEXT: [[TMP79:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP79]], align 8
+// CHECK9-NEXT: [[TMP80:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP80]], align 8
+// CHECK9-NEXT: [[TMP81:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP81]], align 8
+// CHECK9-NEXT: [[TMP82:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP82]], align 8
+// CHECK9-NEXT: [[TMP83:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP83]], align 8
+// CHECK9-NEXT: [[TMP84:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 2
+// CHECK9-NEXT: store i64 [[TMP75]], ptr [[TMP84]], align 8
+// CHECK9-NEXT: [[TMP85:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP85]], align 8
+// CHECK9-NEXT: [[TMP86:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP87:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP88:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP89:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP89]], ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK9-NEXT: [[TMP90:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK9-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP90]], 0
// CHECK9-NEXT: [[DIV27:%.*]] = sdiv i32 [[SUB26]], 1
// CHECK9-NEXT: [[SUB28:%.*]] = sub nsw i32 [[DIV27]], 1
// CHECK9-NEXT: store i32 [[SUB28]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK9-NEXT: [[TMP83:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK9-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP83]], 1
-// CHECK9-NEXT: [[TMP84:%.*]] = zext i32 [[ADD29]] to i64
+// CHECK9-NEXT: [[TMP91:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK9-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP91]], 1
+// CHECK9-NEXT: [[TMP92:%.*]] = zext i32 [[ADD29]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS30:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP85]], align 4
-// CHECK9-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
-// CHECK9-NEXT: store i32 3, ptr [[TMP86]], align 4
-// CHECK9-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP78]], ptr [[TMP87]], align 8
-// CHECK9-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP79]], ptr [[TMP88]], align 8
-// CHECK9-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP80]], ptr [[TMP89]], align 8
-// CHECK9-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP90]], align 8
-// CHECK9-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP91]], align 8
-// CHECK9-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP92]], align 8
-// CHECK9-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP84]], ptr [[TMP93]], align 8
-// CHECK9-NEXT: [[TMP94:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l110.region_id, ptr [[KERNEL_ARGS30]])
-// CHECK9-NEXT: [[TMP95:%.*]] = icmp ne i32 [[TMP94]], 0
-// CHECK9-NEXT: br i1 [[TMP95]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
+// CHECK9-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP93]], align 4
+// CHECK9-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
+// CHECK9-NEXT: store i32 3, ptr [[TMP94]], align 4
+// CHECK9-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP86]], ptr [[TMP95]], align 8
+// CHECK9-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP87]], ptr [[TMP96]], align 8
+// CHECK9-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP88]], ptr [[TMP97]], align 8
+// CHECK9-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP98]], align 8
+// CHECK9-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP99]], align 8
+// CHECK9-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP100]], align 8
+// CHECK9-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP92]], ptr [[TMP101]], align 8
+// CHECK9-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP102]], align 8
+// CHECK9-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP103]], align 4
+// CHECK9-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP104]], align 4
+// CHECK9-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP105]], align 4
+// CHECK9-NEXT: [[TMP106:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l110.region_id, ptr [[KERNEL_ARGS30]])
+// CHECK9-NEXT: [[TMP107:%.*]] = icmp ne i32 [[TMP106]], 0
+// CHECK9-NEXT: br i1 [[TMP107]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
// CHECK9: omp_offload.failed31:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l110(i64 [[TMP66]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l110(i64 [[TMP74]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT32]]
// CHECK9: omp_offload.cont32:
-// CHECK9-NEXT: [[TMP96:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP96]])
+// CHECK9-NEXT: [[TMP108:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP108]])
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK9-NEXT: [[TMP97:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP97]])
-// CHECK9-NEXT: [[TMP98:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP98]]
+// CHECK9-NEXT: [[TMP109:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP109]])
+// CHECK9-NEXT: [[TMP110:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP110]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100
@@ -1547,7 +1619,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK9-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK9-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1564,77 +1636,101 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l79.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK9-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK9-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l79.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK9-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l79(ptr [[A]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP16]], align 8
-// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP17]], align 8
-// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP20]], align 8
+// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP21]], align 8
+// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK9-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.11, ptr [[TMP25]], align 8
-// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP26]], align 8
-// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK9-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK9-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK9-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK9-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK9-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.11, ptr [[TMP29]], align 8
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP30]], align 8
+// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK9-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK9-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK9-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK9-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK9: omp_offload.failed6:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84(ptr [[A]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK9: omp_offload.cont7:
-// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP32]], align 8
-// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP33]], align 8
-// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP34]], align 8
-// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP40]], align 8
+// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP41]], align 8
+// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP42]], align 8
+// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK9-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
-// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
-// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.14, ptr [[TMP41]], align 8
-// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.15, ptr [[TMP42]], align 8
-// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP43]], align 8
-// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP44]], align 8
-// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK9-NEXT: store i64 10, ptr [[TMP45]], align 8
-// CHECK9-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK9-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK9-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK9-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 8
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 8
+// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.14, ptr [[TMP49]], align 8
+// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.15, ptr [[TMP50]], align 8
+// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP51]], align 8
+// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP52]], align 8
+// CHECK9-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK9-NEXT: store i64 10, ptr [[TMP53]], align 8
+// CHECK9-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK9-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK9-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK9-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK9-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK9-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK9-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK9: omp_offload.failed13:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89(ptr [[A]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT14]]
@@ -1980,7 +2076,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1997,150 +2093,174 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK11-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK11-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK11-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100(i32 [[TMP3]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK11-NEXT: [[TMP35:%.*]] = load i32, ptr [[N_CASTED3]], align 4
-// CHECK11-NEXT: [[TMP36:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK11-NEXT: [[TMP37:%.*]] = sext i32 [[TMP36]] to i64
+// CHECK11-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK11-NEXT: [[TMP39:%.*]] = load i32, ptr [[N_CASTED3]], align 4
+// CHECK11-NEXT: [[TMP40:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK11-NEXT: [[TMP41:%.*]] = sext i32 [[TMP40]] to i64
// CHECK11-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES7]], ptr align 4 @.offload_sizes.2, i32 24, i1 false)
-// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP35]], ptr [[TMP38]], align 4
-// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP35]], ptr [[TMP39]], align 4
-// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP40]], align 4
-// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP41]], align 4
-// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP42]], align 4
-// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 4
-// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP45]], align 4
-// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK11-NEXT: store i64 [[TMP37]], ptr [[TMP46]], align 4
-// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP39]], ptr [[TMP42]], align 4
+// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP39]], ptr [[TMP43]], align 4
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP44]], align 4
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
// CHECK11-NEXT: store ptr null, ptr [[TMP47]], align 4
-// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP51:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP51]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK11-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK11-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP52]], 0
+// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 4
+// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP49]], align 4
+// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK11-NEXT: store i64 [[TMP41]], ptr [[TMP50]], align 4
+// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP54:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP55:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP55]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK11-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK11-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP56]], 0
// CHECK11-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK11-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK11-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK11-NEXT: [[TMP53:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK11-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP53]], 1
-// CHECK11-NEXT: [[TMP54:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK11-NEXT: [[TMP57:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK11-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP57]], 1
+// CHECK11-NEXT: [[TMP58:%.*]] = zext i32 [[ADD14]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP55]], align 4
-// CHECK11-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK11-NEXT: store i32 3, ptr [[TMP56]], align 4
-// CHECK11-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 4
-// CHECK11-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 4
-// CHECK11-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP50]], ptr [[TMP59]], align 4
-// CHECK11-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP60]], align 4
-// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP61]], align 4
-// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP62]], align 4
-// CHECK11-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP54]], ptr [[TMP63]], align 8
-// CHECK11-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l105.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK11-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
-// CHECK11-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK11-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP59]], align 4
+// CHECK11-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK11-NEXT: store i32 3, ptr [[TMP60]], align 4
+// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 4
+// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 4
+// CHECK11-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP54]], ptr [[TMP63]], align 4
+// CHECK11-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP64]], align 4
+// CHECK11-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP65]], align 4
+// CHECK11-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP66]], align 4
+// CHECK11-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP58]], ptr [[TMP67]], align 8
+// CHECK11-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP68]], align 8
+// CHECK11-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK11-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP70]], align 4
+// CHECK11-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP71]], align 4
+// CHECK11-NEXT: [[TMP72:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l105.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK11-NEXT: [[TMP73:%.*]] = icmp ne i32 [[TMP72]], 0
+// CHECK11-NEXT: br i1 [[TMP73]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK11: omp_offload.failed16:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l105(i32 [[TMP35]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l105(i32 [[TMP39]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK11: omp_offload.cont17:
-// CHECK11-NEXT: [[TMP66:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP66]], ptr [[N_CASTED18]], align 4
-// CHECK11-NEXT: [[TMP67:%.*]] = load i32, ptr [[N_CASTED18]], align 4
-// CHECK11-NEXT: [[TMP68:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK11-NEXT: [[TMP69:%.*]] = sext i32 [[TMP68]] to i64
+// CHECK11-NEXT: [[TMP74:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP74]], ptr [[N_CASTED18]], align 4
+// CHECK11-NEXT: [[TMP75:%.*]] = load i32, ptr [[N_CASTED18]], align 4
+// CHECK11-NEXT: [[TMP76:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK11-NEXT: [[TMP77:%.*]] = sext i32 [[TMP76]] to i64
// CHECK11-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES22]], ptr align 4 @.offload_sizes.5, i32 24, i1 false)
-// CHECK11-NEXT: [[TMP70:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP67]], ptr [[TMP70]], align 4
-// CHECK11-NEXT: [[TMP71:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP67]], ptr [[TMP71]], align 4
-// CHECK11-NEXT: [[TMP72:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP72]], align 4
-// CHECK11-NEXT: [[TMP73:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP73]], align 4
-// CHECK11-NEXT: [[TMP74:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP74]], align 4
-// CHECK11-NEXT: [[TMP75:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP75]], align 4
-// CHECK11-NEXT: [[TMP76:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP76]], align 4
-// CHECK11-NEXT: [[TMP77:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP77]], align 4
-// CHECK11-NEXT: [[TMP78:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 2
-// CHECK11-NEXT: store i64 [[TMP69]], ptr [[TMP78]], align 4
-// CHECK11-NEXT: [[TMP79:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP79]], align 4
-// CHECK11-NEXT: [[TMP80:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP81:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP82:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP83:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP83]], ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK11-NEXT: [[TMP84:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK11-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP84]], 0
+// CHECK11-NEXT: [[TMP78:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP75]], ptr [[TMP78]], align 4
+// CHECK11-NEXT: [[TMP79:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP75]], ptr [[TMP79]], align 4
+// CHECK11-NEXT: [[TMP80:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP80]], align 4
+// CHECK11-NEXT: [[TMP81:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP81]], align 4
+// CHECK11-NEXT: [[TMP82:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP82]], align 4
+// CHECK11-NEXT: [[TMP83:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP83]], align 4
+// CHECK11-NEXT: [[TMP84:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP84]], align 4
+// CHECK11-NEXT: [[TMP85:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP85]], align 4
+// CHECK11-NEXT: [[TMP86:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 2
+// CHECK11-NEXT: store i64 [[TMP77]], ptr [[TMP86]], align 4
+// CHECK11-NEXT: [[TMP87:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP87]], align 4
+// CHECK11-NEXT: [[TMP88:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP89:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP90:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP91:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP91]], ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK11-NEXT: [[TMP92:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK11-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP92]], 0
// CHECK11-NEXT: [[DIV27:%.*]] = sdiv i32 [[SUB26]], 1
// CHECK11-NEXT: [[SUB28:%.*]] = sub nsw i32 [[DIV27]], 1
// CHECK11-NEXT: store i32 [[SUB28]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK11-NEXT: [[TMP85:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK11-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP85]], 1
-// CHECK11-NEXT: [[TMP86:%.*]] = zext i32 [[ADD29]] to i64
+// CHECK11-NEXT: [[TMP93:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK11-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP93]], 1
+// CHECK11-NEXT: [[TMP94:%.*]] = zext i32 [[ADD29]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS30:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP87]], align 4
-// CHECK11-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
-// CHECK11-NEXT: store i32 3, ptr [[TMP88]], align 4
-// CHECK11-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP80]], ptr [[TMP89]], align 4
-// CHECK11-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP81]], ptr [[TMP90]], align 4
-// CHECK11-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP82]], ptr [[TMP91]], align 4
-// CHECK11-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP92]], align 4
-// CHECK11-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP93]], align 4
-// CHECK11-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP94]], align 4
-// CHECK11-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP86]], ptr [[TMP95]], align 8
-// CHECK11-NEXT: [[TMP96:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l110.region_id, ptr [[KERNEL_ARGS30]])
-// CHECK11-NEXT: [[TMP97:%.*]] = icmp ne i32 [[TMP96]], 0
-// CHECK11-NEXT: br i1 [[TMP97]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
+// CHECK11-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP95]], align 4
+// CHECK11-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
+// CHECK11-NEXT: store i32 3, ptr [[TMP96]], align 4
+// CHECK11-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP88]], ptr [[TMP97]], align 4
+// CHECK11-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP89]], ptr [[TMP98]], align 4
+// CHECK11-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP90]], ptr [[TMP99]], align 4
+// CHECK11-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP100]], align 4
+// CHECK11-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP101]], align 4
+// CHECK11-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP102]], align 4
+// CHECK11-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP94]], ptr [[TMP103]], align 8
+// CHECK11-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP104]], align 8
+// CHECK11-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP105]], align 4
+// CHECK11-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP106]], align 4
+// CHECK11-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP107]], align 4
+// CHECK11-NEXT: [[TMP108:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l110.region_id, ptr [[KERNEL_ARGS30]])
+// CHECK11-NEXT: [[TMP109:%.*]] = icmp ne i32 [[TMP108]], 0
+// CHECK11-NEXT: br i1 [[TMP109]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
// CHECK11: omp_offload.failed31:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l110(i32 [[TMP67]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l110(i32 [[TMP75]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT32]]
// CHECK11: omp_offload.cont32:
-// CHECK11-NEXT: [[TMP98:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP98]])
+// CHECK11-NEXT: [[TMP110:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP110]])
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK11-NEXT: [[TMP99:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP99]])
-// CHECK11-NEXT: [[TMP100:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP100]]
+// CHECK11-NEXT: [[TMP111:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP111]])
+// CHECK11-NEXT: [[TMP112:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP112]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100
@@ -2525,7 +2645,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK11-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK11-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2542,77 +2662,101 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK11-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK11-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l79.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK11-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK11-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l79.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK11-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l79(ptr [[A]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP16]], align 4
-// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP17]], align 4
-// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP20]], align 4
+// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP21]], align 4
+// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK11-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.11, ptr [[TMP25]], align 4
-// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP26]], align 4
-// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK11-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK11-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK11-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK11-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK11-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.11, ptr [[TMP29]], align 4
+// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP30]], align 4
+// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK11-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK11-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK11-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK11-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK11: omp_offload.failed6:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84(ptr [[A]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK11: omp_offload.cont7:
-// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP32]], align 4
-// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP33]], align 4
-// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP34]], align 4
-// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP40]], align 4
+// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP41]], align 4
+// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP42]], align 4
+// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK11-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 4
-// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 4
-// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.14, ptr [[TMP41]], align 4
-// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.15, ptr [[TMP42]], align 4
-// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP44]], align 4
-// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK11-NEXT: store i64 10, ptr [[TMP45]], align 8
-// CHECK11-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK11-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK11-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK11-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 4
+// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 4
+// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.14, ptr [[TMP49]], align 4
+// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.15, ptr [[TMP50]], align 4
+// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP52]], align 4
+// CHECK11-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK11-NEXT: store i64 10, ptr [[TMP53]], align 8
+// CHECK11-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK11-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK11-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK11-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK11-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK11-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK11-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK11: omp_offload.failed13:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89(ptr [[A]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT14]]
diff --git a/clang/test/OpenMP/teams_distribute_firstprivate_codegen.cpp b/clang/test/OpenMP/teams_distribute_firstprivate_codegen.cpp
index a2df2fe99ace3..9c7ead55fd6c5 100644
--- a/clang/test/OpenMP/teams_distribute_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_firstprivate_codegen.cpp
@@ -306,7 +306,7 @@ int main() {
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 5, ptr [[TMP22]], align 4
// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -323,9 +323,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK1-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK1-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94(i64 [[TMP1]], ptr @vec, ptr @s_arr, ptr @var, i64 [[TMP3]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -579,7 +587,7 @@ int main() {
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -596,27 +604,35 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK1-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK1-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK1-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56(i64 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK1-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK1: arraydestroy.body:
-// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK1-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK1-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK1: arraydestroy.done2:
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK1-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK1-NEXT: ret i32 [[TMP32]]
+// CHECK1-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK1-NEXT: ret i32 [[TMP36]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN2StC2Ev
@@ -1093,7 +1109,7 @@ int main() {
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 5, ptr [[TMP22]], align 4
// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1110,9 +1126,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4
// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP29]], align 8
-// CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK3-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK3-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK3-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94(i32 [[TMP1]], ptr @vec, ptr @s_arr, ptr @var, i32 [[TMP3]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1364,7 +1388,7 @@ int main() {
// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1381,27 +1405,35 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP27]], align 4
// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK3-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK3-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK3-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56(i32 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK3-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK3-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK3: arraydestroy.body:
-// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK3-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK3-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK3: arraydestroy.done2:
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK3-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK3-NEXT: ret i32 [[TMP32]]
+// CHECK3-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK3-NEXT: ret i32 [[TMP36]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN2StC2Ev
diff --git a/clang/test/OpenMP/teams_distribute_lastprivate_codegen.cpp b/clang/test/OpenMP/teams_distribute_lastprivate_codegen.cpp
index 6fa37bc5e2037..a3c4723bad37b 100644
--- a/clang/test/OpenMP/teams_distribute_lastprivate_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_lastprivate_codegen.cpp
@@ -532,7 +532,7 @@ int main() {
// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP25]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP25]], align 4
// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 5, ptr [[TMP26]], align 4
// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -549,9 +549,17 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP32]], align 8
// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP33]], align 8
-// CHECK9-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l96.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
-// CHECK9-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK9-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l96.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK9-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l96(i64 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]], i64 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -559,18 +567,18 @@ int main() {
// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiET_v()
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP36]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP40]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP37:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP37]]
+// CHECK9-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP41]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -835,7 +843,7 @@ int main() {
// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -852,27 +860,35 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP27]], align 8
// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK9-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK9-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK9-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK9-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49(i64 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP32]]
+// CHECK9-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP36]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
@@ -1219,7 +1235,7 @@ int main() {
// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP25]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP25]], align 4
// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 5, ptr [[TMP26]], align 4
// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1236,9 +1252,17 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP32]], align 4
// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP33]], align 8
-// CHECK11-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l96.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
-// CHECK11-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK11-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l96.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK11-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l96(i32 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]], i32 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1246,18 +1270,18 @@ int main() {
// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiET_v()
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP36]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP40]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP37:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP37]]
+// CHECK11-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP41]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -1520,7 +1544,7 @@ int main() {
// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1537,27 +1561,35 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP27]], align 4
// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK11-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK11-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK11-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK11-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49(i32 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP32]]
+// CHECK11-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP36]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
diff --git a/clang/test/OpenMP/teams_distribute_parallel_for_codegen.cpp b/clang/test/OpenMP/teams_distribute_parallel_for_codegen.cpp
index af43a1d2453b5..f7046f352bf6c 100644
--- a/clang/test/OpenMP/teams_distribute_parallel_for_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_parallel_for_codegen.cpp
@@ -248,87 +248,104 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP24]], 1
// CHECK1-NEXT: [[TMP25:%.*]] = zext i32 [[ADD]] to i64
+// CHECK1-NEXT: [[TMP26:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP21]], 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP26]], align 4
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK1-NEXT: store i32 4, ptr [[TMP27]], align 4
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP19]], ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP20]], ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP30]], align 8
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP31]], align 8
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP32]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK1-NEXT: store i32 4, ptr [[TMP28]], align 4
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP19]], ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP20]], ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK1-NEXT: store ptr null, ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK1-NEXT: store i64 [[TMP25]], ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 [[TMP21]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP36:%.*]] = icmp ne i32 [[TMP35]], 0
-// CHECK1-NEXT: br i1 [[TMP36]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK1-NEXT: store i64 [[TMP25]], ptr [[TMP35]], align 8
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP36]], align 8
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [[TMP26]], ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP38]], align 4
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP39]], align 4
+// CHECK1-NEXT: [[TMP40:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 [[TMP21]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP41:%.*]] = icmp ne i32 [[TMP40]], 0
+// CHECK1-NEXT: br i1 [[TMP41]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l28(i64 [[TMP2]], i64 [[TMP4]], i64 [[TMP6]], ptr @a) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
-// CHECK1-NEXT: [[TMP37:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: store i32 [[TMP37]], ptr [[N_CASTED4]], align 4
-// CHECK1-NEXT: [[TMP38:%.*]] = load i64, ptr [[N_CASTED4]], align 8
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP38]], ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP38]], ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP41]], align 8
-// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
-// CHECK1-NEXT: store ptr @a, ptr [[TMP42]], align 8
-// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
-// CHECK1-NEXT: store ptr @a, ptr [[TMP43]], align 8
-// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP44]], align 8
-// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP47:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: store i32 [[TMP47]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK1-NEXT: [[TMP48:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK1-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP48]], 0
+// CHECK1-NEXT: [[TMP42:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: store i32 [[TMP42]], ptr [[N_CASTED4]], align 4
+// CHECK1-NEXT: [[TMP43:%.*]] = load i64, ptr [[N_CASTED4]], align 8
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP43]], ptr [[TMP44]], align 8
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP43]], ptr [[TMP45]], align 8
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP46]], align 8
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
+// CHECK1-NEXT: store ptr @a, ptr [[TMP47]], align 8
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
+// CHECK1-NEXT: store ptr @a, ptr [[TMP48]], align 8
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP49]], align 8
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP52:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: store i32 [[TMP52]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK1-NEXT: [[TMP53:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK1-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP53]], 0
// CHECK1-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK1-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK1-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK1-NEXT: [[TMP49:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK1-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP49]], 1
-// CHECK1-NEXT: [[TMP50:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK1-NEXT: [[TMP54:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK1-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP54]], 1
+// CHECK1-NEXT: [[TMP55:%.*]] = zext i32 [[ADD14]] to i64
// CHECK1-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP51]], align 4
-// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK1-NEXT: store i32 2, ptr [[TMP52]], align 4
-// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP45]], ptr [[TMP53]], align 8
-// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP46]], ptr [[TMP54]], align 8
-// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.4, ptr [[TMP55]], align 8
-// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP56]], align 8
-// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP57]], align 8
-// CHECK1-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP58]], align 8
-// CHECK1-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK1-NEXT: store i64 [[TMP50]], ptr [[TMP59]], align 8
-// CHECK1-NEXT: [[TMP60:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l35.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK1-NEXT: [[TMP61:%.*]] = icmp ne i32 [[TMP60]], 0
-// CHECK1-NEXT: br i1 [[TMP61]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP56]], align 4
+// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK1-NEXT: store i32 2, ptr [[TMP57]], align 4
+// CHECK1-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP50]], ptr [[TMP58]], align 8
+// CHECK1-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP51]], ptr [[TMP59]], align 8
+// CHECK1-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.4, ptr [[TMP60]], align 8
+// CHECK1-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP61]], align 8
+// CHECK1-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP62]], align 8
+// CHECK1-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP63]], align 8
+// CHECK1-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK1-NEXT: store i64 [[TMP55]], ptr [[TMP64]], align 8
+// CHECK1-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP65]], align 8
+// CHECK1-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP66]], align 4
+// CHECK1-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP67]], align 4
+// CHECK1-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP68]], align 4
+// CHECK1-NEXT: [[TMP69:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l35.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK1-NEXT: [[TMP70:%.*]] = icmp ne i32 [[TMP69]], 0
+// CHECK1-NEXT: br i1 [[TMP70]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK1: omp_offload.failed16:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l35(i64 [[TMP38]], ptr @a) #[[ATTR2]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l35(i64 [[TMP43]], ptr @a) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK1: omp_offload.cont17:
-// CHECK1-NEXT: [[TMP62:%.*]] = load i32, ptr @a, align 4
-// CHECK1-NEXT: ret i32 [[TMP62]]
+// CHECK1-NEXT: [[TMP71:%.*]] = load i32, ptr @a, align 4
+// CHECK1-NEXT: ret i32 [[TMP71]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l28
@@ -838,87 +855,104 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP24]], 1
// CHECK3-NEXT: [[TMP25:%.*]] = zext i32 [[ADD]] to i64
+// CHECK3-NEXT: [[TMP26:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP21]], 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP26]], align 4
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK3-NEXT: store i32 4, ptr [[TMP27]], align 4
-// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP19]], ptr [[TMP28]], align 4
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP20]], ptr [[TMP29]], align 4
-// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes, ptr [[TMP30]], align 4
-// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP31]], align 4
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP32]], align 4
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK3-NEXT: store i32 4, ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP19]], ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP20]], ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK3-NEXT: store ptr null, ptr [[TMP33]], align 4
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK3-NEXT: store i64 [[TMP25]], ptr [[TMP34]], align 8
-// CHECK3-NEXT: [[TMP35:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 [[TMP21]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP36:%.*]] = icmp ne i32 [[TMP35]], 0
-// CHECK3-NEXT: br i1 [[TMP36]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP34]], align 4
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK3-NEXT: store i64 [[TMP25]], ptr [[TMP35]], align 8
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP36]], align 8
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [[TMP26]], ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP38]], align 4
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP39]], align 4
+// CHECK3-NEXT: [[TMP40:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 [[TMP21]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP41:%.*]] = icmp ne i32 [[TMP40]], 0
+// CHECK3-NEXT: br i1 [[TMP41]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l28(i32 [[TMP2]], i32 [[TMP4]], i32 [[TMP6]], ptr @a) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
-// CHECK3-NEXT: [[TMP37:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: store i32 [[TMP37]], ptr [[N_CASTED4]], align 4
-// CHECK3-NEXT: [[TMP38:%.*]] = load i32, ptr [[N_CASTED4]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP38]], ptr [[TMP39]], align 4
-// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP38]], ptr [[TMP40]], align 4
-// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP41]], align 4
-// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
-// CHECK3-NEXT: store ptr @a, ptr [[TMP42]], align 4
-// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
-// CHECK3-NEXT: store ptr @a, ptr [[TMP43]], align 4
-// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 1
-// CHECK3-NEXT: store ptr null, ptr [[TMP44]], align 4
-// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP47:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: store i32 [[TMP47]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK3-NEXT: [[TMP48:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK3-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP48]], 0
+// CHECK3-NEXT: [[TMP42:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: store i32 [[TMP42]], ptr [[N_CASTED4]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = load i32, ptr [[N_CASTED4]], align 4
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP43]], ptr [[TMP44]], align 4
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP43]], ptr [[TMP45]], align 4
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP46]], align 4
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
+// CHECK3-NEXT: store ptr @a, ptr [[TMP47]], align 4
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
+// CHECK3-NEXT: store ptr @a, ptr [[TMP48]], align 4
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 1
+// CHECK3-NEXT: store ptr null, ptr [[TMP49]], align 4
+// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP52:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: store i32 [[TMP52]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK3-NEXT: [[TMP53:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK3-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP53]], 0
// CHECK3-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK3-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK3-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK3-NEXT: [[TMP49:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK3-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP49]], 1
-// CHECK3-NEXT: [[TMP50:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK3-NEXT: [[TMP54:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK3-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP54]], 1
+// CHECK3-NEXT: [[TMP55:%.*]] = zext i32 [[ADD14]] to i64
// CHECK3-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP51]], align 4
-// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK3-NEXT: store i32 2, ptr [[TMP52]], align 4
-// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP45]], ptr [[TMP53]], align 4
-// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP46]], ptr [[TMP54]], align 4
-// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.4, ptr [[TMP55]], align 4
-// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP56]], align 4
-// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP57]], align 4
-// CHECK3-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP58]], align 4
-// CHECK3-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK3-NEXT: store i64 [[TMP50]], ptr [[TMP59]], align 8
-// CHECK3-NEXT: [[TMP60:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l35.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK3-NEXT: [[TMP61:%.*]] = icmp ne i32 [[TMP60]], 0
-// CHECK3-NEXT: br i1 [[TMP61]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP56]], align 4
+// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK3-NEXT: store i32 2, ptr [[TMP57]], align 4
+// CHECK3-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP50]], ptr [[TMP58]], align 4
+// CHECK3-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP51]], ptr [[TMP59]], align 4
+// CHECK3-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.4, ptr [[TMP60]], align 4
+// CHECK3-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP61]], align 4
+// CHECK3-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP62]], align 4
+// CHECK3-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP63]], align 4
+// CHECK3-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK3-NEXT: store i64 [[TMP55]], ptr [[TMP64]], align 8
+// CHECK3-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP65]], align 8
+// CHECK3-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP66]], align 4
+// CHECK3-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP67]], align 4
+// CHECK3-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP68]], align 4
+// CHECK3-NEXT: [[TMP69:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l35.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK3-NEXT: [[TMP70:%.*]] = icmp ne i32 [[TMP69]], 0
+// CHECK3-NEXT: br i1 [[TMP70]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK3: omp_offload.failed16:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l35(i32 [[TMP38]], ptr @a) #[[ATTR2]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l35(i32 [[TMP43]], ptr @a) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK3: omp_offload.cont17:
-// CHECK3-NEXT: [[TMP62:%.*]] = load i32, ptr @a, align 4
-// CHECK3-NEXT: ret i32 [[TMP62]]
+// CHECK3-NEXT: [[TMP71:%.*]] = load i32, ptr @a, align 4
+// CHECK3-NEXT: ret i32 [[TMP71]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l28
@@ -1406,7 +1440,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1423,18 +1457,26 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK9-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l73.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK9-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l73.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK9-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l73(i64 [[TMP4]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[VLA]], i64 0
-// CHECK9-NEXT: [[TMP34:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK9-NEXT: [[TMP35:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP35]])
-// CHECK9-NEXT: ret i32 [[TMP34]]
+// CHECK9-NEXT: [[TMP38:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK9-NEXT: [[TMP39:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP39]])
+// CHECK9-NEXT: ret i32 [[TMP38]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l73
@@ -1716,7 +1758,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1733,18 +1775,26 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK11-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l73.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK11-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l73.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK11-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l73(i32 [[TMP3]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[VLA]], i32 0
-// CHECK11-NEXT: [[TMP34:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK11-NEXT: [[TMP35:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP35]])
-// CHECK11-NEXT: ret i32 [[TMP34]]
+// CHECK11-NEXT: [[TMP38:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK11-NEXT: [[TMP39:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP39]])
+// CHECK11-NEXT: ret i32 [[TMP38]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l73
@@ -1989,7 +2039,7 @@ int main (int argc, char **argv) {
// CHECK17-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK17-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK17-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK17-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK17-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK17-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2006,17 +2056,25 @@ int main (int argc, char **argv) {
// CHECK17-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK17-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK17-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK17-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l109.region_id, ptr [[KERNEL_ARGS]])
-// CHECK17-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK17-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK17-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK17-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK17-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK17-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK17-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l109.region_id, ptr [[KERNEL_ARGS]])
+// CHECK17-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK17-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK17: omp_offload.failed:
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l109(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK17: omp_offload.cont:
// CHECK17-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK17-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A2]], i64 0, i64 0
-// CHECK17-NEXT: [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK17-NEXT: ret i32 [[TMP16]]
+// CHECK17-NEXT: [[TMP20:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK17-NEXT: ret i32 [[TMP20]]
//
//
// CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l109
@@ -2206,7 +2264,7 @@ int main (int argc, char **argv) {
// CHECK19-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK19-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK19-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK19-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK19-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK19-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2223,17 +2281,25 @@ int main (int argc, char **argv) {
// CHECK19-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK19-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK19-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK19-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l109.region_id, ptr [[KERNEL_ARGS]])
-// CHECK19-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK19-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK19-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK19-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK19-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK19-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK19-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l109.region_id, ptr [[KERNEL_ARGS]])
+// CHECK19-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK19-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK19: omp_offload.failed:
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l109(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK19: omp_offload.cont:
// CHECK19-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK19-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A2]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK19-NEXT: ret i32 [[TMP16]]
+// CHECK19-NEXT: [[TMP20:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK19-NEXT: ret i32 [[TMP20]]
//
//
// CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l109
@@ -2456,7 +2522,7 @@ int main (int argc, char **argv) {
// CHECK25-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK25-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK25-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK25-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK25-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK25-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK25-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK25-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2473,20 +2539,28 @@ int main (int argc, char **argv) {
// CHECK25-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK25-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK25-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK25-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l162.region_id, ptr [[KERNEL_ARGS]])
-// CHECK25-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK25-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK25-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK25-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK25-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK25-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK25-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK25-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK25-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK25-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK25-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l162.region_id, ptr [[KERNEL_ARGS]])
+// CHECK25-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK25-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK25: omp_offload.failed:
// CHECK25-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l162(i64 [[TMP4]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK25-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK25: omp_offload.cont:
-// CHECK25-NEXT: [[TMP34:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK25-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP34]])
+// CHECK25-NEXT: [[TMP38:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK25-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP38]])
// CHECK25-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK25-NEXT: [[TMP35:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK25-NEXT: call void @llvm.stackrestore(ptr [[TMP35]])
-// CHECK25-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK25-NEXT: ret i32 [[TMP36]]
+// CHECK25-NEXT: [[TMP39:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK25-NEXT: call void @llvm.stackrestore(ptr [[TMP39]])
+// CHECK25-NEXT: [[TMP40:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK25-NEXT: ret i32 [[TMP40]]
//
//
// CHECK25-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l162
@@ -2743,28 +2817,37 @@ int main (int argc, char **argv) {
// CHECK25-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK25-NEXT: [[TMP14:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK25-NEXT: [[TMP15:%.*]] = load i32, ptr [[TE]], align 4
+// CHECK25-NEXT: [[TMP16:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP15]], 0
// CHECK25-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK25-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK25-NEXT: store i32 1, ptr [[TMP16]], align 4
-// CHECK25-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK25-NEXT: store i32 3, ptr [[TMP17]], align 4
-// CHECK25-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK25-NEXT: store ptr [[TMP13]], ptr [[TMP18]], align 8
-// CHECK25-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK25-NEXT: store ptr [[TMP14]], ptr [[TMP19]], align 8
-// CHECK25-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK25-NEXT: store ptr @.offload_sizes.4, ptr [[TMP20]], align 8
-// CHECK25-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK25-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP21]], align 8
-// CHECK25-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK25-NEXT: store ptr null, ptr [[TMP22]], align 8
-// CHECK25-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK25-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK25-NEXT: store i32 2, ptr [[TMP17]], align 4
+// CHECK25-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK25-NEXT: store i32 3, ptr [[TMP18]], align 4
+// CHECK25-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK25-NEXT: store ptr [[TMP13]], ptr [[TMP19]], align 8
+// CHECK25-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK25-NEXT: store ptr [[TMP14]], ptr [[TMP20]], align 8
+// CHECK25-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK25-NEXT: store ptr @.offload_sizes.4, ptr [[TMP21]], align 8
+// CHECK25-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK25-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP22]], align 8
+// CHECK25-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK25-NEXT: store ptr null, ptr [[TMP23]], align 8
-// CHECK25-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK25-NEXT: store i64 10, ptr [[TMP24]], align 8
-// CHECK25-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 [[TMP15]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l151.region_id, ptr [[KERNEL_ARGS]])
-// CHECK25-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK25-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK25-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK25-NEXT: store ptr null, ptr [[TMP24]], align 8
+// CHECK25-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK25-NEXT: store i64 10, ptr [[TMP25]], align 8
+// CHECK25-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK25-NEXT: store i64 0, ptr [[TMP26]], align 8
+// CHECK25-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK25-NEXT: store [3 x i32] [[TMP16]], ptr [[TMP27]], align 4
+// CHECK25-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK25-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK25-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK25-NEXT: store i32 0, ptr [[TMP29]], align 4
+// CHECK25-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 [[TMP15]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l151.region_id, ptr [[KERNEL_ARGS]])
+// CHECK25-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
+// CHECK25-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK25: omp_offload.failed:
// CHECK25-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l151(i64 [[TMP1]], i64 [[TMP3]], ptr [[A]]) #[[ATTR3]]
// CHECK25-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -3004,7 +3087,7 @@ int main (int argc, char **argv) {
// CHECK27-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK27-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK27-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK27-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK27-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK27-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK27-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK27-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3021,20 +3104,28 @@ int main (int argc, char **argv) {
// CHECK27-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK27-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK27-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK27-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l162.region_id, ptr [[KERNEL_ARGS]])
-// CHECK27-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK27-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK27-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK27-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK27-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK27-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK27-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK27-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK27-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK27-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK27-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l162.region_id, ptr [[KERNEL_ARGS]])
+// CHECK27-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK27-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK27: omp_offload.failed:
// CHECK27-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l162(i32 [[TMP3]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK27-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK27: omp_offload.cont:
-// CHECK27-NEXT: [[TMP34:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK27-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP34]])
+// CHECK27-NEXT: [[TMP38:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK27-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP38]])
// CHECK27-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK27-NEXT: [[TMP35:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK27-NEXT: call void @llvm.stackrestore(ptr [[TMP35]])
-// CHECK27-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK27-NEXT: ret i32 [[TMP36]]
+// CHECK27-NEXT: [[TMP39:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK27-NEXT: call void @llvm.stackrestore(ptr [[TMP39]])
+// CHECK27-NEXT: [[TMP40:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK27-NEXT: ret i32 [[TMP40]]
//
//
// CHECK27-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l162
@@ -3286,28 +3377,37 @@ int main (int argc, char **argv) {
// CHECK27-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK27-NEXT: [[TMP14:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK27-NEXT: [[TMP15:%.*]] = load i32, ptr [[TE]], align 4
+// CHECK27-NEXT: [[TMP16:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP15]], 0
// CHECK27-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK27-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK27-NEXT: store i32 1, ptr [[TMP16]], align 4
-// CHECK27-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK27-NEXT: store i32 3, ptr [[TMP17]], align 4
-// CHECK27-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK27-NEXT: store ptr [[TMP13]], ptr [[TMP18]], align 4
-// CHECK27-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK27-NEXT: store ptr [[TMP14]], ptr [[TMP19]], align 4
-// CHECK27-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK27-NEXT: store ptr @.offload_sizes.4, ptr [[TMP20]], align 4
-// CHECK27-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK27-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP21]], align 4
-// CHECK27-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK27-NEXT: store ptr null, ptr [[TMP22]], align 4
-// CHECK27-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK27-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK27-NEXT: store i32 2, ptr [[TMP17]], align 4
+// CHECK27-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK27-NEXT: store i32 3, ptr [[TMP18]], align 4
+// CHECK27-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK27-NEXT: store ptr [[TMP13]], ptr [[TMP19]], align 4
+// CHECK27-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK27-NEXT: store ptr [[TMP14]], ptr [[TMP20]], align 4
+// CHECK27-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK27-NEXT: store ptr @.offload_sizes.4, ptr [[TMP21]], align 4
+// CHECK27-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK27-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP22]], align 4
+// CHECK27-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK27-NEXT: store ptr null, ptr [[TMP23]], align 4
-// CHECK27-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK27-NEXT: store i64 10, ptr [[TMP24]], align 8
-// CHECK27-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 [[TMP15]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l151.region_id, ptr [[KERNEL_ARGS]])
-// CHECK27-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK27-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK27-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK27-NEXT: store ptr null, ptr [[TMP24]], align 4
+// CHECK27-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK27-NEXT: store i64 10, ptr [[TMP25]], align 8
+// CHECK27-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK27-NEXT: store i64 0, ptr [[TMP26]], align 8
+// CHECK27-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK27-NEXT: store [3 x i32] [[TMP16]], ptr [[TMP27]], align 4
+// CHECK27-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK27-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK27-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK27-NEXT: store i32 0, ptr [[TMP29]], align 4
+// CHECK27-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 [[TMP15]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l151.region_id, ptr [[KERNEL_ARGS]])
+// CHECK27-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
+// CHECK27-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK27: omp_offload.failed:
// CHECK27-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l151(i32 [[TMP1]], i32 [[TMP3]], ptr [[A]]) #[[ATTR3]]
// CHECK27-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/teams_distribute_parallel_for_collapse_codegen.cpp b/clang/test/OpenMP/teams_distribute_parallel_for_collapse_codegen.cpp
index 4ffdb5103c52d..e4501d975a531 100644
--- a/clang/test/OpenMP/teams_distribute_parallel_for_collapse_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_parallel_for_collapse_codegen.cpp
@@ -132,7 +132,7 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -149,9 +149,17 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 56088, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -159,8 +167,8 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[A3:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x [456 x i32]], ptr [[A3]], i64 0, i64 0
// CHECK1-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds [456 x i32], ptr [[ARRAYIDX]], i64 0, i64 0
-// CHECK1-NEXT: [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
-// CHECK1-NEXT: ret i32 [[TMP16]]
+// CHECK1-NEXT: [[TMP20:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
+// CHECK1-NEXT: ret i32 [[TMP20]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -367,7 +375,7 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -384,9 +392,17 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 56088, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -394,8 +410,8 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[A3:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x [456 x i32]], ptr [[A3]], i32 0, i32 0
// CHECK3-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds [456 x i32], ptr [[ARRAYIDX]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
-// CHECK3-NEXT: ret i32 [[TMP16]]
+// CHECK3-NEXT: [[TMP20:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
+// CHECK3-NEXT: ret i32 [[TMP20]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -666,7 +682,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[ADD:%.*]] = add nsw i64 [[TMP35]], 1
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP36]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP36]], align 4
// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 5, ptr [[TMP37]], align 4
// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -683,20 +699,28 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP43]], align 8
// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 [[ADD]], ptr [[TMP44]], align 8
-// CHECK9-NEXT: [[TMP45:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l83.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP46:%.*]] = icmp ne i32 [[TMP45]], 0
-// CHECK9-NEXT: br i1 [[TMP46]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP46]], align 4
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP47]], align 4
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP48]], align 4
+// CHECK9-NEXT: [[TMP49:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l83.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP50:%.*]] = icmp ne i32 [[TMP49]], 0
+// CHECK9-NEXT: br i1 [[TMP50]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l83(i64 [[TMP7]], i64 [[TMP9]], i64 [[TMP1]], i64 [[TMP3]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP47:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef signext [[TMP47]])
+// CHECK9-NEXT: [[TMP51:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef signext [[TMP51]])
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK9-NEXT: [[TMP48:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP48]])
-// CHECK9-NEXT: [[TMP49:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP49]]
+// CHECK9-NEXT: [[TMP52:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP52]])
+// CHECK9-NEXT: [[TMP53:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP53]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l83
@@ -1008,7 +1032,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK9-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK9-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1025,9 +1049,17 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 20, ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l69.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK9-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK9-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l69.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK9-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l69(ptr [[A]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1307,7 +1339,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[ADD:%.*]] = add nsw i64 [[TMP34]], 1
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP35]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP35]], align 4
// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 5, ptr [[TMP36]], align 4
// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1324,20 +1356,28 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP42]], align 4
// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 [[ADD]], ptr [[TMP43]], align 8
-// CHECK11-NEXT: [[TMP44:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l83.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP45:%.*]] = icmp ne i32 [[TMP44]], 0
-// CHECK11-NEXT: br i1 [[TMP45]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP44]], align 8
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP47]], align 4
+// CHECK11-NEXT: [[TMP48:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l83.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP49:%.*]] = icmp ne i32 [[TMP48]], 0
+// CHECK11-NEXT: br i1 [[TMP49]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l83(i32 [[TMP5]], i32 [[TMP7]], i32 [[TMP0]], i32 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP46:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef [[TMP46]])
+// CHECK11-NEXT: [[TMP50:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef [[TMP50]])
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK11-NEXT: [[TMP47:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP47]])
-// CHECK11-NEXT: [[TMP48:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP48]]
+// CHECK11-NEXT: [[TMP51:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP51]])
+// CHECK11-NEXT: [[TMP52:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP52]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l83
@@ -1651,7 +1691,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK11-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK11-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1668,9 +1708,17 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK11-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 20, ptr [[TMP13]], align 8
-// CHECK11-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l69.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK11-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK11-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l69.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK11-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l69(ptr [[A]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/teams_distribute_parallel_for_copyin_codegen.cpp b/clang/test/OpenMP/teams_distribute_parallel_for_copyin_codegen.cpp
index 73444042d75ed..6df07faa00ee0 100644
--- a/clang/test/OpenMP/teams_distribute_parallel_for_copyin_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_parallel_for_copyin_codegen.cpp
@@ -123,7 +123,7 @@ int main() {
// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP11]], align 4
// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 2, ptr [[TMP12]], align 4
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -140,9 +140,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l64.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP21]], align 4
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP22]], align 4
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP23]], align 4
+// CHECK1-NEXT: [[TMP24:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l64.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP25:%.*]] = icmp ne i32 [[TMP24]], 0
+// CHECK1-NEXT: br i1 [[TMP25]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l64(i64 [[TMP2]], ptr [[A]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -339,7 +347,7 @@ int main() {
// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP11]], align 4
// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 2, ptr [[TMP12]], align 4
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -356,9 +364,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l34.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP21]], align 4
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP22]], align 4
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP23]], align 4
+// CHECK1-NEXT: [[TMP24:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l34.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP25:%.*]] = icmp ne i32 [[TMP24]], 0
+// CHECK1-NEXT: br i1 [[TMP25]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l34(i64 [[TMP2]], ptr [[A]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -569,7 +585,7 @@ int main() {
// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP11]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP11]], align 4
// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 2, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -586,9 +602,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP18]], align 4
// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP19]], align 8
-// CHECK3-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l64.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK3-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP20]], align 8
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP21]], align 4
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP22]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP23]], align 4
+// CHECK3-NEXT: [[TMP24:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l64.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP25:%.*]] = icmp ne i32 [[TMP24]], 0
+// CHECK3-NEXT: br i1 [[TMP25]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l64(i32 [[TMP2]], ptr [[A]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -780,7 +804,7 @@ int main() {
// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP11]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP11]], align 4
// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 2, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -797,9 +821,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP18]], align 4
// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP19]], align 8
-// CHECK3-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l34.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK3-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP20]], align 8
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP21]], align 4
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP22]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP23]], align 4
+// CHECK3-NEXT: [[TMP24:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l34.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP25:%.*]] = icmp ne i32 [[TMP24]], 0
+// CHECK3-NEXT: br i1 [[TMP25]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l34(i32 [[TMP2]], ptr [[A]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/teams_distribute_parallel_for_dist_schedule_codegen.cpp b/clang/test/OpenMP/teams_distribute_parallel_for_dist_schedule_codegen.cpp
index 83da0eb89304e..49667087c8673 100644
--- a/clang/test/OpenMP/teams_distribute_parallel_for_dist_schedule_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_parallel_for_dist_schedule_codegen.cpp
@@ -180,7 +180,7 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -197,87 +197,111 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.4, ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l33.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.4, ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l33.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK1-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK1-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK1: omp_offload.failed8:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l33(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK1: omp_offload.cont9:
// CHECK1-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A10]], ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A10]], ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.8, ptr [[TMP41]], align 8
-// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP42]], align 8
-// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP43]], align 8
-// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP44]], align 8
-// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK1-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l38.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK1-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK1-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 8
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 8
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.8, ptr [[TMP49]], align 8
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP50]], align 8
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP51]], align 8
+// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP52]], align 8
+// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK1-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l38.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK1-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK1-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK1: omp_offload.failed16:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l38(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK1: omp_offload.cont17:
// CHECK1-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A18]], i64 0, i64 0
-// CHECK1-NEXT: [[TMP48:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK1-NEXT: ret i32 [[TMP48]]
+// CHECK1-NEXT: [[TMP60:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK1-NEXT: ret i32 [[TMP60]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -797,7 +821,7 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -814,87 +838,111 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 4
-// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A2]], ptr [[TMP17]], align 4
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 4
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A2]], ptr [[TMP21]], align 4
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.4, ptr [[TMP25]], align 4
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP26]], align 4
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK3-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l33.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK3-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK3-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.4, ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK3-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l33.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK3-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK3-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK3: omp_offload.failed8:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l33(ptr [[THIS1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK3: omp_offload.cont9:
// CHECK3-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 4
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A10]], ptr [[TMP33]], align 4
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP34]], align 4
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A10]], ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP42]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 4
-// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 4
-// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.8, ptr [[TMP41]], align 4
-// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP42]], align 4
-// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP44]], align 4
-// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK3-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK3-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l38.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK3-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK3-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 4
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 4
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.8, ptr [[TMP49]], align 4
+// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP50]], align 4
+// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP52]], align 4
+// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK3-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK3-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l38.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK3-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK3-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK3: omp_offload.failed16:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l38(ptr [[THIS1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK3: omp_offload.cont17:
// CHECK3-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A18]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP48:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK3-NEXT: ret i32 [[TMP48]]
+// CHECK3-NEXT: [[TMP60:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK3-NEXT: ret i32 [[TMP60]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -1448,7 +1496,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1465,157 +1513,181 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK9-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l106.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK9-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l106.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK9-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l106(i64 [[TMP4]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK9-NEXT: [[TMP35:%.*]] = load i64, ptr [[N_CASTED3]], align 8
-// CHECK9-NEXT: [[TMP36:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK9-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK9-NEXT: [[TMP39:%.*]] = load i64, ptr [[N_CASTED3]], align 8
+// CHECK9-NEXT: [[TMP40:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK9-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES7]], ptr align 8 @.offload_sizes.4, i64 24, i1 false)
-// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP35]], ptr [[TMP37]], align 8
-// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP35]], ptr [[TMP38]], align 8
-// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP40]], align 8
-// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP41]], align 8
-// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP42]], align 8
-// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP43]], align 8
-// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 8
-// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK9-NEXT: store i64 [[TMP36]], ptr [[TMP45]], align 8
-// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP39]], ptr [[TMP41]], align 8
+// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP39]], ptr [[TMP42]], align 8
+// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP44]], align 8
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
// CHECK9-NEXT: store ptr null, ptr [[TMP46]], align 8
-// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP50:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP50]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK9-NEXT: [[TMP51:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK9-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP51]], 0
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP47]], align 8
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 8
+// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK9-NEXT: store i64 [[TMP40]], ptr [[TMP49]], align 8
+// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP50]], align 8
+// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP54:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP54]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK9-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK9-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP55]], 0
// CHECK9-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK9-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK9-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK9-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK9-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP52]], 1
-// CHECK9-NEXT: [[TMP53:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK9-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK9-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP56]], 1
+// CHECK9-NEXT: [[TMP57:%.*]] = zext i32 [[ADD14]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK9-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK9-NEXT: store i32 3, ptr [[TMP55]], align 4
-// CHECK9-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP47]], ptr [[TMP56]], align 8
-// CHECK9-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 8
-// CHECK9-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 8
-// CHECK9-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP59]], align 8
-// CHECK9-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP60]], align 8
-// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP61]], align 8
-// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP53]], ptr [[TMP62]], align 8
-// CHECK9-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l111.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK9-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
-// CHECK9-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK9-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP58]], align 4
+// CHECK9-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK9-NEXT: store i32 3, ptr [[TMP59]], align 4
+// CHECK9-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP51]], ptr [[TMP60]], align 8
+// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 8
+// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 8
+// CHECK9-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP63]], align 8
+// CHECK9-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP64]], align 8
+// CHECK9-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP65]], align 8
+// CHECK9-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP57]], ptr [[TMP66]], align 8
+// CHECK9-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP67]], align 8
+// CHECK9-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP68]], align 4
+// CHECK9-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK9-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP70]], align 4
+// CHECK9-NEXT: [[TMP71:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l111.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK9-NEXT: [[TMP72:%.*]] = icmp ne i32 [[TMP71]], 0
+// CHECK9-NEXT: br i1 [[TMP72]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK9: omp_offload.failed16:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l111(i64 [[TMP35]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l111(i64 [[TMP39]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK9: omp_offload.cont17:
-// CHECK9-NEXT: [[TMP65:%.*]] = load i32, ptr [[M]], align 4
-// CHECK9-NEXT: store i32 [[TMP65]], ptr [[M_CASTED]], align 4
-// CHECK9-NEXT: [[TMP66:%.*]] = load i64, ptr [[M_CASTED]], align 8
-// CHECK9-NEXT: [[TMP67:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP67]], ptr [[N_CASTED18]], align 4
-// CHECK9-NEXT: [[TMP68:%.*]] = load i64, ptr [[N_CASTED18]], align 8
-// CHECK9-NEXT: [[TMP69:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK9-NEXT: [[TMP73:%.*]] = load i32, ptr [[M]], align 4
+// CHECK9-NEXT: store i32 [[TMP73]], ptr [[M_CASTED]], align 4
+// CHECK9-NEXT: [[TMP74:%.*]] = load i64, ptr [[M_CASTED]], align 8
+// CHECK9-NEXT: [[TMP75:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP75]], ptr [[N_CASTED18]], align 4
+// CHECK9-NEXT: [[TMP76:%.*]] = load i64, ptr [[N_CASTED18]], align 8
+// CHECK9-NEXT: [[TMP77:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK9-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES22]], ptr align 8 @.offload_sizes.8, i64 32, i1 false)
-// CHECK9-NEXT: [[TMP70:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP66]], ptr [[TMP70]], align 8
-// CHECK9-NEXT: [[TMP71:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP66]], ptr [[TMP71]], align 8
-// CHECK9-NEXT: [[TMP72:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP72]], align 8
-// CHECK9-NEXT: [[TMP73:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP68]], ptr [[TMP73]], align 8
-// CHECK9-NEXT: [[TMP74:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP68]], ptr [[TMP74]], align 8
-// CHECK9-NEXT: [[TMP75:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP75]], align 8
-// CHECK9-NEXT: [[TMP76:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP76]], align 8
-// CHECK9-NEXT: [[TMP77:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP77]], align 8
-// CHECK9-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP78]], align 8
-// CHECK9-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP79]], align 8
-// CHECK9-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP80]], align 8
-// CHECK9-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 3
-// CHECK9-NEXT: store i64 [[TMP69]], ptr [[TMP81]], align 8
-// CHECK9-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP82]], align 8
-// CHECK9-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP86:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP86]], ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK9-NEXT: [[TMP87:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK9-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP87]], 0
+// CHECK9-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP74]], ptr [[TMP78]], align 8
+// CHECK9-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP74]], ptr [[TMP79]], align 8
+// CHECK9-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP80]], align 8
+// CHECK9-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP76]], ptr [[TMP81]], align 8
+// CHECK9-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP76]], ptr [[TMP82]], align 8
+// CHECK9-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP83]], align 8
+// CHECK9-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP84]], align 8
+// CHECK9-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP85]], align 8
+// CHECK9-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP86]], align 8
+// CHECK9-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP87]], align 8
+// CHECK9-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP88]], align 8
+// CHECK9-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 3
+// CHECK9-NEXT: store i64 [[TMP77]], ptr [[TMP89]], align 8
+// CHECK9-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP90]], align 8
+// CHECK9-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP94:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP94]], ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK9-NEXT: [[TMP95:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK9-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP95]], 0
// CHECK9-NEXT: [[DIV27:%.*]] = sdiv i32 [[SUB26]], 1
// CHECK9-NEXT: [[SUB28:%.*]] = sub nsw i32 [[DIV27]], 1
// CHECK9-NEXT: store i32 [[SUB28]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK9-NEXT: [[TMP88:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK9-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP88]], 1
-// CHECK9-NEXT: [[TMP89:%.*]] = zext i32 [[ADD29]] to i64
+// CHECK9-NEXT: [[TMP96:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK9-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP96]], 1
+// CHECK9-NEXT: [[TMP97:%.*]] = zext i32 [[ADD29]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS30:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP90]], align 4
-// CHECK9-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
-// CHECK9-NEXT: store i32 4, ptr [[TMP91]], align 4
-// CHECK9-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP83]], ptr [[TMP92]], align 8
-// CHECK9-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP84]], ptr [[TMP93]], align 8
-// CHECK9-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP85]], ptr [[TMP94]], align 8
-// CHECK9-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP95]], align 8
-// CHECK9-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP96]], align 8
-// CHECK9-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP97]], align 8
-// CHECK9-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP89]], ptr [[TMP98]], align 8
-// CHECK9-NEXT: [[TMP99:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l116.region_id, ptr [[KERNEL_ARGS30]])
-// CHECK9-NEXT: [[TMP100:%.*]] = icmp ne i32 [[TMP99]], 0
-// CHECK9-NEXT: br i1 [[TMP100]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
+// CHECK9-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP98]], align 4
+// CHECK9-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
+// CHECK9-NEXT: store i32 4, ptr [[TMP99]], align 4
+// CHECK9-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP91]], ptr [[TMP100]], align 8
+// CHECK9-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP92]], ptr [[TMP101]], align 8
+// CHECK9-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP93]], ptr [[TMP102]], align 8
+// CHECK9-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP103]], align 8
+// CHECK9-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP104]], align 8
+// CHECK9-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP105]], align 8
+// CHECK9-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP97]], ptr [[TMP106]], align 8
+// CHECK9-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP107]], align 8
+// CHECK9-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP108]], align 4
+// CHECK9-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP109]], align 4
+// CHECK9-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP110]], align 4
+// CHECK9-NEXT: [[TMP111:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l116.region_id, ptr [[KERNEL_ARGS30]])
+// CHECK9-NEXT: [[TMP112:%.*]] = icmp ne i32 [[TMP111]], 0
+// CHECK9-NEXT: br i1 [[TMP112]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
// CHECK9: omp_offload.failed31:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l116(i64 [[TMP66]], i64 [[TMP68]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l116(i64 [[TMP74]], i64 [[TMP76]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT32]]
// CHECK9: omp_offload.cont32:
-// CHECK9-NEXT: [[TMP101:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP101]])
+// CHECK9-NEXT: [[TMP113:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP113]])
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK9-NEXT: [[TMP102:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP102]])
-// CHECK9-NEXT: [[TMP103:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP103]]
+// CHECK9-NEXT: [[TMP114:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP114]])
+// CHECK9-NEXT: [[TMP115:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP115]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l106
@@ -2324,7 +2396,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK9-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK9-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2341,88 +2413,112 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK9-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK9-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK9-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84(ptr [[A]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP16]], align 8
-// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP17]], align 8
-// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP20]], align 8
+// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP21]], align 8
+// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK9-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.16, ptr [[TMP25]], align 8
-// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP26]], align 8
-// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK9-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK9-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK9-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK9-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK9-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.16, ptr [[TMP29]], align 8
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP30]], align 8
+// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK9-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK9-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK9-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK9-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK9: omp_offload.failed6:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89(ptr [[A]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK9: omp_offload.cont7:
-// CHECK9-NEXT: [[TMP32:%.*]] = load i32, ptr [[M]], align 4
-// CHECK9-NEXT: store i32 [[TMP32]], ptr [[M_CASTED]], align 4
-// CHECK9-NEXT: [[TMP33:%.*]] = load i64, ptr [[M_CASTED]], align 8
-// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP33]], ptr [[TMP34]], align 8
-// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP33]], ptr [[TMP35]], align 8
-// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP36]], align 8
-// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP37]], align 8
-// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP38]], align 8
-// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP40:%.*]] = load i32, ptr [[M]], align 4
+// CHECK9-NEXT: store i32 [[TMP40]], ptr [[M_CASTED]], align 4
+// CHECK9-NEXT: [[TMP41:%.*]] = load i64, ptr [[M_CASTED]], align 8
+// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP41]], ptr [[TMP42]], align 8
+// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP41]], ptr [[TMP43]], align 8
+// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP44]], align 8
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP46]], align 8
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP47]], align 8
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP42]], align 4
-// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK9-NEXT: store i32 2, ptr [[TMP43]], align 4
-// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP40]], ptr [[TMP44]], align 8
-// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP41]], ptr [[TMP45]], align 8
-// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.20, ptr [[TMP46]], align 8
-// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.21, ptr [[TMP47]], align 8
-// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP48]], align 8
-// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP49]], align 8
-// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK9-NEXT: store i64 10, ptr [[TMP50]], align 8
-// CHECK9-NEXT: [[TMP51:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l94.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK9-NEXT: [[TMP52:%.*]] = icmp ne i32 [[TMP51]], 0
-// CHECK9-NEXT: br i1 [[TMP52]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP50]], align 4
+// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK9-NEXT: store i32 2, ptr [[TMP51]], align 4
+// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP48]], ptr [[TMP52]], align 8
+// CHECK9-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 8
+// CHECK9-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.20, ptr [[TMP54]], align 8
+// CHECK9-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.21, ptr [[TMP55]], align 8
+// CHECK9-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP56]], align 8
+// CHECK9-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP57]], align 8
+// CHECK9-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK9-NEXT: store i64 10, ptr [[TMP58]], align 8
+// CHECK9-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP59]], align 8
+// CHECK9-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP60]], align 4
+// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4
+// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP62]], align 4
+// CHECK9-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l94.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK9-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
+// CHECK9-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK9: omp_offload.failed13:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l94(i64 [[TMP33]], ptr [[A]]) #[[ATTR3]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l94(i64 [[TMP41]], ptr [[A]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT14]]
// CHECK9: omp_offload.cont14:
// CHECK9-NEXT: ret i32 0
@@ -3009,7 +3105,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3026,159 +3122,183 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK11-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l106.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK11-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l106.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK11-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l106(i32 [[TMP3]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK11-NEXT: [[TMP35:%.*]] = load i32, ptr [[N_CASTED3]], align 4
-// CHECK11-NEXT: [[TMP36:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK11-NEXT: [[TMP37:%.*]] = sext i32 [[TMP36]] to i64
+// CHECK11-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK11-NEXT: [[TMP39:%.*]] = load i32, ptr [[N_CASTED3]], align 4
+// CHECK11-NEXT: [[TMP40:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK11-NEXT: [[TMP41:%.*]] = sext i32 [[TMP40]] to i64
// CHECK11-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES7]], ptr align 4 @.offload_sizes.4, i32 24, i1 false)
-// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP35]], ptr [[TMP38]], align 4
-// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP35]], ptr [[TMP39]], align 4
-// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP40]], align 4
-// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP41]], align 4
-// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP42]], align 4
-// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 4
-// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP45]], align 4
-// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK11-NEXT: store i64 [[TMP37]], ptr [[TMP46]], align 4
-// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP39]], ptr [[TMP42]], align 4
+// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP39]], ptr [[TMP43]], align 4
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP44]], align 4
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
// CHECK11-NEXT: store ptr null, ptr [[TMP47]], align 4
-// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP51:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP51]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK11-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK11-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP52]], 0
+// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 4
+// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP49]], align 4
+// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK11-NEXT: store i64 [[TMP41]], ptr [[TMP50]], align 4
+// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP54:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP55:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP55]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK11-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK11-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP56]], 0
// CHECK11-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK11-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK11-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK11-NEXT: [[TMP53:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK11-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP53]], 1
-// CHECK11-NEXT: [[TMP54:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK11-NEXT: [[TMP57:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK11-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP57]], 1
+// CHECK11-NEXT: [[TMP58:%.*]] = zext i32 [[ADD14]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP55]], align 4
-// CHECK11-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK11-NEXT: store i32 3, ptr [[TMP56]], align 4
-// CHECK11-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 4
-// CHECK11-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 4
-// CHECK11-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP50]], ptr [[TMP59]], align 4
-// CHECK11-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP60]], align 4
-// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP61]], align 4
-// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP62]], align 4
-// CHECK11-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP54]], ptr [[TMP63]], align 8
-// CHECK11-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l111.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK11-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
-// CHECK11-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK11-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP59]], align 4
+// CHECK11-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK11-NEXT: store i32 3, ptr [[TMP60]], align 4
+// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 4
+// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 4
+// CHECK11-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP54]], ptr [[TMP63]], align 4
+// CHECK11-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP64]], align 4
+// CHECK11-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP65]], align 4
+// CHECK11-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP66]], align 4
+// CHECK11-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP58]], ptr [[TMP67]], align 8
+// CHECK11-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP68]], align 8
+// CHECK11-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK11-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP70]], align 4
+// CHECK11-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP71]], align 4
+// CHECK11-NEXT: [[TMP72:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l111.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK11-NEXT: [[TMP73:%.*]] = icmp ne i32 [[TMP72]], 0
+// CHECK11-NEXT: br i1 [[TMP73]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK11: omp_offload.failed16:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l111(i32 [[TMP35]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l111(i32 [[TMP39]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK11: omp_offload.cont17:
-// CHECK11-NEXT: [[TMP66:%.*]] = load i32, ptr [[M]], align 4
-// CHECK11-NEXT: store i32 [[TMP66]], ptr [[M_CASTED]], align 4
-// CHECK11-NEXT: [[TMP67:%.*]] = load i32, ptr [[M_CASTED]], align 4
-// CHECK11-NEXT: [[TMP68:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP68]], ptr [[N_CASTED18]], align 4
-// CHECK11-NEXT: [[TMP69:%.*]] = load i32, ptr [[N_CASTED18]], align 4
-// CHECK11-NEXT: [[TMP70:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK11-NEXT: [[TMP71:%.*]] = sext i32 [[TMP70]] to i64
+// CHECK11-NEXT: [[TMP74:%.*]] = load i32, ptr [[M]], align 4
+// CHECK11-NEXT: store i32 [[TMP74]], ptr [[M_CASTED]], align 4
+// CHECK11-NEXT: [[TMP75:%.*]] = load i32, ptr [[M_CASTED]], align 4
+// CHECK11-NEXT: [[TMP76:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP76]], ptr [[N_CASTED18]], align 4
+// CHECK11-NEXT: [[TMP77:%.*]] = load i32, ptr [[N_CASTED18]], align 4
+// CHECK11-NEXT: [[TMP78:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK11-NEXT: [[TMP79:%.*]] = sext i32 [[TMP78]] to i64
// CHECK11-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES22]], ptr align 4 @.offload_sizes.8, i32 32, i1 false)
-// CHECK11-NEXT: [[TMP72:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP67]], ptr [[TMP72]], align 4
-// CHECK11-NEXT: [[TMP73:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP67]], ptr [[TMP73]], align 4
-// CHECK11-NEXT: [[TMP74:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP74]], align 4
-// CHECK11-NEXT: [[TMP75:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP69]], ptr [[TMP75]], align 4
-// CHECK11-NEXT: [[TMP76:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP69]], ptr [[TMP76]], align 4
-// CHECK11-NEXT: [[TMP77:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP77]], align 4
-// CHECK11-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP78]], align 4
-// CHECK11-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP79]], align 4
-// CHECK11-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP80]], align 4
-// CHECK11-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP81]], align 4
-// CHECK11-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP82]], align 4
-// CHECK11-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 3
-// CHECK11-NEXT: store i64 [[TMP71]], ptr [[TMP83]], align 4
-// CHECK11-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP84]], align 4
-// CHECK11-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP88:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP88]], ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK11-NEXT: [[TMP89:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK11-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP89]], 0
+// CHECK11-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP75]], ptr [[TMP80]], align 4
+// CHECK11-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP75]], ptr [[TMP81]], align 4
+// CHECK11-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP82]], align 4
+// CHECK11-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP77]], ptr [[TMP83]], align 4
+// CHECK11-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP77]], ptr [[TMP84]], align 4
+// CHECK11-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP85]], align 4
+// CHECK11-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP86]], align 4
+// CHECK11-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP87]], align 4
+// CHECK11-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP88]], align 4
+// CHECK11-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP89]], align 4
+// CHECK11-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP90]], align 4
+// CHECK11-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 3
+// CHECK11-NEXT: store i64 [[TMP79]], ptr [[TMP91]], align 4
+// CHECK11-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP92]], align 4
+// CHECK11-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP94:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP95:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP96:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP96]], ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK11-NEXT: [[TMP97:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK11-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP97]], 0
// CHECK11-NEXT: [[DIV27:%.*]] = sdiv i32 [[SUB26]], 1
// CHECK11-NEXT: [[SUB28:%.*]] = sub nsw i32 [[DIV27]], 1
// CHECK11-NEXT: store i32 [[SUB28]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK11-NEXT: [[TMP90:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK11-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP90]], 1
-// CHECK11-NEXT: [[TMP91:%.*]] = zext i32 [[ADD29]] to i64
+// CHECK11-NEXT: [[TMP98:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK11-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP98]], 1
+// CHECK11-NEXT: [[TMP99:%.*]] = zext i32 [[ADD29]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS30:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP92]], align 4
-// CHECK11-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
-// CHECK11-NEXT: store i32 4, ptr [[TMP93]], align 4
-// CHECK11-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP85]], ptr [[TMP94]], align 4
-// CHECK11-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP86]], ptr [[TMP95]], align 4
-// CHECK11-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP87]], ptr [[TMP96]], align 4
-// CHECK11-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP97]], align 4
-// CHECK11-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP98]], align 4
-// CHECK11-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP99]], align 4
-// CHECK11-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP91]], ptr [[TMP100]], align 8
-// CHECK11-NEXT: [[TMP101:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l116.region_id, ptr [[KERNEL_ARGS30]])
-// CHECK11-NEXT: [[TMP102:%.*]] = icmp ne i32 [[TMP101]], 0
-// CHECK11-NEXT: br i1 [[TMP102]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
+// CHECK11-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP100]], align 4
+// CHECK11-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
+// CHECK11-NEXT: store i32 4, ptr [[TMP101]], align 4
+// CHECK11-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP93]], ptr [[TMP102]], align 4
+// CHECK11-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP94]], ptr [[TMP103]], align 4
+// CHECK11-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP95]], ptr [[TMP104]], align 4
+// CHECK11-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP105]], align 4
+// CHECK11-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP106]], align 4
+// CHECK11-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP107]], align 4
+// CHECK11-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP99]], ptr [[TMP108]], align 8
+// CHECK11-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP109]], align 8
+// CHECK11-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP110]], align 4
+// CHECK11-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP111]], align 4
+// CHECK11-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP112]], align 4
+// CHECK11-NEXT: [[TMP113:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l116.region_id, ptr [[KERNEL_ARGS30]])
+// CHECK11-NEXT: [[TMP114:%.*]] = icmp ne i32 [[TMP113]], 0
+// CHECK11-NEXT: br i1 [[TMP114]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
// CHECK11: omp_offload.failed31:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l116(i32 [[TMP67]], i32 [[TMP69]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l116(i32 [[TMP75]], i32 [[TMP77]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT32]]
// CHECK11: omp_offload.cont32:
-// CHECK11-NEXT: [[TMP103:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP103]])
+// CHECK11-NEXT: [[TMP115:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP115]])
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK11-NEXT: [[TMP104:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP104]])
-// CHECK11-NEXT: [[TMP105:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP105]]
+// CHECK11-NEXT: [[TMP116:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP116]])
+// CHECK11-NEXT: [[TMP117:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP117]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l106
@@ -3872,7 +3992,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK11-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK11-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3889,88 +4009,112 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK11-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK11-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK11-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK11-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK11-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84(ptr [[A]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP16]], align 4
-// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP17]], align 4
-// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP20]], align 4
+// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP21]], align 4
+// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK11-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.16, ptr [[TMP25]], align 4
-// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP26]], align 4
-// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK11-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK11-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK11-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK11-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK11-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.16, ptr [[TMP29]], align 4
+// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP30]], align 4
+// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK11-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK11-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK11-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK11-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK11: omp_offload.failed6:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89(ptr [[A]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK11: omp_offload.cont7:
-// CHECK11-NEXT: [[TMP32:%.*]] = load i32, ptr [[M]], align 4
-// CHECK11-NEXT: store i32 [[TMP32]], ptr [[M_CASTED]], align 4
-// CHECK11-NEXT: [[TMP33:%.*]] = load i32, ptr [[M_CASTED]], align 4
-// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP33]], ptr [[TMP34]], align 4
-// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP33]], ptr [[TMP35]], align 4
-// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP36]], align 4
-// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP37]], align 4
-// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP38]], align 4
-// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP39]], align 4
-// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP40:%.*]] = load i32, ptr [[M]], align 4
+// CHECK11-NEXT: store i32 [[TMP40]], ptr [[M_CASTED]], align 4
+// CHECK11-NEXT: [[TMP41:%.*]] = load i32, ptr [[M_CASTED]], align 4
+// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP41]], ptr [[TMP42]], align 4
+// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP41]], ptr [[TMP43]], align 4
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP44]], align 4
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP47]], align 4
+// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP42]], align 4
-// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK11-NEXT: store i32 2, ptr [[TMP43]], align 4
-// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP40]], ptr [[TMP44]], align 4
-// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP41]], ptr [[TMP45]], align 4
-// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.20, ptr [[TMP46]], align 4
-// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.21, ptr [[TMP47]], align 4
-// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP48]], align 4
-// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP49]], align 4
-// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK11-NEXT: store i64 10, ptr [[TMP50]], align 8
-// CHECK11-NEXT: [[TMP51:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l94.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK11-NEXT: [[TMP52:%.*]] = icmp ne i32 [[TMP51]], 0
-// CHECK11-NEXT: br i1 [[TMP52]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP50]], align 4
+// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK11-NEXT: store i32 2, ptr [[TMP51]], align 4
+// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP48]], ptr [[TMP52]], align 4
+// CHECK11-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 4
+// CHECK11-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.20, ptr [[TMP54]], align 4
+// CHECK11-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.21, ptr [[TMP55]], align 4
+// CHECK11-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP56]], align 4
+// CHECK11-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP57]], align 4
+// CHECK11-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK11-NEXT: store i64 10, ptr [[TMP58]], align 8
+// CHECK11-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP59]], align 8
+// CHECK11-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP60]], align 4
+// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4
+// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP62]], align 4
+// CHECK11-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l94.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK11-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
+// CHECK11-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK11: omp_offload.failed13:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l94(i32 [[TMP33]], ptr [[A]]) #[[ATTR3]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l94(i32 [[TMP41]], ptr [[A]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT14]]
// CHECK11: omp_offload.cont14:
// CHECK11-NEXT: ret i32 0
diff --git a/clang/test/OpenMP/teams_distribute_parallel_for_firstprivate_codegen.cpp b/clang/test/OpenMP/teams_distribute_parallel_for_firstprivate_codegen.cpp
index d4090ec4809ea..107f17848761b 100644
--- a/clang/test/OpenMP/teams_distribute_parallel_for_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_parallel_for_firstprivate_codegen.cpp
@@ -340,7 +340,7 @@ int main() {
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 5, ptr [[TMP22]], align 4
// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -357,9 +357,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l99.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK1-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l99.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK1-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l99(i64 [[TMP1]], ptr @vec, ptr @s_arr, ptr @var, i64 [[TMP3]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -743,7 +751,7 @@ int main() {
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -760,27 +768,35 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK1-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK1-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK1-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56(i64 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK1-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK1: arraydestroy.body:
-// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK1-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK1-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK1: arraydestroy.done2:
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK1-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK1-NEXT: ret i32 [[TMP32]]
+// CHECK1-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK1-NEXT: ret i32 [[TMP36]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN2StC2Ev
@@ -1387,7 +1403,7 @@ int main() {
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 5, ptr [[TMP22]], align 4
// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1404,9 +1420,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4
// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP29]], align 8
-// CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l99.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK3-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK3-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l99.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK3-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l99(i32 [[TMP1]], ptr @vec, ptr @s_arr, ptr @var, i32 [[TMP3]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1784,7 +1808,7 @@ int main() {
// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1801,27 +1825,35 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP27]], align 4
// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK3-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK3-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK3-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56(i32 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK3-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK3-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK3: arraydestroy.body:
-// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK3-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK3-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK3: arraydestroy.done2:
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK3-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK3-NEXT: ret i32 [[TMP32]]
+// CHECK3-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK3-NEXT: ret i32 [[TMP36]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN2StC2Ev
diff --git a/clang/test/OpenMP/teams_distribute_parallel_for_if_codegen.cpp b/clang/test/OpenMP/teams_distribute_parallel_for_if_codegen.cpp
index af17f4759ddc0..718d4c5812622 100644
--- a/clang/test/OpenMP/teams_distribute_parallel_for_if_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_parallel_for_if_codegen.cpp
@@ -120,7 +120,7 @@ int main() {
// CHECK1-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -137,35 +137,51 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l48.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l48.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l48() #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l52.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l52.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK1-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l52() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -468,7 +484,7 @@ int main() {
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -485,85 +501,110 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l89.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l89.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK1-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l89() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK1: omp_offload.cont4:
-// CHECK1-NEXT: [[TMP22:%.*]] = load i32, ptr @Arg, align 4
-// CHECK1-NEXT: store i32 [[TMP22]], ptr [[ARG_CASTED]], align 4
-// CHECK1-NEXT: [[TMP23:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP23]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP23]], ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP29:%.*]] = load i32, ptr @Arg, align 4
-// CHECK1-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK1-NEXT: [[TMP30:%.*]] = load i32, ptr @Arg, align 4
+// CHECK1-NEXT: store i32 [[TMP30]], ptr [[ARG_CASTED]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP31]], ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP31]], ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP37:%.*]] = load i32, ptr @Arg, align 4
+// CHECK1-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP37]], 0
// CHECK1-NEXT: [[FROMBOOL:%.*]] = zext i1 [[TOBOOL]] to i8
// CHECK1-NEXT: store i8 [[FROMBOOL]], ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TMP30:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TOBOOL5:%.*]] = trunc i8 [[TMP30]] to i1
-// CHECK1-NEXT: [[TMP31:%.*]] = select i1 [[TOBOOL5]], i32 0, i32 1
+// CHECK1-NEXT: [[TMP38:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK1-NEXT: [[TOBOOL5:%.*]] = trunc i8 [[TMP38]] to i1
+// CHECK1-NEXT: [[TMP39:%.*]] = select i1 [[TOBOOL5]], i32 0, i32 1
+// CHECK1-NEXT: [[TMP40:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP39]], 0
// CHECK1-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP32]], align 4
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP33]], align 4
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP27]], ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP28]], ptr [[TMP35]], align 8
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP36]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP38]], align 8
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP31]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l97.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK1-NEXT: [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
-// CHECK1-NEXT: br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP41]], align 4
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP42]], align 4
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP35]], ptr [[TMP43]], align 8
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP36]], ptr [[TMP44]], align 8
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP45]], align 8
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP46]], align 8
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP47]], align 8
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP48]], align 8
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP49]], align 8
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP50]], align 8
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP51]], align 4
+// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP40]], ptr [[TMP52]], align 4
+// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP53]], align 4
+// CHECK1-NEXT: [[TMP54:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP39]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l97.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK1-NEXT: [[TMP55:%.*]] = icmp ne i32 [[TMP54]], 0
+// CHECK1-NEXT: br i1 [[TMP55]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK1: omp_offload.failed8:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l97(i64 [[TMP23]]) #[[ATTR2]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l97(i64 [[TMP31]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK1: omp_offload.cont9:
-// CHECK1-NEXT: [[TMP43:%.*]] = load i32, ptr @Arg, align 4
-// CHECK1-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP43]])
+// CHECK1-NEXT: [[TMP56:%.*]] = load i32, ptr @Arg, align 4
+// CHECK1-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP56]])
// CHECK1-NEXT: ret i32 [[CALL]]
//
//
@@ -1030,7 +1071,7 @@ int main() {
// CHECK1-NEXT: store i32 [[ARG]], ptr [[ARG_ADDR]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1047,81 +1088,106 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l62.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l62.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l62() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l67.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l67.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK1-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l67() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK1: omp_offload.cont4:
-// CHECK1-NEXT: [[TMP22:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
-// CHECK1-NEXT: store i32 [[TMP22]], ptr [[ARG_CASTED]], align 4
-// CHECK1-NEXT: [[TMP23:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP23]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP23]], ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP29:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
-// CHECK1-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK1-NEXT: [[TMP30:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
+// CHECK1-NEXT: store i32 [[TMP30]], ptr [[ARG_CASTED]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP31]], ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP31]], ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP37:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
+// CHECK1-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP37]], 0
// CHECK1-NEXT: [[FROMBOOL:%.*]] = zext i1 [[TOBOOL]] to i8
// CHECK1-NEXT: store i8 [[FROMBOOL]], ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TMP30:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TOBOOL5:%.*]] = trunc i8 [[TMP30]] to i1
-// CHECK1-NEXT: [[TMP31:%.*]] = select i1 [[TOBOOL5]], i32 0, i32 1
+// CHECK1-NEXT: [[TMP38:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK1-NEXT: [[TOBOOL5:%.*]] = trunc i8 [[TMP38]] to i1
+// CHECK1-NEXT: [[TMP39:%.*]] = select i1 [[TOBOOL5]], i32 0, i32 1
+// CHECK1-NEXT: [[TMP40:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP39]], 0
// CHECK1-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP32]], align 4
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP33]], align 4
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP27]], ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP28]], ptr [[TMP35]], align 8
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.16, ptr [[TMP36]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP38]], align 8
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP31]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l72.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK1-NEXT: [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
-// CHECK1-NEXT: br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP41]], align 4
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP42]], align 4
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP35]], ptr [[TMP43]], align 8
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP36]], ptr [[TMP44]], align 8
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.16, ptr [[TMP45]], align 8
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP46]], align 8
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP47]], align 8
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP48]], align 8
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP49]], align 8
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP50]], align 8
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP51]], align 4
+// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP40]], ptr [[TMP52]], align 4
+// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP53]], align 4
+// CHECK1-NEXT: [[TMP54:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP39]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l72.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK1-NEXT: [[TMP55:%.*]] = icmp ne i32 [[TMP54]], 0
+// CHECK1-NEXT: br i1 [[TMP55]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK1: omp_offload.failed8:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l72(i64 [[TMP23]]) #[[ATTR2]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l72(i64 [[TMP31]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK1: omp_offload.cont9:
// CHECK1-NEXT: ret i32 0
diff --git a/clang/test/OpenMP/teams_distribute_parallel_for_lastprivate_codegen.cpp b/clang/test/OpenMP/teams_distribute_parallel_for_lastprivate_codegen.cpp
index d269884fffb77..25d1bca100ed0 100644
--- a/clang/test/OpenMP/teams_distribute_parallel_for_lastprivate_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_parallel_for_lastprivate_codegen.cpp
@@ -780,7 +780,7 @@ int main() {
// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP25]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP25]], align 4
// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 5, ptr [[TMP26]], align 4
// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -797,9 +797,17 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP32]], align 8
// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP33]], align 8
-// CHECK9-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l106.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
-// CHECK9-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK9-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l106.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK9-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l106(i64 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]], i64 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -807,18 +815,18 @@ int main() {
// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiET_v()
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP36]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP40]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP37:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP37]]
+// CHECK9-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP41]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -1231,7 +1239,7 @@ int main() {
// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1248,27 +1256,35 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP27]], align 8
// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK9-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK9-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK9-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK9-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49(i64 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP32]]
+// CHECK9-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP36]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
@@ -1757,7 +1773,7 @@ int main() {
// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP25]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP25]], align 4
// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 5, ptr [[TMP26]], align 4
// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1774,9 +1790,17 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP32]], align 4
// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP33]], align 8
-// CHECK11-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l106.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
-// CHECK11-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK11-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l106.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK11-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l106(i32 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]], i32 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1784,18 +1808,18 @@ int main() {
// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiET_v()
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP36]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP40]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP37:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP37]]
+// CHECK11-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP41]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -2202,7 +2226,7 @@ int main() {
// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2219,27 +2243,35 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP27]], align 4
// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK11-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK11-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK11-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK11-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49(i32 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP32]]
+// CHECK11-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP36]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
diff --git a/clang/test/OpenMP/teams_distribute_parallel_for_num_threads_codegen.cpp b/clang/test/OpenMP/teams_distribute_parallel_for_num_threads_codegen.cpp
index fca65725473fb..a36a096858c56 100644
--- a/clang/test/OpenMP/teams_distribute_parallel_for_num_threads_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_parallel_for_num_threads_codegen.cpp
@@ -96,7 +96,7 @@ int main() {
// CHECK1-NEXT: store i8 [[CALL]], ptr [[A]], align 1
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -113,65 +113,82 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 2, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l50.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 2, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 2, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l50.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l50() #[[ATTR6:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: lpad:
-// CHECK1-NEXT: [[TMP11:%.*]] = landingpad { ptr, i32 }
+// CHECK1-NEXT: [[TMP15:%.*]] = landingpad { ptr, i32 }
// CHECK1-NEXT: cleanup
-// CHECK1-NEXT: [[TMP12:%.*]] = extractvalue { ptr, i32 } [[TMP11]], 0
-// CHECK1-NEXT: store ptr [[TMP12]], ptr [[EXN_SLOT]], align 8
-// CHECK1-NEXT: [[TMP13:%.*]] = extractvalue { ptr, i32 } [[TMP11]], 1
-// CHECK1-NEXT: store i32 [[TMP13]], ptr [[EHSELECTOR_SLOT]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = extractvalue { ptr, i32 } [[TMP15]], 0
+// CHECK1-NEXT: store ptr [[TMP16]], ptr [[EXN_SLOT]], align 8
+// CHECK1-NEXT: [[TMP17:%.*]] = extractvalue { ptr, i32 } [[TMP15]], 1
+// CHECK1-NEXT: store i32 [[TMP17]], ptr [[EHSELECTOR_SLOT]], align 4
// CHECK1-NEXT: call void @_ZN1SD1Ev(ptr noundef nonnull align 8 dereferenceable(24) [[S]]) #[[ATTR6]]
// CHECK1-NEXT: br label [[EH_RESUME:%.*]]
// CHECK1: omp_offload.cont:
-// CHECK1-NEXT: [[TMP14:%.*]] = load i8, ptr [[A]], align 1
-// CHECK1-NEXT: store i8 [[TMP14]], ptr [[A_CASTED]], align 1
-// CHECK1-NEXT: [[TMP15:%.*]] = load i64, ptr [[A_CASTED]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP15]], ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP15]], ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP21:%.*]] = load i8, ptr [[A]], align 1
-// CHECK1-NEXT: store i8 [[TMP21]], ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TMP22:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TMP23:%.*]] = zext i8 [[TMP22]] to i32
+// CHECK1-NEXT: [[TMP18:%.*]] = load i8, ptr [[A]], align 1
+// CHECK1-NEXT: store i8 [[TMP18]], ptr [[A_CASTED]], align 1
+// CHECK1-NEXT: [[TMP19:%.*]] = load i64, ptr [[A_CASTED]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP19]], ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP19]], ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP25:%.*]] = load i8, ptr [[A]], align 1
+// CHECK1-NEXT: store i8 [[TMP25]], ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK1-NEXT: [[TMP26:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK1-NEXT: [[TMP27:%.*]] = zext i8 [[TMP26]] to i32
+// CHECK1-NEXT: [[TMP28:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP27]], 0
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP24]], align 4
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP25]], align 4
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP19]], ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP20]], ptr [[TMP27]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP30]], align 8
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP31]], align 8
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP32]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP23]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l55.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
-// CHECK1-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP29]], align 4
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP30]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP23]], ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP24]], ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP35]], align 8
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP36]], align 8
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP37]], align 8
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP38]], align 8
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP39]], align 4
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP28]], ptr [[TMP40]], align 4
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP41]], align 4
+// CHECK1-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP27]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l55.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
+// CHECK1-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l55(i64 [[TMP15]]) #[[ATTR6]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l55(i64 [[TMP19]]) #[[ATTR6]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK1: omp_offload.cont4:
-// CHECK1-NEXT: [[TMP35:%.*]] = load i8, ptr [[A]], align 1
-// CHECK1-NEXT: [[CONV:%.*]] = sext i8 [[TMP35]] to i32
+// CHECK1-NEXT: [[TMP44:%.*]] = load i8, ptr [[A]], align 1
+// CHECK1-NEXT: [[CONV:%.*]] = sext i8 [[TMP44]] to i32
// CHECK1-NEXT: [[CALL6:%.*]] = invoke noundef signext i32 @_Z5tmainIcLi5EEiv()
// CHECK1-NEXT: to label [[INVOKE_CONT5:%.*]] unwind label [[LPAD]]
// CHECK1: invoke.cont5:
@@ -182,8 +199,8 @@ int main() {
// CHECK1-NEXT: [[ADD9:%.*]] = add nsw i32 [[ADD]], [[CALL8]]
// CHECK1-NEXT: store i32 [[ADD9]], ptr [[RETVAL]], align 4
// CHECK1-NEXT: call void @_ZN1SD1Ev(ptr noundef nonnull align 8 dereferenceable(24) [[S]]) #[[ATTR6]]
-// CHECK1-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK1-NEXT: ret i32 [[TMP36]]
+// CHECK1-NEXT: [[TMP45:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK1-NEXT: ret i32 [[TMP45]]
// CHECK1: eh.resume:
// CHECK1-NEXT: [[EXN:%.*]] = load ptr, ptr [[EXN_SLOT]], align 8
// CHECK1-NEXT: [[SEL:%.*]] = load i32, ptr [[EHSELECTOR_SLOT]], align 4
@@ -538,7 +555,7 @@ int main() {
// CHECK1-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -555,35 +572,51 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 5, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l36.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 5, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 5, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l36.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l36() #[[ATTR6]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 23, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l40.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 23, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 23, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l40.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK1-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l40() #[[ATTR6]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -600,7 +633,7 @@ int main() {
// CHECK1-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -617,9 +650,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l36.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l36.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l36() #[[ATTR6]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -630,40 +671,49 @@ int main() {
// CHECK1-NEXT: [[CALL:%.*]] = call noundef signext i8 @_ZN1ScvcEv(ptr noundef nonnull align 8 dereferenceable(24) [[REF_TMP]])
// CHECK1-NEXT: call void @_ZN1SD1Ev(ptr noundef nonnull align 8 dereferenceable(24) [[REF_TMP]]) #[[ATTR6]]
// CHECK1-NEXT: store i8 [[CALL]], ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TMP11:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TMP12:%.*]] = zext i8 [[TMP11]] to i32
+// CHECK1-NEXT: [[TMP15:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK1-NEXT: [[TMP16:%.*]] = zext i8 [[TMP15]] to i32
+// CHECK1-NEXT: [[TMP17:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP16]], 0
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP13]], align 4
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP14]], align 4
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP18]], align 4
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP21]], align 8
-// CHECK1-NEXT: [[TMP22:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP12]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l40.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0
-// CHECK1-NEXT: br i1 [[TMP23]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP25]], align 8
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP26]], align 8
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP17]], ptr [[TMP29]], align 4
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP16]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l40.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK1-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l40() #[[ATTR6]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK1: omp_offload.cont4:
// CHECK1-NEXT: ret i32 0
// CHECK1: terminate.lpad:
-// CHECK1-NEXT: [[TMP24:%.*]] = landingpad { ptr, i32 }
+// CHECK1-NEXT: [[TMP33:%.*]] = landingpad { ptr, i32 }
// CHECK1-NEXT: catch ptr null
-// CHECK1-NEXT: [[TMP25:%.*]] = extractvalue { ptr, i32 } [[TMP24]], 0
-// CHECK1-NEXT: call void @__clang_call_terminate(ptr [[TMP25]]) #[[ATTR9]]
+// CHECK1-NEXT: [[TMP34:%.*]] = extractvalue { ptr, i32 } [[TMP33]], 0
+// CHECK1-NEXT: call void @__clang_call_terminate(ptr [[TMP34]]) #[[ATTR9]]
// CHECK1-NEXT: unreachable
//
//
@@ -1340,7 +1390,7 @@ int main() {
// CHECK5-NEXT: store i8 [[CALL]], ptr [[A]], align 1
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK5-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK5-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1357,65 +1407,82 @@ int main() {
// CHECK5-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK5-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 2, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l50.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK5-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK5-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [i32 2, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK5-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK5-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 2, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l50.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK5-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l50() #[[ATTR6:[0-9]+]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK5: lpad:
-// CHECK5-NEXT: [[TMP11:%.*]] = landingpad { ptr, i32 }
+// CHECK5-NEXT: [[TMP15:%.*]] = landingpad { ptr, i32 }
// CHECK5-NEXT: cleanup
-// CHECK5-NEXT: [[TMP12:%.*]] = extractvalue { ptr, i32 } [[TMP11]], 0
-// CHECK5-NEXT: store ptr [[TMP12]], ptr [[EXN_SLOT]], align 8
-// CHECK5-NEXT: [[TMP13:%.*]] = extractvalue { ptr, i32 } [[TMP11]], 1
-// CHECK5-NEXT: store i32 [[TMP13]], ptr [[EHSELECTOR_SLOT]], align 4
+// CHECK5-NEXT: [[TMP16:%.*]] = extractvalue { ptr, i32 } [[TMP15]], 0
+// CHECK5-NEXT: store ptr [[TMP16]], ptr [[EXN_SLOT]], align 8
+// CHECK5-NEXT: [[TMP17:%.*]] = extractvalue { ptr, i32 } [[TMP15]], 1
+// CHECK5-NEXT: store i32 [[TMP17]], ptr [[EHSELECTOR_SLOT]], align 4
// CHECK5-NEXT: call void @_ZN1SD1Ev(ptr noundef nonnull align 8 dereferenceable(24) [[S]]) #[[ATTR6]]
// CHECK5-NEXT: br label [[EH_RESUME:%.*]]
// CHECK5: omp_offload.cont:
-// CHECK5-NEXT: [[TMP14:%.*]] = load i8, ptr [[A]], align 1
-// CHECK5-NEXT: store i8 [[TMP14]], ptr [[A_CASTED]], align 1
-// CHECK5-NEXT: [[TMP15:%.*]] = load i64, ptr [[A_CASTED]], align 8
-// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK5-NEXT: store i64 [[TMP15]], ptr [[TMP16]], align 8
-// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK5-NEXT: store i64 [[TMP15]], ptr [[TMP17]], align 8
-// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK5-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP21:%.*]] = load i8, ptr [[A]], align 1
-// CHECK5-NEXT: store i8 [[TMP21]], ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK5-NEXT: [[TMP22:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK5-NEXT: [[TMP23:%.*]] = zext i8 [[TMP22]] to i32
+// CHECK5-NEXT: [[TMP18:%.*]] = load i8, ptr [[A]], align 1
+// CHECK5-NEXT: store i8 [[TMP18]], ptr [[A_CASTED]], align 1
+// CHECK5-NEXT: [[TMP19:%.*]] = load i64, ptr [[A_CASTED]], align 8
+// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK5-NEXT: store i64 [[TMP19]], ptr [[TMP20]], align 8
+// CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK5-NEXT: store i64 [[TMP19]], ptr [[TMP21]], align 8
+// CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK5-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP25:%.*]] = load i8, ptr [[A]], align 1
+// CHECK5-NEXT: store i8 [[TMP25]], ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK5-NEXT: [[TMP26:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK5-NEXT: [[TMP27:%.*]] = zext i8 [[TMP26]] to i32
+// CHECK5-NEXT: [[TMP28:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP27]], 0
// CHECK5-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP24]], align 4
-// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK5-NEXT: store i32 1, ptr [[TMP25]], align 4
-// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK5-NEXT: store ptr [[TMP19]], ptr [[TMP26]], align 8
-// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[TMP20]], ptr [[TMP27]], align 8
-// CHECK5-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK5-NEXT: store ptr @.offload_sizes, ptr [[TMP28]], align 8
-// CHECK5-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK5-NEXT: store ptr @.offload_maptypes, ptr [[TMP29]], align 8
-// CHECK5-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP30]], align 8
-// CHECK5-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
-// CHECK5-NEXT: store ptr null, ptr [[TMP31]], align 8
-// CHECK5-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK5-NEXT: store i64 100, ptr [[TMP32]], align 8
-// CHECK5-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP23]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l55.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK5-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
-// CHECK5-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK5-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP29]], align 4
+// CHECK5-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK5-NEXT: store i32 1, ptr [[TMP30]], align 4
+// CHECK5-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
+// CHECK5-NEXT: store ptr [[TMP23]], ptr [[TMP31]], align 8
+// CHECK5-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[TMP24]], ptr [[TMP32]], align 8
+// CHECK5-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK5-NEXT: store ptr @.offload_sizes, ptr [[TMP33]], align 8
+// CHECK5-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK5-NEXT: store ptr @.offload_maptypes, ptr [[TMP34]], align 8
+// CHECK5-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP35]], align 8
+// CHECK5-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP36]], align 8
+// CHECK5-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK5-NEXT: store i64 100, ptr [[TMP37]], align 8
+// CHECK5-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP38]], align 8
+// CHECK5-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP39]], align 4
+// CHECK5-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [[TMP28]], ptr [[TMP40]], align 4
+// CHECK5-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP41]], align 4
+// CHECK5-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP27]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l55.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK5-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
+// CHECK5-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK5: omp_offload.failed3:
-// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l55(i64 [[TMP15]]) #[[ATTR6]]
+// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l55(i64 [[TMP19]]) #[[ATTR6]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK5: omp_offload.cont4:
-// CHECK5-NEXT: [[TMP35:%.*]] = load i8, ptr [[A]], align 1
-// CHECK5-NEXT: [[CONV:%.*]] = sext i8 [[TMP35]] to i32
+// CHECK5-NEXT: [[TMP44:%.*]] = load i8, ptr [[A]], align 1
+// CHECK5-NEXT: [[CONV:%.*]] = sext i8 [[TMP44]] to i32
// CHECK5-NEXT: [[CALL6:%.*]] = invoke noundef signext i32 @_Z5tmainIcLi5EEiv()
// CHECK5-NEXT: to label [[INVOKE_CONT5:%.*]] unwind label [[LPAD]]
// CHECK5: invoke.cont5:
@@ -1426,8 +1493,8 @@ int main() {
// CHECK5-NEXT: [[ADD9:%.*]] = add nsw i32 [[ADD]], [[CALL8]]
// CHECK5-NEXT: store i32 [[ADD9]], ptr [[RETVAL]], align 4
// CHECK5-NEXT: call void @_ZN1SD1Ev(ptr noundef nonnull align 8 dereferenceable(24) [[S]]) #[[ATTR6]]
-// CHECK5-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK5-NEXT: ret i32 [[TMP36]]
+// CHECK5-NEXT: [[TMP45:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK5-NEXT: ret i32 [[TMP45]]
// CHECK5: eh.resume:
// CHECK5-NEXT: [[EXN:%.*]] = load ptr, ptr [[EXN_SLOT]], align 8
// CHECK5-NEXT: [[SEL:%.*]] = load i32, ptr [[EHSELECTOR_SLOT]], align 4
@@ -1782,7 +1849,7 @@ int main() {
// CHECK5-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK5-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK5-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1799,35 +1866,51 @@ int main() {
// CHECK5-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK5-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 5, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l36.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK5-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK5-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [i32 5, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK5-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK5-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 5, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l36.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK5-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l36() #[[ATTR6]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK5: omp_offload.cont:
// CHECK5-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK5-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK5-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK5-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK5-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK5-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK5-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK5-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK5-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK5-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK5-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK5-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK5-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK5-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK5-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK5-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 23, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l40.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK5-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK5-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK5-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK5-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK5-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [i32 23, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK5-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 23, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l40.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK5-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK5-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK5: omp_offload.failed3:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l40() #[[ATTR6]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -1844,7 +1927,7 @@ int main() {
// CHECK5-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK5-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK5-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1861,9 +1944,17 @@ int main() {
// CHECK5-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK5-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l36.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK5-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK5-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK5-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK5-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l36.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK5-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l36() #[[ATTR6]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1874,40 +1965,49 @@ int main() {
// CHECK5-NEXT: [[CALL:%.*]] = call noundef signext i8 @_ZN1ScvcEv(ptr noundef nonnull align 8 dereferenceable(24) [[REF_TMP]])
// CHECK5-NEXT: call void @_ZN1SD1Ev(ptr noundef nonnull align 8 dereferenceable(24) [[REF_TMP]]) #[[ATTR6]]
// CHECK5-NEXT: store i8 [[CALL]], ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK5-NEXT: [[TMP11:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK5-NEXT: [[TMP12:%.*]] = zext i8 [[TMP11]] to i32
+// CHECK5-NEXT: [[TMP15:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK5-NEXT: [[TMP16:%.*]] = zext i8 [[TMP15]] to i32
+// CHECK5-NEXT: [[TMP17:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP16]], 0
// CHECK5-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP13]], align 4
-// CHECK5-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK5-NEXT: store i32 0, ptr [[TMP14]], align 4
-// CHECK5-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK5-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK5-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK5-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK5-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP19]], align 8
-// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP18]], align 4
+// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK5-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK5-NEXT: store ptr null, ptr [[TMP20]], align 8
-// CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK5-NEXT: store i64 100, ptr [[TMP21]], align 8
-// CHECK5-NEXT: [[TMP22:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP12]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l40.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK5-NEXT: [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0
-// CHECK5-NEXT: br i1 [[TMP23]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
+// CHECK5-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK5-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK5-NEXT: store ptr null, ptr [[TMP23]], align 8
+// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP24]], align 8
+// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP25]], align 8
+// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK5-NEXT: store i64 100, ptr [[TMP26]], align 8
+// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK5-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK5-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [[TMP17]], ptr [[TMP29]], align 4
+// CHECK5-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK5-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP16]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l40.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK5-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK5-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK5: omp_offload.failed3:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l40() #[[ATTR6]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK5: omp_offload.cont4:
// CHECK5-NEXT: ret i32 0
// CHECK5: terminate.lpad:
-// CHECK5-NEXT: [[TMP24:%.*]] = landingpad { ptr, i32 }
+// CHECK5-NEXT: [[TMP33:%.*]] = landingpad { ptr, i32 }
// CHECK5-NEXT: catch ptr null
-// CHECK5-NEXT: [[TMP25:%.*]] = extractvalue { ptr, i32 } [[TMP24]], 0
-// CHECK5-NEXT: call void @__clang_call_terminate(ptr [[TMP25]]) #[[ATTR9]]
+// CHECK5-NEXT: [[TMP34:%.*]] = extractvalue { ptr, i32 } [[TMP33]], 0
+// CHECK5-NEXT: call void @__clang_call_terminate(ptr [[TMP34]]) #[[ATTR9]]
// CHECK5-NEXT: unreachable
//
//
diff --git a/clang/test/OpenMP/teams_distribute_parallel_for_private_codegen.cpp b/clang/test/OpenMP/teams_distribute_parallel_for_private_codegen.cpp
index 27de9caf3bf4f..c3ba6bc777d8f 100644
--- a/clang/test/OpenMP/teams_distribute_parallel_for_private_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_parallel_for_private_codegen.cpp
@@ -263,7 +263,7 @@ int main() {
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -280,9 +280,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l96.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l96.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l96() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -525,7 +533,7 @@ int main() {
// CHECK1-NEXT: store ptr undef, ptr [[_TMP1]], align 8
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -542,27 +550,35 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK1-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK1: arraydestroy.body:
-// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK1-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK1-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK1: arraydestroy.done2:
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK1-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK1-NEXT: ret i32 [[TMP12]]
+// CHECK1-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK1-NEXT: ret i32 [[TMP16]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN1SIiEC1Ev
@@ -987,7 +1003,7 @@ int main() {
// CHECK3-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1004,9 +1020,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l96.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK3-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l96.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK3-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l96() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1243,7 +1267,7 @@ int main() {
// CHECK3-NEXT: store ptr undef, ptr [[_TMP1]], align 4
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1260,27 +1284,35 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK3-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK3-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK3-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK3-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK3: arraydestroy.body:
-// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK3-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK3-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK3: arraydestroy.done2:
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK3-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK3-NEXT: ret i32 [[TMP12]]
+// CHECK3-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK3-NEXT: ret i32 [[TMP16]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN1SIiEC1Ev
diff --git a/clang/test/OpenMP/teams_distribute_parallel_for_proc_bind_codegen.cpp b/clang/test/OpenMP/teams_distribute_parallel_for_proc_bind_codegen.cpp
index f20158889760a..141adc5c8c9de 100644
--- a/clang/test/OpenMP/teams_distribute_parallel_for_proc_bind_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_parallel_for_proc_bind_codegen.cpp
@@ -59,7 +59,7 @@ int main() {
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -76,35 +76,51 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 1000, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l36.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l36.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l36() #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 1000, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l39.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 1000, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l39.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK1-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l39() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -395,7 +411,7 @@ int main() {
// CHECK1-NEXT: [[TMP:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -412,9 +428,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 1000, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l29.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l29.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l29() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/teams_distribute_parallel_for_reduction_codegen.cpp b/clang/test/OpenMP/teams_distribute_parallel_for_reduction_codegen.cpp
index f6e5898f06ed7..de787727f3fa9 100644
--- a/clang/test/OpenMP/teams_distribute_parallel_for_reduction_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_parallel_for_reduction_codegen.cpp
@@ -116,7 +116,7 @@ int main() {
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -133,9 +133,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK1-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68(i64 [[TMP1]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -402,7 +410,7 @@ int main() {
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -419,9 +427,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK1-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32(i64 [[TMP1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -692,7 +708,7 @@ int main() {
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -709,9 +725,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP14]], align 4
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP15]], align 8
-// CHECK3-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK3-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK3-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK3-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l68(i32 [[TMP1]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -974,7 +998,7 @@ int main() {
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -991,9 +1015,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP14]], align 4
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP15]], align 8
-// CHECK3-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK3-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK3-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK3-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32(i32 [[TMP1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/teams_distribute_parallel_for_schedule_codegen.cpp b/clang/test/OpenMP/teams_distribute_parallel_for_schedule_codegen.cpp
index c2ce2766101db..0b8915954cff8 100644
--- a/clang/test/OpenMP/teams_distribute_parallel_for_schedule_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_parallel_for_schedule_codegen.cpp
@@ -253,7 +253,7 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -270,157 +270,197 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l35.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l35.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l35(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.4, ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.4, ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK1-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK1-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK1: omp_offload.failed8:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK1: omp_offload.cont9:
// CHECK1-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A10]], ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A10]], ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.8, ptr [[TMP41]], align 8
-// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP42]], align 8
-// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP43]], align 8
-// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP44]], align 8
-// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK1-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l45.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK1-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK1-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 8
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 8
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.8, ptr [[TMP49]], align 8
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP50]], align 8
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP51]], align 8
+// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP52]], align 8
+// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK1-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l45.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK1-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK1-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK1: omp_offload.failed16:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l45(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK1: omp_offload.cont17:
// CHECK1-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP48]], align 8
-// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A18]], ptr [[TMP49]], align 8
-// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP50]], align 8
-// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP60:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP60]], align 8
+// CHECK1-NEXT: [[TMP61:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A18]], ptr [[TMP61]], align 8
+// CHECK1-NEXT: [[TMP62:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP62]], align 8
+// CHECK1-NEXT: [[TMP63:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS23:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP53]], align 4
-// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP51]], ptr [[TMP55]], align 8
-// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP52]], ptr [[TMP56]], align 8
-// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.12, ptr [[TMP57]], align 8
-// CHECK1-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP58]], align 8
-// CHECK1-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP59]], align 8
-// CHECK1-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP60]], align 8
-// CHECK1-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP61]], align 8
-// CHECK1-NEXT: [[TMP62:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l51.region_id, ptr [[KERNEL_ARGS23]])
-// CHECK1-NEXT: [[TMP63:%.*]] = icmp ne i32 [[TMP62]], 0
-// CHECK1-NEXT: br i1 [[TMP63]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
+// CHECK1-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP65]], align 4
+// CHECK1-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP66]], align 4
+// CHECK1-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP63]], ptr [[TMP67]], align 8
+// CHECK1-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP64]], ptr [[TMP68]], align 8
+// CHECK1-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.12, ptr [[TMP69]], align 8
+// CHECK1-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP70]], align 8
+// CHECK1-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP71]], align 8
+// CHECK1-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP72]], align 8
+// CHECK1-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP73]], align 8
+// CHECK1-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP74]], align 8
+// CHECK1-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP75]], align 4
+// CHECK1-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP76]], align 4
+// CHECK1-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP77]], align 4
+// CHECK1-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l51.region_id, ptr [[KERNEL_ARGS23]])
+// CHECK1-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
+// CHECK1-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
// CHECK1: omp_offload.failed24:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l51(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT25]]
// CHECK1: omp_offload.cont25:
// CHECK1-NEXT: [[A26:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP64]], align 8
-// CHECK1-NEXT: [[TMP65:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A26]], ptr [[TMP65]], align 8
-// CHECK1-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP66]], align 8
-// CHECK1-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP80:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP80]], align 8
+// CHECK1-NEXT: [[TMP81:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A26]], ptr [[TMP81]], align 8
+// CHECK1-NEXT: [[TMP82:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP82]], align 8
+// CHECK1-NEXT: [[TMP83:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP84:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP69]], align 4
-// CHECK1-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP70]], align 4
-// CHECK1-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP67]], ptr [[TMP71]], align 8
-// CHECK1-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 8
-// CHECK1-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.16, ptr [[TMP73]], align 8
-// CHECK1-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP74]], align 8
-// CHECK1-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP75]], align 8
-// CHECK1-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP76]], align 8
-// CHECK1-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP77]], align 8
-// CHECK1-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l57.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK1-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
-// CHECK1-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK1-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP85]], align 4
+// CHECK1-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP86]], align 4
+// CHECK1-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP83]], ptr [[TMP87]], align 8
+// CHECK1-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP84]], ptr [[TMP88]], align 8
+// CHECK1-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.16, ptr [[TMP89]], align 8
+// CHECK1-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP90]], align 8
+// CHECK1-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP91]], align 8
+// CHECK1-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP92]], align 8
+// CHECK1-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP93]], align 8
+// CHECK1-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP94]], align 8
+// CHECK1-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP95]], align 4
+// CHECK1-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP96]], align 4
+// CHECK1-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP97]], align 4
+// CHECK1-NEXT: [[TMP98:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l57.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK1-NEXT: [[TMP99:%.*]] = icmp ne i32 [[TMP98]], 0
+// CHECK1-NEXT: br i1 [[TMP99]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK1: omp_offload.failed32:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l57(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK1: omp_offload.cont33:
// CHECK1-NEXT: [[A34:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A34]], i64 0, i64 0
-// CHECK1-NEXT: [[TMP80:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK1-NEXT: ret i32 [[TMP80]]
+// CHECK1-NEXT: [[TMP100:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK1-NEXT: ret i32 [[TMP100]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l35
@@ -1247,7 +1287,7 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1264,157 +1304,197 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l35.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l35.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l35(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 4
-// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A2]], ptr [[TMP17]], align 4
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 4
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A2]], ptr [[TMP21]], align 4
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.4, ptr [[TMP25]], align 4
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP26]], align 4
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK3-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK3-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK3-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.4, ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK3-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK3-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK3-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK3: omp_offload.failed8:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40(ptr [[THIS1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK3: omp_offload.cont9:
// CHECK3-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 4
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A10]], ptr [[TMP33]], align 4
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP34]], align 4
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A10]], ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP42]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 4
-// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 4
-// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.8, ptr [[TMP41]], align 4
-// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP42]], align 4
-// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP44]], align 4
-// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK3-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK3-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l45.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK3-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK3-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 4
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 4
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.8, ptr [[TMP49]], align 4
+// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP50]], align 4
+// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP52]], align 4
+// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK3-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK3-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l45.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK3-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK3-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK3: omp_offload.failed16:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l45(ptr [[THIS1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK3: omp_offload.cont17:
// CHECK3-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP48]], align 4
-// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A18]], ptr [[TMP49]], align 4
-// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP50]], align 4
-// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP60:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP60]], align 4
+// CHECK3-NEXT: [[TMP61:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A18]], ptr [[TMP61]], align 4
+// CHECK3-NEXT: [[TMP62:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP62]], align 4
+// CHECK3-NEXT: [[TMP63:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS23:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP53]], align 4
-// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP51]], ptr [[TMP55]], align 4
-// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP52]], ptr [[TMP56]], align 4
-// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.12, ptr [[TMP57]], align 4
-// CHECK3-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP58]], align 4
-// CHECK3-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP59]], align 4
-// CHECK3-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP60]], align 4
-// CHECK3-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
-// CHECK3-NEXT: store i64 123, ptr [[TMP61]], align 8
-// CHECK3-NEXT: [[TMP62:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l51.region_id, ptr [[KERNEL_ARGS23]])
-// CHECK3-NEXT: [[TMP63:%.*]] = icmp ne i32 [[TMP62]], 0
-// CHECK3-NEXT: br i1 [[TMP63]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
+// CHECK3-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP65]], align 4
+// CHECK3-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP66]], align 4
+// CHECK3-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP63]], ptr [[TMP67]], align 4
+// CHECK3-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP64]], ptr [[TMP68]], align 4
+// CHECK3-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.12, ptr [[TMP69]], align 4
+// CHECK3-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP70]], align 4
+// CHECK3-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP71]], align 4
+// CHECK3-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP72]], align 4
+// CHECK3-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
+// CHECK3-NEXT: store i64 123, ptr [[TMP73]], align 8
+// CHECK3-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP74]], align 8
+// CHECK3-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP75]], align 4
+// CHECK3-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP76]], align 4
+// CHECK3-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP77]], align 4
+// CHECK3-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l51.region_id, ptr [[KERNEL_ARGS23]])
+// CHECK3-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
+// CHECK3-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
// CHECK3: omp_offload.failed24:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l51(ptr [[THIS1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT25]]
// CHECK3: omp_offload.cont25:
// CHECK3-NEXT: [[A26:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP64]], align 4
-// CHECK3-NEXT: [[TMP65:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A26]], ptr [[TMP65]], align 4
-// CHECK3-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP66]], align 4
-// CHECK3-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP80:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP80]], align 4
+// CHECK3-NEXT: [[TMP81:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A26]], ptr [[TMP81]], align 4
+// CHECK3-NEXT: [[TMP82:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP82]], align 4
+// CHECK3-NEXT: [[TMP83:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP84:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP69]], align 4
-// CHECK3-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP70]], align 4
-// CHECK3-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP67]], ptr [[TMP71]], align 4
-// CHECK3-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 4
-// CHECK3-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.16, ptr [[TMP73]], align 4
-// CHECK3-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP74]], align 4
-// CHECK3-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP75]], align 4
-// CHECK3-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP76]], align 4
-// CHECK3-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK3-NEXT: store i64 123, ptr [[TMP77]], align 8
-// CHECK3-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l57.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK3-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
-// CHECK3-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK3-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP85]], align 4
+// CHECK3-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP86]], align 4
+// CHECK3-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP83]], ptr [[TMP87]], align 4
+// CHECK3-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP84]], ptr [[TMP88]], align 4
+// CHECK3-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.16, ptr [[TMP89]], align 4
+// CHECK3-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP90]], align 4
+// CHECK3-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP91]], align 4
+// CHECK3-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP92]], align 4
+// CHECK3-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK3-NEXT: store i64 123, ptr [[TMP93]], align 8
+// CHECK3-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP94]], align 8
+// CHECK3-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP95]], align 4
+// CHECK3-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP96]], align 4
+// CHECK3-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP97]], align 4
+// CHECK3-NEXT: [[TMP98:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l57.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK3-NEXT: [[TMP99:%.*]] = icmp ne i32 [[TMP98]], 0
+// CHECK3-NEXT: br i1 [[TMP99]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK3: omp_offload.failed32:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l57(ptr [[THIS1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK3: omp_offload.cont33:
// CHECK3-NEXT: [[A34:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A34]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP80:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK3-NEXT: ret i32 [[TMP80]]
+// CHECK3-NEXT: [[TMP100:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK3-NEXT: ret i32 [[TMP100]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l35
@@ -2214,7 +2294,7 @@ int main (int argc, char **argv) {
// CHECK5-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK5-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK5-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2231,157 +2311,197 @@ int main (int argc, char **argv) {
// CHECK5-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK5-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK5-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l35.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK5-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK5-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK5-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l35.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK5-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l35(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK5: omp_offload.cont:
// CHECK5-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 8
-// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[A2]], ptr [[TMP17]], align 8
-// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
-// CHECK5-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 8
+// CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[A2]], ptr [[TMP21]], align 8
+// CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
+// CHECK5-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK5-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK5-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK5-NEXT: store ptr @.offload_sizes.4, ptr [[TMP25]], align 8
-// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK5-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP26]], align 8
-// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK5-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK5-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK5-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK5-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK5-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK5-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK5-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK5-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK5-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK5-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK5-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK5-NEXT: store ptr @.offload_sizes.4, ptr [[TMP29]], align 8
+// CHECK5-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK5-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP30]], align 8
+// CHECK5-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK5-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK5-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK5-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK5-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK5-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK5-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK5-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK5-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK5-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK5-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK5: omp_offload.failed8:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40(ptr [[THIS1]]) #[[ATTR2]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK5: omp_offload.cont9:
// CHECK5-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 8
-// CHECK5-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[A10]], ptr [[TMP33]], align 8
-// CHECK5-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
-// CHECK5-NEXT: store ptr null, ptr [[TMP34]], align 8
-// CHECK5-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 8
+// CHECK5-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[A10]], ptr [[TMP41]], align 8
+// CHECK5-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
+// CHECK5-NEXT: store ptr null, ptr [[TMP42]], align 8
+// CHECK5-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK5-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK5-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK5-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK5-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
-// CHECK5-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
-// CHECK5-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK5-NEXT: store ptr @.offload_sizes.8, ptr [[TMP41]], align 8
-// CHECK5-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK5-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP42]], align 8
-// CHECK5-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP43]], align 8
-// CHECK5-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK5-NEXT: store ptr null, ptr [[TMP44]], align 8
-// CHECK5-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK5-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK5-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l45.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK5-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK5-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK5-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK5-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK5-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK5-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK5-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 8
+// CHECK5-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 8
+// CHECK5-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK5-NEXT: store ptr @.offload_sizes.8, ptr [[TMP49]], align 8
+// CHECK5-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK5-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP50]], align 8
+// CHECK5-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP51]], align 8
+// CHECK5-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP52]], align 8
+// CHECK5-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK5-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK5-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK5-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK5-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK5-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK5-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l45.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK5-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK5-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK5: omp_offload.failed16:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l45(ptr [[THIS1]]) #[[ATTR2]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK5: omp_offload.cont17:
// CHECK5-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP48:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP48]], align 8
-// CHECK5-NEXT: [[TMP49:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[A18]], ptr [[TMP49]], align 8
-// CHECK5-NEXT: [[TMP50:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
-// CHECK5-NEXT: store ptr null, ptr [[TMP50]], align 8
-// CHECK5-NEXT: [[TMP51:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP52:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP60:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP60]], align 8
+// CHECK5-NEXT: [[TMP61:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[A18]], ptr [[TMP61]], align 8
+// CHECK5-NEXT: [[TMP62:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
+// CHECK5-NEXT: store ptr null, ptr [[TMP62]], align 8
+// CHECK5-NEXT: [[TMP63:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS23:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP53]], align 4
-// CHECK5-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
-// CHECK5-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK5-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
-// CHECK5-NEXT: store ptr [[TMP51]], ptr [[TMP55]], align 8
-// CHECK5-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[TMP52]], ptr [[TMP56]], align 8
-// CHECK5-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
-// CHECK5-NEXT: store ptr @.offload_sizes.12, ptr [[TMP57]], align 8
-// CHECK5-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
-// CHECK5-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP58]], align 8
-// CHECK5-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP59]], align 8
-// CHECK5-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
-// CHECK5-NEXT: store ptr null, ptr [[TMP60]], align 8
-// CHECK5-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
-// CHECK5-NEXT: store i64 123, ptr [[TMP61]], align 8
-// CHECK5-NEXT: [[TMP62:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l51.region_id, ptr [[KERNEL_ARGS23]])
-// CHECK5-NEXT: [[TMP63:%.*]] = icmp ne i32 [[TMP62]], 0
-// CHECK5-NEXT: br i1 [[TMP63]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
+// CHECK5-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP65]], align 4
+// CHECK5-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
+// CHECK5-NEXT: store i32 1, ptr [[TMP66]], align 4
+// CHECK5-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
+// CHECK5-NEXT: store ptr [[TMP63]], ptr [[TMP67]], align 8
+// CHECK5-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[TMP64]], ptr [[TMP68]], align 8
+// CHECK5-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
+// CHECK5-NEXT: store ptr @.offload_sizes.12, ptr [[TMP69]], align 8
+// CHECK5-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
+// CHECK5-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP70]], align 8
+// CHECK5-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP71]], align 8
+// CHECK5-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP72]], align 8
+// CHECK5-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
+// CHECK5-NEXT: store i64 123, ptr [[TMP73]], align 8
+// CHECK5-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP74]], align 8
+// CHECK5-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP75]], align 4
+// CHECK5-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP76]], align 4
+// CHECK5-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP77]], align 4
+// CHECK5-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l51.region_id, ptr [[KERNEL_ARGS23]])
+// CHECK5-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
+// CHECK5-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
// CHECK5: omp_offload.failed24:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l51(ptr [[THIS1]]) #[[ATTR2]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT25]]
// CHECK5: omp_offload.cont25:
// CHECK5-NEXT: [[A26:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP64]], align 8
-// CHECK5-NEXT: [[TMP65:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[A26]], ptr [[TMP65]], align 8
-// CHECK5-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i64 0, i64 0
-// CHECK5-NEXT: store ptr null, ptr [[TMP66]], align 8
-// CHECK5-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP80:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP80]], align 8
+// CHECK5-NEXT: [[TMP81:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[A26]], ptr [[TMP81]], align 8
+// CHECK5-NEXT: [[TMP82:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i64 0, i64 0
+// CHECK5-NEXT: store ptr null, ptr [[TMP82]], align 8
+// CHECK5-NEXT: [[TMP83:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP84:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP69]], align 4
-// CHECK5-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK5-NEXT: store i32 1, ptr [[TMP70]], align 4
-// CHECK5-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK5-NEXT: store ptr [[TMP67]], ptr [[TMP71]], align 8
-// CHECK5-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 8
-// CHECK5-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK5-NEXT: store ptr @.offload_sizes.16, ptr [[TMP73]], align 8
-// CHECK5-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK5-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP74]], align 8
-// CHECK5-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP75]], align 8
-// CHECK5-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK5-NEXT: store ptr null, ptr [[TMP76]], align 8
-// CHECK5-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK5-NEXT: store i64 123, ptr [[TMP77]], align 8
-// CHECK5-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l57.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK5-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
-// CHECK5-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK5-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP85]], align 4
+// CHECK5-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK5-NEXT: store i32 1, ptr [[TMP86]], align 4
+// CHECK5-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK5-NEXT: store ptr [[TMP83]], ptr [[TMP87]], align 8
+// CHECK5-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[TMP84]], ptr [[TMP88]], align 8
+// CHECK5-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK5-NEXT: store ptr @.offload_sizes.16, ptr [[TMP89]], align 8
+// CHECK5-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK5-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP90]], align 8
+// CHECK5-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP91]], align 8
+// CHECK5-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP92]], align 8
+// CHECK5-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK5-NEXT: store i64 123, ptr [[TMP93]], align 8
+// CHECK5-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP94]], align 8
+// CHECK5-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP95]], align 4
+// CHECK5-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP96]], align 4
+// CHECK5-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP97]], align 4
+// CHECK5-NEXT: [[TMP98:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l57.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK5-NEXT: [[TMP99:%.*]] = icmp ne i32 [[TMP98]], 0
+// CHECK5-NEXT: br i1 [[TMP99]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK5: omp_offload.failed32:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l57(ptr [[THIS1]]) #[[ATTR2]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK5: omp_offload.cont33:
// CHECK5-NEXT: [[A34:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK5-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A34]], i64 0, i64 0
-// CHECK5-NEXT: [[TMP80:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK5-NEXT: ret i32 [[TMP80]]
+// CHECK5-NEXT: [[TMP100:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK5-NEXT: ret i32 [[TMP100]]
//
//
// CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l35
@@ -3208,7 +3328,7 @@ int main (int argc, char **argv) {
// CHECK7-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK7-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK7-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK7-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK7-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK7-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3225,157 +3345,197 @@ int main (int argc, char **argv) {
// CHECK7-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK7-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK7-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK7-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l35.region_id, ptr [[KERNEL_ARGS]])
-// CHECK7-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK7-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK7-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK7-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK7-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK7-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK7-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l35.region_id, ptr [[KERNEL_ARGS]])
+// CHECK7-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK7-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK7: omp_offload.failed:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l35(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK7: omp_offload.cont:
// CHECK7-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 4
-// CHECK7-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK7-NEXT: store ptr [[A2]], ptr [[TMP17]], align 4
-// CHECK7-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
-// CHECK7-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK7-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 4
+// CHECK7-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK7-NEXT: store ptr [[A2]], ptr [[TMP21]], align 4
+// CHECK7-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
+// CHECK7-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK7-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK7-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK7-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK7-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK7-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK7-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK7-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK7-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK7-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK7-NEXT: store ptr @.offload_sizes.4, ptr [[TMP25]], align 4
-// CHECK7-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK7-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP26]], align 4
-// CHECK7-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK7-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK7-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK7-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK7-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK7-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK7-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK7-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK7-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK7-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK7-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK7-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK7-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK7-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK7-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK7-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK7-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK7-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK7-NEXT: store ptr @.offload_sizes.4, ptr [[TMP29]], align 4
+// CHECK7-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK7-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP30]], align 4
+// CHECK7-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK7-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK7-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK7-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK7-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK7-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK7-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK7-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK7-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK7-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK7-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK7-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK7-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK7: omp_offload.failed8:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40(ptr [[THIS1]]) #[[ATTR2]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK7: omp_offload.cont9:
// CHECK7-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 4
-// CHECK7-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK7-NEXT: store ptr [[A10]], ptr [[TMP33]], align 4
-// CHECK7-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
-// CHECK7-NEXT: store ptr null, ptr [[TMP34]], align 4
-// CHECK7-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 4
+// CHECK7-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK7-NEXT: store ptr [[A10]], ptr [[TMP41]], align 4
+// CHECK7-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
+// CHECK7-NEXT: store ptr null, ptr [[TMP42]], align 4
+// CHECK7-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK7-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK7-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK7-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK7-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK7-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 4
-// CHECK7-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK7-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 4
-// CHECK7-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK7-NEXT: store ptr @.offload_sizes.8, ptr [[TMP41]], align 4
-// CHECK7-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK7-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP42]], align 4
-// CHECK7-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK7-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK7-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK7-NEXT: store ptr null, ptr [[TMP44]], align 4
-// CHECK7-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK7-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK7-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l45.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK7-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK7-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK7-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK7-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK7-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK7-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK7-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK7-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 4
+// CHECK7-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK7-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 4
+// CHECK7-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK7-NEXT: store ptr @.offload_sizes.8, ptr [[TMP49]], align 4
+// CHECK7-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK7-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP50]], align 4
+// CHECK7-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK7-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK7-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK7-NEXT: store ptr null, ptr [[TMP52]], align 4
+// CHECK7-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK7-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK7-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK7-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK7-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK7-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK7-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l45.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK7-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK7-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK7: omp_offload.failed16:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l45(ptr [[THIS1]]) #[[ATTR2]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK7: omp_offload.cont17:
// CHECK7-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP48:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP48]], align 4
-// CHECK7-NEXT: [[TMP49:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK7-NEXT: store ptr [[A18]], ptr [[TMP49]], align 4
-// CHECK7-NEXT: [[TMP50:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
-// CHECK7-NEXT: store ptr null, ptr [[TMP50]], align 4
-// CHECK7-NEXT: [[TMP51:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP52:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP60:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP60]], align 4
+// CHECK7-NEXT: [[TMP61:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK7-NEXT: store ptr [[A18]], ptr [[TMP61]], align 4
+// CHECK7-NEXT: [[TMP62:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
+// CHECK7-NEXT: store ptr null, ptr [[TMP62]], align 4
+// CHECK7-NEXT: [[TMP63:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS23:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK7-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP53]], align 4
-// CHECK7-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
-// CHECK7-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK7-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
-// CHECK7-NEXT: store ptr [[TMP51]], ptr [[TMP55]], align 4
-// CHECK7-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
-// CHECK7-NEXT: store ptr [[TMP52]], ptr [[TMP56]], align 4
-// CHECK7-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
-// CHECK7-NEXT: store ptr @.offload_sizes.12, ptr [[TMP57]], align 4
-// CHECK7-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
-// CHECK7-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP58]], align 4
-// CHECK7-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
-// CHECK7-NEXT: store ptr null, ptr [[TMP59]], align 4
-// CHECK7-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
-// CHECK7-NEXT: store ptr null, ptr [[TMP60]], align 4
-// CHECK7-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
-// CHECK7-NEXT: store i64 123, ptr [[TMP61]], align 8
-// CHECK7-NEXT: [[TMP62:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l51.region_id, ptr [[KERNEL_ARGS23]])
-// CHECK7-NEXT: [[TMP63:%.*]] = icmp ne i32 [[TMP62]], 0
-// CHECK7-NEXT: br i1 [[TMP63]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
+// CHECK7-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
+// CHECK7-NEXT: store i32 2, ptr [[TMP65]], align 4
+// CHECK7-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
+// CHECK7-NEXT: store i32 1, ptr [[TMP66]], align 4
+// CHECK7-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
+// CHECK7-NEXT: store ptr [[TMP63]], ptr [[TMP67]], align 4
+// CHECK7-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
+// CHECK7-NEXT: store ptr [[TMP64]], ptr [[TMP68]], align 4
+// CHECK7-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
+// CHECK7-NEXT: store ptr @.offload_sizes.12, ptr [[TMP69]], align 4
+// CHECK7-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
+// CHECK7-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP70]], align 4
+// CHECK7-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
+// CHECK7-NEXT: store ptr null, ptr [[TMP71]], align 4
+// CHECK7-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
+// CHECK7-NEXT: store ptr null, ptr [[TMP72]], align 4
+// CHECK7-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
+// CHECK7-NEXT: store i64 123, ptr [[TMP73]], align 8
+// CHECK7-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP74]], align 8
+// CHECK7-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP75]], align 4
+// CHECK7-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP76]], align 4
+// CHECK7-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP77]], align 4
+// CHECK7-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l51.region_id, ptr [[KERNEL_ARGS23]])
+// CHECK7-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
+// CHECK7-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
// CHECK7: omp_offload.failed24:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l51(ptr [[THIS1]]) #[[ATTR2]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT25]]
// CHECK7: omp_offload.cont25:
// CHECK7-NEXT: [[A26:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP64]], align 4
-// CHECK7-NEXT: [[TMP65:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
-// CHECK7-NEXT: store ptr [[A26]], ptr [[TMP65]], align 4
-// CHECK7-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i32 0, i32 0
-// CHECK7-NEXT: store ptr null, ptr [[TMP66]], align 4
-// CHECK7-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP80:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP80]], align 4
+// CHECK7-NEXT: [[TMP81:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK7-NEXT: store ptr [[A26]], ptr [[TMP81]], align 4
+// CHECK7-NEXT: [[TMP82:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i32 0, i32 0
+// CHECK7-NEXT: store ptr null, ptr [[TMP82]], align 4
+// CHECK7-NEXT: [[TMP83:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK7-NEXT: [[TMP84:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
// CHECK7-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK7-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK7-NEXT: store i32 1, ptr [[TMP69]], align 4
-// CHECK7-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK7-NEXT: store i32 1, ptr [[TMP70]], align 4
-// CHECK7-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK7-NEXT: store ptr [[TMP67]], ptr [[TMP71]], align 4
-// CHECK7-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK7-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 4
-// CHECK7-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK7-NEXT: store ptr @.offload_sizes.16, ptr [[TMP73]], align 4
-// CHECK7-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK7-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP74]], align 4
-// CHECK7-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK7-NEXT: store ptr null, ptr [[TMP75]], align 4
-// CHECK7-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK7-NEXT: store ptr null, ptr [[TMP76]], align 4
-// CHECK7-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK7-NEXT: store i64 123, ptr [[TMP77]], align 8
-// CHECK7-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l57.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK7-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
-// CHECK7-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK7-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK7-NEXT: store i32 2, ptr [[TMP85]], align 4
+// CHECK7-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK7-NEXT: store i32 1, ptr [[TMP86]], align 4
+// CHECK7-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK7-NEXT: store ptr [[TMP83]], ptr [[TMP87]], align 4
+// CHECK7-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK7-NEXT: store ptr [[TMP84]], ptr [[TMP88]], align 4
+// CHECK7-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK7-NEXT: store ptr @.offload_sizes.16, ptr [[TMP89]], align 4
+// CHECK7-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK7-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP90]], align 4
+// CHECK7-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK7-NEXT: store ptr null, ptr [[TMP91]], align 4
+// CHECK7-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK7-NEXT: store ptr null, ptr [[TMP92]], align 4
+// CHECK7-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK7-NEXT: store i64 123, ptr [[TMP93]], align 8
+// CHECK7-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK7-NEXT: store i64 0, ptr [[TMP94]], align 8
+// CHECK7-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP95]], align 4
+// CHECK7-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP96]], align 4
+// CHECK7-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK7-NEXT: store i32 0, ptr [[TMP97]], align 4
+// CHECK7-NEXT: [[TMP98:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l57.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK7-NEXT: [[TMP99:%.*]] = icmp ne i32 [[TMP98]], 0
+// CHECK7-NEXT: br i1 [[TMP99]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK7: omp_offload.failed32:
// CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l57(ptr [[THIS1]]) #[[ATTR2]]
// CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK7: omp_offload.cont33:
// CHECK7-NEXT: [[A34:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK7-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A34]], i32 0, i32 0
-// CHECK7-NEXT: [[TMP80:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK7-NEXT: ret i32 [[TMP80]]
+// CHECK7-NEXT: [[TMP100:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK7-NEXT: ret i32 [[TMP100]]
//
//
// CHECK7-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l35
@@ -4233,7 +4393,7 @@ int main (int argc, char **argv) {
// CHECK13-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK13-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK13-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK13-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK13-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK13-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK13-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -4250,294 +4410,334 @@ int main (int argc, char **argv) {
// CHECK13-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK13-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK13-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK13-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l148.region_id, ptr [[KERNEL_ARGS]])
-// CHECK13-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK13-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK13-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK13-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK13-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK13-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK13-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l148.region_id, ptr [[KERNEL_ARGS]])
+// CHECK13-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK13-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK13: omp_offload.failed:
// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l148(i64 [[TMP4]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK13: omp_offload.cont:
-// CHECK13-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK13-NEXT: [[TMP35:%.*]] = load i64, ptr [[N_CASTED3]], align 8
-// CHECK13-NEXT: [[TMP36:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK13-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK13-NEXT: [[TMP39:%.*]] = load i64, ptr [[N_CASTED3]], align 8
+// CHECK13-NEXT: [[TMP40:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK13-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES7]], ptr align 8 @.offload_sizes.4, i64 24, i1 false)
-// CHECK13-NEXT: [[TMP37:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP35]], ptr [[TMP37]], align 8
-// CHECK13-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP35]], ptr [[TMP38]], align 8
-// CHECK13-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK13-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP40]], align 8
-// CHECK13-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP41]], align 8
-// CHECK13-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
-// CHECK13-NEXT: store ptr null, ptr [[TMP42]], align 8
-// CHECK13-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP43]], align 8
-// CHECK13-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 8
-// CHECK13-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK13-NEXT: store i64 [[TMP36]], ptr [[TMP45]], align 8
-// CHECK13-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK13-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP39]], ptr [[TMP41]], align 8
+// CHECK13-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP39]], ptr [[TMP42]], align 8
+// CHECK13-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK13-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP44]], align 8
+// CHECK13-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP45]], align 8
+// CHECK13-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
// CHECK13-NEXT: store ptr null, ptr [[TMP46]], align 8
-// CHECK13-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP50:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP50]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK13-NEXT: [[TMP51:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK13-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP51]], 0
+// CHECK13-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP47]], align 8
+// CHECK13-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 8
+// CHECK13-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK13-NEXT: store i64 [[TMP40]], ptr [[TMP49]], align 8
+// CHECK13-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK13-NEXT: store ptr null, ptr [[TMP50]], align 8
+// CHECK13-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP54:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP54]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK13-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK13-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP55]], 0
// CHECK13-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK13-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK13-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK13-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK13-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP52]], 1
-// CHECK13-NEXT: [[TMP53:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK13-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK13-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP56]], 1
+// CHECK13-NEXT: [[TMP57:%.*]] = zext i32 [[ADD14]] to i64
// CHECK13-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK13-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK13-NEXT: store i32 3, ptr [[TMP55]], align 4
-// CHECK13-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP47]], ptr [[TMP56]], align 8
-// CHECK13-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 8
-// CHECK13-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK13-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 8
-// CHECK13-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP59]], align 8
-// CHECK13-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP60]], align 8
-// CHECK13-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP61]], align 8
-// CHECK13-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK13-NEXT: store i64 [[TMP53]], ptr [[TMP62]], align 8
-// CHECK13-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l153.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK13-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
-// CHECK13-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK13-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP58]], align 4
+// CHECK13-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK13-NEXT: store i32 3, ptr [[TMP59]], align 4
+// CHECK13-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP51]], ptr [[TMP60]], align 8
+// CHECK13-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 8
+// CHECK13-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK13-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 8
+// CHECK13-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP63]], align 8
+// CHECK13-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP64]], align 8
+// CHECK13-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP65]], align 8
+// CHECK13-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK13-NEXT: store i64 [[TMP57]], ptr [[TMP66]], align 8
+// CHECK13-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP67]], align 8
+// CHECK13-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP68]], align 4
+// CHECK13-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK13-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP70]], align 4
+// CHECK13-NEXT: [[TMP71:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l153.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK13-NEXT: [[TMP72:%.*]] = icmp ne i32 [[TMP71]], 0
+// CHECK13-NEXT: br i1 [[TMP72]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK13: omp_offload.failed16:
-// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l153(i64 [[TMP35]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l153(i64 [[TMP39]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK13: omp_offload.cont17:
-// CHECK13-NEXT: [[TMP65:%.*]] = load i32, ptr [[M]], align 4
-// CHECK13-NEXT: store i32 [[TMP65]], ptr [[M_CASTED]], align 4
-// CHECK13-NEXT: [[TMP66:%.*]] = load i64, ptr [[M_CASTED]], align 8
-// CHECK13-NEXT: [[TMP67:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP67]], ptr [[N_CASTED18]], align 4
-// CHECK13-NEXT: [[TMP68:%.*]] = load i64, ptr [[N_CASTED18]], align 8
-// CHECK13-NEXT: [[TMP69:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK13-NEXT: [[TMP73:%.*]] = load i32, ptr [[M]], align 4
+// CHECK13-NEXT: store i32 [[TMP73]], ptr [[M_CASTED]], align 4
+// CHECK13-NEXT: [[TMP74:%.*]] = load i64, ptr [[M_CASTED]], align 8
+// CHECK13-NEXT: [[TMP75:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP75]], ptr [[N_CASTED18]], align 4
+// CHECK13-NEXT: [[TMP76:%.*]] = load i64, ptr [[N_CASTED18]], align 8
+// CHECK13-NEXT: [[TMP77:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK13-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES22]], ptr align 8 @.offload_sizes.8, i64 32, i1 false)
-// CHECK13-NEXT: [[TMP70:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP66]], ptr [[TMP70]], align 8
-// CHECK13-NEXT: [[TMP71:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP66]], ptr [[TMP71]], align 8
-// CHECK13-NEXT: [[TMP72:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP72]], align 8
-// CHECK13-NEXT: [[TMP73:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP68]], ptr [[TMP73]], align 8
-// CHECK13-NEXT: [[TMP74:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP68]], ptr [[TMP74]], align 8
-// CHECK13-NEXT: [[TMP75:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 1
-// CHECK13-NEXT: store ptr null, ptr [[TMP75]], align 8
-// CHECK13-NEXT: [[TMP76:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP76]], align 8
-// CHECK13-NEXT: [[TMP77:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP77]], align 8
-// CHECK13-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 2
-// CHECK13-NEXT: store ptr null, ptr [[TMP78]], align 8
-// CHECK13-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP79]], align 8
-// CHECK13-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP80]], align 8
-// CHECK13-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 3
-// CHECK13-NEXT: store i64 [[TMP69]], ptr [[TMP81]], align 8
-// CHECK13-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 3
-// CHECK13-NEXT: store ptr null, ptr [[TMP82]], align 8
-// CHECK13-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP86:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP86]], ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK13-NEXT: [[TMP87:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK13-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP87]], 0
+// CHECK13-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP74]], ptr [[TMP78]], align 8
+// CHECK13-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP74]], ptr [[TMP79]], align 8
+// CHECK13-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP80]], align 8
+// CHECK13-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP76]], ptr [[TMP81]], align 8
+// CHECK13-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP76]], ptr [[TMP82]], align 8
+// CHECK13-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 1
+// CHECK13-NEXT: store ptr null, ptr [[TMP83]], align 8
+// CHECK13-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP84]], align 8
+// CHECK13-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP85]], align 8
+// CHECK13-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 2
+// CHECK13-NEXT: store ptr null, ptr [[TMP86]], align 8
+// CHECK13-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP87]], align 8
+// CHECK13-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP88]], align 8
+// CHECK13-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 3
+// CHECK13-NEXT: store i64 [[TMP77]], ptr [[TMP89]], align 8
+// CHECK13-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 3
+// CHECK13-NEXT: store ptr null, ptr [[TMP90]], align 8
+// CHECK13-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP94:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP94]], ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK13-NEXT: [[TMP95:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK13-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP95]], 0
// CHECK13-NEXT: [[DIV27:%.*]] = sdiv i32 [[SUB26]], 1
// CHECK13-NEXT: [[SUB28:%.*]] = sub nsw i32 [[DIV27]], 1
// CHECK13-NEXT: store i32 [[SUB28]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK13-NEXT: [[TMP88:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK13-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP88]], 1
-// CHECK13-NEXT: [[TMP89:%.*]] = zext i32 [[ADD29]] to i64
+// CHECK13-NEXT: [[TMP96:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK13-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP96]], 1
+// CHECK13-NEXT: [[TMP97:%.*]] = zext i32 [[ADD29]] to i64
// CHECK13-NEXT: [[KERNEL_ARGS30:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP90]], align 4
-// CHECK13-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
-// CHECK13-NEXT: store i32 4, ptr [[TMP91]], align 4
-// CHECK13-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP83]], ptr [[TMP92]], align 8
-// CHECK13-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP84]], ptr [[TMP93]], align 8
-// CHECK13-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
-// CHECK13-NEXT: store ptr [[TMP85]], ptr [[TMP94]], align 8
-// CHECK13-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP95]], align 8
-// CHECK13-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP96]], align 8
-// CHECK13-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP97]], align 8
-// CHECK13-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
-// CHECK13-NEXT: store i64 [[TMP89]], ptr [[TMP98]], align 8
-// CHECK13-NEXT: [[TMP99:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l158.region_id, ptr [[KERNEL_ARGS30]])
-// CHECK13-NEXT: [[TMP100:%.*]] = icmp ne i32 [[TMP99]], 0
-// CHECK13-NEXT: br i1 [[TMP100]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
+// CHECK13-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP98]], align 4
+// CHECK13-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
+// CHECK13-NEXT: store i32 4, ptr [[TMP99]], align 4
+// CHECK13-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP91]], ptr [[TMP100]], align 8
+// CHECK13-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP92]], ptr [[TMP101]], align 8
+// CHECK13-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
+// CHECK13-NEXT: store ptr [[TMP93]], ptr [[TMP102]], align 8
+// CHECK13-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP103]], align 8
+// CHECK13-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP104]], align 8
+// CHECK13-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP105]], align 8
+// CHECK13-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
+// CHECK13-NEXT: store i64 [[TMP97]], ptr [[TMP106]], align 8
+// CHECK13-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP107]], align 8
+// CHECK13-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP108]], align 4
+// CHECK13-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP109]], align 4
+// CHECK13-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP110]], align 4
+// CHECK13-NEXT: [[TMP111:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l158.region_id, ptr [[KERNEL_ARGS30]])
+// CHECK13-NEXT: [[TMP112:%.*]] = icmp ne i32 [[TMP111]], 0
+// CHECK13-NEXT: br i1 [[TMP112]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
// CHECK13: omp_offload.failed31:
-// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l158(i64 [[TMP66]], i64 [[TMP68]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l158(i64 [[TMP74]], i64 [[TMP76]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT32]]
// CHECK13: omp_offload.cont32:
-// CHECK13-NEXT: [[TMP101:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP101]], ptr [[N_CASTED33]], align 4
-// CHECK13-NEXT: [[TMP102:%.*]] = load i64, ptr [[N_CASTED33]], align 8
-// CHECK13-NEXT: [[TMP103:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK13-NEXT: [[TMP113:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP113]], ptr [[N_CASTED33]], align 4
+// CHECK13-NEXT: [[TMP114:%.*]] = load i64, ptr [[N_CASTED33]], align 8
+// CHECK13-NEXT: [[TMP115:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK13-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES37]], ptr align 8 @.offload_sizes.12, i64 24, i1 false)
-// CHECK13-NEXT: [[TMP104:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP102]], ptr [[TMP104]], align 8
-// CHECK13-NEXT: [[TMP105:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP102]], ptr [[TMP105]], align 8
-// CHECK13-NEXT: [[TMP106:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP106]], align 8
-// CHECK13-NEXT: [[TMP107:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP107]], align 8
-// CHECK13-NEXT: [[TMP108:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP108]], align 8
-// CHECK13-NEXT: [[TMP109:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i64 0, i64 1
-// CHECK13-NEXT: store ptr null, ptr [[TMP109]], align 8
-// CHECK13-NEXT: [[TMP110:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP110]], align 8
-// CHECK13-NEXT: [[TMP111:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP111]], align 8
-// CHECK13-NEXT: [[TMP112:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 2
-// CHECK13-NEXT: store i64 [[TMP103]], ptr [[TMP112]], align 8
-// CHECK13-NEXT: [[TMP113:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i64 0, i64 2
-// CHECK13-NEXT: store ptr null, ptr [[TMP113]], align 8
-// CHECK13-NEXT: [[TMP114:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP115:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP116:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP117:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP117]], ptr [[DOTCAPTURE_EXPR_39]], align 4
-// CHECK13-NEXT: [[TMP118:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_39]], align 4
-// CHECK13-NEXT: [[SUB41:%.*]] = sub nsw i32 [[TMP118]], 0
+// CHECK13-NEXT: [[TMP116:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP114]], ptr [[TMP116]], align 8
+// CHECK13-NEXT: [[TMP117:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP114]], ptr [[TMP117]], align 8
+// CHECK13-NEXT: [[TMP118:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP118]], align 8
+// CHECK13-NEXT: [[TMP119:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP119]], align 8
+// CHECK13-NEXT: [[TMP120:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP120]], align 8
+// CHECK13-NEXT: [[TMP121:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i64 0, i64 1
+// CHECK13-NEXT: store ptr null, ptr [[TMP121]], align 8
+// CHECK13-NEXT: [[TMP122:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP122]], align 8
+// CHECK13-NEXT: [[TMP123:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP123]], align 8
+// CHECK13-NEXT: [[TMP124:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 2
+// CHECK13-NEXT: store i64 [[TMP115]], ptr [[TMP124]], align 8
+// CHECK13-NEXT: [[TMP125:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i64 0, i64 2
+// CHECK13-NEXT: store ptr null, ptr [[TMP125]], align 8
+// CHECK13-NEXT: [[TMP126:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP127:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP128:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP129:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP129]], ptr [[DOTCAPTURE_EXPR_39]], align 4
+// CHECK13-NEXT: [[TMP130:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_39]], align 4
+// CHECK13-NEXT: [[SUB41:%.*]] = sub nsw i32 [[TMP130]], 0
// CHECK13-NEXT: [[DIV42:%.*]] = sdiv i32 [[SUB41]], 1
// CHECK13-NEXT: [[SUB43:%.*]] = sub nsw i32 [[DIV42]], 1
// CHECK13-NEXT: store i32 [[SUB43]], ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK13-NEXT: [[TMP119:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK13-NEXT: [[ADD44:%.*]] = add nsw i32 [[TMP119]], 1
-// CHECK13-NEXT: [[TMP120:%.*]] = zext i32 [[ADD44]] to i64
+// CHECK13-NEXT: [[TMP131:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
+// CHECK13-NEXT: [[ADD44:%.*]] = add nsw i32 [[TMP131]], 1
+// CHECK13-NEXT: [[TMP132:%.*]] = zext i32 [[ADD44]] to i64
// CHECK13-NEXT: [[KERNEL_ARGS45:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP121:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP121]], align 4
-// CHECK13-NEXT: [[TMP122:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 1
-// CHECK13-NEXT: store i32 3, ptr [[TMP122]], align 4
-// CHECK13-NEXT: [[TMP123:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP114]], ptr [[TMP123]], align 8
-// CHECK13-NEXT: [[TMP124:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP115]], ptr [[TMP124]], align 8
-// CHECK13-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 4
-// CHECK13-NEXT: store ptr [[TMP116]], ptr [[TMP125]], align 8
-// CHECK13-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP126]], align 8
-// CHECK13-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP127]], align 8
-// CHECK13-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP128]], align 8
-// CHECK13-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 8
-// CHECK13-NEXT: store i64 [[TMP120]], ptr [[TMP129]], align 8
-// CHECK13-NEXT: [[TMP130:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l163.region_id, ptr [[KERNEL_ARGS45]])
-// CHECK13-NEXT: [[TMP131:%.*]] = icmp ne i32 [[TMP130]], 0
-// CHECK13-NEXT: br i1 [[TMP131]], label [[OMP_OFFLOAD_FAILED46:%.*]], label [[OMP_OFFLOAD_CONT47:%.*]]
+// CHECK13-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP133]], align 4
+// CHECK13-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 1
+// CHECK13-NEXT: store i32 3, ptr [[TMP134]], align 4
+// CHECK13-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP126]], ptr [[TMP135]], align 8
+// CHECK13-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP127]], ptr [[TMP136]], align 8
+// CHECK13-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 4
+// CHECK13-NEXT: store ptr [[TMP128]], ptr [[TMP137]], align 8
+// CHECK13-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP138]], align 8
+// CHECK13-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP139]], align 8
+// CHECK13-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP140]], align 8
+// CHECK13-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 8
+// CHECK13-NEXT: store i64 [[TMP132]], ptr [[TMP141]], align 8
+// CHECK13-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP142]], align 8
+// CHECK13-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP143]], align 4
+// CHECK13-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP144]], align 4
+// CHECK13-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP145]], align 4
+// CHECK13-NEXT: [[TMP146:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l163.region_id, ptr [[KERNEL_ARGS45]])
+// CHECK13-NEXT: [[TMP147:%.*]] = icmp ne i32 [[TMP146]], 0
+// CHECK13-NEXT: br i1 [[TMP147]], label [[OMP_OFFLOAD_FAILED46:%.*]], label [[OMP_OFFLOAD_CONT47:%.*]]
// CHECK13: omp_offload.failed46:
-// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l163(i64 [[TMP102]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l163(i64 [[TMP114]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT47]]
// CHECK13: omp_offload.cont47:
-// CHECK13-NEXT: [[TMP132:%.*]] = load i32, ptr [[M]], align 4
-// CHECK13-NEXT: store i32 [[TMP132]], ptr [[M_CASTED48]], align 4
-// CHECK13-NEXT: [[TMP133:%.*]] = load i64, ptr [[M_CASTED48]], align 8
-// CHECK13-NEXT: [[TMP134:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP134]], ptr [[N_CASTED49]], align 4
-// CHECK13-NEXT: [[TMP135:%.*]] = load i64, ptr [[N_CASTED49]], align 8
-// CHECK13-NEXT: [[TMP136:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK13-NEXT: [[TMP148:%.*]] = load i32, ptr [[M]], align 4
+// CHECK13-NEXT: store i32 [[TMP148]], ptr [[M_CASTED48]], align 4
+// CHECK13-NEXT: [[TMP149:%.*]] = load i64, ptr [[M_CASTED48]], align 8
+// CHECK13-NEXT: [[TMP150:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP150]], ptr [[N_CASTED49]], align 4
+// CHECK13-NEXT: [[TMP151:%.*]] = load i64, ptr [[N_CASTED49]], align 8
+// CHECK13-NEXT: [[TMP152:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK13-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES53]], ptr align 8 @.offload_sizes.16, i64 32, i1 false)
-// CHECK13-NEXT: [[TMP137:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP133]], ptr [[TMP137]], align 8
-// CHECK13-NEXT: [[TMP138:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP133]], ptr [[TMP138]], align 8
-// CHECK13-NEXT: [[TMP139:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP139]], align 8
-// CHECK13-NEXT: [[TMP140:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP135]], ptr [[TMP140]], align 8
-// CHECK13-NEXT: [[TMP141:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP135]], ptr [[TMP141]], align 8
-// CHECK13-NEXT: [[TMP142:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 1
-// CHECK13-NEXT: store ptr null, ptr [[TMP142]], align 8
-// CHECK13-NEXT: [[TMP143:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 2
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP143]], align 8
-// CHECK13-NEXT: [[TMP144:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 2
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP144]], align 8
-// CHECK13-NEXT: [[TMP145:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 2
-// CHECK13-NEXT: store ptr null, ptr [[TMP145]], align 8
-// CHECK13-NEXT: [[TMP146:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP146]], align 8
-// CHECK13-NEXT: [[TMP147:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP147]], align 8
-// CHECK13-NEXT: [[TMP148:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 3
-// CHECK13-NEXT: store i64 [[TMP136]], ptr [[TMP148]], align 8
-// CHECK13-NEXT: [[TMP149:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 3
-// CHECK13-NEXT: store ptr null, ptr [[TMP149]], align 8
-// CHECK13-NEXT: [[TMP150:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP151:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP152:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP153:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP153]], ptr [[DOTCAPTURE_EXPR_55]], align 4
-// CHECK13-NEXT: [[TMP154:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_55]], align 4
-// CHECK13-NEXT: [[SUB57:%.*]] = sub nsw i32 [[TMP154]], 0
+// CHECK13-NEXT: [[TMP153:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP149]], ptr [[TMP153]], align 8
+// CHECK13-NEXT: [[TMP154:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP149]], ptr [[TMP154]], align 8
+// CHECK13-NEXT: [[TMP155:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP155]], align 8
+// CHECK13-NEXT: [[TMP156:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP151]], ptr [[TMP156]], align 8
+// CHECK13-NEXT: [[TMP157:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP151]], ptr [[TMP157]], align 8
+// CHECK13-NEXT: [[TMP158:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 1
+// CHECK13-NEXT: store ptr null, ptr [[TMP158]], align 8
+// CHECK13-NEXT: [[TMP159:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 2
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP159]], align 8
+// CHECK13-NEXT: [[TMP160:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 2
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP160]], align 8
+// CHECK13-NEXT: [[TMP161:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 2
+// CHECK13-NEXT: store ptr null, ptr [[TMP161]], align 8
+// CHECK13-NEXT: [[TMP162:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP162]], align 8
+// CHECK13-NEXT: [[TMP163:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP163]], align 8
+// CHECK13-NEXT: [[TMP164:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 3
+// CHECK13-NEXT: store i64 [[TMP152]], ptr [[TMP164]], align 8
+// CHECK13-NEXT: [[TMP165:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 3
+// CHECK13-NEXT: store ptr null, ptr [[TMP165]], align 8
+// CHECK13-NEXT: [[TMP166:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP167:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP168:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP169:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP169]], ptr [[DOTCAPTURE_EXPR_55]], align 4
+// CHECK13-NEXT: [[TMP170:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_55]], align 4
+// CHECK13-NEXT: [[SUB57:%.*]] = sub nsw i32 [[TMP170]], 0
// CHECK13-NEXT: [[DIV58:%.*]] = sdiv i32 [[SUB57]], 1
// CHECK13-NEXT: [[SUB59:%.*]] = sub nsw i32 [[DIV58]], 1
// CHECK13-NEXT: store i32 [[SUB59]], ptr [[DOTCAPTURE_EXPR_56]], align 4
-// CHECK13-NEXT: [[TMP155:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_56]], align 4
-// CHECK13-NEXT: [[ADD60:%.*]] = add nsw i32 [[TMP155]], 1
-// CHECK13-NEXT: [[TMP156:%.*]] = zext i32 [[ADD60]] to i64
+// CHECK13-NEXT: [[TMP171:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_56]], align 4
+// CHECK13-NEXT: [[ADD60:%.*]] = add nsw i32 [[TMP171]], 1
+// CHECK13-NEXT: [[TMP172:%.*]] = zext i32 [[ADD60]] to i64
// CHECK13-NEXT: [[KERNEL_ARGS61:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP157:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP157]], align 4
-// CHECK13-NEXT: [[TMP158:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 1
-// CHECK13-NEXT: store i32 4, ptr [[TMP158]], align 4
-// CHECK13-NEXT: [[TMP159:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP150]], ptr [[TMP159]], align 8
-// CHECK13-NEXT: [[TMP160:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP151]], ptr [[TMP160]], align 8
-// CHECK13-NEXT: [[TMP161:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 4
-// CHECK13-NEXT: store ptr [[TMP152]], ptr [[TMP161]], align 8
-// CHECK13-NEXT: [[TMP162:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP162]], align 8
-// CHECK13-NEXT: [[TMP163:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP163]], align 8
-// CHECK13-NEXT: [[TMP164:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP164]], align 8
-// CHECK13-NEXT: [[TMP165:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 8
-// CHECK13-NEXT: store i64 [[TMP156]], ptr [[TMP165]], align 8
-// CHECK13-NEXT: [[TMP166:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l168.region_id, ptr [[KERNEL_ARGS61]])
-// CHECK13-NEXT: [[TMP167:%.*]] = icmp ne i32 [[TMP166]], 0
-// CHECK13-NEXT: br i1 [[TMP167]], label [[OMP_OFFLOAD_FAILED62:%.*]], label [[OMP_OFFLOAD_CONT63:%.*]]
+// CHECK13-NEXT: [[TMP173:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP173]], align 4
+// CHECK13-NEXT: [[TMP174:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 1
+// CHECK13-NEXT: store i32 4, ptr [[TMP174]], align 4
+// CHECK13-NEXT: [[TMP175:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP166]], ptr [[TMP175]], align 8
+// CHECK13-NEXT: [[TMP176:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP167]], ptr [[TMP176]], align 8
+// CHECK13-NEXT: [[TMP177:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 4
+// CHECK13-NEXT: store ptr [[TMP168]], ptr [[TMP177]], align 8
+// CHECK13-NEXT: [[TMP178:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP178]], align 8
+// CHECK13-NEXT: [[TMP179:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP179]], align 8
+// CHECK13-NEXT: [[TMP180:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP180]], align 8
+// CHECK13-NEXT: [[TMP181:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 8
+// CHECK13-NEXT: store i64 [[TMP172]], ptr [[TMP181]], align 8
+// CHECK13-NEXT: [[TMP182:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP182]], align 8
+// CHECK13-NEXT: [[TMP183:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP183]], align 4
+// CHECK13-NEXT: [[TMP184:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP184]], align 4
+// CHECK13-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP185]], align 4
+// CHECK13-NEXT: [[TMP186:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l168.region_id, ptr [[KERNEL_ARGS61]])
+// CHECK13-NEXT: [[TMP187:%.*]] = icmp ne i32 [[TMP186]], 0
+// CHECK13-NEXT: br i1 [[TMP187]], label [[OMP_OFFLOAD_FAILED62:%.*]], label [[OMP_OFFLOAD_CONT63:%.*]]
// CHECK13: omp_offload.failed62:
-// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l168(i64 [[TMP133]], i64 [[TMP135]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l168(i64 [[TMP149]], i64 [[TMP151]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT63]]
// CHECK13: omp_offload.cont63:
-// CHECK13-NEXT: [[TMP168:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK13-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP168]])
+// CHECK13-NEXT: [[TMP188:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK13-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP188]])
// CHECK13-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK13-NEXT: [[TMP169:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK13-NEXT: call void @llvm.stackrestore(ptr [[TMP169]])
-// CHECK13-NEXT: [[TMP170:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK13-NEXT: ret i32 [[TMP170]]
+// CHECK13-NEXT: [[TMP189:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK13-NEXT: call void @llvm.stackrestore(ptr [[TMP189]])
+// CHECK13-NEXT: [[TMP190:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK13-NEXT: ret i32 [[TMP190]]
//
//
// CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l148
@@ -5687,7 +5887,7 @@ int main (int argc, char **argv) {
// CHECK13-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK13-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK13-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK13-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK13-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK13-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK13-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -5704,165 +5904,205 @@ int main (int argc, char **argv) {
// CHECK13-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK13-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK13-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK13-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.region_id, ptr [[KERNEL_ARGS]])
-// CHECK13-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK13-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK13-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK13-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK13-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK13-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK13-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.region_id, ptr [[KERNEL_ARGS]])
+// CHECK13-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK13-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK13: omp_offload.failed:
// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116(ptr [[A]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK13: omp_offload.cont:
-// CHECK13-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP16]], align 8
-// CHECK13-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP17]], align 8
-// CHECK13-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK13-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP20]], align 8
+// CHECK13-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP21]], align 8
+// CHECK13-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK13-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK13-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK13-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK13-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK13-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK13-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK13-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK13-NEXT: store ptr @.offload_sizes.24, ptr [[TMP25]], align 8
-// CHECK13-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP26]], align 8
-// CHECK13-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK13-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK13-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK13-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK13-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l121.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK13-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK13-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK13-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK13-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK13-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK13-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK13-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK13-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK13-NEXT: store ptr @.offload_sizes.24, ptr [[TMP29]], align 8
+// CHECK13-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP30]], align 8
+// CHECK13-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK13-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK13-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK13-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK13-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK13-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK13-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK13-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK13-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l121.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK13-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK13-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK13: omp_offload.failed6:
// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l121(ptr [[A]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK13: omp_offload.cont7:
-// CHECK13-NEXT: [[TMP32:%.*]] = load i32, ptr [[M]], align 4
-// CHECK13-NEXT: store i32 [[TMP32]], ptr [[M_CASTED]], align 4
-// CHECK13-NEXT: [[TMP33:%.*]] = load i64, ptr [[M_CASTED]], align 8
-// CHECK13-NEXT: [[TMP34:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP33]], ptr [[TMP34]], align 8
-// CHECK13-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP33]], ptr [[TMP35]], align 8
-// CHECK13-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP36]], align 8
-// CHECK13-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP37]], align 8
-// CHECK13-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP38]], align 8
-// CHECK13-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
-// CHECK13-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK13-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP40:%.*]] = load i32, ptr [[M]], align 4
+// CHECK13-NEXT: store i32 [[TMP40]], ptr [[M_CASTED]], align 4
+// CHECK13-NEXT: [[TMP41:%.*]] = load i64, ptr [[M_CASTED]], align 8
+// CHECK13-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP41]], ptr [[TMP42]], align 8
+// CHECK13-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP41]], ptr [[TMP43]], align 8
+// CHECK13-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP44]], align 8
+// CHECK13-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP45]], align 8
+// CHECK13-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP46]], align 8
+// CHECK13-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
+// CHECK13-NEXT: store ptr null, ptr [[TMP47]], align 8
+// CHECK13-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK13-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP42]], align 4
-// CHECK13-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK13-NEXT: store i32 2, ptr [[TMP43]], align 4
-// CHECK13-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP40]], ptr [[TMP44]], align 8
-// CHECK13-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP41]], ptr [[TMP45]], align 8
-// CHECK13-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK13-NEXT: store ptr @.offload_sizes.28, ptr [[TMP46]], align 8
-// CHECK13-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP47]], align 8
-// CHECK13-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP48]], align 8
-// CHECK13-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP49]], align 8
-// CHECK13-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK13-NEXT: store i64 10, ptr [[TMP50]], align 8
-// CHECK13-NEXT: [[TMP51:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l126.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK13-NEXT: [[TMP52:%.*]] = icmp ne i32 [[TMP51]], 0
-// CHECK13-NEXT: br i1 [[TMP52]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK13-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP50]], align 4
+// CHECK13-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK13-NEXT: store i32 2, ptr [[TMP51]], align 4
+// CHECK13-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP48]], ptr [[TMP52]], align 8
+// CHECK13-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 8
+// CHECK13-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK13-NEXT: store ptr @.offload_sizes.28, ptr [[TMP54]], align 8
+// CHECK13-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP55]], align 8
+// CHECK13-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP56]], align 8
+// CHECK13-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP57]], align 8
+// CHECK13-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK13-NEXT: store i64 10, ptr [[TMP58]], align 8
+// CHECK13-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP59]], align 8
+// CHECK13-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP60]], align 4
+// CHECK13-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4
+// CHECK13-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP62]], align 4
+// CHECK13-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l126.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK13-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
+// CHECK13-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK13: omp_offload.failed13:
-// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l126(i64 [[TMP33]], ptr [[A]]) #[[ATTR3]]
+// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l126(i64 [[TMP41]], ptr [[A]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT14]]
// CHECK13: omp_offload.cont14:
-// CHECK13-NEXT: [[TMP53:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP53]], align 8
-// CHECK13-NEXT: [[TMP54:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP54]], align 8
-// CHECK13-NEXT: [[TMP55:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP55]], align 8
-// CHECK13-NEXT: [[TMP56:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP57:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP65:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP65]], align 8
+// CHECK13-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP66]], align 8
+// CHECK13-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP67]], align 8
+// CHECK13-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP69:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
// CHECK13-NEXT: [[KERNEL_ARGS19:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP58]], align 4
-// CHECK13-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
-// CHECK13-NEXT: store i32 1, ptr [[TMP59]], align 4
-// CHECK13-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP56]], ptr [[TMP60]], align 8
-// CHECK13-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP57]], ptr [[TMP61]], align 8
-// CHECK13-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
-// CHECK13-NEXT: store ptr @.offload_sizes.32, ptr [[TMP62]], align 8
-// CHECK13-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP63]], align 8
-// CHECK13-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP64]], align 8
-// CHECK13-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP65]], align 8
-// CHECK13-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
-// CHECK13-NEXT: store i64 10, ptr [[TMP66]], align 8
-// CHECK13-NEXT: [[TMP67:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l131.region_id, ptr [[KERNEL_ARGS19]])
-// CHECK13-NEXT: [[TMP68:%.*]] = icmp ne i32 [[TMP67]], 0
-// CHECK13-NEXT: br i1 [[TMP68]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
+// CHECK13-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP70]], align 4
+// CHECK13-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
+// CHECK13-NEXT: store i32 1, ptr [[TMP71]], align 4
+// CHECK13-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 8
+// CHECK13-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP69]], ptr [[TMP73]], align 8
+// CHECK13-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
+// CHECK13-NEXT: store ptr @.offload_sizes.32, ptr [[TMP74]], align 8
+// CHECK13-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP75]], align 8
+// CHECK13-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP76]], align 8
+// CHECK13-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP77]], align 8
+// CHECK13-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
+// CHECK13-NEXT: store i64 10, ptr [[TMP78]], align 8
+// CHECK13-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP79]], align 8
+// CHECK13-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP80]], align 4
+// CHECK13-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP81]], align 4
+// CHECK13-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP82]], align 4
+// CHECK13-NEXT: [[TMP83:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l131.region_id, ptr [[KERNEL_ARGS19]])
+// CHECK13-NEXT: [[TMP84:%.*]] = icmp ne i32 [[TMP83]], 0
+// CHECK13-NEXT: br i1 [[TMP84]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
// CHECK13: omp_offload.failed20:
// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l131(ptr [[A]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT21]]
// CHECK13: omp_offload.cont21:
-// CHECK13-NEXT: [[TMP69:%.*]] = load i32, ptr [[M]], align 4
-// CHECK13-NEXT: store i32 [[TMP69]], ptr [[M_CASTED22]], align 4
-// CHECK13-NEXT: [[TMP70:%.*]] = load i64, ptr [[M_CASTED22]], align 8
-// CHECK13-NEXT: [[TMP71:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP70]], ptr [[TMP71]], align 8
-// CHECK13-NEXT: [[TMP72:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP70]], ptr [[TMP72]], align 8
-// CHECK13-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP73]], align 8
-// CHECK13-NEXT: [[TMP74:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 1
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP74]], align 8
-// CHECK13-NEXT: [[TMP75:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 1
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP75]], align 8
-// CHECK13-NEXT: [[TMP76:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i64 0, i64 1
-// CHECK13-NEXT: store ptr null, ptr [[TMP76]], align 8
-// CHECK13-NEXT: [[TMP77:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP78:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP85:%.*]] = load i32, ptr [[M]], align 4
+// CHECK13-NEXT: store i32 [[TMP85]], ptr [[M_CASTED22]], align 4
+// CHECK13-NEXT: [[TMP86:%.*]] = load i64, ptr [[M_CASTED22]], align 8
+// CHECK13-NEXT: [[TMP87:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP86]], ptr [[TMP87]], align 8
+// CHECK13-NEXT: [[TMP88:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP86]], ptr [[TMP88]], align 8
+// CHECK13-NEXT: [[TMP89:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP89]], align 8
+// CHECK13-NEXT: [[TMP90:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 1
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP90]], align 8
+// CHECK13-NEXT: [[TMP91:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 1
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP91]], align 8
+// CHECK13-NEXT: [[TMP92:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i64 0, i64 1
+// CHECK13-NEXT: store ptr null, ptr [[TMP92]], align 8
+// CHECK13-NEXT: [[TMP93:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP94:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
// CHECK13-NEXT: [[KERNEL_ARGS27:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP79]], align 4
-// CHECK13-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 1
-// CHECK13-NEXT: store i32 2, ptr [[TMP80]], align 4
-// CHECK13-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP77]], ptr [[TMP81]], align 8
-// CHECK13-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP78]], ptr [[TMP82]], align 8
-// CHECK13-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 4
-// CHECK13-NEXT: store ptr @.offload_sizes.36, ptr [[TMP83]], align 8
-// CHECK13-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP84]], align 8
-// CHECK13-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP85]], align 8
-// CHECK13-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP86]], align 8
-// CHECK13-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 8
-// CHECK13-NEXT: store i64 10, ptr [[TMP87]], align 8
-// CHECK13-NEXT: [[TMP88:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l136.region_id, ptr [[KERNEL_ARGS27]])
-// CHECK13-NEXT: [[TMP89:%.*]] = icmp ne i32 [[TMP88]], 0
-// CHECK13-NEXT: br i1 [[TMP89]], label [[OMP_OFFLOAD_FAILED28:%.*]], label [[OMP_OFFLOAD_CONT29:%.*]]
+// CHECK13-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP95]], align 4
+// CHECK13-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 1
+// CHECK13-NEXT: store i32 2, ptr [[TMP96]], align 4
+// CHECK13-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP93]], ptr [[TMP97]], align 8
+// CHECK13-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP94]], ptr [[TMP98]], align 8
+// CHECK13-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 4
+// CHECK13-NEXT: store ptr @.offload_sizes.36, ptr [[TMP99]], align 8
+// CHECK13-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP100]], align 8
+// CHECK13-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP101]], align 8
+// CHECK13-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP102]], align 8
+// CHECK13-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 8
+// CHECK13-NEXT: store i64 10, ptr [[TMP103]], align 8
+// CHECK13-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP104]], align 8
+// CHECK13-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP105]], align 4
+// CHECK13-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP106]], align 4
+// CHECK13-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP107]], align 4
+// CHECK13-NEXT: [[TMP108:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l136.region_id, ptr [[KERNEL_ARGS27]])
+// CHECK13-NEXT: [[TMP109:%.*]] = icmp ne i32 [[TMP108]], 0
+// CHECK13-NEXT: br i1 [[TMP109]], label [[OMP_OFFLOAD_FAILED28:%.*]], label [[OMP_OFFLOAD_CONT29:%.*]]
// CHECK13: omp_offload.failed28:
-// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l136(i64 [[TMP70]], ptr [[A]]) #[[ATTR3]]
+// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l136(i64 [[TMP86]], ptr [[A]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT29]]
// CHECK13: omp_offload.cont29:
// CHECK13-NEXT: ret i32 0
@@ -6781,7 +7021,7 @@ int main (int argc, char **argv) {
// CHECK15-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK15-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK15-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK15-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK15-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK15-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK15-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -6798,298 +7038,338 @@ int main (int argc, char **argv) {
// CHECK15-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK15-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK15-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK15-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l148.region_id, ptr [[KERNEL_ARGS]])
-// CHECK15-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK15-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK15-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK15-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK15-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK15-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK15-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l148.region_id, ptr [[KERNEL_ARGS]])
+// CHECK15-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK15-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK15: omp_offload.failed:
// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l148(i32 [[TMP3]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK15: omp_offload.cont:
-// CHECK15-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK15-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK15-NEXT: [[TMP35:%.*]] = load i32, ptr [[N_CASTED3]], align 4
-// CHECK15-NEXT: [[TMP36:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK15-NEXT: [[TMP37:%.*]] = sext i32 [[TMP36]] to i64
+// CHECK15-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK15-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK15-NEXT: [[TMP39:%.*]] = load i32, ptr [[N_CASTED3]], align 4
+// CHECK15-NEXT: [[TMP40:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK15-NEXT: [[TMP41:%.*]] = sext i32 [[TMP40]] to i64
// CHECK15-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES7]], ptr align 4 @.offload_sizes.4, i32 24, i1 false)
-// CHECK15-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK15-NEXT: store i32 [[TMP35]], ptr [[TMP38]], align 4
-// CHECK15-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK15-NEXT: store i32 [[TMP35]], ptr [[TMP39]], align 4
-// CHECK15-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
-// CHECK15-NEXT: store ptr null, ptr [[TMP40]], align 4
-// CHECK15-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP41]], align 4
-// CHECK15-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP42]], align 4
-// CHECK15-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
-// CHECK15-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK15-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 4
-// CHECK15-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP45]], align 4
-// CHECK15-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK15-NEXT: store i64 [[TMP37]], ptr [[TMP46]], align 4
-// CHECK15-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK15-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK15-NEXT: store i32 [[TMP39]], ptr [[TMP42]], align 4
+// CHECK15-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK15-NEXT: store i32 [[TMP39]], ptr [[TMP43]], align 4
+// CHECK15-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
+// CHECK15-NEXT: store ptr null, ptr [[TMP44]], align 4
+// CHECK15-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP45]], align 4
+// CHECK15-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP46]], align 4
+// CHECK15-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
// CHECK15-NEXT: store ptr null, ptr [[TMP47]], align 4
-// CHECK15-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP51:%.*]] = load i32, ptr [[N]], align 4
-// CHECK15-NEXT: store i32 [[TMP51]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK15-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK15-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP52]], 0
+// CHECK15-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 4
+// CHECK15-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP49]], align 4
+// CHECK15-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK15-NEXT: store i64 [[TMP41]], ptr [[TMP50]], align 4
+// CHECK15-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK15-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK15-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP54:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP55:%.*]] = load i32, ptr [[N]], align 4
+// CHECK15-NEXT: store i32 [[TMP55]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK15-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK15-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP56]], 0
// CHECK15-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK15-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK15-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK15-NEXT: [[TMP53:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK15-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP53]], 1
-// CHECK15-NEXT: [[TMP54:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK15-NEXT: [[TMP57:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK15-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP57]], 1
+// CHECK15-NEXT: [[TMP58:%.*]] = zext i32 [[ADD14]] to i64
// CHECK15-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK15-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP55]], align 4
-// CHECK15-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK15-NEXT: store i32 3, ptr [[TMP56]], align 4
-// CHECK15-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 4
-// CHECK15-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK15-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 4
-// CHECK15-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK15-NEXT: store ptr [[TMP50]], ptr [[TMP59]], align 4
-// CHECK15-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK15-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP60]], align 4
-// CHECK15-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK15-NEXT: store ptr null, ptr [[TMP61]], align 4
-// CHECK15-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK15-NEXT: store ptr null, ptr [[TMP62]], align 4
-// CHECK15-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK15-NEXT: store i64 [[TMP54]], ptr [[TMP63]], align 8
-// CHECK15-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l153.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK15-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
-// CHECK15-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK15-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK15-NEXT: store i32 2, ptr [[TMP59]], align 4
+// CHECK15-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK15-NEXT: store i32 3, ptr [[TMP60]], align 4
+// CHECK15-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 4
+// CHECK15-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK15-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 4
+// CHECK15-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK15-NEXT: store ptr [[TMP54]], ptr [[TMP63]], align 4
+// CHECK15-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK15-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP64]], align 4
+// CHECK15-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK15-NEXT: store ptr null, ptr [[TMP65]], align 4
+// CHECK15-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK15-NEXT: store ptr null, ptr [[TMP66]], align 4
+// CHECK15-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK15-NEXT: store i64 [[TMP58]], ptr [[TMP67]], align 8
+// CHECK15-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP68]], align 8
+// CHECK15-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK15-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP70]], align 4
+// CHECK15-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP71]], align 4
+// CHECK15-NEXT: [[TMP72:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l153.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK15-NEXT: [[TMP73:%.*]] = icmp ne i32 [[TMP72]], 0
+// CHECK15-NEXT: br i1 [[TMP73]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK15: omp_offload.failed16:
-// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l153(i32 [[TMP35]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l153(i32 [[TMP39]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK15: omp_offload.cont17:
-// CHECK15-NEXT: [[TMP66:%.*]] = load i32, ptr [[M]], align 4
-// CHECK15-NEXT: store i32 [[TMP66]], ptr [[M_CASTED]], align 4
-// CHECK15-NEXT: [[TMP67:%.*]] = load i32, ptr [[M_CASTED]], align 4
-// CHECK15-NEXT: [[TMP68:%.*]] = load i32, ptr [[N]], align 4
-// CHECK15-NEXT: store i32 [[TMP68]], ptr [[N_CASTED18]], align 4
-// CHECK15-NEXT: [[TMP69:%.*]] = load i32, ptr [[N_CASTED18]], align 4
-// CHECK15-NEXT: [[TMP70:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK15-NEXT: [[TMP71:%.*]] = sext i32 [[TMP70]] to i64
+// CHECK15-NEXT: [[TMP74:%.*]] = load i32, ptr [[M]], align 4
+// CHECK15-NEXT: store i32 [[TMP74]], ptr [[M_CASTED]], align 4
+// CHECK15-NEXT: [[TMP75:%.*]] = load i32, ptr [[M_CASTED]], align 4
+// CHECK15-NEXT: [[TMP76:%.*]] = load i32, ptr [[N]], align 4
+// CHECK15-NEXT: store i32 [[TMP76]], ptr [[N_CASTED18]], align 4
+// CHECK15-NEXT: [[TMP77:%.*]] = load i32, ptr [[N_CASTED18]], align 4
+// CHECK15-NEXT: [[TMP78:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK15-NEXT: [[TMP79:%.*]] = sext i32 [[TMP78]] to i64
// CHECK15-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES22]], ptr align 4 @.offload_sizes.8, i32 32, i1 false)
-// CHECK15-NEXT: [[TMP72:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK15-NEXT: store i32 [[TMP67]], ptr [[TMP72]], align 4
-// CHECK15-NEXT: [[TMP73:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK15-NEXT: store i32 [[TMP67]], ptr [[TMP73]], align 4
-// CHECK15-NEXT: [[TMP74:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
-// CHECK15-NEXT: store ptr null, ptr [[TMP74]], align 4
-// CHECK15-NEXT: [[TMP75:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP69]], ptr [[TMP75]], align 4
-// CHECK15-NEXT: [[TMP76:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP69]], ptr [[TMP76]], align 4
-// CHECK15-NEXT: [[TMP77:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 1
-// CHECK15-NEXT: store ptr null, ptr [[TMP77]], align 4
-// CHECK15-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
-// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP78]], align 4
-// CHECK15-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
-// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP79]], align 4
-// CHECK15-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 2
-// CHECK15-NEXT: store ptr null, ptr [[TMP80]], align 4
-// CHECK15-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 3
-// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP81]], align 4
-// CHECK15-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 3
-// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP82]], align 4
-// CHECK15-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 3
-// CHECK15-NEXT: store i64 [[TMP71]], ptr [[TMP83]], align 4
-// CHECK15-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 3
-// CHECK15-NEXT: store ptr null, ptr [[TMP84]], align 4
-// CHECK15-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP88:%.*]] = load i32, ptr [[N]], align 4
-// CHECK15-NEXT: store i32 [[TMP88]], ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK15-NEXT: [[TMP89:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK15-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP89]], 0
+// CHECK15-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK15-NEXT: store i32 [[TMP75]], ptr [[TMP80]], align 4
+// CHECK15-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK15-NEXT: store i32 [[TMP75]], ptr [[TMP81]], align 4
+// CHECK15-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
+// CHECK15-NEXT: store ptr null, ptr [[TMP82]], align 4
+// CHECK15-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP77]], ptr [[TMP83]], align 4
+// CHECK15-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP77]], ptr [[TMP84]], align 4
+// CHECK15-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 1
+// CHECK15-NEXT: store ptr null, ptr [[TMP85]], align 4
+// CHECK15-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
+// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP86]], align 4
+// CHECK15-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
+// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP87]], align 4
+// CHECK15-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 2
+// CHECK15-NEXT: store ptr null, ptr [[TMP88]], align 4
+// CHECK15-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 3
+// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP89]], align 4
+// CHECK15-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 3
+// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP90]], align 4
+// CHECK15-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 3
+// CHECK15-NEXT: store i64 [[TMP79]], ptr [[TMP91]], align 4
+// CHECK15-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 3
+// CHECK15-NEXT: store ptr null, ptr [[TMP92]], align 4
+// CHECK15-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP94:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP95:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP96:%.*]] = load i32, ptr [[N]], align 4
+// CHECK15-NEXT: store i32 [[TMP96]], ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK15-NEXT: [[TMP97:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK15-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP97]], 0
// CHECK15-NEXT: [[DIV27:%.*]] = sdiv i32 [[SUB26]], 1
// CHECK15-NEXT: [[SUB28:%.*]] = sub nsw i32 [[DIV27]], 1
// CHECK15-NEXT: store i32 [[SUB28]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK15-NEXT: [[TMP90:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK15-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP90]], 1
-// CHECK15-NEXT: [[TMP91:%.*]] = zext i32 [[ADD29]] to i64
+// CHECK15-NEXT: [[TMP98:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK15-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP98]], 1
+// CHECK15-NEXT: [[TMP99:%.*]] = zext i32 [[ADD29]] to i64
// CHECK15-NEXT: [[KERNEL_ARGS30:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK15-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP92]], align 4
-// CHECK15-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
-// CHECK15-NEXT: store i32 4, ptr [[TMP93]], align 4
-// CHECK15-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[TMP85]], ptr [[TMP94]], align 4
-// CHECK15-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
-// CHECK15-NEXT: store ptr [[TMP86]], ptr [[TMP95]], align 4
-// CHECK15-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
-// CHECK15-NEXT: store ptr [[TMP87]], ptr [[TMP96]], align 4
-// CHECK15-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
-// CHECK15-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP97]], align 4
-// CHECK15-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
-// CHECK15-NEXT: store ptr null, ptr [[TMP98]], align 4
-// CHECK15-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
-// CHECK15-NEXT: store ptr null, ptr [[TMP99]], align 4
-// CHECK15-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
-// CHECK15-NEXT: store i64 [[TMP91]], ptr [[TMP100]], align 8
-// CHECK15-NEXT: [[TMP101:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l158.region_id, ptr [[KERNEL_ARGS30]])
-// CHECK15-NEXT: [[TMP102:%.*]] = icmp ne i32 [[TMP101]], 0
-// CHECK15-NEXT: br i1 [[TMP102]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
+// CHECK15-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
+// CHECK15-NEXT: store i32 2, ptr [[TMP100]], align 4
+// CHECK15-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
+// CHECK15-NEXT: store i32 4, ptr [[TMP101]], align 4
+// CHECK15-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[TMP93]], ptr [[TMP102]], align 4
+// CHECK15-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
+// CHECK15-NEXT: store ptr [[TMP94]], ptr [[TMP103]], align 4
+// CHECK15-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
+// CHECK15-NEXT: store ptr [[TMP95]], ptr [[TMP104]], align 4
+// CHECK15-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
+// CHECK15-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP105]], align 4
+// CHECK15-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
+// CHECK15-NEXT: store ptr null, ptr [[TMP106]], align 4
+// CHECK15-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
+// CHECK15-NEXT: store ptr null, ptr [[TMP107]], align 4
+// CHECK15-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
+// CHECK15-NEXT: store i64 [[TMP99]], ptr [[TMP108]], align 8
+// CHECK15-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP109]], align 8
+// CHECK15-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP110]], align 4
+// CHECK15-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP111]], align 4
+// CHECK15-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP112]], align 4
+// CHECK15-NEXT: [[TMP113:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l158.region_id, ptr [[KERNEL_ARGS30]])
+// CHECK15-NEXT: [[TMP114:%.*]] = icmp ne i32 [[TMP113]], 0
+// CHECK15-NEXT: br i1 [[TMP114]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
// CHECK15: omp_offload.failed31:
-// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l158(i32 [[TMP67]], i32 [[TMP69]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l158(i32 [[TMP75]], i32 [[TMP77]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT32]]
// CHECK15: omp_offload.cont32:
-// CHECK15-NEXT: [[TMP103:%.*]] = load i32, ptr [[N]], align 4
-// CHECK15-NEXT: store i32 [[TMP103]], ptr [[N_CASTED33]], align 4
-// CHECK15-NEXT: [[TMP104:%.*]] = load i32, ptr [[N_CASTED33]], align 4
-// CHECK15-NEXT: [[TMP105:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK15-NEXT: [[TMP106:%.*]] = sext i32 [[TMP105]] to i64
+// CHECK15-NEXT: [[TMP115:%.*]] = load i32, ptr [[N]], align 4
+// CHECK15-NEXT: store i32 [[TMP115]], ptr [[N_CASTED33]], align 4
+// CHECK15-NEXT: [[TMP116:%.*]] = load i32, ptr [[N_CASTED33]], align 4
+// CHECK15-NEXT: [[TMP117:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK15-NEXT: [[TMP118:%.*]] = sext i32 [[TMP117]] to i64
// CHECK15-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES37]], ptr align 4 @.offload_sizes.12, i32 24, i1 false)
-// CHECK15-NEXT: [[TMP107:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
-// CHECK15-NEXT: store i32 [[TMP104]], ptr [[TMP107]], align 4
-// CHECK15-NEXT: [[TMP108:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
-// CHECK15-NEXT: store i32 [[TMP104]], ptr [[TMP108]], align 4
-// CHECK15-NEXT: [[TMP109:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i32 0, i32 0
-// CHECK15-NEXT: store ptr null, ptr [[TMP109]], align 4
-// CHECK15-NEXT: [[TMP110:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP110]], align 4
-// CHECK15-NEXT: [[TMP111:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP111]], align 4
-// CHECK15-NEXT: [[TMP112:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i32 0, i32 1
-// CHECK15-NEXT: store ptr null, ptr [[TMP112]], align 4
-// CHECK15-NEXT: [[TMP113:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP113]], align 4
-// CHECK15-NEXT: [[TMP114:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP114]], align 4
-// CHECK15-NEXT: [[TMP115:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 2
-// CHECK15-NEXT: store i64 [[TMP106]], ptr [[TMP115]], align 4
-// CHECK15-NEXT: [[TMP116:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i32 0, i32 2
-// CHECK15-NEXT: store ptr null, ptr [[TMP116]], align 4
-// CHECK15-NEXT: [[TMP117:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP118:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP119:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP120:%.*]] = load i32, ptr [[N]], align 4
-// CHECK15-NEXT: store i32 [[TMP120]], ptr [[DOTCAPTURE_EXPR_39]], align 4
-// CHECK15-NEXT: [[TMP121:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_39]], align 4
-// CHECK15-NEXT: [[SUB41:%.*]] = sub nsw i32 [[TMP121]], 0
+// CHECK15-NEXT: [[TMP119:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
+// CHECK15-NEXT: store i32 [[TMP116]], ptr [[TMP119]], align 4
+// CHECK15-NEXT: [[TMP120:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
+// CHECK15-NEXT: store i32 [[TMP116]], ptr [[TMP120]], align 4
+// CHECK15-NEXT: [[TMP121:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i32 0, i32 0
+// CHECK15-NEXT: store ptr null, ptr [[TMP121]], align 4
+// CHECK15-NEXT: [[TMP122:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP122]], align 4
+// CHECK15-NEXT: [[TMP123:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP123]], align 4
+// CHECK15-NEXT: [[TMP124:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i32 0, i32 1
+// CHECK15-NEXT: store ptr null, ptr [[TMP124]], align 4
+// CHECK15-NEXT: [[TMP125:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP125]], align 4
+// CHECK15-NEXT: [[TMP126:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP126]], align 4
+// CHECK15-NEXT: [[TMP127:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 2
+// CHECK15-NEXT: store i64 [[TMP118]], ptr [[TMP127]], align 4
+// CHECK15-NEXT: [[TMP128:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i32 0, i32 2
+// CHECK15-NEXT: store ptr null, ptr [[TMP128]], align 4
+// CHECK15-NEXT: [[TMP129:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP130:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP131:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP132:%.*]] = load i32, ptr [[N]], align 4
+// CHECK15-NEXT: store i32 [[TMP132]], ptr [[DOTCAPTURE_EXPR_39]], align 4
+// CHECK15-NEXT: [[TMP133:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_39]], align 4
+// CHECK15-NEXT: [[SUB41:%.*]] = sub nsw i32 [[TMP133]], 0
// CHECK15-NEXT: [[DIV42:%.*]] = sdiv i32 [[SUB41]], 1
// CHECK15-NEXT: [[SUB43:%.*]] = sub nsw i32 [[DIV42]], 1
// CHECK15-NEXT: store i32 [[SUB43]], ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK15-NEXT: [[TMP122:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK15-NEXT: [[ADD44:%.*]] = add nsw i32 [[TMP122]], 1
-// CHECK15-NEXT: [[TMP123:%.*]] = zext i32 [[ADD44]] to i64
+// CHECK15-NEXT: [[TMP134:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
+// CHECK15-NEXT: [[ADD44:%.*]] = add nsw i32 [[TMP134]], 1
+// CHECK15-NEXT: [[TMP135:%.*]] = zext i32 [[ADD44]] to i64
// CHECK15-NEXT: [[KERNEL_ARGS45:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK15-NEXT: [[TMP124:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP124]], align 4
-// CHECK15-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 1
-// CHECK15-NEXT: store i32 3, ptr [[TMP125]], align 4
-// CHECK15-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[TMP117]], ptr [[TMP126]], align 4
-// CHECK15-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 3
-// CHECK15-NEXT: store ptr [[TMP118]], ptr [[TMP127]], align 4
-// CHECK15-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 4
-// CHECK15-NEXT: store ptr [[TMP119]], ptr [[TMP128]], align 4
-// CHECK15-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 5
-// CHECK15-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP129]], align 4
-// CHECK15-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 6
-// CHECK15-NEXT: store ptr null, ptr [[TMP130]], align 4
-// CHECK15-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 7
-// CHECK15-NEXT: store ptr null, ptr [[TMP131]], align 4
-// CHECK15-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 8
-// CHECK15-NEXT: store i64 [[TMP123]], ptr [[TMP132]], align 8
-// CHECK15-NEXT: [[TMP133:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l163.region_id, ptr [[KERNEL_ARGS45]])
-// CHECK15-NEXT: [[TMP134:%.*]] = icmp ne i32 [[TMP133]], 0
-// CHECK15-NEXT: br i1 [[TMP134]], label [[OMP_OFFLOAD_FAILED46:%.*]], label [[OMP_OFFLOAD_CONT47:%.*]]
+// CHECK15-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 0
+// CHECK15-NEXT: store i32 2, ptr [[TMP136]], align 4
+// CHECK15-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 1
+// CHECK15-NEXT: store i32 3, ptr [[TMP137]], align 4
+// CHECK15-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[TMP129]], ptr [[TMP138]], align 4
+// CHECK15-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 3
+// CHECK15-NEXT: store ptr [[TMP130]], ptr [[TMP139]], align 4
+// CHECK15-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 4
+// CHECK15-NEXT: store ptr [[TMP131]], ptr [[TMP140]], align 4
+// CHECK15-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 5
+// CHECK15-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP141]], align 4
+// CHECK15-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 6
+// CHECK15-NEXT: store ptr null, ptr [[TMP142]], align 4
+// CHECK15-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 7
+// CHECK15-NEXT: store ptr null, ptr [[TMP143]], align 4
+// CHECK15-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 8
+// CHECK15-NEXT: store i64 [[TMP135]], ptr [[TMP144]], align 8
+// CHECK15-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP145]], align 8
+// CHECK15-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP146]], align 4
+// CHECK15-NEXT: [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP147]], align 4
+// CHECK15-NEXT: [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP148]], align 4
+// CHECK15-NEXT: [[TMP149:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l163.region_id, ptr [[KERNEL_ARGS45]])
+// CHECK15-NEXT: [[TMP150:%.*]] = icmp ne i32 [[TMP149]], 0
+// CHECK15-NEXT: br i1 [[TMP150]], label [[OMP_OFFLOAD_FAILED46:%.*]], label [[OMP_OFFLOAD_CONT47:%.*]]
// CHECK15: omp_offload.failed46:
-// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l163(i32 [[TMP104]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l163(i32 [[TMP116]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT47]]
// CHECK15: omp_offload.cont47:
-// CHECK15-NEXT: [[TMP135:%.*]] = load i32, ptr [[M]], align 4
-// CHECK15-NEXT: store i32 [[TMP135]], ptr [[M_CASTED48]], align 4
-// CHECK15-NEXT: [[TMP136:%.*]] = load i32, ptr [[M_CASTED48]], align 4
-// CHECK15-NEXT: [[TMP137:%.*]] = load i32, ptr [[N]], align 4
-// CHECK15-NEXT: store i32 [[TMP137]], ptr [[N_CASTED49]], align 4
-// CHECK15-NEXT: [[TMP138:%.*]] = load i32, ptr [[N_CASTED49]], align 4
-// CHECK15-NEXT: [[TMP139:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK15-NEXT: [[TMP140:%.*]] = sext i32 [[TMP139]] to i64
+// CHECK15-NEXT: [[TMP151:%.*]] = load i32, ptr [[M]], align 4
+// CHECK15-NEXT: store i32 [[TMP151]], ptr [[M_CASTED48]], align 4
+// CHECK15-NEXT: [[TMP152:%.*]] = load i32, ptr [[M_CASTED48]], align 4
+// CHECK15-NEXT: [[TMP153:%.*]] = load i32, ptr [[N]], align 4
+// CHECK15-NEXT: store i32 [[TMP153]], ptr [[N_CASTED49]], align 4
+// CHECK15-NEXT: [[TMP154:%.*]] = load i32, ptr [[N_CASTED49]], align 4
+// CHECK15-NEXT: [[TMP155:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK15-NEXT: [[TMP156:%.*]] = sext i32 [[TMP155]] to i64
// CHECK15-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES53]], ptr align 4 @.offload_sizes.16, i32 32, i1 false)
-// CHECK15-NEXT: [[TMP141:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
-// CHECK15-NEXT: store i32 [[TMP136]], ptr [[TMP141]], align 4
-// CHECK15-NEXT: [[TMP142:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
-// CHECK15-NEXT: store i32 [[TMP136]], ptr [[TMP142]], align 4
-// CHECK15-NEXT: [[TMP143:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 0
-// CHECK15-NEXT: store ptr null, ptr [[TMP143]], align 4
-// CHECK15-NEXT: [[TMP144:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP138]], ptr [[TMP144]], align 4
-// CHECK15-NEXT: [[TMP145:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 1
-// CHECK15-NEXT: store i32 [[TMP138]], ptr [[TMP145]], align 4
-// CHECK15-NEXT: [[TMP146:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 1
-// CHECK15-NEXT: store ptr null, ptr [[TMP146]], align 4
-// CHECK15-NEXT: [[TMP147:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 2
-// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP147]], align 4
-// CHECK15-NEXT: [[TMP148:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 2
-// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP148]], align 4
-// CHECK15-NEXT: [[TMP149:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 2
-// CHECK15-NEXT: store ptr null, ptr [[TMP149]], align 4
-// CHECK15-NEXT: [[TMP150:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 3
-// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP150]], align 4
-// CHECK15-NEXT: [[TMP151:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 3
-// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP151]], align 4
-// CHECK15-NEXT: [[TMP152:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 3
-// CHECK15-NEXT: store i64 [[TMP140]], ptr [[TMP152]], align 4
-// CHECK15-NEXT: [[TMP153:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 3
-// CHECK15-NEXT: store ptr null, ptr [[TMP153]], align 4
-// CHECK15-NEXT: [[TMP154:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP155:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP156:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP157:%.*]] = load i32, ptr [[N]], align 4
-// CHECK15-NEXT: store i32 [[TMP157]], ptr [[DOTCAPTURE_EXPR_55]], align 4
-// CHECK15-NEXT: [[TMP158:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_55]], align 4
-// CHECK15-NEXT: [[SUB57:%.*]] = sub nsw i32 [[TMP158]], 0
+// CHECK15-NEXT: [[TMP157:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
+// CHECK15-NEXT: store i32 [[TMP152]], ptr [[TMP157]], align 4
+// CHECK15-NEXT: [[TMP158:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
+// CHECK15-NEXT: store i32 [[TMP152]], ptr [[TMP158]], align 4
+// CHECK15-NEXT: [[TMP159:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 0
+// CHECK15-NEXT: store ptr null, ptr [[TMP159]], align 4
+// CHECK15-NEXT: [[TMP160:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP154]], ptr [[TMP160]], align 4
+// CHECK15-NEXT: [[TMP161:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 1
+// CHECK15-NEXT: store i32 [[TMP154]], ptr [[TMP161]], align 4
+// CHECK15-NEXT: [[TMP162:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 1
+// CHECK15-NEXT: store ptr null, ptr [[TMP162]], align 4
+// CHECK15-NEXT: [[TMP163:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 2
+// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP163]], align 4
+// CHECK15-NEXT: [[TMP164:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 2
+// CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP164]], align 4
+// CHECK15-NEXT: [[TMP165:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 2
+// CHECK15-NEXT: store ptr null, ptr [[TMP165]], align 4
+// CHECK15-NEXT: [[TMP166:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 3
+// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP166]], align 4
+// CHECK15-NEXT: [[TMP167:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 3
+// CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP167]], align 4
+// CHECK15-NEXT: [[TMP168:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 3
+// CHECK15-NEXT: store i64 [[TMP156]], ptr [[TMP168]], align 4
+// CHECK15-NEXT: [[TMP169:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 3
+// CHECK15-NEXT: store ptr null, ptr [[TMP169]], align 4
+// CHECK15-NEXT: [[TMP170:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP171:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP172:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP173:%.*]] = load i32, ptr [[N]], align 4
+// CHECK15-NEXT: store i32 [[TMP173]], ptr [[DOTCAPTURE_EXPR_55]], align 4
+// CHECK15-NEXT: [[TMP174:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_55]], align 4
+// CHECK15-NEXT: [[SUB57:%.*]] = sub nsw i32 [[TMP174]], 0
// CHECK15-NEXT: [[DIV58:%.*]] = sdiv i32 [[SUB57]], 1
// CHECK15-NEXT: [[SUB59:%.*]] = sub nsw i32 [[DIV58]], 1
// CHECK15-NEXT: store i32 [[SUB59]], ptr [[DOTCAPTURE_EXPR_56]], align 4
-// CHECK15-NEXT: [[TMP159:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_56]], align 4
-// CHECK15-NEXT: [[ADD60:%.*]] = add nsw i32 [[TMP159]], 1
-// CHECK15-NEXT: [[TMP160:%.*]] = zext i32 [[ADD60]] to i64
+// CHECK15-NEXT: [[TMP175:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_56]], align 4
+// CHECK15-NEXT: [[ADD60:%.*]] = add nsw i32 [[TMP175]], 1
+// CHECK15-NEXT: [[TMP176:%.*]] = zext i32 [[ADD60]] to i64
// CHECK15-NEXT: [[KERNEL_ARGS61:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK15-NEXT: [[TMP161:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP161]], align 4
-// CHECK15-NEXT: [[TMP162:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 1
-// CHECK15-NEXT: store i32 4, ptr [[TMP162]], align 4
-// CHECK15-NEXT: [[TMP163:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[TMP154]], ptr [[TMP163]], align 4
-// CHECK15-NEXT: [[TMP164:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 3
-// CHECK15-NEXT: store ptr [[TMP155]], ptr [[TMP164]], align 4
-// CHECK15-NEXT: [[TMP165:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 4
-// CHECK15-NEXT: store ptr [[TMP156]], ptr [[TMP165]], align 4
-// CHECK15-NEXT: [[TMP166:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 5
-// CHECK15-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP166]], align 4
-// CHECK15-NEXT: [[TMP167:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 6
-// CHECK15-NEXT: store ptr null, ptr [[TMP167]], align 4
-// CHECK15-NEXT: [[TMP168:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 7
-// CHECK15-NEXT: store ptr null, ptr [[TMP168]], align 4
-// CHECK15-NEXT: [[TMP169:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 8
-// CHECK15-NEXT: store i64 [[TMP160]], ptr [[TMP169]], align 8
-// CHECK15-NEXT: [[TMP170:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l168.region_id, ptr [[KERNEL_ARGS61]])
-// CHECK15-NEXT: [[TMP171:%.*]] = icmp ne i32 [[TMP170]], 0
-// CHECK15-NEXT: br i1 [[TMP171]], label [[OMP_OFFLOAD_FAILED62:%.*]], label [[OMP_OFFLOAD_CONT63:%.*]]
+// CHECK15-NEXT: [[TMP177:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 0
+// CHECK15-NEXT: store i32 2, ptr [[TMP177]], align 4
+// CHECK15-NEXT: [[TMP178:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 1
+// CHECK15-NEXT: store i32 4, ptr [[TMP178]], align 4
+// CHECK15-NEXT: [[TMP179:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[TMP170]], ptr [[TMP179]], align 4
+// CHECK15-NEXT: [[TMP180:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 3
+// CHECK15-NEXT: store ptr [[TMP171]], ptr [[TMP180]], align 4
+// CHECK15-NEXT: [[TMP181:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 4
+// CHECK15-NEXT: store ptr [[TMP172]], ptr [[TMP181]], align 4
+// CHECK15-NEXT: [[TMP182:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 5
+// CHECK15-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP182]], align 4
+// CHECK15-NEXT: [[TMP183:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 6
+// CHECK15-NEXT: store ptr null, ptr [[TMP183]], align 4
+// CHECK15-NEXT: [[TMP184:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 7
+// CHECK15-NEXT: store ptr null, ptr [[TMP184]], align 4
+// CHECK15-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 8
+// CHECK15-NEXT: store i64 [[TMP176]], ptr [[TMP185]], align 8
+// CHECK15-NEXT: [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP186]], align 8
+// CHECK15-NEXT: [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP187]], align 4
+// CHECK15-NEXT: [[TMP188:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP188]], align 4
+// CHECK15-NEXT: [[TMP189:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP189]], align 4
+// CHECK15-NEXT: [[TMP190:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l168.region_id, ptr [[KERNEL_ARGS61]])
+// CHECK15-NEXT: [[TMP191:%.*]] = icmp ne i32 [[TMP190]], 0
+// CHECK15-NEXT: br i1 [[TMP191]], label [[OMP_OFFLOAD_FAILED62:%.*]], label [[OMP_OFFLOAD_CONT63:%.*]]
// CHECK15: omp_offload.failed62:
-// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l168(i32 [[TMP136]], i32 [[TMP138]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l168(i32 [[TMP152]], i32 [[TMP154]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT63]]
// CHECK15: omp_offload.cont63:
-// CHECK15-NEXT: [[TMP172:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK15-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP172]])
+// CHECK15-NEXT: [[TMP192:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK15-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP192]])
// CHECK15-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK15-NEXT: [[TMP173:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK15-NEXT: call void @llvm.stackrestore(ptr [[TMP173]])
-// CHECK15-NEXT: [[TMP174:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK15-NEXT: ret i32 [[TMP174]]
+// CHECK15-NEXT: [[TMP193:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK15-NEXT: call void @llvm.stackrestore(ptr [[TMP193]])
+// CHECK15-NEXT: [[TMP194:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK15-NEXT: ret i32 [[TMP194]]
//
//
// CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l148
@@ -8214,7 +8494,7 @@ int main (int argc, char **argv) {
// CHECK15-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK15-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK15-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK15-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK15-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK15-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK15-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -8231,165 +8511,205 @@ int main (int argc, char **argv) {
// CHECK15-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK15-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK15-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK15-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.region_id, ptr [[KERNEL_ARGS]])
-// CHECK15-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK15-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK15-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK15-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK15-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK15-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK15-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.region_id, ptr [[KERNEL_ARGS]])
+// CHECK15-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK15-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK15: omp_offload.failed:
// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116(ptr [[A]]) #[[ATTR3]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK15: omp_offload.cont:
-// CHECK15-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK15-NEXT: store ptr [[A]], ptr [[TMP16]], align 4
-// CHECK15-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK15-NEXT: store ptr [[A]], ptr [[TMP17]], align 4
-// CHECK15-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
-// CHECK15-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK15-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK15-NEXT: store ptr [[A]], ptr [[TMP20]], align 4
+// CHECK15-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK15-NEXT: store ptr [[A]], ptr [[TMP21]], align 4
+// CHECK15-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
+// CHECK15-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK15-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK15-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK15-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK15-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK15-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK15-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK15-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK15-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK15-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK15-NEXT: store ptr @.offload_sizes.24, ptr [[TMP25]], align 4
-// CHECK15-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK15-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP26]], align 4
-// CHECK15-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK15-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK15-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK15-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK15-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK15-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK15-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l121.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK15-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK15-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK15-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK15-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK15-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK15-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK15-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK15-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK15-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK15-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK15-NEXT: store ptr @.offload_sizes.24, ptr [[TMP29]], align 4
+// CHECK15-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK15-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP30]], align 4
+// CHECK15-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK15-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK15-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK15-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK15-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK15-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK15-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK15-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK15-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK15-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK15-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l121.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK15-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK15-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK15: omp_offload.failed6:
// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l121(ptr [[A]]) #[[ATTR3]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK15: omp_offload.cont7:
-// CHECK15-NEXT: [[TMP32:%.*]] = load i32, ptr [[M]], align 4
-// CHECK15-NEXT: store i32 [[TMP32]], ptr [[M_CASTED]], align 4
-// CHECK15-NEXT: [[TMP33:%.*]] = load i32, ptr [[M_CASTED]], align 4
-// CHECK15-NEXT: [[TMP34:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK15-NEXT: store i32 [[TMP33]], ptr [[TMP34]], align 4
-// CHECK15-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK15-NEXT: store i32 [[TMP33]], ptr [[TMP35]], align 4
-// CHECK15-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
-// CHECK15-NEXT: store ptr null, ptr [[TMP36]], align 4
-// CHECK15-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
-// CHECK15-NEXT: store ptr [[A]], ptr [[TMP37]], align 4
-// CHECK15-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
-// CHECK15-NEXT: store ptr [[A]], ptr [[TMP38]], align 4
-// CHECK15-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1
-// CHECK15-NEXT: store ptr null, ptr [[TMP39]], align 4
-// CHECK15-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP40:%.*]] = load i32, ptr [[M]], align 4
+// CHECK15-NEXT: store i32 [[TMP40]], ptr [[M_CASTED]], align 4
+// CHECK15-NEXT: [[TMP41:%.*]] = load i32, ptr [[M_CASTED]], align 4
+// CHECK15-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK15-NEXT: store i32 [[TMP41]], ptr [[TMP42]], align 4
+// CHECK15-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK15-NEXT: store i32 [[TMP41]], ptr [[TMP43]], align 4
+// CHECK15-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
+// CHECK15-NEXT: store ptr null, ptr [[TMP44]], align 4
+// CHECK15-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
+// CHECK15-NEXT: store ptr [[A]], ptr [[TMP45]], align 4
+// CHECK15-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
+// CHECK15-NEXT: store ptr [[A]], ptr [[TMP46]], align 4
+// CHECK15-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1
+// CHECK15-NEXT: store ptr null, ptr [[TMP47]], align 4
+// CHECK15-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK15-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK15-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP42]], align 4
-// CHECK15-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK15-NEXT: store i32 2, ptr [[TMP43]], align 4
-// CHECK15-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[TMP40]], ptr [[TMP44]], align 4
-// CHECK15-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK15-NEXT: store ptr [[TMP41]], ptr [[TMP45]], align 4
-// CHECK15-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK15-NEXT: store ptr @.offload_sizes.28, ptr [[TMP46]], align 4
-// CHECK15-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK15-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP47]], align 4
-// CHECK15-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK15-NEXT: store ptr null, ptr [[TMP48]], align 4
-// CHECK15-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK15-NEXT: store ptr null, ptr [[TMP49]], align 4
-// CHECK15-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK15-NEXT: store i64 10, ptr [[TMP50]], align 8
-// CHECK15-NEXT: [[TMP51:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l126.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK15-NEXT: [[TMP52:%.*]] = icmp ne i32 [[TMP51]], 0
-// CHECK15-NEXT: br i1 [[TMP52]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK15-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK15-NEXT: store i32 2, ptr [[TMP50]], align 4
+// CHECK15-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK15-NEXT: store i32 2, ptr [[TMP51]], align 4
+// CHECK15-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[TMP48]], ptr [[TMP52]], align 4
+// CHECK15-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK15-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 4
+// CHECK15-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK15-NEXT: store ptr @.offload_sizes.28, ptr [[TMP54]], align 4
+// CHECK15-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK15-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP55]], align 4
+// CHECK15-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK15-NEXT: store ptr null, ptr [[TMP56]], align 4
+// CHECK15-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK15-NEXT: store ptr null, ptr [[TMP57]], align 4
+// CHECK15-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK15-NEXT: store i64 10, ptr [[TMP58]], align 8
+// CHECK15-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP59]], align 8
+// CHECK15-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP60]], align 4
+// CHECK15-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4
+// CHECK15-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP62]], align 4
+// CHECK15-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l126.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK15-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
+// CHECK15-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK15: omp_offload.failed13:
-// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l126(i32 [[TMP33]], ptr [[A]]) #[[ATTR3]]
+// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l126(i32 [[TMP41]], ptr [[A]]) #[[ATTR3]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT14]]
// CHECK15: omp_offload.cont14:
-// CHECK15-NEXT: [[TMP53:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK15-NEXT: store ptr [[A]], ptr [[TMP53]], align 4
-// CHECK15-NEXT: [[TMP54:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
-// CHECK15-NEXT: store ptr [[A]], ptr [[TMP54]], align 4
-// CHECK15-NEXT: [[TMP55:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i32 0, i32 0
-// CHECK15-NEXT: store ptr null, ptr [[TMP55]], align 4
-// CHECK15-NEXT: [[TMP56:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP57:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP65:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK15-NEXT: store ptr [[A]], ptr [[TMP65]], align 4
+// CHECK15-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK15-NEXT: store ptr [[A]], ptr [[TMP66]], align 4
+// CHECK15-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i32 0, i32 0
+// CHECK15-NEXT: store ptr null, ptr [[TMP67]], align 4
+// CHECK15-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP69:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
// CHECK15-NEXT: [[KERNEL_ARGS19:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK15-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP58]], align 4
-// CHECK15-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
-// CHECK15-NEXT: store i32 1, ptr [[TMP59]], align 4
-// CHECK15-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[TMP56]], ptr [[TMP60]], align 4
-// CHECK15-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
-// CHECK15-NEXT: store ptr [[TMP57]], ptr [[TMP61]], align 4
-// CHECK15-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
-// CHECK15-NEXT: store ptr @.offload_sizes.32, ptr [[TMP62]], align 4
-// CHECK15-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
-// CHECK15-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP63]], align 4
-// CHECK15-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
-// CHECK15-NEXT: store ptr null, ptr [[TMP64]], align 4
-// CHECK15-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
-// CHECK15-NEXT: store ptr null, ptr [[TMP65]], align 4
-// CHECK15-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
-// CHECK15-NEXT: store i64 10, ptr [[TMP66]], align 8
-// CHECK15-NEXT: [[TMP67:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l131.region_id, ptr [[KERNEL_ARGS19]])
-// CHECK15-NEXT: [[TMP68:%.*]] = icmp ne i32 [[TMP67]], 0
-// CHECK15-NEXT: br i1 [[TMP68]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
+// CHECK15-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
+// CHECK15-NEXT: store i32 2, ptr [[TMP70]], align 4
+// CHECK15-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
+// CHECK15-NEXT: store i32 1, ptr [[TMP71]], align 4
+// CHECK15-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 4
+// CHECK15-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
+// CHECK15-NEXT: store ptr [[TMP69]], ptr [[TMP73]], align 4
+// CHECK15-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
+// CHECK15-NEXT: store ptr @.offload_sizes.32, ptr [[TMP74]], align 4
+// CHECK15-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
+// CHECK15-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP75]], align 4
+// CHECK15-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
+// CHECK15-NEXT: store ptr null, ptr [[TMP76]], align 4
+// CHECK15-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
+// CHECK15-NEXT: store ptr null, ptr [[TMP77]], align 4
+// CHECK15-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
+// CHECK15-NEXT: store i64 10, ptr [[TMP78]], align 8
+// CHECK15-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP79]], align 8
+// CHECK15-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP80]], align 4
+// CHECK15-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP81]], align 4
+// CHECK15-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP82]], align 4
+// CHECK15-NEXT: [[TMP83:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l131.region_id, ptr [[KERNEL_ARGS19]])
+// CHECK15-NEXT: [[TMP84:%.*]] = icmp ne i32 [[TMP83]], 0
+// CHECK15-NEXT: br i1 [[TMP84]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
// CHECK15: omp_offload.failed20:
// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l131(ptr [[A]]) #[[ATTR3]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT21]]
// CHECK15: omp_offload.cont21:
-// CHECK15-NEXT: [[TMP69:%.*]] = load i32, ptr [[M]], align 4
-// CHECK15-NEXT: store i32 [[TMP69]], ptr [[M_CASTED22]], align 4
-// CHECK15-NEXT: [[TMP70:%.*]] = load i32, ptr [[M_CASTED22]], align 4
-// CHECK15-NEXT: [[TMP71:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
-// CHECK15-NEXT: store i32 [[TMP70]], ptr [[TMP71]], align 4
-// CHECK15-NEXT: [[TMP72:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
-// CHECK15-NEXT: store i32 [[TMP70]], ptr [[TMP72]], align 4
-// CHECK15-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i32 0, i32 0
-// CHECK15-NEXT: store ptr null, ptr [[TMP73]], align 4
-// CHECK15-NEXT: [[TMP74:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 1
-// CHECK15-NEXT: store ptr [[A]], ptr [[TMP74]], align 4
-// CHECK15-NEXT: [[TMP75:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 1
-// CHECK15-NEXT: store ptr [[A]], ptr [[TMP75]], align 4
-// CHECK15-NEXT: [[TMP76:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i32 0, i32 1
-// CHECK15-NEXT: store ptr null, ptr [[TMP76]], align 4
-// CHECK15-NEXT: [[TMP77:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
-// CHECK15-NEXT: [[TMP78:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP85:%.*]] = load i32, ptr [[M]], align 4
+// CHECK15-NEXT: store i32 [[TMP85]], ptr [[M_CASTED22]], align 4
+// CHECK15-NEXT: [[TMP86:%.*]] = load i32, ptr [[M_CASTED22]], align 4
+// CHECK15-NEXT: [[TMP87:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
+// CHECK15-NEXT: store i32 [[TMP86]], ptr [[TMP87]], align 4
+// CHECK15-NEXT: [[TMP88:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
+// CHECK15-NEXT: store i32 [[TMP86]], ptr [[TMP88]], align 4
+// CHECK15-NEXT: [[TMP89:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i32 0, i32 0
+// CHECK15-NEXT: store ptr null, ptr [[TMP89]], align 4
+// CHECK15-NEXT: [[TMP90:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 1
+// CHECK15-NEXT: store ptr [[A]], ptr [[TMP90]], align 4
+// CHECK15-NEXT: [[TMP91:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 1
+// CHECK15-NEXT: store ptr [[A]], ptr [[TMP91]], align 4
+// CHECK15-NEXT: [[TMP92:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i32 0, i32 1
+// CHECK15-NEXT: store ptr null, ptr [[TMP92]], align 4
+// CHECK15-NEXT: [[TMP93:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
+// CHECK15-NEXT: [[TMP94:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
// CHECK15-NEXT: [[KERNEL_ARGS27:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK15-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 0
-// CHECK15-NEXT: store i32 1, ptr [[TMP79]], align 4
-// CHECK15-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 1
-// CHECK15-NEXT: store i32 2, ptr [[TMP80]], align 4
-// CHECK15-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 2
-// CHECK15-NEXT: store ptr [[TMP77]], ptr [[TMP81]], align 4
-// CHECK15-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 3
-// CHECK15-NEXT: store ptr [[TMP78]], ptr [[TMP82]], align 4
-// CHECK15-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 4
-// CHECK15-NEXT: store ptr @.offload_sizes.36, ptr [[TMP83]], align 4
-// CHECK15-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 5
-// CHECK15-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP84]], align 4
-// CHECK15-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 6
-// CHECK15-NEXT: store ptr null, ptr [[TMP85]], align 4
-// CHECK15-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 7
-// CHECK15-NEXT: store ptr null, ptr [[TMP86]], align 4
-// CHECK15-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 8
-// CHECK15-NEXT: store i64 10, ptr [[TMP87]], align 8
-// CHECK15-NEXT: [[TMP88:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l136.region_id, ptr [[KERNEL_ARGS27]])
-// CHECK15-NEXT: [[TMP89:%.*]] = icmp ne i32 [[TMP88]], 0
-// CHECK15-NEXT: br i1 [[TMP89]], label [[OMP_OFFLOAD_FAILED28:%.*]], label [[OMP_OFFLOAD_CONT29:%.*]]
+// CHECK15-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 0
+// CHECK15-NEXT: store i32 2, ptr [[TMP95]], align 4
+// CHECK15-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 1
+// CHECK15-NEXT: store i32 2, ptr [[TMP96]], align 4
+// CHECK15-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 2
+// CHECK15-NEXT: store ptr [[TMP93]], ptr [[TMP97]], align 4
+// CHECK15-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 3
+// CHECK15-NEXT: store ptr [[TMP94]], ptr [[TMP98]], align 4
+// CHECK15-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 4
+// CHECK15-NEXT: store ptr @.offload_sizes.36, ptr [[TMP99]], align 4
+// CHECK15-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 5
+// CHECK15-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP100]], align 4
+// CHECK15-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 6
+// CHECK15-NEXT: store ptr null, ptr [[TMP101]], align 4
+// CHECK15-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 7
+// CHECK15-NEXT: store ptr null, ptr [[TMP102]], align 4
+// CHECK15-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 8
+// CHECK15-NEXT: store i64 10, ptr [[TMP103]], align 8
+// CHECK15-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 9
+// CHECK15-NEXT: store i64 0, ptr [[TMP104]], align 8
+// CHECK15-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 10
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP105]], align 4
+// CHECK15-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 11
+// CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP106]], align 4
+// CHECK15-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 12
+// CHECK15-NEXT: store i32 0, ptr [[TMP107]], align 4
+// CHECK15-NEXT: [[TMP108:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l136.region_id, ptr [[KERNEL_ARGS27]])
+// CHECK15-NEXT: [[TMP109:%.*]] = icmp ne i32 [[TMP108]], 0
+// CHECK15-NEXT: br i1 [[TMP109]], label [[OMP_OFFLOAD_FAILED28:%.*]], label [[OMP_OFFLOAD_CONT29:%.*]]
// CHECK15: omp_offload.failed28:
-// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l136(i32 [[TMP70]], ptr [[A]]) #[[ATTR3]]
+// CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l136(i32 [[TMP86]], ptr [[A]]) #[[ATTR3]]
// CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT29]]
// CHECK15: omp_offload.cont29:
// CHECK15-NEXT: ret i32 0
@@ -9281,7 +9601,7 @@ int main (int argc, char **argv) {
// CHECK17-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK17-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK17-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK17-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK17-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK17-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK17-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -9298,294 +9618,334 @@ int main (int argc, char **argv) {
// CHECK17-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK17-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK17-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK17-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l148.region_id, ptr [[KERNEL_ARGS]])
-// CHECK17-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK17-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK17-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK17-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK17-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK17-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK17-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l148.region_id, ptr [[KERNEL_ARGS]])
+// CHECK17-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK17-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK17: omp_offload.failed:
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l148(i64 [[TMP4]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK17: omp_offload.cont:
-// CHECK17-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK17-NEXT: [[TMP35:%.*]] = load i64, ptr [[N_CASTED3]], align 8
-// CHECK17-NEXT: [[TMP36:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK17-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK17-NEXT: [[TMP39:%.*]] = load i64, ptr [[N_CASTED3]], align 8
+// CHECK17-NEXT: [[TMP40:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES7]], ptr align 8 @.offload_sizes.4, i64 24, i1 false)
-// CHECK17-NEXT: [[TMP37:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP35]], ptr [[TMP37]], align 8
-// CHECK17-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP35]], ptr [[TMP38]], align 8
-// CHECK17-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK17-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP40]], align 8
-// CHECK17-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP41]], align 8
-// CHECK17-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
-// CHECK17-NEXT: store ptr null, ptr [[TMP42]], align 8
-// CHECK17-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP43]], align 8
-// CHECK17-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 8
-// CHECK17-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK17-NEXT: store i64 [[TMP36]], ptr [[TMP45]], align 8
-// CHECK17-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK17-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP39]], ptr [[TMP41]], align 8
+// CHECK17-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP39]], ptr [[TMP42]], align 8
+// CHECK17-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK17-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP44]], align 8
+// CHECK17-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP45]], align 8
+// CHECK17-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
// CHECK17-NEXT: store ptr null, ptr [[TMP46]], align 8
-// CHECK17-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP50:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP50]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK17-NEXT: [[TMP51:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK17-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP51]], 0
+// CHECK17-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP47]], align 8
+// CHECK17-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 8
+// CHECK17-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK17-NEXT: store i64 [[TMP40]], ptr [[TMP49]], align 8
+// CHECK17-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK17-NEXT: store ptr null, ptr [[TMP50]], align 8
+// CHECK17-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP54:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP54]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK17-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK17-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP55]], 0
// CHECK17-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK17-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK17-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK17-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK17-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP52]], 1
-// CHECK17-NEXT: [[TMP53:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK17-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK17-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP56]], 1
+// CHECK17-NEXT: [[TMP57:%.*]] = zext i32 [[ADD14]] to i64
// CHECK17-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK17-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK17-NEXT: store i32 3, ptr [[TMP55]], align 4
-// CHECK17-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP47]], ptr [[TMP56]], align 8
-// CHECK17-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 8
-// CHECK17-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK17-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 8
-// CHECK17-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP59]], align 8
-// CHECK17-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP60]], align 8
-// CHECK17-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP61]], align 8
-// CHECK17-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK17-NEXT: store i64 [[TMP53]], ptr [[TMP62]], align 8
-// CHECK17-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l153.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK17-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
-// CHECK17-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK17-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP58]], align 4
+// CHECK17-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK17-NEXT: store i32 3, ptr [[TMP59]], align 4
+// CHECK17-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP51]], ptr [[TMP60]], align 8
+// CHECK17-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 8
+// CHECK17-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK17-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 8
+// CHECK17-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP63]], align 8
+// CHECK17-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP64]], align 8
+// CHECK17-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP65]], align 8
+// CHECK17-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK17-NEXT: store i64 [[TMP57]], ptr [[TMP66]], align 8
+// CHECK17-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP67]], align 8
+// CHECK17-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP68]], align 4
+// CHECK17-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK17-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP70]], align 4
+// CHECK17-NEXT: [[TMP71:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l153.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK17-NEXT: [[TMP72:%.*]] = icmp ne i32 [[TMP71]], 0
+// CHECK17-NEXT: br i1 [[TMP72]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK17: omp_offload.failed16:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l153(i64 [[TMP35]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l153(i64 [[TMP39]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK17: omp_offload.cont17:
-// CHECK17-NEXT: [[TMP65:%.*]] = load i32, ptr [[M]], align 4
-// CHECK17-NEXT: store i32 [[TMP65]], ptr [[M_CASTED]], align 4
-// CHECK17-NEXT: [[TMP66:%.*]] = load i64, ptr [[M_CASTED]], align 8
-// CHECK17-NEXT: [[TMP67:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP67]], ptr [[N_CASTED18]], align 4
-// CHECK17-NEXT: [[TMP68:%.*]] = load i64, ptr [[N_CASTED18]], align 8
-// CHECK17-NEXT: [[TMP69:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK17-NEXT: [[TMP73:%.*]] = load i32, ptr [[M]], align 4
+// CHECK17-NEXT: store i32 [[TMP73]], ptr [[M_CASTED]], align 4
+// CHECK17-NEXT: [[TMP74:%.*]] = load i64, ptr [[M_CASTED]], align 8
+// CHECK17-NEXT: [[TMP75:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP75]], ptr [[N_CASTED18]], align 4
+// CHECK17-NEXT: [[TMP76:%.*]] = load i64, ptr [[N_CASTED18]], align 8
+// CHECK17-NEXT: [[TMP77:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES22]], ptr align 8 @.offload_sizes.8, i64 32, i1 false)
-// CHECK17-NEXT: [[TMP70:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP66]], ptr [[TMP70]], align 8
-// CHECK17-NEXT: [[TMP71:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP66]], ptr [[TMP71]], align 8
-// CHECK17-NEXT: [[TMP72:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP72]], align 8
-// CHECK17-NEXT: [[TMP73:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP68]], ptr [[TMP73]], align 8
-// CHECK17-NEXT: [[TMP74:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP68]], ptr [[TMP74]], align 8
-// CHECK17-NEXT: [[TMP75:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 1
-// CHECK17-NEXT: store ptr null, ptr [[TMP75]], align 8
-// CHECK17-NEXT: [[TMP76:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
-// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP76]], align 8
-// CHECK17-NEXT: [[TMP77:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
-// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP77]], align 8
-// CHECK17-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 2
-// CHECK17-NEXT: store ptr null, ptr [[TMP78]], align 8
-// CHECK17-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP79]], align 8
-// CHECK17-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP80]], align 8
-// CHECK17-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 3
-// CHECK17-NEXT: store i64 [[TMP69]], ptr [[TMP81]], align 8
-// CHECK17-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 3
-// CHECK17-NEXT: store ptr null, ptr [[TMP82]], align 8
-// CHECK17-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP86:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP86]], ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK17-NEXT: [[TMP87:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK17-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP87]], 0
+// CHECK17-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP74]], ptr [[TMP78]], align 8
+// CHECK17-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP74]], ptr [[TMP79]], align 8
+// CHECK17-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP80]], align 8
+// CHECK17-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP76]], ptr [[TMP81]], align 8
+// CHECK17-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP76]], ptr [[TMP82]], align 8
+// CHECK17-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 1
+// CHECK17-NEXT: store ptr null, ptr [[TMP83]], align 8
+// CHECK17-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
+// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP84]], align 8
+// CHECK17-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
+// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP85]], align 8
+// CHECK17-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 2
+// CHECK17-NEXT: store ptr null, ptr [[TMP86]], align 8
+// CHECK17-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP87]], align 8
+// CHECK17-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP88]], align 8
+// CHECK17-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 3
+// CHECK17-NEXT: store i64 [[TMP77]], ptr [[TMP89]], align 8
+// CHECK17-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 3
+// CHECK17-NEXT: store ptr null, ptr [[TMP90]], align 8
+// CHECK17-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP94:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP94]], ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK17-NEXT: [[TMP95:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK17-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP95]], 0
// CHECK17-NEXT: [[DIV27:%.*]] = sdiv i32 [[SUB26]], 1
// CHECK17-NEXT: [[SUB28:%.*]] = sub nsw i32 [[DIV27]], 1
// CHECK17-NEXT: store i32 [[SUB28]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK17-NEXT: [[TMP88:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK17-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP88]], 1
-// CHECK17-NEXT: [[TMP89:%.*]] = zext i32 [[ADD29]] to i64
+// CHECK17-NEXT: [[TMP96:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK17-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP96]], 1
+// CHECK17-NEXT: [[TMP97:%.*]] = zext i32 [[ADD29]] to i64
// CHECK17-NEXT: [[KERNEL_ARGS30:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP90]], align 4
-// CHECK17-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
-// CHECK17-NEXT: store i32 4, ptr [[TMP91]], align 4
-// CHECK17-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP83]], ptr [[TMP92]], align 8
-// CHECK17-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP84]], ptr [[TMP93]], align 8
-// CHECK17-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
-// CHECK17-NEXT: store ptr [[TMP85]], ptr [[TMP94]], align 8
-// CHECK17-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP95]], align 8
-// CHECK17-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP96]], align 8
-// CHECK17-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP97]], align 8
-// CHECK17-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
-// CHECK17-NEXT: store i64 [[TMP89]], ptr [[TMP98]], align 8
-// CHECK17-NEXT: [[TMP99:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l158.region_id, ptr [[KERNEL_ARGS30]])
-// CHECK17-NEXT: [[TMP100:%.*]] = icmp ne i32 [[TMP99]], 0
-// CHECK17-NEXT: br i1 [[TMP100]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
+// CHECK17-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP98]], align 4
+// CHECK17-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
+// CHECK17-NEXT: store i32 4, ptr [[TMP99]], align 4
+// CHECK17-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP91]], ptr [[TMP100]], align 8
+// CHECK17-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP92]], ptr [[TMP101]], align 8
+// CHECK17-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
+// CHECK17-NEXT: store ptr [[TMP93]], ptr [[TMP102]], align 8
+// CHECK17-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP103]], align 8
+// CHECK17-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP104]], align 8
+// CHECK17-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP105]], align 8
+// CHECK17-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
+// CHECK17-NEXT: store i64 [[TMP97]], ptr [[TMP106]], align 8
+// CHECK17-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP107]], align 8
+// CHECK17-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP108]], align 4
+// CHECK17-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP109]], align 4
+// CHECK17-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP110]], align 4
+// CHECK17-NEXT: [[TMP111:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l158.region_id, ptr [[KERNEL_ARGS30]])
+// CHECK17-NEXT: [[TMP112:%.*]] = icmp ne i32 [[TMP111]], 0
+// CHECK17-NEXT: br i1 [[TMP112]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
// CHECK17: omp_offload.failed31:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l158(i64 [[TMP66]], i64 [[TMP68]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l158(i64 [[TMP74]], i64 [[TMP76]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT32]]
// CHECK17: omp_offload.cont32:
-// CHECK17-NEXT: [[TMP101:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP101]], ptr [[N_CASTED33]], align 4
-// CHECK17-NEXT: [[TMP102:%.*]] = load i64, ptr [[N_CASTED33]], align 8
-// CHECK17-NEXT: [[TMP103:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK17-NEXT: [[TMP113:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP113]], ptr [[N_CASTED33]], align 4
+// CHECK17-NEXT: [[TMP114:%.*]] = load i64, ptr [[N_CASTED33]], align 8
+// CHECK17-NEXT: [[TMP115:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES37]], ptr align 8 @.offload_sizes.12, i64 24, i1 false)
-// CHECK17-NEXT: [[TMP104:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP102]], ptr [[TMP104]], align 8
-// CHECK17-NEXT: [[TMP105:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP102]], ptr [[TMP105]], align 8
-// CHECK17-NEXT: [[TMP106:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i64 0, i64 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP106]], align 8
-// CHECK17-NEXT: [[TMP107:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP107]], align 8
-// CHECK17-NEXT: [[TMP108:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP108]], align 8
-// CHECK17-NEXT: [[TMP109:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i64 0, i64 1
-// CHECK17-NEXT: store ptr null, ptr [[TMP109]], align 8
-// CHECK17-NEXT: [[TMP110:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP110]], align 8
-// CHECK17-NEXT: [[TMP111:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP111]], align 8
-// CHECK17-NEXT: [[TMP112:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 2
-// CHECK17-NEXT: store i64 [[TMP103]], ptr [[TMP112]], align 8
-// CHECK17-NEXT: [[TMP113:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i64 0, i64 2
-// CHECK17-NEXT: store ptr null, ptr [[TMP113]], align 8
-// CHECK17-NEXT: [[TMP114:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP115:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP116:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP117:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP117]], ptr [[DOTCAPTURE_EXPR_39]], align 4
-// CHECK17-NEXT: [[TMP118:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_39]], align 4
-// CHECK17-NEXT: [[SUB41:%.*]] = sub nsw i32 [[TMP118]], 0
+// CHECK17-NEXT: [[TMP116:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP114]], ptr [[TMP116]], align 8
+// CHECK17-NEXT: [[TMP117:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP114]], ptr [[TMP117]], align 8
+// CHECK17-NEXT: [[TMP118:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i64 0, i64 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP118]], align 8
+// CHECK17-NEXT: [[TMP119:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP119]], align 8
+// CHECK17-NEXT: [[TMP120:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP120]], align 8
+// CHECK17-NEXT: [[TMP121:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i64 0, i64 1
+// CHECK17-NEXT: store ptr null, ptr [[TMP121]], align 8
+// CHECK17-NEXT: [[TMP122:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP122]], align 8
+// CHECK17-NEXT: [[TMP123:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP123]], align 8
+// CHECK17-NEXT: [[TMP124:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 2
+// CHECK17-NEXT: store i64 [[TMP115]], ptr [[TMP124]], align 8
+// CHECK17-NEXT: [[TMP125:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i64 0, i64 2
+// CHECK17-NEXT: store ptr null, ptr [[TMP125]], align 8
+// CHECK17-NEXT: [[TMP126:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP127:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP128:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP129:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP129]], ptr [[DOTCAPTURE_EXPR_39]], align 4
+// CHECK17-NEXT: [[TMP130:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_39]], align 4
+// CHECK17-NEXT: [[SUB41:%.*]] = sub nsw i32 [[TMP130]], 0
// CHECK17-NEXT: [[DIV42:%.*]] = sdiv i32 [[SUB41]], 1
// CHECK17-NEXT: [[SUB43:%.*]] = sub nsw i32 [[DIV42]], 1
// CHECK17-NEXT: store i32 [[SUB43]], ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK17-NEXT: [[TMP119:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK17-NEXT: [[ADD44:%.*]] = add nsw i32 [[TMP119]], 1
-// CHECK17-NEXT: [[TMP120:%.*]] = zext i32 [[ADD44]] to i64
+// CHECK17-NEXT: [[TMP131:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
+// CHECK17-NEXT: [[ADD44:%.*]] = add nsw i32 [[TMP131]], 1
+// CHECK17-NEXT: [[TMP132:%.*]] = zext i32 [[ADD44]] to i64
// CHECK17-NEXT: [[KERNEL_ARGS45:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP121:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP121]], align 4
-// CHECK17-NEXT: [[TMP122:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 1
-// CHECK17-NEXT: store i32 3, ptr [[TMP122]], align 4
-// CHECK17-NEXT: [[TMP123:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP114]], ptr [[TMP123]], align 8
-// CHECK17-NEXT: [[TMP124:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP115]], ptr [[TMP124]], align 8
-// CHECK17-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 4
-// CHECK17-NEXT: store ptr [[TMP116]], ptr [[TMP125]], align 8
-// CHECK17-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP126]], align 8
-// CHECK17-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP127]], align 8
-// CHECK17-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP128]], align 8
-// CHECK17-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 8
-// CHECK17-NEXT: store i64 [[TMP120]], ptr [[TMP129]], align 8
-// CHECK17-NEXT: [[TMP130:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l163.region_id, ptr [[KERNEL_ARGS45]])
-// CHECK17-NEXT: [[TMP131:%.*]] = icmp ne i32 [[TMP130]], 0
-// CHECK17-NEXT: br i1 [[TMP131]], label [[OMP_OFFLOAD_FAILED46:%.*]], label [[OMP_OFFLOAD_CONT47:%.*]]
+// CHECK17-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP133]], align 4
+// CHECK17-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 1
+// CHECK17-NEXT: store i32 3, ptr [[TMP134]], align 4
+// CHECK17-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP126]], ptr [[TMP135]], align 8
+// CHECK17-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP127]], ptr [[TMP136]], align 8
+// CHECK17-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 4
+// CHECK17-NEXT: store ptr [[TMP128]], ptr [[TMP137]], align 8
+// CHECK17-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP138]], align 8
+// CHECK17-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP139]], align 8
+// CHECK17-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP140]], align 8
+// CHECK17-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 8
+// CHECK17-NEXT: store i64 [[TMP132]], ptr [[TMP141]], align 8
+// CHECK17-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP142]], align 8
+// CHECK17-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP143]], align 4
+// CHECK17-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP144]], align 4
+// CHECK17-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP145]], align 4
+// CHECK17-NEXT: [[TMP146:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l163.region_id, ptr [[KERNEL_ARGS45]])
+// CHECK17-NEXT: [[TMP147:%.*]] = icmp ne i32 [[TMP146]], 0
+// CHECK17-NEXT: br i1 [[TMP147]], label [[OMP_OFFLOAD_FAILED46:%.*]], label [[OMP_OFFLOAD_CONT47:%.*]]
// CHECK17: omp_offload.failed46:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l163(i64 [[TMP102]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l163(i64 [[TMP114]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT47]]
// CHECK17: omp_offload.cont47:
-// CHECK17-NEXT: [[TMP132:%.*]] = load i32, ptr [[M]], align 4
-// CHECK17-NEXT: store i32 [[TMP132]], ptr [[M_CASTED48]], align 4
-// CHECK17-NEXT: [[TMP133:%.*]] = load i64, ptr [[M_CASTED48]], align 8
-// CHECK17-NEXT: [[TMP134:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP134]], ptr [[N_CASTED49]], align 4
-// CHECK17-NEXT: [[TMP135:%.*]] = load i64, ptr [[N_CASTED49]], align 8
-// CHECK17-NEXT: [[TMP136:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK17-NEXT: [[TMP148:%.*]] = load i32, ptr [[M]], align 4
+// CHECK17-NEXT: store i32 [[TMP148]], ptr [[M_CASTED48]], align 4
+// CHECK17-NEXT: [[TMP149:%.*]] = load i64, ptr [[M_CASTED48]], align 8
+// CHECK17-NEXT: [[TMP150:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP150]], ptr [[N_CASTED49]], align 4
+// CHECK17-NEXT: [[TMP151:%.*]] = load i64, ptr [[N_CASTED49]], align 8
+// CHECK17-NEXT: [[TMP152:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES53]], ptr align 8 @.offload_sizes.16, i64 32, i1 false)
-// CHECK17-NEXT: [[TMP137:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP133]], ptr [[TMP137]], align 8
-// CHECK17-NEXT: [[TMP138:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP133]], ptr [[TMP138]], align 8
-// CHECK17-NEXT: [[TMP139:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP139]], align 8
-// CHECK17-NEXT: [[TMP140:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP135]], ptr [[TMP140]], align 8
-// CHECK17-NEXT: [[TMP141:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 1
-// CHECK17-NEXT: store i64 [[TMP135]], ptr [[TMP141]], align 8
-// CHECK17-NEXT: [[TMP142:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 1
-// CHECK17-NEXT: store ptr null, ptr [[TMP142]], align 8
-// CHECK17-NEXT: [[TMP143:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 2
-// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP143]], align 8
-// CHECK17-NEXT: [[TMP144:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 2
-// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP144]], align 8
-// CHECK17-NEXT: [[TMP145:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 2
-// CHECK17-NEXT: store ptr null, ptr [[TMP145]], align 8
-// CHECK17-NEXT: [[TMP146:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP146]], align 8
-// CHECK17-NEXT: [[TMP147:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP147]], align 8
-// CHECK17-NEXT: [[TMP148:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 3
-// CHECK17-NEXT: store i64 [[TMP136]], ptr [[TMP148]], align 8
-// CHECK17-NEXT: [[TMP149:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 3
-// CHECK17-NEXT: store ptr null, ptr [[TMP149]], align 8
-// CHECK17-NEXT: [[TMP150:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP151:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP152:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP153:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP153]], ptr [[DOTCAPTURE_EXPR_55]], align 4
-// CHECK17-NEXT: [[TMP154:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_55]], align 4
-// CHECK17-NEXT: [[SUB57:%.*]] = sub nsw i32 [[TMP154]], 0
+// CHECK17-NEXT: [[TMP153:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP149]], ptr [[TMP153]], align 8
+// CHECK17-NEXT: [[TMP154:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP149]], ptr [[TMP154]], align 8
+// CHECK17-NEXT: [[TMP155:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP155]], align 8
+// CHECK17-NEXT: [[TMP156:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP151]], ptr [[TMP156]], align 8
+// CHECK17-NEXT: [[TMP157:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 1
+// CHECK17-NEXT: store i64 [[TMP151]], ptr [[TMP157]], align 8
+// CHECK17-NEXT: [[TMP158:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 1
+// CHECK17-NEXT: store ptr null, ptr [[TMP158]], align 8
+// CHECK17-NEXT: [[TMP159:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 2
+// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP159]], align 8
+// CHECK17-NEXT: [[TMP160:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 2
+// CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP160]], align 8
+// CHECK17-NEXT: [[TMP161:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 2
+// CHECK17-NEXT: store ptr null, ptr [[TMP161]], align 8
+// CHECK17-NEXT: [[TMP162:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP162]], align 8
+// CHECK17-NEXT: [[TMP163:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP163]], align 8
+// CHECK17-NEXT: [[TMP164:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 3
+// CHECK17-NEXT: store i64 [[TMP152]], ptr [[TMP164]], align 8
+// CHECK17-NEXT: [[TMP165:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 3
+// CHECK17-NEXT: store ptr null, ptr [[TMP165]], align 8
+// CHECK17-NEXT: [[TMP166:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP167:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP168:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP169:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP169]], ptr [[DOTCAPTURE_EXPR_55]], align 4
+// CHECK17-NEXT: [[TMP170:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_55]], align 4
+// CHECK17-NEXT: [[SUB57:%.*]] = sub nsw i32 [[TMP170]], 0
// CHECK17-NEXT: [[DIV58:%.*]] = sdiv i32 [[SUB57]], 1
// CHECK17-NEXT: [[SUB59:%.*]] = sub nsw i32 [[DIV58]], 1
// CHECK17-NEXT: store i32 [[SUB59]], ptr [[DOTCAPTURE_EXPR_56]], align 4
-// CHECK17-NEXT: [[TMP155:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_56]], align 4
-// CHECK17-NEXT: [[ADD60:%.*]] = add nsw i32 [[TMP155]], 1
-// CHECK17-NEXT: [[TMP156:%.*]] = zext i32 [[ADD60]] to i64
+// CHECK17-NEXT: [[TMP171:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_56]], align 4
+// CHECK17-NEXT: [[ADD60:%.*]] = add nsw i32 [[TMP171]], 1
+// CHECK17-NEXT: [[TMP172:%.*]] = zext i32 [[ADD60]] to i64
// CHECK17-NEXT: [[KERNEL_ARGS61:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP157:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP157]], align 4
-// CHECK17-NEXT: [[TMP158:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 1
-// CHECK17-NEXT: store i32 4, ptr [[TMP158]], align 4
-// CHECK17-NEXT: [[TMP159:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP150]], ptr [[TMP159]], align 8
-// CHECK17-NEXT: [[TMP160:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP151]], ptr [[TMP160]], align 8
-// CHECK17-NEXT: [[TMP161:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 4
-// CHECK17-NEXT: store ptr [[TMP152]], ptr [[TMP161]], align 8
-// CHECK17-NEXT: [[TMP162:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP162]], align 8
-// CHECK17-NEXT: [[TMP163:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP163]], align 8
-// CHECK17-NEXT: [[TMP164:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP164]], align 8
-// CHECK17-NEXT: [[TMP165:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 8
-// CHECK17-NEXT: store i64 [[TMP156]], ptr [[TMP165]], align 8
-// CHECK17-NEXT: [[TMP166:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l168.region_id, ptr [[KERNEL_ARGS61]])
-// CHECK17-NEXT: [[TMP167:%.*]] = icmp ne i32 [[TMP166]], 0
-// CHECK17-NEXT: br i1 [[TMP167]], label [[OMP_OFFLOAD_FAILED62:%.*]], label [[OMP_OFFLOAD_CONT63:%.*]]
+// CHECK17-NEXT: [[TMP173:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP173]], align 4
+// CHECK17-NEXT: [[TMP174:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 1
+// CHECK17-NEXT: store i32 4, ptr [[TMP174]], align 4
+// CHECK17-NEXT: [[TMP175:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP166]], ptr [[TMP175]], align 8
+// CHECK17-NEXT: [[TMP176:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP167]], ptr [[TMP176]], align 8
+// CHECK17-NEXT: [[TMP177:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 4
+// CHECK17-NEXT: store ptr [[TMP168]], ptr [[TMP177]], align 8
+// CHECK17-NEXT: [[TMP178:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP178]], align 8
+// CHECK17-NEXT: [[TMP179:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP179]], align 8
+// CHECK17-NEXT: [[TMP180:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP180]], align 8
+// CHECK17-NEXT: [[TMP181:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 8
+// CHECK17-NEXT: store i64 [[TMP172]], ptr [[TMP181]], align 8
+// CHECK17-NEXT: [[TMP182:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP182]], align 8
+// CHECK17-NEXT: [[TMP183:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP183]], align 4
+// CHECK17-NEXT: [[TMP184:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP184]], align 4
+// CHECK17-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP185]], align 4
+// CHECK17-NEXT: [[TMP186:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l168.region_id, ptr [[KERNEL_ARGS61]])
+// CHECK17-NEXT: [[TMP187:%.*]] = icmp ne i32 [[TMP186]], 0
+// CHECK17-NEXT: br i1 [[TMP187]], label [[OMP_OFFLOAD_FAILED62:%.*]], label [[OMP_OFFLOAD_CONT63:%.*]]
// CHECK17: omp_offload.failed62:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l168(i64 [[TMP133]], i64 [[TMP135]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l168(i64 [[TMP149]], i64 [[TMP151]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT63]]
// CHECK17: omp_offload.cont63:
-// CHECK17-NEXT: [[TMP168:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK17-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP168]])
+// CHECK17-NEXT: [[TMP188:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK17-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP188]])
// CHECK17-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK17-NEXT: [[TMP169:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK17-NEXT: call void @llvm.stackrestore(ptr [[TMP169]])
-// CHECK17-NEXT: [[TMP170:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK17-NEXT: ret i32 [[TMP170]]
+// CHECK17-NEXT: [[TMP189:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK17-NEXT: call void @llvm.stackrestore(ptr [[TMP189]])
+// CHECK17-NEXT: [[TMP190:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK17-NEXT: ret i32 [[TMP190]]
//
//
// CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l148
@@ -10735,7 +11095,7 @@ int main (int argc, char **argv) {
// CHECK17-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK17-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK17-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK17-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK17-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK17-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -10752,165 +11112,205 @@ int main (int argc, char **argv) {
// CHECK17-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK17-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK17-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK17-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.region_id, ptr [[KERNEL_ARGS]])
-// CHECK17-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK17-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK17-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK17-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK17-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK17-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK17-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.region_id, ptr [[KERNEL_ARGS]])
+// CHECK17-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK17-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK17: omp_offload.failed:
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116(ptr [[A]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK17: omp_offload.cont:
-// CHECK17-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP16]], align 8
-// CHECK17-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP17]], align 8
-// CHECK17-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK17-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP20]], align 8
+// CHECK17-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP21]], align 8
+// CHECK17-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK17-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK17-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK17-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK17-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK17-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK17-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK17-NEXT: store ptr @.offload_sizes.24, ptr [[TMP25]], align 8
-// CHECK17-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP26]], align 8
-// CHECK17-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK17-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK17-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK17-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK17-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l121.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK17-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK17-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK17-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK17-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK17-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK17-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK17-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK17-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK17-NEXT: store ptr @.offload_sizes.24, ptr [[TMP29]], align 8
+// CHECK17-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP30]], align 8
+// CHECK17-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK17-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK17-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK17-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK17-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK17-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK17-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK17-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK17-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l121.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK17-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK17-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK17: omp_offload.failed6:
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l121(ptr [[A]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK17: omp_offload.cont7:
-// CHECK17-NEXT: [[TMP32:%.*]] = load i32, ptr [[M]], align 4
-// CHECK17-NEXT: store i32 [[TMP32]], ptr [[M_CASTED]], align 4
-// CHECK17-NEXT: [[TMP33:%.*]] = load i64, ptr [[M_CASTED]], align 8
-// CHECK17-NEXT: [[TMP34:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP33]], ptr [[TMP34]], align 8
-// CHECK17-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP33]], ptr [[TMP35]], align 8
-// CHECK17-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP36]], align 8
-// CHECK17-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP37]], align 8
-// CHECK17-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP38]], align 8
-// CHECK17-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
-// CHECK17-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK17-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP40:%.*]] = load i32, ptr [[M]], align 4
+// CHECK17-NEXT: store i32 [[TMP40]], ptr [[M_CASTED]], align 4
+// CHECK17-NEXT: [[TMP41:%.*]] = load i64, ptr [[M_CASTED]], align 8
+// CHECK17-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP41]], ptr [[TMP42]], align 8
+// CHECK17-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP41]], ptr [[TMP43]], align 8
+// CHECK17-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP44]], align 8
+// CHECK17-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP45]], align 8
+// CHECK17-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP46]], align 8
+// CHECK17-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
+// CHECK17-NEXT: store ptr null, ptr [[TMP47]], align 8
+// CHECK17-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP42]], align 4
-// CHECK17-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK17-NEXT: store i32 2, ptr [[TMP43]], align 4
-// CHECK17-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP40]], ptr [[TMP44]], align 8
-// CHECK17-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP41]], ptr [[TMP45]], align 8
-// CHECK17-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK17-NEXT: store ptr @.offload_sizes.28, ptr [[TMP46]], align 8
-// CHECK17-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP47]], align 8
-// CHECK17-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP48]], align 8
-// CHECK17-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP49]], align 8
-// CHECK17-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK17-NEXT: store i64 10, ptr [[TMP50]], align 8
-// CHECK17-NEXT: [[TMP51:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l126.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK17-NEXT: [[TMP52:%.*]] = icmp ne i32 [[TMP51]], 0
-// CHECK17-NEXT: br i1 [[TMP52]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK17-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP50]], align 4
+// CHECK17-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK17-NEXT: store i32 2, ptr [[TMP51]], align 4
+// CHECK17-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP48]], ptr [[TMP52]], align 8
+// CHECK17-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 8
+// CHECK17-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK17-NEXT: store ptr @.offload_sizes.28, ptr [[TMP54]], align 8
+// CHECK17-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP55]], align 8
+// CHECK17-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP56]], align 8
+// CHECK17-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP57]], align 8
+// CHECK17-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK17-NEXT: store i64 10, ptr [[TMP58]], align 8
+// CHECK17-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP59]], align 8
+// CHECK17-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP60]], align 4
+// CHECK17-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4
+// CHECK17-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP62]], align 4
+// CHECK17-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l126.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK17-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
+// CHECK17-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK17: omp_offload.failed13:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l126(i64 [[TMP33]], ptr [[A]]) #[[ATTR3]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l126(i64 [[TMP41]], ptr [[A]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT14]]
// CHECK17: omp_offload.cont14:
-// CHECK17-NEXT: [[TMP53:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP53]], align 8
-// CHECK17-NEXT: [[TMP54:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP54]], align 8
-// CHECK17-NEXT: [[TMP55:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i64 0, i64 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP55]], align 8
-// CHECK17-NEXT: [[TMP56:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP57:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP65:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP65]], align 8
+// CHECK17-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP66]], align 8
+// CHECK17-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i64 0, i64 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP67]], align 8
+// CHECK17-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP69:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS19:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP58]], align 4
-// CHECK17-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
-// CHECK17-NEXT: store i32 1, ptr [[TMP59]], align 4
-// CHECK17-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP56]], ptr [[TMP60]], align 8
-// CHECK17-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP57]], ptr [[TMP61]], align 8
-// CHECK17-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
-// CHECK17-NEXT: store ptr @.offload_sizes.32, ptr [[TMP62]], align 8
-// CHECK17-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP63]], align 8
-// CHECK17-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP64]], align 8
-// CHECK17-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP65]], align 8
-// CHECK17-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
-// CHECK17-NEXT: store i64 10, ptr [[TMP66]], align 8
-// CHECK17-NEXT: [[TMP67:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l131.region_id, ptr [[KERNEL_ARGS19]])
-// CHECK17-NEXT: [[TMP68:%.*]] = icmp ne i32 [[TMP67]], 0
-// CHECK17-NEXT: br i1 [[TMP68]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
+// CHECK17-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP70]], align 4
+// CHECK17-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
+// CHECK17-NEXT: store i32 1, ptr [[TMP71]], align 4
+// CHECK17-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 8
+// CHECK17-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP69]], ptr [[TMP73]], align 8
+// CHECK17-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
+// CHECK17-NEXT: store ptr @.offload_sizes.32, ptr [[TMP74]], align 8
+// CHECK17-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP75]], align 8
+// CHECK17-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP76]], align 8
+// CHECK17-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP77]], align 8
+// CHECK17-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
+// CHECK17-NEXT: store i64 10, ptr [[TMP78]], align 8
+// CHECK17-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP79]], align 8
+// CHECK17-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP80]], align 4
+// CHECK17-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP81]], align 4
+// CHECK17-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP82]], align 4
+// CHECK17-NEXT: [[TMP83:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l131.region_id, ptr [[KERNEL_ARGS19]])
+// CHECK17-NEXT: [[TMP84:%.*]] = icmp ne i32 [[TMP83]], 0
+// CHECK17-NEXT: br i1 [[TMP84]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
// CHECK17: omp_offload.failed20:
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l131(ptr [[A]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT21]]
// CHECK17: omp_offload.cont21:
-// CHECK17-NEXT: [[TMP69:%.*]] = load i32, ptr [[M]], align 4
-// CHECK17-NEXT: store i32 [[TMP69]], ptr [[M_CASTED22]], align 4
-// CHECK17-NEXT: [[TMP70:%.*]] = load i64, ptr [[M_CASTED22]], align 8
-// CHECK17-NEXT: [[TMP71:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP70]], ptr [[TMP71]], align 8
-// CHECK17-NEXT: [[TMP72:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
-// CHECK17-NEXT: store i64 [[TMP70]], ptr [[TMP72]], align 8
-// CHECK17-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i64 0, i64 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP73]], align 8
-// CHECK17-NEXT: [[TMP74:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 1
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP74]], align 8
-// CHECK17-NEXT: [[TMP75:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 1
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP75]], align 8
-// CHECK17-NEXT: [[TMP76:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i64 0, i64 1
-// CHECK17-NEXT: store ptr null, ptr [[TMP76]], align 8
-// CHECK17-NEXT: [[TMP77:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP78:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP85:%.*]] = load i32, ptr [[M]], align 4
+// CHECK17-NEXT: store i32 [[TMP85]], ptr [[M_CASTED22]], align 4
+// CHECK17-NEXT: [[TMP86:%.*]] = load i64, ptr [[M_CASTED22]], align 8
+// CHECK17-NEXT: [[TMP87:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP86]], ptr [[TMP87]], align 8
+// CHECK17-NEXT: [[TMP88:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
+// CHECK17-NEXT: store i64 [[TMP86]], ptr [[TMP88]], align 8
+// CHECK17-NEXT: [[TMP89:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i64 0, i64 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP89]], align 8
+// CHECK17-NEXT: [[TMP90:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 1
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP90]], align 8
+// CHECK17-NEXT: [[TMP91:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 1
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP91]], align 8
+// CHECK17-NEXT: [[TMP92:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i64 0, i64 1
+// CHECK17-NEXT: store ptr null, ptr [[TMP92]], align 8
+// CHECK17-NEXT: [[TMP93:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP94:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS27:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP79]], align 4
-// CHECK17-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 1
-// CHECK17-NEXT: store i32 2, ptr [[TMP80]], align 4
-// CHECK17-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP77]], ptr [[TMP81]], align 8
-// CHECK17-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP78]], ptr [[TMP82]], align 8
-// CHECK17-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 4
-// CHECK17-NEXT: store ptr @.offload_sizes.36, ptr [[TMP83]], align 8
-// CHECK17-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP84]], align 8
-// CHECK17-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP85]], align 8
-// CHECK17-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP86]], align 8
-// CHECK17-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 8
-// CHECK17-NEXT: store i64 10, ptr [[TMP87]], align 8
-// CHECK17-NEXT: [[TMP88:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l136.region_id, ptr [[KERNEL_ARGS27]])
-// CHECK17-NEXT: [[TMP89:%.*]] = icmp ne i32 [[TMP88]], 0
-// CHECK17-NEXT: br i1 [[TMP89]], label [[OMP_OFFLOAD_FAILED28:%.*]], label [[OMP_OFFLOAD_CONT29:%.*]]
+// CHECK17-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP95]], align 4
+// CHECK17-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 1
+// CHECK17-NEXT: store i32 2, ptr [[TMP96]], align 4
+// CHECK17-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP93]], ptr [[TMP97]], align 8
+// CHECK17-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP94]], ptr [[TMP98]], align 8
+// CHECK17-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 4
+// CHECK17-NEXT: store ptr @.offload_sizes.36, ptr [[TMP99]], align 8
+// CHECK17-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP100]], align 8
+// CHECK17-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP101]], align 8
+// CHECK17-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP102]], align 8
+// CHECK17-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 8
+// CHECK17-NEXT: store i64 10, ptr [[TMP103]], align 8
+// CHECK17-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP104]], align 8
+// CHECK17-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP105]], align 4
+// CHECK17-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP106]], align 4
+// CHECK17-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP107]], align 4
+// CHECK17-NEXT: [[TMP108:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l136.region_id, ptr [[KERNEL_ARGS27]])
+// CHECK17-NEXT: [[TMP109:%.*]] = icmp ne i32 [[TMP108]], 0
+// CHECK17-NEXT: br i1 [[TMP109]], label [[OMP_OFFLOAD_FAILED28:%.*]], label [[OMP_OFFLOAD_CONT29:%.*]]
// CHECK17: omp_offload.failed28:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l136(i64 [[TMP70]], ptr [[A]]) #[[ATTR3]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l136(i64 [[TMP86]], ptr [[A]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT29]]
// CHECK17: omp_offload.cont29:
// CHECK17-NEXT: ret i32 0
@@ -11829,7 +12229,7 @@ int main (int argc, char **argv) {
// CHECK19-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK19-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK19-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK19-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK19-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK19-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK19-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -11846,298 +12246,338 @@ int main (int argc, char **argv) {
// CHECK19-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK19-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK19-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK19-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l148.region_id, ptr [[KERNEL_ARGS]])
-// CHECK19-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK19-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK19-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK19-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK19-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK19-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK19-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l148.region_id, ptr [[KERNEL_ARGS]])
+// CHECK19-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK19-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK19: omp_offload.failed:
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l148(i32 [[TMP3]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK19: omp_offload.cont:
-// CHECK19-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK19-NEXT: [[TMP35:%.*]] = load i32, ptr [[N_CASTED3]], align 4
-// CHECK19-NEXT: [[TMP36:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK19-NEXT: [[TMP37:%.*]] = sext i32 [[TMP36]] to i64
+// CHECK19-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK19-NEXT: [[TMP39:%.*]] = load i32, ptr [[N_CASTED3]], align 4
+// CHECK19-NEXT: [[TMP40:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK19-NEXT: [[TMP41:%.*]] = sext i32 [[TMP40]] to i64
// CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES7]], ptr align 4 @.offload_sizes.4, i32 24, i1 false)
-// CHECK19-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP35]], ptr [[TMP38]], align 4
-// CHECK19-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP35]], ptr [[TMP39]], align 4
-// CHECK19-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP40]], align 4
-// CHECK19-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP41]], align 4
-// CHECK19-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP42]], align 4
-// CHECK19-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
-// CHECK19-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK19-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 4
-// CHECK19-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP45]], align 4
-// CHECK19-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK19-NEXT: store i64 [[TMP37]], ptr [[TMP46]], align 4
-// CHECK19-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK19-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP39]], ptr [[TMP42]], align 4
+// CHECK19-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP39]], ptr [[TMP43]], align 4
+// CHECK19-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP44]], align 4
+// CHECK19-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP45]], align 4
+// CHECK19-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP46]], align 4
+// CHECK19-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
// CHECK19-NEXT: store ptr null, ptr [[TMP47]], align 4
-// CHECK19-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP51:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP51]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK19-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK19-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP52]], 0
+// CHECK19-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 4
+// CHECK19-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP49]], align 4
+// CHECK19-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK19-NEXT: store i64 [[TMP41]], ptr [[TMP50]], align 4
+// CHECK19-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK19-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK19-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP54:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP55:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP55]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK19-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK19-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP56]], 0
// CHECK19-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK19-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK19-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK19-NEXT: [[TMP53:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK19-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP53]], 1
-// CHECK19-NEXT: [[TMP54:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK19-NEXT: [[TMP57:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK19-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP57]], 1
+// CHECK19-NEXT: [[TMP58:%.*]] = zext i32 [[ADD14]] to i64
// CHECK19-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP55]], align 4
-// CHECK19-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK19-NEXT: store i32 3, ptr [[TMP56]], align 4
-// CHECK19-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 4
-// CHECK19-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 4
-// CHECK19-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK19-NEXT: store ptr [[TMP50]], ptr [[TMP59]], align 4
-// CHECK19-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP60]], align 4
-// CHECK19-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP61]], align 4
-// CHECK19-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP62]], align 4
-// CHECK19-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK19-NEXT: store i64 [[TMP54]], ptr [[TMP63]], align 8
-// CHECK19-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l153.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK19-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
-// CHECK19-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK19-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP59]], align 4
+// CHECK19-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK19-NEXT: store i32 3, ptr [[TMP60]], align 4
+// CHECK19-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 4
+// CHECK19-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 4
+// CHECK19-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK19-NEXT: store ptr [[TMP54]], ptr [[TMP63]], align 4
+// CHECK19-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP64]], align 4
+// CHECK19-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP65]], align 4
+// CHECK19-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP66]], align 4
+// CHECK19-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK19-NEXT: store i64 [[TMP58]], ptr [[TMP67]], align 8
+// CHECK19-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP68]], align 8
+// CHECK19-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK19-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP70]], align 4
+// CHECK19-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP71]], align 4
+// CHECK19-NEXT: [[TMP72:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l153.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK19-NEXT: [[TMP73:%.*]] = icmp ne i32 [[TMP72]], 0
+// CHECK19-NEXT: br i1 [[TMP73]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK19: omp_offload.failed16:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l153(i32 [[TMP35]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l153(i32 [[TMP39]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK19: omp_offload.cont17:
-// CHECK19-NEXT: [[TMP66:%.*]] = load i32, ptr [[M]], align 4
-// CHECK19-NEXT: store i32 [[TMP66]], ptr [[M_CASTED]], align 4
-// CHECK19-NEXT: [[TMP67:%.*]] = load i32, ptr [[M_CASTED]], align 4
-// CHECK19-NEXT: [[TMP68:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP68]], ptr [[N_CASTED18]], align 4
-// CHECK19-NEXT: [[TMP69:%.*]] = load i32, ptr [[N_CASTED18]], align 4
-// CHECK19-NEXT: [[TMP70:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK19-NEXT: [[TMP71:%.*]] = sext i32 [[TMP70]] to i64
+// CHECK19-NEXT: [[TMP74:%.*]] = load i32, ptr [[M]], align 4
+// CHECK19-NEXT: store i32 [[TMP74]], ptr [[M_CASTED]], align 4
+// CHECK19-NEXT: [[TMP75:%.*]] = load i32, ptr [[M_CASTED]], align 4
+// CHECK19-NEXT: [[TMP76:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP76]], ptr [[N_CASTED18]], align 4
+// CHECK19-NEXT: [[TMP77:%.*]] = load i32, ptr [[N_CASTED18]], align 4
+// CHECK19-NEXT: [[TMP78:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK19-NEXT: [[TMP79:%.*]] = sext i32 [[TMP78]] to i64
// CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES22]], ptr align 4 @.offload_sizes.8, i32 32, i1 false)
-// CHECK19-NEXT: [[TMP72:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP67]], ptr [[TMP72]], align 4
-// CHECK19-NEXT: [[TMP73:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP67]], ptr [[TMP73]], align 4
-// CHECK19-NEXT: [[TMP74:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP74]], align 4
-// CHECK19-NEXT: [[TMP75:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP69]], ptr [[TMP75]], align 4
-// CHECK19-NEXT: [[TMP76:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP69]], ptr [[TMP76]], align 4
-// CHECK19-NEXT: [[TMP77:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 1
-// CHECK19-NEXT: store ptr null, ptr [[TMP77]], align 4
-// CHECK19-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP78]], align 4
-// CHECK19-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP79]], align 4
-// CHECK19-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 2
-// CHECK19-NEXT: store ptr null, ptr [[TMP80]], align 4
-// CHECK19-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP81]], align 4
-// CHECK19-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP82]], align 4
-// CHECK19-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 3
-// CHECK19-NEXT: store i64 [[TMP71]], ptr [[TMP83]], align 4
-// CHECK19-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 3
-// CHECK19-NEXT: store ptr null, ptr [[TMP84]], align 4
-// CHECK19-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP88:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP88]], ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK19-NEXT: [[TMP89:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK19-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP89]], 0
+// CHECK19-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP75]], ptr [[TMP80]], align 4
+// CHECK19-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP75]], ptr [[TMP81]], align 4
+// CHECK19-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP82]], align 4
+// CHECK19-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP77]], ptr [[TMP83]], align 4
+// CHECK19-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP77]], ptr [[TMP84]], align 4
+// CHECK19-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 1
+// CHECK19-NEXT: store ptr null, ptr [[TMP85]], align 4
+// CHECK19-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP86]], align 4
+// CHECK19-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP87]], align 4
+// CHECK19-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 2
+// CHECK19-NEXT: store ptr null, ptr [[TMP88]], align 4
+// CHECK19-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP89]], align 4
+// CHECK19-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP90]], align 4
+// CHECK19-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 3
+// CHECK19-NEXT: store i64 [[TMP79]], ptr [[TMP91]], align 4
+// CHECK19-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 3
+// CHECK19-NEXT: store ptr null, ptr [[TMP92]], align 4
+// CHECK19-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP94:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP95:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP96:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP96]], ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK19-NEXT: [[TMP97:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK19-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP97]], 0
// CHECK19-NEXT: [[DIV27:%.*]] = sdiv i32 [[SUB26]], 1
// CHECK19-NEXT: [[SUB28:%.*]] = sub nsw i32 [[DIV27]], 1
// CHECK19-NEXT: store i32 [[SUB28]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK19-NEXT: [[TMP90:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK19-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP90]], 1
-// CHECK19-NEXT: [[TMP91:%.*]] = zext i32 [[ADD29]] to i64
+// CHECK19-NEXT: [[TMP98:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK19-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP98]], 1
+// CHECK19-NEXT: [[TMP99:%.*]] = zext i32 [[ADD29]] to i64
// CHECK19-NEXT: [[KERNEL_ARGS30:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP92]], align 4
-// CHECK19-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
-// CHECK19-NEXT: store i32 4, ptr [[TMP93]], align 4
-// CHECK19-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP85]], ptr [[TMP94]], align 4
-// CHECK19-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP86]], ptr [[TMP95]], align 4
-// CHECK19-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
-// CHECK19-NEXT: store ptr [[TMP87]], ptr [[TMP96]], align 4
-// CHECK19-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP97]], align 4
-// CHECK19-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP98]], align 4
-// CHECK19-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP99]], align 4
-// CHECK19-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
-// CHECK19-NEXT: store i64 [[TMP91]], ptr [[TMP100]], align 8
-// CHECK19-NEXT: [[TMP101:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l158.region_id, ptr [[KERNEL_ARGS30]])
-// CHECK19-NEXT: [[TMP102:%.*]] = icmp ne i32 [[TMP101]], 0
-// CHECK19-NEXT: br i1 [[TMP102]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
+// CHECK19-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP100]], align 4
+// CHECK19-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
+// CHECK19-NEXT: store i32 4, ptr [[TMP101]], align 4
+// CHECK19-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP93]], ptr [[TMP102]], align 4
+// CHECK19-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP94]], ptr [[TMP103]], align 4
+// CHECK19-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
+// CHECK19-NEXT: store ptr [[TMP95]], ptr [[TMP104]], align 4
+// CHECK19-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP105]], align 4
+// CHECK19-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP106]], align 4
+// CHECK19-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP107]], align 4
+// CHECK19-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
+// CHECK19-NEXT: store i64 [[TMP99]], ptr [[TMP108]], align 8
+// CHECK19-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP109]], align 8
+// CHECK19-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP110]], align 4
+// CHECK19-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP111]], align 4
+// CHECK19-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP112]], align 4
+// CHECK19-NEXT: [[TMP113:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l158.region_id, ptr [[KERNEL_ARGS30]])
+// CHECK19-NEXT: [[TMP114:%.*]] = icmp ne i32 [[TMP113]], 0
+// CHECK19-NEXT: br i1 [[TMP114]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
// CHECK19: omp_offload.failed31:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l158(i32 [[TMP67]], i32 [[TMP69]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l158(i32 [[TMP75]], i32 [[TMP77]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT32]]
// CHECK19: omp_offload.cont32:
-// CHECK19-NEXT: [[TMP103:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP103]], ptr [[N_CASTED33]], align 4
-// CHECK19-NEXT: [[TMP104:%.*]] = load i32, ptr [[N_CASTED33]], align 4
-// CHECK19-NEXT: [[TMP105:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK19-NEXT: [[TMP106:%.*]] = sext i32 [[TMP105]] to i64
+// CHECK19-NEXT: [[TMP115:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP115]], ptr [[N_CASTED33]], align 4
+// CHECK19-NEXT: [[TMP116:%.*]] = load i32, ptr [[N_CASTED33]], align 4
+// CHECK19-NEXT: [[TMP117:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK19-NEXT: [[TMP118:%.*]] = sext i32 [[TMP117]] to i64
// CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES37]], ptr align 4 @.offload_sizes.12, i32 24, i1 false)
-// CHECK19-NEXT: [[TMP107:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP104]], ptr [[TMP107]], align 4
-// CHECK19-NEXT: [[TMP108:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP104]], ptr [[TMP108]], align 4
-// CHECK19-NEXT: [[TMP109:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP109]], align 4
-// CHECK19-NEXT: [[TMP110:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP110]], align 4
-// CHECK19-NEXT: [[TMP111:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP111]], align 4
-// CHECK19-NEXT: [[TMP112:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i32 0, i32 1
-// CHECK19-NEXT: store ptr null, ptr [[TMP112]], align 4
-// CHECK19-NEXT: [[TMP113:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP113]], align 4
-// CHECK19-NEXT: [[TMP114:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP114]], align 4
-// CHECK19-NEXT: [[TMP115:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 2
-// CHECK19-NEXT: store i64 [[TMP106]], ptr [[TMP115]], align 4
-// CHECK19-NEXT: [[TMP116:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i32 0, i32 2
-// CHECK19-NEXT: store ptr null, ptr [[TMP116]], align 4
-// CHECK19-NEXT: [[TMP117:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP118:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP119:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP120:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP120]], ptr [[DOTCAPTURE_EXPR_39]], align 4
-// CHECK19-NEXT: [[TMP121:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_39]], align 4
-// CHECK19-NEXT: [[SUB41:%.*]] = sub nsw i32 [[TMP121]], 0
+// CHECK19-NEXT: [[TMP119:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP116]], ptr [[TMP119]], align 4
+// CHECK19-NEXT: [[TMP120:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP116]], ptr [[TMP120]], align 4
+// CHECK19-NEXT: [[TMP121:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP121]], align 4
+// CHECK19-NEXT: [[TMP122:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP122]], align 4
+// CHECK19-NEXT: [[TMP123:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP123]], align 4
+// CHECK19-NEXT: [[TMP124:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i32 0, i32 1
+// CHECK19-NEXT: store ptr null, ptr [[TMP124]], align 4
+// CHECK19-NEXT: [[TMP125:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP125]], align 4
+// CHECK19-NEXT: [[TMP126:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP126]], align 4
+// CHECK19-NEXT: [[TMP127:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 2
+// CHECK19-NEXT: store i64 [[TMP118]], ptr [[TMP127]], align 4
+// CHECK19-NEXT: [[TMP128:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i32 0, i32 2
+// CHECK19-NEXT: store ptr null, ptr [[TMP128]], align 4
+// CHECK19-NEXT: [[TMP129:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP130:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP131:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP132:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP132]], ptr [[DOTCAPTURE_EXPR_39]], align 4
+// CHECK19-NEXT: [[TMP133:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_39]], align 4
+// CHECK19-NEXT: [[SUB41:%.*]] = sub nsw i32 [[TMP133]], 0
// CHECK19-NEXT: [[DIV42:%.*]] = sdiv i32 [[SUB41]], 1
// CHECK19-NEXT: [[SUB43:%.*]] = sub nsw i32 [[DIV42]], 1
// CHECK19-NEXT: store i32 [[SUB43]], ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK19-NEXT: [[TMP122:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK19-NEXT: [[ADD44:%.*]] = add nsw i32 [[TMP122]], 1
-// CHECK19-NEXT: [[TMP123:%.*]] = zext i32 [[ADD44]] to i64
+// CHECK19-NEXT: [[TMP134:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
+// CHECK19-NEXT: [[ADD44:%.*]] = add nsw i32 [[TMP134]], 1
+// CHECK19-NEXT: [[TMP135:%.*]] = zext i32 [[ADD44]] to i64
// CHECK19-NEXT: [[KERNEL_ARGS45:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP124:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP124]], align 4
-// CHECK19-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 1
-// CHECK19-NEXT: store i32 3, ptr [[TMP125]], align 4
-// CHECK19-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP117]], ptr [[TMP126]], align 4
-// CHECK19-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP118]], ptr [[TMP127]], align 4
-// CHECK19-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 4
-// CHECK19-NEXT: store ptr [[TMP119]], ptr [[TMP128]], align 4
-// CHECK19-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP129]], align 4
-// CHECK19-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP130]], align 4
-// CHECK19-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP131]], align 4
-// CHECK19-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 8
-// CHECK19-NEXT: store i64 [[TMP123]], ptr [[TMP132]], align 8
-// CHECK19-NEXT: [[TMP133:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l163.region_id, ptr [[KERNEL_ARGS45]])
-// CHECK19-NEXT: [[TMP134:%.*]] = icmp ne i32 [[TMP133]], 0
-// CHECK19-NEXT: br i1 [[TMP134]], label [[OMP_OFFLOAD_FAILED46:%.*]], label [[OMP_OFFLOAD_CONT47:%.*]]
+// CHECK19-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP136]], align 4
+// CHECK19-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 1
+// CHECK19-NEXT: store i32 3, ptr [[TMP137]], align 4
+// CHECK19-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP129]], ptr [[TMP138]], align 4
+// CHECK19-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP130]], ptr [[TMP139]], align 4
+// CHECK19-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 4
+// CHECK19-NEXT: store ptr [[TMP131]], ptr [[TMP140]], align 4
+// CHECK19-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP141]], align 4
+// CHECK19-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP142]], align 4
+// CHECK19-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP143]], align 4
+// CHECK19-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 8
+// CHECK19-NEXT: store i64 [[TMP135]], ptr [[TMP144]], align 8
+// CHECK19-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP145]], align 8
+// CHECK19-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP146]], align 4
+// CHECK19-NEXT: [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP147]], align 4
+// CHECK19-NEXT: [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP148]], align 4
+// CHECK19-NEXT: [[TMP149:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l163.region_id, ptr [[KERNEL_ARGS45]])
+// CHECK19-NEXT: [[TMP150:%.*]] = icmp ne i32 [[TMP149]], 0
+// CHECK19-NEXT: br i1 [[TMP150]], label [[OMP_OFFLOAD_FAILED46:%.*]], label [[OMP_OFFLOAD_CONT47:%.*]]
// CHECK19: omp_offload.failed46:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l163(i32 [[TMP104]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l163(i32 [[TMP116]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT47]]
// CHECK19: omp_offload.cont47:
-// CHECK19-NEXT: [[TMP135:%.*]] = load i32, ptr [[M]], align 4
-// CHECK19-NEXT: store i32 [[TMP135]], ptr [[M_CASTED48]], align 4
-// CHECK19-NEXT: [[TMP136:%.*]] = load i32, ptr [[M_CASTED48]], align 4
-// CHECK19-NEXT: [[TMP137:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP137]], ptr [[N_CASTED49]], align 4
-// CHECK19-NEXT: [[TMP138:%.*]] = load i32, ptr [[N_CASTED49]], align 4
-// CHECK19-NEXT: [[TMP139:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK19-NEXT: [[TMP140:%.*]] = sext i32 [[TMP139]] to i64
+// CHECK19-NEXT: [[TMP151:%.*]] = load i32, ptr [[M]], align 4
+// CHECK19-NEXT: store i32 [[TMP151]], ptr [[M_CASTED48]], align 4
+// CHECK19-NEXT: [[TMP152:%.*]] = load i32, ptr [[M_CASTED48]], align 4
+// CHECK19-NEXT: [[TMP153:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP153]], ptr [[N_CASTED49]], align 4
+// CHECK19-NEXT: [[TMP154:%.*]] = load i32, ptr [[N_CASTED49]], align 4
+// CHECK19-NEXT: [[TMP155:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK19-NEXT: [[TMP156:%.*]] = sext i32 [[TMP155]] to i64
// CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES53]], ptr align 4 @.offload_sizes.16, i32 32, i1 false)
-// CHECK19-NEXT: [[TMP141:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP136]], ptr [[TMP141]], align 4
-// CHECK19-NEXT: [[TMP142:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP136]], ptr [[TMP142]], align 4
-// CHECK19-NEXT: [[TMP143:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP143]], align 4
-// CHECK19-NEXT: [[TMP144:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP138]], ptr [[TMP144]], align 4
-// CHECK19-NEXT: [[TMP145:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP138]], ptr [[TMP145]], align 4
-// CHECK19-NEXT: [[TMP146:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 1
-// CHECK19-NEXT: store ptr null, ptr [[TMP146]], align 4
-// CHECK19-NEXT: [[TMP147:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 2
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP147]], align 4
-// CHECK19-NEXT: [[TMP148:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 2
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP148]], align 4
-// CHECK19-NEXT: [[TMP149:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 2
-// CHECK19-NEXT: store ptr null, ptr [[TMP149]], align 4
-// CHECK19-NEXT: [[TMP150:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP150]], align 4
-// CHECK19-NEXT: [[TMP151:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP151]], align 4
-// CHECK19-NEXT: [[TMP152:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 3
-// CHECK19-NEXT: store i64 [[TMP140]], ptr [[TMP152]], align 4
-// CHECK19-NEXT: [[TMP153:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 3
-// CHECK19-NEXT: store ptr null, ptr [[TMP153]], align 4
-// CHECK19-NEXT: [[TMP154:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP155:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP156:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP157:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP157]], ptr [[DOTCAPTURE_EXPR_55]], align 4
-// CHECK19-NEXT: [[TMP158:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_55]], align 4
-// CHECK19-NEXT: [[SUB57:%.*]] = sub nsw i32 [[TMP158]], 0
+// CHECK19-NEXT: [[TMP157:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP152]], ptr [[TMP157]], align 4
+// CHECK19-NEXT: [[TMP158:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP152]], ptr [[TMP158]], align 4
+// CHECK19-NEXT: [[TMP159:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP159]], align 4
+// CHECK19-NEXT: [[TMP160:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP154]], ptr [[TMP160]], align 4
+// CHECK19-NEXT: [[TMP161:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP154]], ptr [[TMP161]], align 4
+// CHECK19-NEXT: [[TMP162:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 1
+// CHECK19-NEXT: store ptr null, ptr [[TMP162]], align 4
+// CHECK19-NEXT: [[TMP163:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 2
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP163]], align 4
+// CHECK19-NEXT: [[TMP164:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 2
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP164]], align 4
+// CHECK19-NEXT: [[TMP165:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 2
+// CHECK19-NEXT: store ptr null, ptr [[TMP165]], align 4
+// CHECK19-NEXT: [[TMP166:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP166]], align 4
+// CHECK19-NEXT: [[TMP167:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP167]], align 4
+// CHECK19-NEXT: [[TMP168:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 3
+// CHECK19-NEXT: store i64 [[TMP156]], ptr [[TMP168]], align 4
+// CHECK19-NEXT: [[TMP169:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 3
+// CHECK19-NEXT: store ptr null, ptr [[TMP169]], align 4
+// CHECK19-NEXT: [[TMP170:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP171:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP172:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP173:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP173]], ptr [[DOTCAPTURE_EXPR_55]], align 4
+// CHECK19-NEXT: [[TMP174:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_55]], align 4
+// CHECK19-NEXT: [[SUB57:%.*]] = sub nsw i32 [[TMP174]], 0
// CHECK19-NEXT: [[DIV58:%.*]] = sdiv i32 [[SUB57]], 1
// CHECK19-NEXT: [[SUB59:%.*]] = sub nsw i32 [[DIV58]], 1
// CHECK19-NEXT: store i32 [[SUB59]], ptr [[DOTCAPTURE_EXPR_56]], align 4
-// CHECK19-NEXT: [[TMP159:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_56]], align 4
-// CHECK19-NEXT: [[ADD60:%.*]] = add nsw i32 [[TMP159]], 1
-// CHECK19-NEXT: [[TMP160:%.*]] = zext i32 [[ADD60]] to i64
+// CHECK19-NEXT: [[TMP175:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_56]], align 4
+// CHECK19-NEXT: [[ADD60:%.*]] = add nsw i32 [[TMP175]], 1
+// CHECK19-NEXT: [[TMP176:%.*]] = zext i32 [[ADD60]] to i64
// CHECK19-NEXT: [[KERNEL_ARGS61:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP161:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP161]], align 4
-// CHECK19-NEXT: [[TMP162:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 1
-// CHECK19-NEXT: store i32 4, ptr [[TMP162]], align 4
-// CHECK19-NEXT: [[TMP163:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP154]], ptr [[TMP163]], align 4
-// CHECK19-NEXT: [[TMP164:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP155]], ptr [[TMP164]], align 4
-// CHECK19-NEXT: [[TMP165:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 4
-// CHECK19-NEXT: store ptr [[TMP156]], ptr [[TMP165]], align 4
-// CHECK19-NEXT: [[TMP166:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP166]], align 4
-// CHECK19-NEXT: [[TMP167:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP167]], align 4
-// CHECK19-NEXT: [[TMP168:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP168]], align 4
-// CHECK19-NEXT: [[TMP169:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 8
-// CHECK19-NEXT: store i64 [[TMP160]], ptr [[TMP169]], align 8
-// CHECK19-NEXT: [[TMP170:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l168.region_id, ptr [[KERNEL_ARGS61]])
-// CHECK19-NEXT: [[TMP171:%.*]] = icmp ne i32 [[TMP170]], 0
-// CHECK19-NEXT: br i1 [[TMP171]], label [[OMP_OFFLOAD_FAILED62:%.*]], label [[OMP_OFFLOAD_CONT63:%.*]]
+// CHECK19-NEXT: [[TMP177:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP177]], align 4
+// CHECK19-NEXT: [[TMP178:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 1
+// CHECK19-NEXT: store i32 4, ptr [[TMP178]], align 4
+// CHECK19-NEXT: [[TMP179:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP170]], ptr [[TMP179]], align 4
+// CHECK19-NEXT: [[TMP180:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP171]], ptr [[TMP180]], align 4
+// CHECK19-NEXT: [[TMP181:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 4
+// CHECK19-NEXT: store ptr [[TMP172]], ptr [[TMP181]], align 4
+// CHECK19-NEXT: [[TMP182:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP182]], align 4
+// CHECK19-NEXT: [[TMP183:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP183]], align 4
+// CHECK19-NEXT: [[TMP184:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP184]], align 4
+// CHECK19-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 8
+// CHECK19-NEXT: store i64 [[TMP176]], ptr [[TMP185]], align 8
+// CHECK19-NEXT: [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP186]], align 8
+// CHECK19-NEXT: [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP187]], align 4
+// CHECK19-NEXT: [[TMP188:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP188]], align 4
+// CHECK19-NEXT: [[TMP189:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP189]], align 4
+// CHECK19-NEXT: [[TMP190:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l168.region_id, ptr [[KERNEL_ARGS61]])
+// CHECK19-NEXT: [[TMP191:%.*]] = icmp ne i32 [[TMP190]], 0
+// CHECK19-NEXT: br i1 [[TMP191]], label [[OMP_OFFLOAD_FAILED62:%.*]], label [[OMP_OFFLOAD_CONT63:%.*]]
// CHECK19: omp_offload.failed62:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l168(i32 [[TMP136]], i32 [[TMP138]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l168(i32 [[TMP152]], i32 [[TMP154]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT63]]
// CHECK19: omp_offload.cont63:
-// CHECK19-NEXT: [[TMP172:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK19-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP172]])
+// CHECK19-NEXT: [[TMP192:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK19-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP192]])
// CHECK19-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK19-NEXT: [[TMP173:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK19-NEXT: call void @llvm.stackrestore(ptr [[TMP173]])
-// CHECK19-NEXT: [[TMP174:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK19-NEXT: ret i32 [[TMP174]]
+// CHECK19-NEXT: [[TMP193:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK19-NEXT: call void @llvm.stackrestore(ptr [[TMP193]])
+// CHECK19-NEXT: [[TMP194:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK19-NEXT: ret i32 [[TMP194]]
//
//
// CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l148
@@ -13262,7 +13702,7 @@ int main (int argc, char **argv) {
// CHECK19-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK19-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK19-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK19-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK19-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK19-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -13279,165 +13719,205 @@ int main (int argc, char **argv) {
// CHECK19-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK19-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK19-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK19-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.region_id, ptr [[KERNEL_ARGS]])
-// CHECK19-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK19-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK19-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK19-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK19-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK19-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK19-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.region_id, ptr [[KERNEL_ARGS]])
+// CHECK19-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK19-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK19: omp_offload.failed:
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116(ptr [[A]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK19: omp_offload.cont:
-// CHECK19-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP16]], align 4
-// CHECK19-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP17]], align 4
-// CHECK19-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK19-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP20]], align 4
+// CHECK19-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP21]], align 4
+// CHECK19-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK19-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK19-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK19-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK19-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK19-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK19-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK19-NEXT: store ptr @.offload_sizes.24, ptr [[TMP25]], align 4
-// CHECK19-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP26]], align 4
-// CHECK19-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK19-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK19-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK19-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK19-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l121.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK19-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK19-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK19-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK19-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK19-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK19-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK19-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK19-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK19-NEXT: store ptr @.offload_sizes.24, ptr [[TMP29]], align 4
+// CHECK19-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP30]], align 4
+// CHECK19-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK19-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK19-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK19-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK19-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK19-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK19-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK19-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK19-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l121.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK19-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK19-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK19: omp_offload.failed6:
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l121(ptr [[A]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK19: omp_offload.cont7:
-// CHECK19-NEXT: [[TMP32:%.*]] = load i32, ptr [[M]], align 4
-// CHECK19-NEXT: store i32 [[TMP32]], ptr [[M_CASTED]], align 4
-// CHECK19-NEXT: [[TMP33:%.*]] = load i32, ptr [[M_CASTED]], align 4
-// CHECK19-NEXT: [[TMP34:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP33]], ptr [[TMP34]], align 4
-// CHECK19-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP33]], ptr [[TMP35]], align 4
-// CHECK19-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP36]], align 4
-// CHECK19-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP37]], align 4
-// CHECK19-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP38]], align 4
-// CHECK19-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1
-// CHECK19-NEXT: store ptr null, ptr [[TMP39]], align 4
-// CHECK19-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP40:%.*]] = load i32, ptr [[M]], align 4
+// CHECK19-NEXT: store i32 [[TMP40]], ptr [[M_CASTED]], align 4
+// CHECK19-NEXT: [[TMP41:%.*]] = load i32, ptr [[M_CASTED]], align 4
+// CHECK19-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP41]], ptr [[TMP42]], align 4
+// CHECK19-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP41]], ptr [[TMP43]], align 4
+// CHECK19-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP44]], align 4
+// CHECK19-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP45]], align 4
+// CHECK19-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP46]], align 4
+// CHECK19-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1
+// CHECK19-NEXT: store ptr null, ptr [[TMP47]], align 4
+// CHECK19-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP42]], align 4
-// CHECK19-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK19-NEXT: store i32 2, ptr [[TMP43]], align 4
-// CHECK19-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP40]], ptr [[TMP44]], align 4
-// CHECK19-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP41]], ptr [[TMP45]], align 4
-// CHECK19-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK19-NEXT: store ptr @.offload_sizes.28, ptr [[TMP46]], align 4
-// CHECK19-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP47]], align 4
-// CHECK19-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP48]], align 4
-// CHECK19-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP49]], align 4
-// CHECK19-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK19-NEXT: store i64 10, ptr [[TMP50]], align 8
-// CHECK19-NEXT: [[TMP51:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l126.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK19-NEXT: [[TMP52:%.*]] = icmp ne i32 [[TMP51]], 0
-// CHECK19-NEXT: br i1 [[TMP52]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK19-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP50]], align 4
+// CHECK19-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK19-NEXT: store i32 2, ptr [[TMP51]], align 4
+// CHECK19-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP48]], ptr [[TMP52]], align 4
+// CHECK19-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 4
+// CHECK19-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK19-NEXT: store ptr @.offload_sizes.28, ptr [[TMP54]], align 4
+// CHECK19-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP55]], align 4
+// CHECK19-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP56]], align 4
+// CHECK19-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP57]], align 4
+// CHECK19-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK19-NEXT: store i64 10, ptr [[TMP58]], align 8
+// CHECK19-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP59]], align 8
+// CHECK19-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP60]], align 4
+// CHECK19-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4
+// CHECK19-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP62]], align 4
+// CHECK19-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l126.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK19-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
+// CHECK19-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK19: omp_offload.failed13:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l126(i32 [[TMP33]], ptr [[A]]) #[[ATTR3]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l126(i32 [[TMP41]], ptr [[A]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT14]]
// CHECK19: omp_offload.cont14:
-// CHECK19-NEXT: [[TMP53:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP53]], align 4
-// CHECK19-NEXT: [[TMP54:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP54]], align 4
-// CHECK19-NEXT: [[TMP55:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP55]], align 4
-// CHECK19-NEXT: [[TMP56:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP57:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP65:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP65]], align 4
+// CHECK19-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP66]], align 4
+// CHECK19-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP67]], align 4
+// CHECK19-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP69:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS19:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP58]], align 4
-// CHECK19-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
-// CHECK19-NEXT: store i32 1, ptr [[TMP59]], align 4
-// CHECK19-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP56]], ptr [[TMP60]], align 4
-// CHECK19-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP57]], ptr [[TMP61]], align 4
-// CHECK19-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
-// CHECK19-NEXT: store ptr @.offload_sizes.32, ptr [[TMP62]], align 4
-// CHECK19-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP63]], align 4
-// CHECK19-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP64]], align 4
-// CHECK19-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP65]], align 4
-// CHECK19-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
-// CHECK19-NEXT: store i64 10, ptr [[TMP66]], align 8
-// CHECK19-NEXT: [[TMP67:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l131.region_id, ptr [[KERNEL_ARGS19]])
-// CHECK19-NEXT: [[TMP68:%.*]] = icmp ne i32 [[TMP67]], 0
-// CHECK19-NEXT: br i1 [[TMP68]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
+// CHECK19-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP70]], align 4
+// CHECK19-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
+// CHECK19-NEXT: store i32 1, ptr [[TMP71]], align 4
+// CHECK19-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 4
+// CHECK19-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP69]], ptr [[TMP73]], align 4
+// CHECK19-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
+// CHECK19-NEXT: store ptr @.offload_sizes.32, ptr [[TMP74]], align 4
+// CHECK19-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP75]], align 4
+// CHECK19-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP76]], align 4
+// CHECK19-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP77]], align 4
+// CHECK19-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
+// CHECK19-NEXT: store i64 10, ptr [[TMP78]], align 8
+// CHECK19-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP79]], align 8
+// CHECK19-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP80]], align 4
+// CHECK19-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP81]], align 4
+// CHECK19-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP82]], align 4
+// CHECK19-NEXT: [[TMP83:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l131.region_id, ptr [[KERNEL_ARGS19]])
+// CHECK19-NEXT: [[TMP84:%.*]] = icmp ne i32 [[TMP83]], 0
+// CHECK19-NEXT: br i1 [[TMP84]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
// CHECK19: omp_offload.failed20:
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l131(ptr [[A]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT21]]
// CHECK19: omp_offload.cont21:
-// CHECK19-NEXT: [[TMP69:%.*]] = load i32, ptr [[M]], align 4
-// CHECK19-NEXT: store i32 [[TMP69]], ptr [[M_CASTED22]], align 4
-// CHECK19-NEXT: [[TMP70:%.*]] = load i32, ptr [[M_CASTED22]], align 4
-// CHECK19-NEXT: [[TMP71:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP70]], ptr [[TMP71]], align 4
-// CHECK19-NEXT: [[TMP72:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP70]], ptr [[TMP72]], align 4
-// CHECK19-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP73]], align 4
-// CHECK19-NEXT: [[TMP74:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 1
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP74]], align 4
-// CHECK19-NEXT: [[TMP75:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 1
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP75]], align 4
-// CHECK19-NEXT: [[TMP76:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i32 0, i32 1
-// CHECK19-NEXT: store ptr null, ptr [[TMP76]], align 4
-// CHECK19-NEXT: [[TMP77:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP78:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP85:%.*]] = load i32, ptr [[M]], align 4
+// CHECK19-NEXT: store i32 [[TMP85]], ptr [[M_CASTED22]], align 4
+// CHECK19-NEXT: [[TMP86:%.*]] = load i32, ptr [[M_CASTED22]], align 4
+// CHECK19-NEXT: [[TMP87:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP86]], ptr [[TMP87]], align 4
+// CHECK19-NEXT: [[TMP88:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP86]], ptr [[TMP88]], align 4
+// CHECK19-NEXT: [[TMP89:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP89]], align 4
+// CHECK19-NEXT: [[TMP90:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 1
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP90]], align 4
+// CHECK19-NEXT: [[TMP91:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 1
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP91]], align 4
+// CHECK19-NEXT: [[TMP92:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i32 0, i32 1
+// CHECK19-NEXT: store ptr null, ptr [[TMP92]], align 4
+// CHECK19-NEXT: [[TMP93:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP94:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS27:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP79]], align 4
-// CHECK19-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 1
-// CHECK19-NEXT: store i32 2, ptr [[TMP80]], align 4
-// CHECK19-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP77]], ptr [[TMP81]], align 4
-// CHECK19-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP78]], ptr [[TMP82]], align 4
-// CHECK19-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 4
-// CHECK19-NEXT: store ptr @.offload_sizes.36, ptr [[TMP83]], align 4
-// CHECK19-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP84]], align 4
-// CHECK19-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP85]], align 4
-// CHECK19-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP86]], align 4
-// CHECK19-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 8
-// CHECK19-NEXT: store i64 10, ptr [[TMP87]], align 8
-// CHECK19-NEXT: [[TMP88:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l136.region_id, ptr [[KERNEL_ARGS27]])
-// CHECK19-NEXT: [[TMP89:%.*]] = icmp ne i32 [[TMP88]], 0
-// CHECK19-NEXT: br i1 [[TMP89]], label [[OMP_OFFLOAD_FAILED28:%.*]], label [[OMP_OFFLOAD_CONT29:%.*]]
+// CHECK19-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP95]], align 4
+// CHECK19-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 1
+// CHECK19-NEXT: store i32 2, ptr [[TMP96]], align 4
+// CHECK19-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP93]], ptr [[TMP97]], align 4
+// CHECK19-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP94]], ptr [[TMP98]], align 4
+// CHECK19-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 4
+// CHECK19-NEXT: store ptr @.offload_sizes.36, ptr [[TMP99]], align 4
+// CHECK19-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP100]], align 4
+// CHECK19-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP101]], align 4
+// CHECK19-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP102]], align 4
+// CHECK19-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 8
+// CHECK19-NEXT: store i64 10, ptr [[TMP103]], align 8
+// CHECK19-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP104]], align 8
+// CHECK19-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP105]], align 4
+// CHECK19-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP106]], align 4
+// CHECK19-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP107]], align 4
+// CHECK19-NEXT: [[TMP108:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l136.region_id, ptr [[KERNEL_ARGS27]])
+// CHECK19-NEXT: [[TMP109:%.*]] = icmp ne i32 [[TMP108]], 0
+// CHECK19-NEXT: br i1 [[TMP109]], label [[OMP_OFFLOAD_FAILED28:%.*]], label [[OMP_OFFLOAD_CONT29:%.*]]
// CHECK19: omp_offload.failed28:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l136(i32 [[TMP70]], ptr [[A]]) #[[ATTR3]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l136(i32 [[TMP86]], ptr [[A]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT29]]
// CHECK19: omp_offload.cont29:
// CHECK19-NEXT: ret i32 0
diff --git a/clang/test/OpenMP/teams_distribute_parallel_for_simd_codegen.cpp b/clang/test/OpenMP/teams_distribute_parallel_for_simd_codegen.cpp
index 44acbf76e0c54..c9a831352c50a 100644
--- a/clang/test/OpenMP/teams_distribute_parallel_for_simd_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_parallel_for_simd_codegen.cpp
@@ -256,96 +256,113 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP24]], 1
// CHECK1-NEXT: [[TMP25:%.*]] = zext i32 [[ADD]] to i64
+// CHECK1-NEXT: [[TMP26:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP21]], 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP26]], align 4
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK1-NEXT: store i32 4, ptr [[TMP27]], align 4
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP19]], ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP20]], ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP30]], align 8
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP31]], align 8
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP32]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK1-NEXT: store i32 4, ptr [[TMP28]], align 4
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP19]], ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP20]], ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK1-NEXT: store ptr null, ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK1-NEXT: store i64 [[TMP25]], ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 [[TMP21]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l29.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP36:%.*]] = icmp ne i32 [[TMP35]], 0
-// CHECK1-NEXT: br i1 [[TMP36]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK1-NEXT: store i64 [[TMP25]], ptr [[TMP35]], align 8
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP36]], align 8
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [[TMP26]], ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP38]], align 4
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP39]], align 4
+// CHECK1-NEXT: [[TMP40:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 [[TMP21]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l29.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP41:%.*]] = icmp ne i32 [[TMP40]], 0
+// CHECK1-NEXT: br i1 [[TMP41]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l29(i64 [[TMP2]], i64 [[TMP4]], i64 [[TMP6]], ptr @a) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
-// CHECK1-NEXT: [[TMP37:%.*]] = load i32, ptr [[I]], align 4
-// CHECK1-NEXT: store i32 [[TMP37]], ptr [[I_CASTED]], align 4
-// CHECK1-NEXT: [[TMP38:%.*]] = load i64, ptr [[I_CASTED]], align 8
-// CHECK1-NEXT: [[TMP39:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: store i32 [[TMP39]], ptr [[N_CASTED4]], align 4
-// CHECK1-NEXT: [[TMP40:%.*]] = load i64, ptr [[N_CASTED4]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK1-NEXT: store ptr @a, ptr [[TMP41]], align 8
-// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK1-NEXT: store ptr @a, ptr [[TMP42]], align 8
-// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP43]], align 8
-// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP38]], ptr [[TMP44]], align 8
-// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
-// CHECK1-NEXT: store i64 [[TMP38]], ptr [[TMP45]], align 8
-// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP46]], align 8
-// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 2
-// CHECK1-NEXT: store i64 [[TMP40]], ptr [[TMP47]], align 8
-// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 2
-// CHECK1-NEXT: store i64 [[TMP40]], ptr [[TMP48]], align 8
-// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP49]], align 8
-// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP52:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: store i32 [[TMP52]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK1-NEXT: [[TMP53:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK1-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP53]], 0
+// CHECK1-NEXT: [[TMP42:%.*]] = load i32, ptr [[I]], align 4
+// CHECK1-NEXT: store i32 [[TMP42]], ptr [[I_CASTED]], align 4
+// CHECK1-NEXT: [[TMP43:%.*]] = load i64, ptr [[I_CASTED]], align 8
+// CHECK1-NEXT: [[TMP44:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: store i32 [[TMP44]], ptr [[N_CASTED4]], align 4
+// CHECK1-NEXT: [[TMP45:%.*]] = load i64, ptr [[N_CASTED4]], align 8
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK1-NEXT: store ptr @a, ptr [[TMP46]], align 8
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK1-NEXT: store ptr @a, ptr [[TMP47]], align 8
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP48]], align 8
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP43]], ptr [[TMP49]], align 8
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
+// CHECK1-NEXT: store i64 [[TMP43]], ptr [[TMP50]], align 8
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP51]], align 8
+// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 2
+// CHECK1-NEXT: store i64 [[TMP45]], ptr [[TMP52]], align 8
+// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 2
+// CHECK1-NEXT: store i64 [[TMP45]], ptr [[TMP53]], align 8
+// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 2
+// CHECK1-NEXT: store ptr null, ptr [[TMP54]], align 8
+// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP57:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: store i32 [[TMP57]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK1-NEXT: [[TMP58:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK1-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP58]], 0
// CHECK1-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK1-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK1-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK1-NEXT: [[TMP54:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK1-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP54]], 1
-// CHECK1-NEXT: [[TMP55:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK1-NEXT: [[TMP59:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK1-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP59]], 1
+// CHECK1-NEXT: [[TMP60:%.*]] = zext i32 [[ADD14]] to i64
// CHECK1-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP56]], align 4
-// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK1-NEXT: store i32 3, ptr [[TMP57]], align 4
-// CHECK1-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP50]], ptr [[TMP58]], align 8
-// CHECK1-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP51]], ptr [[TMP59]], align 8
-// CHECK1-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.4, ptr [[TMP60]], align 8
-// CHECK1-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP61]], align 8
-// CHECK1-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP62]], align 8
-// CHECK1-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP63]], align 8
-// CHECK1-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK1-NEXT: store i64 [[TMP55]], ptr [[TMP64]], align 8
-// CHECK1-NEXT: [[TMP65:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l36.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK1-NEXT: [[TMP66:%.*]] = icmp ne i32 [[TMP65]], 0
-// CHECK1-NEXT: br i1 [[TMP66]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK1-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP61]], align 4
+// CHECK1-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK1-NEXT: store i32 3, ptr [[TMP62]], align 4
+// CHECK1-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP55]], ptr [[TMP63]], align 8
+// CHECK1-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP56]], ptr [[TMP64]], align 8
+// CHECK1-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.4, ptr [[TMP65]], align 8
+// CHECK1-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP66]], align 8
+// CHECK1-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP67]], align 8
+// CHECK1-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP68]], align 8
+// CHECK1-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK1-NEXT: store i64 [[TMP60]], ptr [[TMP69]], align 8
+// CHECK1-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP70]], align 8
+// CHECK1-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP71]], align 4
+// CHECK1-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP72]], align 4
+// CHECK1-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP73]], align 4
+// CHECK1-NEXT: [[TMP74:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l36.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK1-NEXT: [[TMP75:%.*]] = icmp ne i32 [[TMP74]], 0
+// CHECK1-NEXT: br i1 [[TMP75]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK1: omp_offload.failed16:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l36(ptr @a, i64 [[TMP38]], i64 [[TMP40]]) #[[ATTR2]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l36(ptr @a, i64 [[TMP43]], i64 [[TMP45]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK1: omp_offload.cont17:
-// CHECK1-NEXT: [[TMP67:%.*]] = load i32, ptr @a, align 4
-// CHECK1-NEXT: ret i32 [[TMP67]]
+// CHECK1-NEXT: [[TMP76:%.*]] = load i32, ptr @a, align 4
+// CHECK1-NEXT: ret i32 [[TMP76]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l29
@@ -915,96 +932,113 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP24]], 1
// CHECK3-NEXT: [[TMP25:%.*]] = zext i32 [[ADD]] to i64
+// CHECK3-NEXT: [[TMP26:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP21]], 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP26]], align 4
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK3-NEXT: store i32 4, ptr [[TMP27]], align 4
-// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP19]], ptr [[TMP28]], align 4
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP20]], ptr [[TMP29]], align 4
-// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes, ptr [[TMP30]], align 4
-// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP31]], align 4
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP32]], align 4
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK3-NEXT: store i32 4, ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP19]], ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP20]], ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK3-NEXT: store ptr null, ptr [[TMP33]], align 4
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK3-NEXT: store i64 [[TMP25]], ptr [[TMP34]], align 8
-// CHECK3-NEXT: [[TMP35:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 [[TMP21]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l29.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP36:%.*]] = icmp ne i32 [[TMP35]], 0
-// CHECK3-NEXT: br i1 [[TMP36]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP34]], align 4
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK3-NEXT: store i64 [[TMP25]], ptr [[TMP35]], align 8
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP36]], align 8
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [[TMP26]], ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP38]], align 4
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP39]], align 4
+// CHECK3-NEXT: [[TMP40:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 [[TMP21]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l29.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP41:%.*]] = icmp ne i32 [[TMP40]], 0
+// CHECK3-NEXT: br i1 [[TMP41]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l29(i32 [[TMP2]], i32 [[TMP4]], i32 [[TMP6]], ptr @a) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
-// CHECK3-NEXT: [[TMP37:%.*]] = load i32, ptr [[I]], align 4
-// CHECK3-NEXT: store i32 [[TMP37]], ptr [[I_CASTED]], align 4
-// CHECK3-NEXT: [[TMP38:%.*]] = load i32, ptr [[I_CASTED]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: store i32 [[TMP39]], ptr [[N_CASTED4]], align 4
-// CHECK3-NEXT: [[TMP40:%.*]] = load i32, ptr [[N_CASTED4]], align 4
-// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK3-NEXT: store ptr @a, ptr [[TMP41]], align 4
-// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK3-NEXT: store ptr @a, ptr [[TMP42]], align 4
-// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP38]], ptr [[TMP44]], align 4
-// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
-// CHECK3-NEXT: store i32 [[TMP38]], ptr [[TMP45]], align 4
-// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 1
-// CHECK3-NEXT: store ptr null, ptr [[TMP46]], align 4
-// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 2
-// CHECK3-NEXT: store i32 [[TMP40]], ptr [[TMP47]], align 4
-// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 2
-// CHECK3-NEXT: store i32 [[TMP40]], ptr [[TMP48]], align 4
-// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 2
-// CHECK3-NEXT: store ptr null, ptr [[TMP49]], align 4
-// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP52:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: store i32 [[TMP52]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK3-NEXT: [[TMP53:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK3-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP53]], 0
+// CHECK3-NEXT: [[TMP42:%.*]] = load i32, ptr [[I]], align 4
+// CHECK3-NEXT: store i32 [[TMP42]], ptr [[I_CASTED]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = load i32, ptr [[I_CASTED]], align 4
+// CHECK3-NEXT: [[TMP44:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: store i32 [[TMP44]], ptr [[N_CASTED4]], align 4
+// CHECK3-NEXT: [[TMP45:%.*]] = load i32, ptr [[N_CASTED4]], align 4
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK3-NEXT: store ptr @a, ptr [[TMP46]], align 4
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK3-NEXT: store ptr @a, ptr [[TMP47]], align 4
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP48]], align 4
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP43]], ptr [[TMP49]], align 4
+// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
+// CHECK3-NEXT: store i32 [[TMP43]], ptr [[TMP50]], align 4
+// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 1
+// CHECK3-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 2
+// CHECK3-NEXT: store i32 [[TMP45]], ptr [[TMP52]], align 4
+// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 2
+// CHECK3-NEXT: store i32 [[TMP45]], ptr [[TMP53]], align 4
+// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 2
+// CHECK3-NEXT: store ptr null, ptr [[TMP54]], align 4
+// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP57:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: store i32 [[TMP57]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK3-NEXT: [[TMP58:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK3-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP58]], 0
// CHECK3-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK3-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK3-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK3-NEXT: [[TMP54:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK3-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP54]], 1
-// CHECK3-NEXT: [[TMP55:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK3-NEXT: [[TMP59:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK3-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP59]], 1
+// CHECK3-NEXT: [[TMP60:%.*]] = zext i32 [[ADD14]] to i64
// CHECK3-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP56]], align 4
-// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK3-NEXT: store i32 3, ptr [[TMP57]], align 4
-// CHECK3-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP50]], ptr [[TMP58]], align 4
-// CHECK3-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP51]], ptr [[TMP59]], align 4
-// CHECK3-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.4, ptr [[TMP60]], align 4
-// CHECK3-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP61]], align 4
-// CHECK3-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP62]], align 4
-// CHECK3-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP63]], align 4
-// CHECK3-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK3-NEXT: store i64 [[TMP55]], ptr [[TMP64]], align 8
-// CHECK3-NEXT: [[TMP65:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l36.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK3-NEXT: [[TMP66:%.*]] = icmp ne i32 [[TMP65]], 0
-// CHECK3-NEXT: br i1 [[TMP66]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK3-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP61]], align 4
+// CHECK3-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK3-NEXT: store i32 3, ptr [[TMP62]], align 4
+// CHECK3-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP55]], ptr [[TMP63]], align 4
+// CHECK3-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP56]], ptr [[TMP64]], align 4
+// CHECK3-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.4, ptr [[TMP65]], align 4
+// CHECK3-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP66]], align 4
+// CHECK3-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP67]], align 4
+// CHECK3-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP68]], align 4
+// CHECK3-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK3-NEXT: store i64 [[TMP60]], ptr [[TMP69]], align 8
+// CHECK3-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP70]], align 8
+// CHECK3-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP71]], align 4
+// CHECK3-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP72]], align 4
+// CHECK3-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP73]], align 4
+// CHECK3-NEXT: [[TMP74:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l36.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK3-NEXT: [[TMP75:%.*]] = icmp ne i32 [[TMP74]], 0
+// CHECK3-NEXT: br i1 [[TMP75]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK3: omp_offload.failed16:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l36(ptr @a, i32 [[TMP38]], i32 [[TMP40]]) #[[ATTR2]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l36(ptr @a, i32 [[TMP43]], i32 [[TMP45]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK3: omp_offload.cont17:
-// CHECK3-NEXT: [[TMP67:%.*]] = load i32, ptr @a, align 4
-// CHECK3-NEXT: ret i32 [[TMP67]]
+// CHECK3-NEXT: [[TMP76:%.*]] = load i32, ptr @a, align 4
+// CHECK3-NEXT: ret i32 [[TMP76]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l29
@@ -1829,7 +1863,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[TMP27:%.*]] = zext i32 [[ADD]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP28]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP28]], align 4
// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 4, ptr [[TMP29]], align 4
// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1846,18 +1880,26 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP35]], align 8
// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 [[TMP27]], ptr [[TMP36]], align 8
-// CHECK9-NEXT: [[TMP37:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l74.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP38:%.*]] = icmp ne i32 [[TMP37]], 0
-// CHECK9-NEXT: br i1 [[TMP38]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP37]], align 8
+// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP38]], align 4
+// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP39]], align 4
+// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP40]], align 4
+// CHECK9-NEXT: [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l74.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
+// CHECK9-NEXT: br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l74(i64 [[TMP1]], ptr [[VLA]], i64 [[TMP4]], i64 [[TMP6]]) #[[ATTR5:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[VLA]], i64 0
-// CHECK9-NEXT: [[TMP39:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK9-NEXT: [[TMP40:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP40]])
-// CHECK9-NEXT: ret i32 [[TMP39]]
+// CHECK9-NEXT: [[TMP43:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK9-NEXT: [[TMP44:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP44]])
+// CHECK9-NEXT: ret i32 [[TMP43]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l74
@@ -2197,7 +2239,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[TMP27:%.*]] = zext i32 [[ADD]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP28]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP28]], align 4
// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 4, ptr [[TMP29]], align 4
// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2214,18 +2256,26 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP35]], align 4
// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 [[TMP27]], ptr [[TMP36]], align 8
-// CHECK11-NEXT: [[TMP37:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l74.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP38:%.*]] = icmp ne i32 [[TMP37]], 0
-// CHECK11-NEXT: br i1 [[TMP38]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP37]], align 8
+// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP38]], align 4
+// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP39]], align 4
+// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP40]], align 4
+// CHECK11-NEXT: [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l74.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
+// CHECK11-NEXT: br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l74(i32 [[TMP0]], ptr [[VLA]], i32 [[TMP3]], i32 [[TMP5]]) #[[ATTR5:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[VLA]], i32 0
-// CHECK11-NEXT: [[TMP39:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK11-NEXT: [[TMP40:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP40]])
-// CHECK11-NEXT: ret i32 [[TMP39]]
+// CHECK11-NEXT: [[TMP43:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK11-NEXT: [[TMP44:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP44]])
+// CHECK11-NEXT: ret i32 [[TMP43]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l74
@@ -2692,7 +2742,7 @@ int main (int argc, char **argv) {
// CHECK17-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK17-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP10]], align 4
+// CHECK17-NEXT: store i32 2, ptr [[TMP10]], align 4
// CHECK17-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK17-NEXT: store i32 2, ptr [[TMP11]], align 4
// CHECK17-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2709,17 +2759,25 @@ int main (int argc, char **argv) {
// CHECK17-NEXT: store ptr null, ptr [[TMP17]], align 8
// CHECK17-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK17-NEXT: store i64 123, ptr [[TMP18]], align 8
-// CHECK17-NEXT: [[TMP19:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l112.region_id, ptr [[KERNEL_ARGS]])
-// CHECK17-NEXT: [[TMP20:%.*]] = icmp ne i32 [[TMP19]], 0
-// CHECK17-NEXT: br i1 [[TMP20]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK17-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP19]], align 8
+// CHECK17-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP20]], align 4
+// CHECK17-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP21]], align 4
+// CHECK17-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP22]], align 4
+// CHECK17-NEXT: [[TMP23:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l112.region_id, ptr [[KERNEL_ARGS]])
+// CHECK17-NEXT: [[TMP24:%.*]] = icmp ne i32 [[TMP23]], 0
+// CHECK17-NEXT: br i1 [[TMP24]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK17: omp_offload.failed:
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l112(ptr [[THIS1]], i64 [[TMP1]]) #[[ATTR4:[0-9]+]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK17: omp_offload.cont:
// CHECK17-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK17-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A2]], i64 0, i64 0
-// CHECK17-NEXT: [[TMP21:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK17-NEXT: ret i32 [[TMP21]]
+// CHECK17-NEXT: [[TMP25:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK17-NEXT: ret i32 [[TMP25]]
//
//
// CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l112
@@ -2959,7 +3017,7 @@ int main (int argc, char **argv) {
// CHECK19-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK19-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP10]], align 4
+// CHECK19-NEXT: store i32 2, ptr [[TMP10]], align 4
// CHECK19-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK19-NEXT: store i32 2, ptr [[TMP11]], align 4
// CHECK19-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2976,17 +3034,25 @@ int main (int argc, char **argv) {
// CHECK19-NEXT: store ptr null, ptr [[TMP17]], align 4
// CHECK19-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK19-NEXT: store i64 123, ptr [[TMP18]], align 8
-// CHECK19-NEXT: [[TMP19:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l112.region_id, ptr [[KERNEL_ARGS]])
-// CHECK19-NEXT: [[TMP20:%.*]] = icmp ne i32 [[TMP19]], 0
-// CHECK19-NEXT: br i1 [[TMP20]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK19-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP19]], align 8
+// CHECK19-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP20]], align 4
+// CHECK19-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP21]], align 4
+// CHECK19-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP22]], align 4
+// CHECK19-NEXT: [[TMP23:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l112.region_id, ptr [[KERNEL_ARGS]])
+// CHECK19-NEXT: [[TMP24:%.*]] = icmp ne i32 [[TMP23]], 0
+// CHECK19-NEXT: br i1 [[TMP24]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK19: omp_offload.failed:
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l112(ptr [[THIS1]], i32 [[TMP1]]) #[[ATTR4:[0-9]+]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK19: omp_offload.cont:
// CHECK19-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK19-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A2]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP21:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK19-NEXT: ret i32 [[TMP21]]
+// CHECK19-NEXT: [[TMP25:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK19-NEXT: ret i32 [[TMP25]]
//
//
// CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l112
@@ -3384,7 +3450,7 @@ int main (int argc, char **argv) {
// CHECK25-NEXT: [[TMP27:%.*]] = zext i32 [[ADD]] to i64
// CHECK25-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK25-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK25-NEXT: store i32 1, ptr [[TMP28]], align 4
+// CHECK25-NEXT: store i32 2, ptr [[TMP28]], align 4
// CHECK25-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK25-NEXT: store i32 4, ptr [[TMP29]], align 4
// CHECK25-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3401,20 +3467,28 @@ int main (int argc, char **argv) {
// CHECK25-NEXT: store ptr null, ptr [[TMP35]], align 8
// CHECK25-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK25-NEXT: store i64 [[TMP27]], ptr [[TMP36]], align 8
-// CHECK25-NEXT: [[TMP37:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l166.region_id, ptr [[KERNEL_ARGS]])
-// CHECK25-NEXT: [[TMP38:%.*]] = icmp ne i32 [[TMP37]], 0
-// CHECK25-NEXT: br i1 [[TMP38]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK25-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK25-NEXT: store i64 0, ptr [[TMP37]], align 8
+// CHECK25-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK25-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP38]], align 4
+// CHECK25-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK25-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP39]], align 4
+// CHECK25-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK25-NEXT: store i32 0, ptr [[TMP40]], align 4
+// CHECK25-NEXT: [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l166.region_id, ptr [[KERNEL_ARGS]])
+// CHECK25-NEXT: [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
+// CHECK25-NEXT: br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK25: omp_offload.failed:
// CHECK25-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l166(i64 [[TMP1]], ptr [[VLA]], i64 [[TMP4]], i64 [[TMP6]]) #[[ATTR5:[0-9]+]]
// CHECK25-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK25: omp_offload.cont:
-// CHECK25-NEXT: [[TMP39:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK25-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP39]])
+// CHECK25-NEXT: [[TMP43:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK25-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP43]])
// CHECK25-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK25-NEXT: [[TMP40:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK25-NEXT: call void @llvm.stackrestore(ptr [[TMP40]])
-// CHECK25-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK25-NEXT: ret i32 [[TMP41]]
+// CHECK25-NEXT: [[TMP44:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK25-NEXT: call void @llvm.stackrestore(ptr [[TMP44]])
+// CHECK25-NEXT: [[TMP45:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK25-NEXT: ret i32 [[TMP45]]
//
//
// CHECK25-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l166
@@ -3718,28 +3792,37 @@ int main (int argc, char **argv) {
// CHECK25-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK25-NEXT: [[TMP14:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK25-NEXT: [[TMP15:%.*]] = load i32, ptr [[TE]], align 4
+// CHECK25-NEXT: [[TMP16:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP15]], 0
// CHECK25-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK25-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK25-NEXT: store i32 1, ptr [[TMP16]], align 4
-// CHECK25-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK25-NEXT: store i32 3, ptr [[TMP17]], align 4
-// CHECK25-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK25-NEXT: store ptr [[TMP13]], ptr [[TMP18]], align 8
-// CHECK25-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK25-NEXT: store ptr [[TMP14]], ptr [[TMP19]], align 8
-// CHECK25-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK25-NEXT: store ptr @.offload_sizes.4, ptr [[TMP20]], align 8
-// CHECK25-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK25-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP21]], align 8
-// CHECK25-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK25-NEXT: store ptr null, ptr [[TMP22]], align 8
-// CHECK25-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK25-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK25-NEXT: store i32 2, ptr [[TMP17]], align 4
+// CHECK25-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK25-NEXT: store i32 3, ptr [[TMP18]], align 4
+// CHECK25-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK25-NEXT: store ptr [[TMP13]], ptr [[TMP19]], align 8
+// CHECK25-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK25-NEXT: store ptr [[TMP14]], ptr [[TMP20]], align 8
+// CHECK25-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK25-NEXT: store ptr @.offload_sizes.4, ptr [[TMP21]], align 8
+// CHECK25-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK25-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP22]], align 8
+// CHECK25-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK25-NEXT: store ptr null, ptr [[TMP23]], align 8
-// CHECK25-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK25-NEXT: store i64 10, ptr [[TMP24]], align 8
-// CHECK25-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 [[TMP15]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l155.region_id, ptr [[KERNEL_ARGS]])
-// CHECK25-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK25-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK25-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK25-NEXT: store ptr null, ptr [[TMP24]], align 8
+// CHECK25-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK25-NEXT: store i64 10, ptr [[TMP25]], align 8
+// CHECK25-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK25-NEXT: store i64 0, ptr [[TMP26]], align 8
+// CHECK25-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK25-NEXT: store [3 x i32] [[TMP16]], ptr [[TMP27]], align 4
+// CHECK25-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK25-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK25-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK25-NEXT: store i32 0, ptr [[TMP29]], align 4
+// CHECK25-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 [[TMP15]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l155.region_id, ptr [[KERNEL_ARGS]])
+// CHECK25-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
+// CHECK25-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK25: omp_offload.failed:
// CHECK25-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l155(i64 [[TMP1]], i64 [[TMP3]], ptr [[A]]) #[[ATTR5]]
// CHECK25-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -4004,7 +4087,7 @@ int main (int argc, char **argv) {
// CHECK27-NEXT: [[TMP27:%.*]] = zext i32 [[ADD]] to i64
// CHECK27-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK27-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK27-NEXT: store i32 1, ptr [[TMP28]], align 4
+// CHECK27-NEXT: store i32 2, ptr [[TMP28]], align 4
// CHECK27-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK27-NEXT: store i32 4, ptr [[TMP29]], align 4
// CHECK27-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -4021,20 +4104,28 @@ int main (int argc, char **argv) {
// CHECK27-NEXT: store ptr null, ptr [[TMP35]], align 4
// CHECK27-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK27-NEXT: store i64 [[TMP27]], ptr [[TMP36]], align 8
-// CHECK27-NEXT: [[TMP37:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l166.region_id, ptr [[KERNEL_ARGS]])
-// CHECK27-NEXT: [[TMP38:%.*]] = icmp ne i32 [[TMP37]], 0
-// CHECK27-NEXT: br i1 [[TMP38]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK27-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK27-NEXT: store i64 0, ptr [[TMP37]], align 8
+// CHECK27-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK27-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP38]], align 4
+// CHECK27-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK27-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP39]], align 4
+// CHECK27-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK27-NEXT: store i32 0, ptr [[TMP40]], align 4
+// CHECK27-NEXT: [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l166.region_id, ptr [[KERNEL_ARGS]])
+// CHECK27-NEXT: [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
+// CHECK27-NEXT: br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK27: omp_offload.failed:
// CHECK27-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l166(i32 [[TMP0]], ptr [[VLA]], i32 [[TMP3]], i32 [[TMP5]]) #[[ATTR5:[0-9]+]]
// CHECK27-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK27: omp_offload.cont:
-// CHECK27-NEXT: [[TMP39:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK27-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP39]])
+// CHECK27-NEXT: [[TMP43:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK27-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP43]])
// CHECK27-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK27-NEXT: [[TMP40:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK27-NEXT: call void @llvm.stackrestore(ptr [[TMP40]])
-// CHECK27-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK27-NEXT: ret i32 [[TMP41]]
+// CHECK27-NEXT: [[TMP44:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK27-NEXT: call void @llvm.stackrestore(ptr [[TMP44]])
+// CHECK27-NEXT: [[TMP45:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK27-NEXT: ret i32 [[TMP45]]
//
//
// CHECK27-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l166
@@ -4333,28 +4424,37 @@ int main (int argc, char **argv) {
// CHECK27-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK27-NEXT: [[TMP14:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK27-NEXT: [[TMP15:%.*]] = load i32, ptr [[TE]], align 4
+// CHECK27-NEXT: [[TMP16:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP15]], 0
// CHECK27-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK27-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK27-NEXT: store i32 1, ptr [[TMP16]], align 4
-// CHECK27-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK27-NEXT: store i32 3, ptr [[TMP17]], align 4
-// CHECK27-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK27-NEXT: store ptr [[TMP13]], ptr [[TMP18]], align 4
-// CHECK27-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK27-NEXT: store ptr [[TMP14]], ptr [[TMP19]], align 4
-// CHECK27-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK27-NEXT: store ptr @.offload_sizes.4, ptr [[TMP20]], align 4
-// CHECK27-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK27-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP21]], align 4
-// CHECK27-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK27-NEXT: store ptr null, ptr [[TMP22]], align 4
-// CHECK27-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK27-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK27-NEXT: store i32 2, ptr [[TMP17]], align 4
+// CHECK27-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK27-NEXT: store i32 3, ptr [[TMP18]], align 4
+// CHECK27-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK27-NEXT: store ptr [[TMP13]], ptr [[TMP19]], align 4
+// CHECK27-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK27-NEXT: store ptr [[TMP14]], ptr [[TMP20]], align 4
+// CHECK27-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK27-NEXT: store ptr @.offload_sizes.4, ptr [[TMP21]], align 4
+// CHECK27-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK27-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP22]], align 4
+// CHECK27-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK27-NEXT: store ptr null, ptr [[TMP23]], align 4
-// CHECK27-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK27-NEXT: store i64 10, ptr [[TMP24]], align 8
-// CHECK27-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 [[TMP15]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l155.region_id, ptr [[KERNEL_ARGS]])
-// CHECK27-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK27-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK27-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK27-NEXT: store ptr null, ptr [[TMP24]], align 4
+// CHECK27-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK27-NEXT: store i64 10, ptr [[TMP25]], align 8
+// CHECK27-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK27-NEXT: store i64 0, ptr [[TMP26]], align 8
+// CHECK27-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK27-NEXT: store [3 x i32] [[TMP16]], ptr [[TMP27]], align 4
+// CHECK27-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK27-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK27-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK27-NEXT: store i32 0, ptr [[TMP29]], align 4
+// CHECK27-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 [[TMP15]], i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l155.region_id, ptr [[KERNEL_ARGS]])
+// CHECK27-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
+// CHECK27-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK27: omp_offload.failed:
// CHECK27-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l155(i32 [[TMP1]], i32 [[TMP3]], ptr [[A]]) #[[ATTR5]]
// CHECK27-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/teams_distribute_parallel_for_simd_collapse_codegen.cpp b/clang/test/OpenMP/teams_distribute_parallel_for_simd_collapse_codegen.cpp
index c161adf0d3f68..3d971c5e92491 100644
--- a/clang/test/OpenMP/teams_distribute_parallel_for_simd_collapse_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_parallel_for_simd_collapse_codegen.cpp
@@ -137,7 +137,7 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -154,9 +154,17 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 56088, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -164,8 +172,8 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[A3:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x [456 x i32]], ptr [[A3]], i64 0, i64 0
// CHECK1-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds [456 x i32], ptr [[ARRAYIDX]], i64 0, i64 0
-// CHECK1-NEXT: [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
-// CHECK1-NEXT: ret i32 [[TMP16]]
+// CHECK1-NEXT: [[TMP20:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
+// CHECK1-NEXT: ret i32 [[TMP20]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -388,7 +396,7 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -405,9 +413,17 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 56088, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -415,8 +431,8 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[A3:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x [456 x i32]], ptr [[A3]], i32 0, i32 0
// CHECK3-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds [456 x i32], ptr [[ARRAYIDX]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
-// CHECK3-NEXT: ret i32 [[TMP16]]
+// CHECK3-NEXT: [[TMP20:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
+// CHECK3-NEXT: ret i32 [[TMP20]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -843,7 +859,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[ADD:%.*]] = add nsw i64 [[TMP35]], 1
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP36]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP36]], align 4
// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 5, ptr [[TMP37]], align 4
// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -860,20 +876,28 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP43]], align 8
// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 [[ADD]], ptr [[TMP44]], align 8
-// CHECK9-NEXT: [[TMP45:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l86.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP46:%.*]] = icmp ne i32 [[TMP45]], 0
-// CHECK9-NEXT: br i1 [[TMP46]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP46]], align 4
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP47]], align 4
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP48]], align 4
+// CHECK9-NEXT: [[TMP49:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l86.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP50:%.*]] = icmp ne i32 [[TMP49]], 0
+// CHECK9-NEXT: br i1 [[TMP50]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l86(i64 [[TMP7]], i64 [[TMP9]], i64 [[TMP1]], i64 [[TMP3]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP47:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef signext [[TMP47]])
+// CHECK9-NEXT: [[TMP51:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef signext [[TMP51]])
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK9-NEXT: [[TMP48:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP48]])
-// CHECK9-NEXT: [[TMP49:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP49]]
+// CHECK9-NEXT: [[TMP52:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP52]])
+// CHECK9-NEXT: [[TMP53:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP53]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l86
@@ -1221,7 +1245,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK9-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK9-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1238,9 +1262,17 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 20, ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l72.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK9-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK9-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l72.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK9-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l72(ptr [[A]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1536,7 +1568,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[ADD:%.*]] = add nsw i64 [[TMP34]], 1
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP35]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP35]], align 4
// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 5, ptr [[TMP36]], align 4
// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1553,20 +1585,28 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP42]], align 4
// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 [[ADD]], ptr [[TMP43]], align 8
-// CHECK11-NEXT: [[TMP44:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l86.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP45:%.*]] = icmp ne i32 [[TMP44]], 0
-// CHECK11-NEXT: br i1 [[TMP45]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP44]], align 8
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP47]], align 4
+// CHECK11-NEXT: [[TMP48:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l86.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP49:%.*]] = icmp ne i32 [[TMP48]], 0
+// CHECK11-NEXT: br i1 [[TMP49]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l86(i32 [[TMP5]], i32 [[TMP7]], i32 [[TMP0]], i32 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP46:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef [[TMP46]])
+// CHECK11-NEXT: [[TMP50:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef [[TMP50]])
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK11-NEXT: [[TMP47:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP47]])
-// CHECK11-NEXT: [[TMP48:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP48]]
+// CHECK11-NEXT: [[TMP51:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP51]])
+// CHECK11-NEXT: [[TMP52:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP52]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l86
@@ -1916,7 +1956,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK11-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK11-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1933,9 +1973,17 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK11-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 20, ptr [[TMP13]], align 8
-// CHECK11-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l72.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK11-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK11-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l72.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK11-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l72(ptr [[A]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/teams_distribute_parallel_for_simd_dist_schedule_codegen.cpp b/clang/test/OpenMP/teams_distribute_parallel_for_simd_dist_schedule_codegen.cpp
index fb1fd5bf59aa3..c2579ccf475b2 100644
--- a/clang/test/OpenMP/teams_distribute_parallel_for_simd_dist_schedule_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_parallel_for_simd_dist_schedule_codegen.cpp
@@ -183,7 +183,7 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -200,87 +200,111 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.4, ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l33.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.4, ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l33.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK1-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK1-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK1: omp_offload.failed8:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l33(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK1: omp_offload.cont9:
// CHECK1-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A10]], ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A10]], ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.8, ptr [[TMP41]], align 8
-// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP42]], align 8
-// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP43]], align 8
-// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP44]], align 8
-// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK1-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l38.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK1-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK1-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 8
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 8
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.8, ptr [[TMP49]], align 8
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP50]], align 8
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP51]], align 8
+// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP52]], align 8
+// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK1-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l38.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK1-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK1-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK1: omp_offload.failed16:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l38(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK1: omp_offload.cont17:
// CHECK1-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A18]], i64 0, i64 0
-// CHECK1-NEXT: [[TMP48:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK1-NEXT: ret i32 [[TMP48]]
+// CHECK1-NEXT: [[TMP60:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK1-NEXT: ret i32 [[TMP60]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -842,7 +866,7 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -859,87 +883,111 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 4
-// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A2]], ptr [[TMP17]], align 4
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 4
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A2]], ptr [[TMP21]], align 4
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.4, ptr [[TMP25]], align 4
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP26]], align 4
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK3-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l33.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK3-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK3-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.4, ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK3-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l33.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK3-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK3-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK3: omp_offload.failed8:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l33(ptr [[THIS1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK3: omp_offload.cont9:
// CHECK3-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 4
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A10]], ptr [[TMP33]], align 4
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP34]], align 4
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A10]], ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP42]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 4
-// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 4
-// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.8, ptr [[TMP41]], align 4
-// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP42]], align 4
-// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP44]], align 4
-// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK3-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK3-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l38.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK3-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK3-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 4
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 4
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.8, ptr [[TMP49]], align 4
+// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP50]], align 4
+// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP52]], align 4
+// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK3-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK3-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l38.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK3-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK3-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK3: omp_offload.failed16:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l38(ptr [[THIS1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK3: omp_offload.cont17:
// CHECK3-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A18]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP48:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK3-NEXT: ret i32 [[TMP48]]
+// CHECK3-NEXT: [[TMP60:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK3-NEXT: ret i32 [[TMP60]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -1782,7 +1830,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1799,157 +1847,181 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK9-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l108.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK9-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l108.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK9-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l108(i64 [[TMP4]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK9-NEXT: [[TMP35:%.*]] = load i64, ptr [[N_CASTED3]], align 8
-// CHECK9-NEXT: [[TMP36:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK9-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK9-NEXT: [[TMP39:%.*]] = load i64, ptr [[N_CASTED3]], align 8
+// CHECK9-NEXT: [[TMP40:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK9-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES7]], ptr align 8 @.offload_sizes.4, i64 24, i1 false)
-// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP35]], ptr [[TMP37]], align 8
-// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP35]], ptr [[TMP38]], align 8
-// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP40]], align 8
-// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP41]], align 8
-// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP42]], align 8
-// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP43]], align 8
-// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 8
-// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK9-NEXT: store i64 [[TMP36]], ptr [[TMP45]], align 8
-// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP39]], ptr [[TMP41]], align 8
+// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP39]], ptr [[TMP42]], align 8
+// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP44]], align 8
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
// CHECK9-NEXT: store ptr null, ptr [[TMP46]], align 8
-// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP50:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP50]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK9-NEXT: [[TMP51:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK9-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP51]], 0
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP47]], align 8
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 8
+// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK9-NEXT: store i64 [[TMP40]], ptr [[TMP49]], align 8
+// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP50]], align 8
+// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP54:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP54]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK9-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK9-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP55]], 0
// CHECK9-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK9-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK9-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK9-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK9-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP52]], 1
-// CHECK9-NEXT: [[TMP53:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK9-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK9-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP56]], 1
+// CHECK9-NEXT: [[TMP57:%.*]] = zext i32 [[ADD14]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK9-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK9-NEXT: store i32 3, ptr [[TMP55]], align 4
-// CHECK9-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP47]], ptr [[TMP56]], align 8
-// CHECK9-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 8
-// CHECK9-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 8
-// CHECK9-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP59]], align 8
-// CHECK9-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP60]], align 8
-// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP61]], align 8
-// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP53]], ptr [[TMP62]], align 8
-// CHECK9-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l113.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK9-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
-// CHECK9-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK9-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP58]], align 4
+// CHECK9-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK9-NEXT: store i32 3, ptr [[TMP59]], align 4
+// CHECK9-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP51]], ptr [[TMP60]], align 8
+// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 8
+// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 8
+// CHECK9-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP63]], align 8
+// CHECK9-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP64]], align 8
+// CHECK9-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP65]], align 8
+// CHECK9-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP57]], ptr [[TMP66]], align 8
+// CHECK9-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP67]], align 8
+// CHECK9-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP68]], align 4
+// CHECK9-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK9-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP70]], align 4
+// CHECK9-NEXT: [[TMP71:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l113.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK9-NEXT: [[TMP72:%.*]] = icmp ne i32 [[TMP71]], 0
+// CHECK9-NEXT: br i1 [[TMP72]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK9: omp_offload.failed16:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l113(i64 [[TMP35]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l113(i64 [[TMP39]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK9: omp_offload.cont17:
-// CHECK9-NEXT: [[TMP65:%.*]] = load i32, ptr [[M]], align 4
-// CHECK9-NEXT: store i32 [[TMP65]], ptr [[M_CASTED]], align 4
-// CHECK9-NEXT: [[TMP66:%.*]] = load i64, ptr [[M_CASTED]], align 8
-// CHECK9-NEXT: [[TMP67:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP67]], ptr [[N_CASTED18]], align 4
-// CHECK9-NEXT: [[TMP68:%.*]] = load i64, ptr [[N_CASTED18]], align 8
-// CHECK9-NEXT: [[TMP69:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK9-NEXT: [[TMP73:%.*]] = load i32, ptr [[M]], align 4
+// CHECK9-NEXT: store i32 [[TMP73]], ptr [[M_CASTED]], align 4
+// CHECK9-NEXT: [[TMP74:%.*]] = load i64, ptr [[M_CASTED]], align 8
+// CHECK9-NEXT: [[TMP75:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP75]], ptr [[N_CASTED18]], align 4
+// CHECK9-NEXT: [[TMP76:%.*]] = load i64, ptr [[N_CASTED18]], align 8
+// CHECK9-NEXT: [[TMP77:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK9-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES22]], ptr align 8 @.offload_sizes.8, i64 32, i1 false)
-// CHECK9-NEXT: [[TMP70:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP66]], ptr [[TMP70]], align 8
-// CHECK9-NEXT: [[TMP71:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP66]], ptr [[TMP71]], align 8
-// CHECK9-NEXT: [[TMP72:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP72]], align 8
-// CHECK9-NEXT: [[TMP73:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP68]], ptr [[TMP73]], align 8
-// CHECK9-NEXT: [[TMP74:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP68]], ptr [[TMP74]], align 8
-// CHECK9-NEXT: [[TMP75:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP75]], align 8
-// CHECK9-NEXT: [[TMP76:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP76]], align 8
-// CHECK9-NEXT: [[TMP77:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP77]], align 8
-// CHECK9-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP78]], align 8
-// CHECK9-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP79]], align 8
-// CHECK9-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP80]], align 8
-// CHECK9-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 3
-// CHECK9-NEXT: store i64 [[TMP69]], ptr [[TMP81]], align 8
-// CHECK9-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP82]], align 8
-// CHECK9-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP86:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP86]], ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK9-NEXT: [[TMP87:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK9-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP87]], 0
+// CHECK9-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP74]], ptr [[TMP78]], align 8
+// CHECK9-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP74]], ptr [[TMP79]], align 8
+// CHECK9-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP80]], align 8
+// CHECK9-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP76]], ptr [[TMP81]], align 8
+// CHECK9-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP76]], ptr [[TMP82]], align 8
+// CHECK9-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP83]], align 8
+// CHECK9-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP84]], align 8
+// CHECK9-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP85]], align 8
+// CHECK9-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP86]], align 8
+// CHECK9-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP87]], align 8
+// CHECK9-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP88]], align 8
+// CHECK9-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 3
+// CHECK9-NEXT: store i64 [[TMP77]], ptr [[TMP89]], align 8
+// CHECK9-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 3
+// CHECK9-NEXT: store ptr null, ptr [[TMP90]], align 8
+// CHECK9-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP94:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP94]], ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK9-NEXT: [[TMP95:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK9-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP95]], 0
// CHECK9-NEXT: [[DIV27:%.*]] = sdiv i32 [[SUB26]], 1
// CHECK9-NEXT: [[SUB28:%.*]] = sub nsw i32 [[DIV27]], 1
// CHECK9-NEXT: store i32 [[SUB28]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK9-NEXT: [[TMP88:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK9-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP88]], 1
-// CHECK9-NEXT: [[TMP89:%.*]] = zext i32 [[ADD29]] to i64
+// CHECK9-NEXT: [[TMP96:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK9-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP96]], 1
+// CHECK9-NEXT: [[TMP97:%.*]] = zext i32 [[ADD29]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS30:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP90]], align 4
-// CHECK9-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
-// CHECK9-NEXT: store i32 4, ptr [[TMP91]], align 4
-// CHECK9-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP83]], ptr [[TMP92]], align 8
-// CHECK9-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP84]], ptr [[TMP93]], align 8
-// CHECK9-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP85]], ptr [[TMP94]], align 8
-// CHECK9-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP95]], align 8
-// CHECK9-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP96]], align 8
-// CHECK9-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP97]], align 8
-// CHECK9-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP89]], ptr [[TMP98]], align 8
-// CHECK9-NEXT: [[TMP99:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l118.region_id, ptr [[KERNEL_ARGS30]])
-// CHECK9-NEXT: [[TMP100:%.*]] = icmp ne i32 [[TMP99]], 0
-// CHECK9-NEXT: br i1 [[TMP100]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
+// CHECK9-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP98]], align 4
+// CHECK9-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
+// CHECK9-NEXT: store i32 4, ptr [[TMP99]], align 4
+// CHECK9-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP91]], ptr [[TMP100]], align 8
+// CHECK9-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP92]], ptr [[TMP101]], align 8
+// CHECK9-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP93]], ptr [[TMP102]], align 8
+// CHECK9-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP103]], align 8
+// CHECK9-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP104]], align 8
+// CHECK9-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP105]], align 8
+// CHECK9-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP97]], ptr [[TMP106]], align 8
+// CHECK9-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP107]], align 8
+// CHECK9-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP108]], align 4
+// CHECK9-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP109]], align 4
+// CHECK9-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP110]], align 4
+// CHECK9-NEXT: [[TMP111:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l118.region_id, ptr [[KERNEL_ARGS30]])
+// CHECK9-NEXT: [[TMP112:%.*]] = icmp ne i32 [[TMP111]], 0
+// CHECK9-NEXT: br i1 [[TMP112]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
// CHECK9: omp_offload.failed31:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l118(i64 [[TMP66]], i64 [[TMP68]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l118(i64 [[TMP74]], i64 [[TMP76]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT32]]
// CHECK9: omp_offload.cont32:
-// CHECK9-NEXT: [[TMP101:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP101]])
+// CHECK9-NEXT: [[TMP113:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP113]])
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK9-NEXT: [[TMP102:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP102]])
-// CHECK9-NEXT: [[TMP103:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP103]]
+// CHECK9-NEXT: [[TMP114:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP114]])
+// CHECK9-NEXT: [[TMP115:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP115]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l108
@@ -2730,7 +2802,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK9-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK9-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2747,88 +2819,112 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l86.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK9-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK9-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l86.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK9-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l86(ptr [[A]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP16]], align 8
-// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP17]], align 8
-// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP20]], align 8
+// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP21]], align 8
+// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK9-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.16, ptr [[TMP25]], align 8
-// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP26]], align 8
-// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK9-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK9-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l91.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK9-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK9-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK9-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.16, ptr [[TMP29]], align 8
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP30]], align 8
+// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK9-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK9-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l91.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK9-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK9-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK9: omp_offload.failed6:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l91(ptr [[A]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK9: omp_offload.cont7:
-// CHECK9-NEXT: [[TMP32:%.*]] = load i32, ptr [[M]], align 4
-// CHECK9-NEXT: store i32 [[TMP32]], ptr [[M_CASTED]], align 4
-// CHECK9-NEXT: [[TMP33:%.*]] = load i64, ptr [[M_CASTED]], align 8
-// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP33]], ptr [[TMP34]], align 8
-// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP33]], ptr [[TMP35]], align 8
-// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP36]], align 8
-// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP37]], align 8
-// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP38]], align 8
-// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP40:%.*]] = load i32, ptr [[M]], align 4
+// CHECK9-NEXT: store i32 [[TMP40]], ptr [[M_CASTED]], align 4
+// CHECK9-NEXT: [[TMP41:%.*]] = load i64, ptr [[M_CASTED]], align 8
+// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP41]], ptr [[TMP42]], align 8
+// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP41]], ptr [[TMP43]], align 8
+// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP44]], align 8
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP46]], align 8
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP47]], align 8
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP42]], align 4
-// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK9-NEXT: store i32 2, ptr [[TMP43]], align 4
-// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP40]], ptr [[TMP44]], align 8
-// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP41]], ptr [[TMP45]], align 8
-// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.20, ptr [[TMP46]], align 8
-// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.21, ptr [[TMP47]], align 8
-// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP48]], align 8
-// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP49]], align 8
-// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK9-NEXT: store i64 10, ptr [[TMP50]], align 8
-// CHECK9-NEXT: [[TMP51:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l96.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK9-NEXT: [[TMP52:%.*]] = icmp ne i32 [[TMP51]], 0
-// CHECK9-NEXT: br i1 [[TMP52]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP50]], align 4
+// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK9-NEXT: store i32 2, ptr [[TMP51]], align 4
+// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP48]], ptr [[TMP52]], align 8
+// CHECK9-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 8
+// CHECK9-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.20, ptr [[TMP54]], align 8
+// CHECK9-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.21, ptr [[TMP55]], align 8
+// CHECK9-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP56]], align 8
+// CHECK9-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP57]], align 8
+// CHECK9-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK9-NEXT: store i64 10, ptr [[TMP58]], align 8
+// CHECK9-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP59]], align 8
+// CHECK9-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP60]], align 4
+// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4
+// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP62]], align 4
+// CHECK9-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l96.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK9-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
+// CHECK9-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK9: omp_offload.failed13:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l96(i64 [[TMP33]], ptr [[A]]) #[[ATTR3]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l96(i64 [[TMP41]], ptr [[A]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT14]]
// CHECK9: omp_offload.cont14:
// CHECK9-NEXT: ret i32 0
@@ -3457,7 +3553,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3474,159 +3570,183 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK11-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l108.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK11-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l108.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK11-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l108(i32 [[TMP3]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK11-NEXT: [[TMP35:%.*]] = load i32, ptr [[N_CASTED3]], align 4
-// CHECK11-NEXT: [[TMP36:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK11-NEXT: [[TMP37:%.*]] = sext i32 [[TMP36]] to i64
+// CHECK11-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK11-NEXT: [[TMP39:%.*]] = load i32, ptr [[N_CASTED3]], align 4
+// CHECK11-NEXT: [[TMP40:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK11-NEXT: [[TMP41:%.*]] = sext i32 [[TMP40]] to i64
// CHECK11-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES7]], ptr align 4 @.offload_sizes.4, i32 24, i1 false)
-// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP35]], ptr [[TMP38]], align 4
-// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP35]], ptr [[TMP39]], align 4
-// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP40]], align 4
-// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP41]], align 4
-// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP42]], align 4
-// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 4
-// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP45]], align 4
-// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK11-NEXT: store i64 [[TMP37]], ptr [[TMP46]], align 4
-// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP39]], ptr [[TMP42]], align 4
+// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP39]], ptr [[TMP43]], align 4
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP44]], align 4
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
// CHECK11-NEXT: store ptr null, ptr [[TMP47]], align 4
-// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP51:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP51]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK11-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK11-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP52]], 0
+// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 4
+// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP49]], align 4
+// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK11-NEXT: store i64 [[TMP41]], ptr [[TMP50]], align 4
+// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP54:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP55:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP55]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK11-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK11-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP56]], 0
// CHECK11-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK11-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK11-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK11-NEXT: [[TMP53:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK11-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP53]], 1
-// CHECK11-NEXT: [[TMP54:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK11-NEXT: [[TMP57:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK11-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP57]], 1
+// CHECK11-NEXT: [[TMP58:%.*]] = zext i32 [[ADD14]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP55]], align 4
-// CHECK11-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK11-NEXT: store i32 3, ptr [[TMP56]], align 4
-// CHECK11-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 4
-// CHECK11-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 4
-// CHECK11-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP50]], ptr [[TMP59]], align 4
-// CHECK11-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP60]], align 4
-// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP61]], align 4
-// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP62]], align 4
-// CHECK11-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP54]], ptr [[TMP63]], align 8
-// CHECK11-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l113.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK11-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
-// CHECK11-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK11-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP59]], align 4
+// CHECK11-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK11-NEXT: store i32 3, ptr [[TMP60]], align 4
+// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 4
+// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 4
+// CHECK11-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP54]], ptr [[TMP63]], align 4
+// CHECK11-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP64]], align 4
+// CHECK11-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP65]], align 4
+// CHECK11-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP66]], align 4
+// CHECK11-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP58]], ptr [[TMP67]], align 8
+// CHECK11-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP68]], align 8
+// CHECK11-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK11-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP70]], align 4
+// CHECK11-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP71]], align 4
+// CHECK11-NEXT: [[TMP72:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l113.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK11-NEXT: [[TMP73:%.*]] = icmp ne i32 [[TMP72]], 0
+// CHECK11-NEXT: br i1 [[TMP73]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK11: omp_offload.failed16:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l113(i32 [[TMP35]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l113(i32 [[TMP39]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK11: omp_offload.cont17:
-// CHECK11-NEXT: [[TMP66:%.*]] = load i32, ptr [[M]], align 4
-// CHECK11-NEXT: store i32 [[TMP66]], ptr [[M_CASTED]], align 4
-// CHECK11-NEXT: [[TMP67:%.*]] = load i32, ptr [[M_CASTED]], align 4
-// CHECK11-NEXT: [[TMP68:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP68]], ptr [[N_CASTED18]], align 4
-// CHECK11-NEXT: [[TMP69:%.*]] = load i32, ptr [[N_CASTED18]], align 4
-// CHECK11-NEXT: [[TMP70:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK11-NEXT: [[TMP71:%.*]] = sext i32 [[TMP70]] to i64
+// CHECK11-NEXT: [[TMP74:%.*]] = load i32, ptr [[M]], align 4
+// CHECK11-NEXT: store i32 [[TMP74]], ptr [[M_CASTED]], align 4
+// CHECK11-NEXT: [[TMP75:%.*]] = load i32, ptr [[M_CASTED]], align 4
+// CHECK11-NEXT: [[TMP76:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP76]], ptr [[N_CASTED18]], align 4
+// CHECK11-NEXT: [[TMP77:%.*]] = load i32, ptr [[N_CASTED18]], align 4
+// CHECK11-NEXT: [[TMP78:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK11-NEXT: [[TMP79:%.*]] = sext i32 [[TMP78]] to i64
// CHECK11-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES22]], ptr align 4 @.offload_sizes.8, i32 32, i1 false)
-// CHECK11-NEXT: [[TMP72:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP67]], ptr [[TMP72]], align 4
-// CHECK11-NEXT: [[TMP73:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP67]], ptr [[TMP73]], align 4
-// CHECK11-NEXT: [[TMP74:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP74]], align 4
-// CHECK11-NEXT: [[TMP75:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP69]], ptr [[TMP75]], align 4
-// CHECK11-NEXT: [[TMP76:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP69]], ptr [[TMP76]], align 4
-// CHECK11-NEXT: [[TMP77:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP77]], align 4
-// CHECK11-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP78]], align 4
-// CHECK11-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP79]], align 4
-// CHECK11-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP80]], align 4
-// CHECK11-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP81]], align 4
-// CHECK11-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP82]], align 4
-// CHECK11-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 3
-// CHECK11-NEXT: store i64 [[TMP71]], ptr [[TMP83]], align 4
-// CHECK11-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP84]], align 4
-// CHECK11-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP88:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP88]], ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK11-NEXT: [[TMP89:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK11-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP89]], 0
+// CHECK11-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP75]], ptr [[TMP80]], align 4
+// CHECK11-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP75]], ptr [[TMP81]], align 4
+// CHECK11-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP82]], align 4
+// CHECK11-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP77]], ptr [[TMP83]], align 4
+// CHECK11-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP77]], ptr [[TMP84]], align 4
+// CHECK11-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP85]], align 4
+// CHECK11-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP86]], align 4
+// CHECK11-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP87]], align 4
+// CHECK11-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP88]], align 4
+// CHECK11-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP89]], align 4
+// CHECK11-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP90]], align 4
+// CHECK11-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 3
+// CHECK11-NEXT: store i64 [[TMP79]], ptr [[TMP91]], align 4
+// CHECK11-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 3
+// CHECK11-NEXT: store ptr null, ptr [[TMP92]], align 4
+// CHECK11-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP94:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP95:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP96:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP96]], ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK11-NEXT: [[TMP97:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK11-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP97]], 0
// CHECK11-NEXT: [[DIV27:%.*]] = sdiv i32 [[SUB26]], 1
// CHECK11-NEXT: [[SUB28:%.*]] = sub nsw i32 [[DIV27]], 1
// CHECK11-NEXT: store i32 [[SUB28]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK11-NEXT: [[TMP90:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK11-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP90]], 1
-// CHECK11-NEXT: [[TMP91:%.*]] = zext i32 [[ADD29]] to i64
+// CHECK11-NEXT: [[TMP98:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK11-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP98]], 1
+// CHECK11-NEXT: [[TMP99:%.*]] = zext i32 [[ADD29]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS30:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP92]], align 4
-// CHECK11-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
-// CHECK11-NEXT: store i32 4, ptr [[TMP93]], align 4
-// CHECK11-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP85]], ptr [[TMP94]], align 4
-// CHECK11-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP86]], ptr [[TMP95]], align 4
-// CHECK11-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP87]], ptr [[TMP96]], align 4
-// CHECK11-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP97]], align 4
-// CHECK11-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP98]], align 4
-// CHECK11-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP99]], align 4
-// CHECK11-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP91]], ptr [[TMP100]], align 8
-// CHECK11-NEXT: [[TMP101:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l118.region_id, ptr [[KERNEL_ARGS30]])
-// CHECK11-NEXT: [[TMP102:%.*]] = icmp ne i32 [[TMP101]], 0
-// CHECK11-NEXT: br i1 [[TMP102]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
+// CHECK11-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP100]], align 4
+// CHECK11-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
+// CHECK11-NEXT: store i32 4, ptr [[TMP101]], align 4
+// CHECK11-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP93]], ptr [[TMP102]], align 4
+// CHECK11-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP94]], ptr [[TMP103]], align 4
+// CHECK11-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP95]], ptr [[TMP104]], align 4
+// CHECK11-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP105]], align 4
+// CHECK11-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP106]], align 4
+// CHECK11-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP107]], align 4
+// CHECK11-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP99]], ptr [[TMP108]], align 8
+// CHECK11-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP109]], align 8
+// CHECK11-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP110]], align 4
+// CHECK11-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP111]], align 4
+// CHECK11-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP112]], align 4
+// CHECK11-NEXT: [[TMP113:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l118.region_id, ptr [[KERNEL_ARGS30]])
+// CHECK11-NEXT: [[TMP114:%.*]] = icmp ne i32 [[TMP113]], 0
+// CHECK11-NEXT: br i1 [[TMP114]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
// CHECK11: omp_offload.failed31:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l118(i32 [[TMP67]], i32 [[TMP69]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l118(i32 [[TMP75]], i32 [[TMP77]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT32]]
// CHECK11: omp_offload.cont32:
-// CHECK11-NEXT: [[TMP103:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP103]])
+// CHECK11-NEXT: [[TMP115:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP115]])
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK11-NEXT: [[TMP104:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP104]])
-// CHECK11-NEXT: [[TMP105:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP105]]
+// CHECK11-NEXT: [[TMP116:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP116]])
+// CHECK11-NEXT: [[TMP117:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP117]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l108
@@ -4392,7 +4512,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK11-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK11-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -4409,88 +4529,112 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK11-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK11-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l86.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK11-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK11-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l86.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK11-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l86(ptr [[A]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP16]], align 4
-// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP17]], align 4
-// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP20]], align 4
+// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP21]], align 4
+// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK11-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.16, ptr [[TMP25]], align 4
-// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP26]], align 4
-// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK11-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK11-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l91.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK11-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK11-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK11-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.16, ptr [[TMP29]], align 4
+// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP30]], align 4
+// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK11-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK11-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l91.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK11-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK11-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK11: omp_offload.failed6:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l91(ptr [[A]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK11: omp_offload.cont7:
-// CHECK11-NEXT: [[TMP32:%.*]] = load i32, ptr [[M]], align 4
-// CHECK11-NEXT: store i32 [[TMP32]], ptr [[M_CASTED]], align 4
-// CHECK11-NEXT: [[TMP33:%.*]] = load i32, ptr [[M_CASTED]], align 4
-// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP33]], ptr [[TMP34]], align 4
-// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP33]], ptr [[TMP35]], align 4
-// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP36]], align 4
-// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP37]], align 4
-// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP38]], align 4
-// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP39]], align 4
-// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP40:%.*]] = load i32, ptr [[M]], align 4
+// CHECK11-NEXT: store i32 [[TMP40]], ptr [[M_CASTED]], align 4
+// CHECK11-NEXT: [[TMP41:%.*]] = load i32, ptr [[M_CASTED]], align 4
+// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP41]], ptr [[TMP42]], align 4
+// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP41]], ptr [[TMP43]], align 4
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP44]], align 4
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP47]], align 4
+// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP42]], align 4
-// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK11-NEXT: store i32 2, ptr [[TMP43]], align 4
-// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP40]], ptr [[TMP44]], align 4
-// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP41]], ptr [[TMP45]], align 4
-// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.20, ptr [[TMP46]], align 4
-// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.21, ptr [[TMP47]], align 4
-// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP48]], align 4
-// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP49]], align 4
-// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK11-NEXT: store i64 10, ptr [[TMP50]], align 8
-// CHECK11-NEXT: [[TMP51:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l96.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK11-NEXT: [[TMP52:%.*]] = icmp ne i32 [[TMP51]], 0
-// CHECK11-NEXT: br i1 [[TMP52]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP50]], align 4
+// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK11-NEXT: store i32 2, ptr [[TMP51]], align 4
+// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP48]], ptr [[TMP52]], align 4
+// CHECK11-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 4
+// CHECK11-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.20, ptr [[TMP54]], align 4
+// CHECK11-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.21, ptr [[TMP55]], align 4
+// CHECK11-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP56]], align 4
+// CHECK11-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP57]], align 4
+// CHECK11-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK11-NEXT: store i64 10, ptr [[TMP58]], align 8
+// CHECK11-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP59]], align 8
+// CHECK11-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP60]], align 4
+// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4
+// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP62]], align 4
+// CHECK11-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l96.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK11-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
+// CHECK11-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK11: omp_offload.failed13:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l96(i32 [[TMP33]], ptr [[A]]) #[[ATTR3]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l96(i32 [[TMP41]], ptr [[A]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT14]]
// CHECK11: omp_offload.cont14:
// CHECK11-NEXT: ret i32 0
diff --git a/clang/test/OpenMP/teams_distribute_parallel_for_simd_firstprivate_codegen.cpp b/clang/test/OpenMP/teams_distribute_parallel_for_simd_firstprivate_codegen.cpp
index 42e64c537b6fa..c36ba04201257 100644
--- a/clang/test/OpenMP/teams_distribute_parallel_for_simd_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_parallel_for_simd_firstprivate_codegen.cpp
@@ -343,7 +343,7 @@ int main() {
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 5, ptr [[TMP22]], align 4
// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -360,9 +360,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l101.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK1-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l101.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK1-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l101(i64 [[TMP1]], ptr @vec, ptr @s_arr, ptr @var, i64 [[TMP3]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -760,7 +768,7 @@ int main() {
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -777,27 +785,35 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK1-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK1-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK1-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56(i64 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK1-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK1: arraydestroy.body:
-// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK1-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK1-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK1: arraydestroy.done2:
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK1-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK1-NEXT: ret i32 [[TMP32]]
+// CHECK1-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK1-NEXT: ret i32 [[TMP36]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN2StC2Ev
@@ -1418,7 +1434,7 @@ int main() {
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 5, ptr [[TMP22]], align 4
// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1435,9 +1451,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4
// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP29]], align 8
-// CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l101.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK3-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK3-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l101.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK3-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l101(i32 [[TMP1]], ptr @vec, ptr @s_arr, ptr @var, i32 [[TMP3]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1829,7 +1853,7 @@ int main() {
// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1846,27 +1870,35 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP27]], align 4
// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK3-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK3-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK3-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56(i32 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK3-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK3-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK3: arraydestroy.body:
-// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK3-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK3-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK3: arraydestroy.done2:
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK3-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK3-NEXT: ret i32 [[TMP32]]
+// CHECK3-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK3-NEXT: ret i32 [[TMP36]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN2StC2Ev
diff --git a/clang/test/OpenMP/teams_distribute_parallel_for_simd_if_codegen.cpp b/clang/test/OpenMP/teams_distribute_parallel_for_simd_if_codegen.cpp
index aec1d9d0cd574..7d8692db7efb7 100644
--- a/clang/test/OpenMP/teams_distribute_parallel_for_simd_if_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_parallel_for_simd_if_codegen.cpp
@@ -117,7 +117,7 @@ int main() {
// CHECK1-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -134,35 +134,51 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l43.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l43.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l43() #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l47.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l47.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK1-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l47() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -493,7 +509,7 @@ int main() {
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -510,85 +526,110 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l76.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l76.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l76() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l84.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l84.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK1-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l84() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK1: omp_offload.cont4:
-// CHECK1-NEXT: [[TMP22:%.*]] = load i32, ptr @Arg, align 4
-// CHECK1-NEXT: store i32 [[TMP22]], ptr [[ARG_CASTED]], align 4
-// CHECK1-NEXT: [[TMP23:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP23]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP23]], ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP29:%.*]] = load i32, ptr @Arg, align 4
-// CHECK1-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK1-NEXT: [[TMP30:%.*]] = load i32, ptr @Arg, align 4
+// CHECK1-NEXT: store i32 [[TMP30]], ptr [[ARG_CASTED]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP31]], ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP31]], ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP37:%.*]] = load i32, ptr @Arg, align 4
+// CHECK1-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP37]], 0
// CHECK1-NEXT: [[FROMBOOL:%.*]] = zext i1 [[TOBOOL]] to i8
// CHECK1-NEXT: store i8 [[FROMBOOL]], ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TMP30:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TOBOOL5:%.*]] = trunc i8 [[TMP30]] to i1
-// CHECK1-NEXT: [[TMP31:%.*]] = select i1 [[TOBOOL5]], i32 0, i32 1
+// CHECK1-NEXT: [[TMP38:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK1-NEXT: [[TOBOOL5:%.*]] = trunc i8 [[TMP38]] to i1
+// CHECK1-NEXT: [[TMP39:%.*]] = select i1 [[TOBOOL5]], i32 0, i32 1
+// CHECK1-NEXT: [[TMP40:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP39]], 0
// CHECK1-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP32]], align 4
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP33]], align 4
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP27]], ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP28]], ptr [[TMP35]], align 8
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP36]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP38]], align 8
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP31]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK1-NEXT: [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
-// CHECK1-NEXT: br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP41]], align 4
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP42]], align 4
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP35]], ptr [[TMP43]], align 8
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP36]], ptr [[TMP44]], align 8
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP45]], align 8
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP46]], align 8
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP47]], align 8
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP48]], align 8
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP49]], align 8
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP50]], align 8
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP51]], align 4
+// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP40]], ptr [[TMP52]], align 4
+// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP53]], align 4
+// CHECK1-NEXT: [[TMP54:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP39]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK1-NEXT: [[TMP55:%.*]] = icmp ne i32 [[TMP54]], 0
+// CHECK1-NEXT: br i1 [[TMP55]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK1: omp_offload.failed8:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92(i64 [[TMP23]]) #[[ATTR2]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92(i64 [[TMP31]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK1: omp_offload.cont9:
-// CHECK1-NEXT: [[TMP43:%.*]] = load i32, ptr @Arg, align 4
-// CHECK1-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP43]])
+// CHECK1-NEXT: [[TMP56:%.*]] = load i32, ptr @Arg, align 4
+// CHECK1-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP56]])
// CHECK1-NEXT: ret i32 [[CALL]]
//
//
@@ -1097,7 +1138,7 @@ int main() {
// CHECK1-NEXT: store i32 [[ARG]], ptr [[ARG_ADDR]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1114,81 +1155,106 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l57.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l57.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l57() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l62.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l62.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK1-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l62() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK1: omp_offload.cont4:
-// CHECK1-NEXT: [[TMP22:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
-// CHECK1-NEXT: store i32 [[TMP22]], ptr [[ARG_CASTED]], align 4
-// CHECK1-NEXT: [[TMP23:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP23]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP23]], ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP29:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
-// CHECK1-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK1-NEXT: [[TMP30:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
+// CHECK1-NEXT: store i32 [[TMP30]], ptr [[ARG_CASTED]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP31]], ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP31]], ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP37:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
+// CHECK1-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP37]], 0
// CHECK1-NEXT: [[FROMBOOL:%.*]] = zext i1 [[TOBOOL]] to i8
// CHECK1-NEXT: store i8 [[FROMBOOL]], ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TMP30:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TOBOOL5:%.*]] = trunc i8 [[TMP30]] to i1
-// CHECK1-NEXT: [[TMP31:%.*]] = select i1 [[TOBOOL5]], i32 0, i32 1
+// CHECK1-NEXT: [[TMP38:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK1-NEXT: [[TOBOOL5:%.*]] = trunc i8 [[TMP38]] to i1
+// CHECK1-NEXT: [[TMP39:%.*]] = select i1 [[TOBOOL5]], i32 0, i32 1
+// CHECK1-NEXT: [[TMP40:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP39]], 0
// CHECK1-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP32]], align 4
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP33]], align 4
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP27]], ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP28]], ptr [[TMP35]], align 8
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.16, ptr [[TMP36]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP38]], align 8
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP31]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l67.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK1-NEXT: [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
-// CHECK1-NEXT: br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP41]], align 4
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP42]], align 4
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP35]], ptr [[TMP43]], align 8
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP36]], ptr [[TMP44]], align 8
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.16, ptr [[TMP45]], align 8
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP46]], align 8
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP47]], align 8
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP48]], align 8
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP49]], align 8
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP50]], align 8
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP51]], align 4
+// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP40]], ptr [[TMP52]], align 4
+// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP53]], align 4
+// CHECK1-NEXT: [[TMP54:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP39]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l67.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK1-NEXT: [[TMP55:%.*]] = icmp ne i32 [[TMP54]], 0
+// CHECK1-NEXT: br i1 [[TMP55]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK1: omp_offload.failed8:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l67(i64 [[TMP23]]) #[[ATTR2]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l67(i64 [[TMP31]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK1: omp_offload.cont9:
// CHECK1-NEXT: ret i32 0
@@ -1698,7 +1764,7 @@ int main() {
// CHECK3-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1715,35 +1781,51 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l43.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK3-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l43.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK3-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l43() #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK3-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK3-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK3-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK3-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK3-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK3-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK3-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK3-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK3-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l47.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK3-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK3-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK3-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK3-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK3-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l47.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK3-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK3-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK3: omp_offload.failed3:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l47() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -2074,7 +2156,7 @@ int main() {
// CHECK3-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2091,85 +2173,110 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l76.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK3-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l76.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK3-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l76() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK3-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK3-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK3-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK3-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK3-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK3-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK3-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK3-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK3-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l84.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK3-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK3-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK3-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK3-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK3-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l84.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK3-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK3-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK3: omp_offload.failed3:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l84() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK3: omp_offload.cont4:
-// CHECK3-NEXT: [[TMP22:%.*]] = load i32, ptr @Arg, align 4
-// CHECK3-NEXT: store i32 [[TMP22]], ptr [[ARG_CASTED]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK3-NEXT: store i64 [[TMP23]], ptr [[TMP24]], align 8
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK3-NEXT: store i64 [[TMP23]], ptr [[TMP25]], align 8
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP26]], align 8
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP29:%.*]] = load i32, ptr @Arg, align 4
-// CHECK3-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK3-NEXT: [[TMP30:%.*]] = load i32, ptr @Arg, align 4
+// CHECK3-NEXT: store i32 [[TMP30]], ptr [[ARG_CASTED]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK3-NEXT: store i64 [[TMP31]], ptr [[TMP32]], align 8
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK3-NEXT: store i64 [[TMP31]], ptr [[TMP33]], align 8
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP37:%.*]] = load i32, ptr @Arg, align 4
+// CHECK3-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP37]], 0
// CHECK3-NEXT: [[FROMBOOL:%.*]] = zext i1 [[TOBOOL]] to i8
// CHECK3-NEXT: store i8 [[FROMBOOL]], ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK3-NEXT: [[TMP30:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK3-NEXT: [[TOBOOL5:%.*]] = trunc i8 [[TMP30]] to i1
-// CHECK3-NEXT: [[TMP31:%.*]] = select i1 [[TOBOOL5]], i32 0, i32 1
+// CHECK3-NEXT: [[TMP38:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK3-NEXT: [[TOBOOL5:%.*]] = trunc i8 [[TMP38]] to i1
+// CHECK3-NEXT: [[TMP39:%.*]] = select i1 [[TOBOOL5]], i32 0, i32 1
+// CHECK3-NEXT: [[TMP40:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP39]], 0
// CHECK3-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP32]], align 4
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP33]], align 4
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP27]], ptr [[TMP34]], align 8
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP28]], ptr [[TMP35]], align 8
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes, ptr [[TMP36]], align 8
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP37]], align 8
-// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP38]], align 8
-// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK3-NEXT: store i64 100, ptr [[TMP40]], align 8
-// CHECK3-NEXT: [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP31]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK3-NEXT: [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
-// CHECK3-NEXT: br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP42]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP35]], ptr [[TMP43]], align 8
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP36]], ptr [[TMP44]], align 8
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes, ptr [[TMP45]], align 8
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP46]], align 8
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP47]], align 8
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP48]], align 8
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK3-NEXT: store i64 100, ptr [[TMP49]], align 8
+// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP50]], align 8
+// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP51]], align 4
+// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [[TMP40]], ptr [[TMP52]], align 4
+// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP53]], align 4
+// CHECK3-NEXT: [[TMP54:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP39]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK3-NEXT: [[TMP55:%.*]] = icmp ne i32 [[TMP54]], 0
+// CHECK3-NEXT: br i1 [[TMP55]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK3: omp_offload.failed8:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92(i64 [[TMP23]]) #[[ATTR2]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92(i64 [[TMP31]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK3: omp_offload.cont9:
-// CHECK3-NEXT: [[TMP43:%.*]] = load i32, ptr @Arg, align 4
-// CHECK3-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP43]])
+// CHECK3-NEXT: [[TMP56:%.*]] = load i32, ptr @Arg, align 4
+// CHECK3-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP56]])
// CHECK3-NEXT: ret i32 [[CALL]]
//
//
@@ -2908,7 +3015,7 @@ int main() {
// CHECK3-NEXT: store i32 [[ARG]], ptr [[ARG_ADDR]], align 4
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2925,81 +3032,106 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l57.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK3-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l57.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK3-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l57() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK3-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK3-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK3-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK3-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK3-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK3-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK3-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK3-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK3-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l62.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK3-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK3-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK3-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK3-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK3-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l62.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK3-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK3-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK3: omp_offload.failed3:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l62() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK3: omp_offload.cont4:
-// CHECK3-NEXT: [[TMP22:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
-// CHECK3-NEXT: store i32 [[TMP22]], ptr [[ARG_CASTED]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK3-NEXT: store i64 [[TMP23]], ptr [[TMP24]], align 8
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK3-NEXT: store i64 [[TMP23]], ptr [[TMP25]], align 8
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP26]], align 8
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP29:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
-// CHECK3-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK3-NEXT: [[TMP30:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
+// CHECK3-NEXT: store i32 [[TMP30]], ptr [[ARG_CASTED]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK3-NEXT: store i64 [[TMP31]], ptr [[TMP32]], align 8
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK3-NEXT: store i64 [[TMP31]], ptr [[TMP33]], align 8
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP37:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
+// CHECK3-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP37]], 0
// CHECK3-NEXT: [[FROMBOOL:%.*]] = zext i1 [[TOBOOL]] to i8
// CHECK3-NEXT: store i8 [[FROMBOOL]], ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK3-NEXT: [[TMP30:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK3-NEXT: [[TOBOOL5:%.*]] = trunc i8 [[TMP30]] to i1
-// CHECK3-NEXT: [[TMP31:%.*]] = select i1 [[TOBOOL5]], i32 0, i32 1
+// CHECK3-NEXT: [[TMP38:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK3-NEXT: [[TOBOOL5:%.*]] = trunc i8 [[TMP38]] to i1
+// CHECK3-NEXT: [[TMP39:%.*]] = select i1 [[TOBOOL5]], i32 0, i32 1
+// CHECK3-NEXT: [[TMP40:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP39]], 0
// CHECK3-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP32]], align 4
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP33]], align 4
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP27]], ptr [[TMP34]], align 8
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP28]], ptr [[TMP35]], align 8
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.17, ptr [[TMP36]], align 8
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.18, ptr [[TMP37]], align 8
-// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP38]], align 8
-// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK3-NEXT: store i64 100, ptr [[TMP40]], align 8
-// CHECK3-NEXT: [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP31]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l67.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK3-NEXT: [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
-// CHECK3-NEXT: br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP42]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP35]], ptr [[TMP43]], align 8
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP36]], ptr [[TMP44]], align 8
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.17, ptr [[TMP45]], align 8
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.18, ptr [[TMP46]], align 8
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP47]], align 8
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP48]], align 8
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK3-NEXT: store i64 100, ptr [[TMP49]], align 8
+// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP50]], align 8
+// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP51]], align 4
+// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [[TMP40]], ptr [[TMP52]], align 4
+// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP53]], align 4
+// CHECK3-NEXT: [[TMP54:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP39]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l67.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK3-NEXT: [[TMP55:%.*]] = icmp ne i32 [[TMP54]], 0
+// CHECK3-NEXT: br i1 [[TMP55]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK3: omp_offload.failed8:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l67(i64 [[TMP23]]) #[[ATTR2]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l67(i64 [[TMP31]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK3: omp_offload.cont9:
// CHECK3-NEXT: ret i32 0
@@ -4100,7 +4232,7 @@ int main() {
// CHECK9-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK9-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK9-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -4117,35 +4249,51 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK9-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK9-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l43.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK9-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK9-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK9-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l43.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK9-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l43() #[[ATTR2:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK9-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK9-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK9-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK9-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK9-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK9-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK9-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l47.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK9-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK9-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK9-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK9-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK9-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK9-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l47.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK9-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK9-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK9: omp_offload.failed3:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l47() #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -4476,7 +4624,7 @@ int main() {
// CHECK9-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK9-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK9-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -4493,85 +4641,110 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK9-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK9-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l76.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK9-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK9-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK9-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l76.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK9-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l76() #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK9-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK9-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK9-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK9-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK9-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK9-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK9-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l84.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK9-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK9-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK9-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK9-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK9-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK9-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l84.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK9-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK9-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK9: omp_offload.failed3:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l84() #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK9: omp_offload.cont4:
-// CHECK9-NEXT: [[TMP22:%.*]] = load i32, ptr @Arg, align 4
-// CHECK9-NEXT: store i32 [[TMP22]], ptr [[ARG_CASTED]], align 4
-// CHECK9-NEXT: [[TMP23:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
-// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP23]], ptr [[TMP24]], align 8
-// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP23]], ptr [[TMP25]], align 8
-// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP26]], align 8
-// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP29:%.*]] = load i32, ptr @Arg, align 4
-// CHECK9-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK9-NEXT: [[TMP30:%.*]] = load i32, ptr @Arg, align 4
+// CHECK9-NEXT: store i32 [[TMP30]], ptr [[ARG_CASTED]], align 4
+// CHECK9-NEXT: [[TMP31:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP31]], ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP31]], ptr [[TMP33]], align 8
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP37:%.*]] = load i32, ptr @Arg, align 4
+// CHECK9-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP37]], 0
// CHECK9-NEXT: [[FROMBOOL:%.*]] = zext i1 [[TOBOOL]] to i8
// CHECK9-NEXT: store i8 [[FROMBOOL]], ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK9-NEXT: [[TMP30:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK9-NEXT: [[TOBOOL5:%.*]] = trunc i8 [[TMP30]] to i1
-// CHECK9-NEXT: [[TMP31:%.*]] = select i1 [[TOBOOL5]], i32 0, i32 1
+// CHECK9-NEXT: [[TMP38:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK9-NEXT: [[TOBOOL5:%.*]] = trunc i8 [[TMP38]] to i1
+// CHECK9-NEXT: [[TMP39:%.*]] = select i1 [[TOBOOL5]], i32 0, i32 1
+// CHECK9-NEXT: [[TMP40:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP39]], 0
// CHECK9-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP32]], align 4
-// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK9-NEXT: store i32 1, ptr [[TMP33]], align 4
-// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP27]], ptr [[TMP34]], align 8
-// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP28]], ptr [[TMP35]], align 8
-// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes, ptr [[TMP36]], align 8
-// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes, ptr [[TMP37]], align 8
-// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP38]], align 8
-// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK9-NEXT: store i64 100, ptr [[TMP40]], align 8
-// CHECK9-NEXT: [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP31]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK9-NEXT: [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
-// CHECK9-NEXT: br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP41]], align 4
+// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK9-NEXT: store i32 1, ptr [[TMP42]], align 4
+// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP35]], ptr [[TMP43]], align 8
+// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP36]], ptr [[TMP44]], align 8
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes, ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes, ptr [[TMP46]], align 8
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP47]], align 8
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP48]], align 8
+// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK9-NEXT: store i64 100, ptr [[TMP49]], align 8
+// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP50]], align 8
+// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP51]], align 4
+// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [[TMP40]], ptr [[TMP52]], align 4
+// CHECK9-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP53]], align 4
+// CHECK9-NEXT: [[TMP54:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP39]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK9-NEXT: [[TMP55:%.*]] = icmp ne i32 [[TMP54]], 0
+// CHECK9-NEXT: br i1 [[TMP55]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK9: omp_offload.failed8:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92(i64 [[TMP23]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92(i64 [[TMP31]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK9: omp_offload.cont9:
-// CHECK9-NEXT: [[TMP43:%.*]] = load i32, ptr @Arg, align 4
-// CHECK9-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP43]])
+// CHECK9-NEXT: [[TMP56:%.*]] = load i32, ptr @Arg, align 4
+// CHECK9-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP56]])
// CHECK9-NEXT: ret i32 [[CALL]]
//
//
@@ -5080,7 +5253,7 @@ int main() {
// CHECK9-NEXT: store i32 [[ARG]], ptr [[ARG_ADDR]], align 4
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK9-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK9-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -5097,81 +5270,106 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK9-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK9-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l57.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK9-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK9-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK9-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l57.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK9-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l57() #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK9-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK9-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK9-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK9-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK9-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK9-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK9-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK9-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l62.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK9-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK9-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK9-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK9-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK9-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK9-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l62.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK9-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK9-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK9: omp_offload.failed3:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l62() #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK9: omp_offload.cont4:
-// CHECK9-NEXT: [[TMP22:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
-// CHECK9-NEXT: store i32 [[TMP22]], ptr [[ARG_CASTED]], align 4
-// CHECK9-NEXT: [[TMP23:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
-// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP23]], ptr [[TMP24]], align 8
-// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP23]], ptr [[TMP25]], align 8
-// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP26]], align 8
-// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP29:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
-// CHECK9-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK9-NEXT: [[TMP30:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
+// CHECK9-NEXT: store i32 [[TMP30]], ptr [[ARG_CASTED]], align 4
+// CHECK9-NEXT: [[TMP31:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP31]], ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP31]], ptr [[TMP33]], align 8
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP37:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
+// CHECK9-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP37]], 0
// CHECK9-NEXT: [[FROMBOOL:%.*]] = zext i1 [[TOBOOL]] to i8
// CHECK9-NEXT: store i8 [[FROMBOOL]], ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK9-NEXT: [[TMP30:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK9-NEXT: [[TOBOOL5:%.*]] = trunc i8 [[TMP30]] to i1
-// CHECK9-NEXT: [[TMP31:%.*]] = select i1 [[TOBOOL5]], i32 0, i32 1
+// CHECK9-NEXT: [[TMP38:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK9-NEXT: [[TOBOOL5:%.*]] = trunc i8 [[TMP38]] to i1
+// CHECK9-NEXT: [[TMP39:%.*]] = select i1 [[TOBOOL5]], i32 0, i32 1
+// CHECK9-NEXT: [[TMP40:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP39]], 0
// CHECK9-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP32]], align 4
-// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK9-NEXT: store i32 1, ptr [[TMP33]], align 4
-// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP27]], ptr [[TMP34]], align 8
-// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP28]], ptr [[TMP35]], align 8
-// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.16, ptr [[TMP36]], align 8
-// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP37]], align 8
-// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP38]], align 8
-// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK9-NEXT: store i64 100, ptr [[TMP40]], align 8
-// CHECK9-NEXT: [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP31]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l67.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK9-NEXT: [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
-// CHECK9-NEXT: br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP41]], align 4
+// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK9-NEXT: store i32 1, ptr [[TMP42]], align 4
+// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP35]], ptr [[TMP43]], align 8
+// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP36]], ptr [[TMP44]], align 8
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.16, ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP46]], align 8
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP47]], align 8
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP48]], align 8
+// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK9-NEXT: store i64 100, ptr [[TMP49]], align 8
+// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP50]], align 8
+// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP51]], align 4
+// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [[TMP40]], ptr [[TMP52]], align 4
+// CHECK9-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP53]], align 4
+// CHECK9-NEXT: [[TMP54:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP39]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l67.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK9-NEXT: [[TMP55:%.*]] = icmp ne i32 [[TMP54]], 0
+// CHECK9-NEXT: br i1 [[TMP55]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK9: omp_offload.failed8:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l67(i64 [[TMP23]]) #[[ATTR2]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l67(i64 [[TMP31]]) #[[ATTR2]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK9: omp_offload.cont9:
// CHECK9-NEXT: ret i32 0
@@ -5681,7 +5879,7 @@ int main() {
// CHECK11-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK11-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK11-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -5698,35 +5896,51 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK11-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK11-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l43.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK11-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK11-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK11-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK11-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l43.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK11-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l43() #[[ATTR2:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK11-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK11-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK11-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK11-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK11-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK11-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK11-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK11-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK11-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK11-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK11-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l47.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK11-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK11-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK11-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK11-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK11-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK11-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l47.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK11-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK11-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK11: omp_offload.failed3:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9gtid_testv_l47() #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -6057,7 +6271,7 @@ int main() {
// CHECK11-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK11-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK11-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -6074,85 +6288,110 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK11-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK11-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l76.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK11-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK11-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK11-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK11-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l76.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK11-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l76() #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK11-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK11-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK11-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK11-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK11-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK11-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK11-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK11-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK11-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK11-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK11-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l84.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK11-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK11-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK11-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK11-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK11-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK11-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l84.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK11-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK11-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK11: omp_offload.failed3:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l84() #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK11: omp_offload.cont4:
-// CHECK11-NEXT: [[TMP22:%.*]] = load i32, ptr @Arg, align 4
-// CHECK11-NEXT: store i32 [[TMP22]], ptr [[ARG_CASTED]], align 4
-// CHECK11-NEXT: [[TMP23:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
-// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK11-NEXT: store i64 [[TMP23]], ptr [[TMP24]], align 8
-// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK11-NEXT: store i64 [[TMP23]], ptr [[TMP25]], align 8
-// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP26]], align 8
-// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP29:%.*]] = load i32, ptr @Arg, align 4
-// CHECK11-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK11-NEXT: [[TMP30:%.*]] = load i32, ptr @Arg, align 4
+// CHECK11-NEXT: store i32 [[TMP30]], ptr [[ARG_CASTED]], align 4
+// CHECK11-NEXT: [[TMP31:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK11-NEXT: store i64 [[TMP31]], ptr [[TMP32]], align 8
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK11-NEXT: store i64 [[TMP31]], ptr [[TMP33]], align 8
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP37:%.*]] = load i32, ptr @Arg, align 4
+// CHECK11-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP37]], 0
// CHECK11-NEXT: [[FROMBOOL:%.*]] = zext i1 [[TOBOOL]] to i8
// CHECK11-NEXT: store i8 [[FROMBOOL]], ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK11-NEXT: [[TMP30:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK11-NEXT: [[TOBOOL5:%.*]] = trunc i8 [[TMP30]] to i1
-// CHECK11-NEXT: [[TMP31:%.*]] = select i1 [[TOBOOL5]], i32 0, i32 1
+// CHECK11-NEXT: [[TMP38:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK11-NEXT: [[TOBOOL5:%.*]] = trunc i8 [[TMP38]] to i1
+// CHECK11-NEXT: [[TMP39:%.*]] = select i1 [[TOBOOL5]], i32 0, i32 1
+// CHECK11-NEXT: [[TMP40:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP39]], 0
// CHECK11-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP32]], align 4
-// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK11-NEXT: store i32 1, ptr [[TMP33]], align 4
-// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP27]], ptr [[TMP34]], align 8
-// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP28]], ptr [[TMP35]], align 8
-// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes, ptr [[TMP36]], align 8
-// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes, ptr [[TMP37]], align 8
-// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP38]], align 8
-// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK11-NEXT: store i64 100, ptr [[TMP40]], align 8
-// CHECK11-NEXT: [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP31]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK11-NEXT: [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
-// CHECK11-NEXT: br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP41]], align 4
+// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK11-NEXT: store i32 1, ptr [[TMP42]], align 4
+// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP35]], ptr [[TMP43]], align 8
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP36]], ptr [[TMP44]], align 8
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes, ptr [[TMP45]], align 8
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes, ptr [[TMP46]], align 8
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP47]], align 8
+// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP48]], align 8
+// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK11-NEXT: store i64 100, ptr [[TMP49]], align 8
+// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP50]], align 8
+// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP51]], align 4
+// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [[TMP40]], ptr [[TMP52]], align 4
+// CHECK11-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP53]], align 4
+// CHECK11-NEXT: [[TMP54:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP39]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK11-NEXT: [[TMP55:%.*]] = icmp ne i32 [[TMP54]], 0
+// CHECK11-NEXT: br i1 [[TMP55]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK11: omp_offload.failed8:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92(i64 [[TMP23]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l92(i64 [[TMP31]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK11: omp_offload.cont9:
-// CHECK11-NEXT: [[TMP43:%.*]] = load i32, ptr @Arg, align 4
-// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP43]])
+// CHECK11-NEXT: [[TMP56:%.*]] = load i32, ptr @Arg, align 4
+// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP56]])
// CHECK11-NEXT: ret i32 [[CALL]]
//
//
@@ -6891,7 +7130,7 @@ int main() {
// CHECK11-NEXT: store i32 [[ARG]], ptr [[ARG_ADDR]], align 4
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK11-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK11-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -6908,81 +7147,106 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK11-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK11-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l57.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK11-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK11-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK11-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK11-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l57.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK11-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l57() #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK11-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK11-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK11-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK11-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK11-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK11-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK11-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK11-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK11-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK11-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK11-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK11-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l62.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK11-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK11-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK11-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK11-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK11-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK11-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l62.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK11-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK11-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK11: omp_offload.failed3:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l62() #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK11: omp_offload.cont4:
-// CHECK11-NEXT: [[TMP22:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
-// CHECK11-NEXT: store i32 [[TMP22]], ptr [[ARG_CASTED]], align 4
-// CHECK11-NEXT: [[TMP23:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
-// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK11-NEXT: store i64 [[TMP23]], ptr [[TMP24]], align 8
-// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK11-NEXT: store i64 [[TMP23]], ptr [[TMP25]], align 8
-// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP26]], align 8
-// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP29:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
-// CHECK11-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP29]], 0
+// CHECK11-NEXT: [[TMP30:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
+// CHECK11-NEXT: store i32 [[TMP30]], ptr [[ARG_CASTED]], align 4
+// CHECK11-NEXT: [[TMP31:%.*]] = load i64, ptr [[ARG_CASTED]], align 8
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK11-NEXT: store i64 [[TMP31]], ptr [[TMP32]], align 8
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK11-NEXT: store i64 [[TMP31]], ptr [[TMP33]], align 8
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP34]], align 8
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP37:%.*]] = load i32, ptr [[ARG_ADDR]], align 4
+// CHECK11-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP37]], 0
// CHECK11-NEXT: [[FROMBOOL:%.*]] = zext i1 [[TOBOOL]] to i8
// CHECK11-NEXT: store i8 [[FROMBOOL]], ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK11-NEXT: [[TMP30:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK11-NEXT: [[TOBOOL5:%.*]] = trunc i8 [[TMP30]] to i1
-// CHECK11-NEXT: [[TMP31:%.*]] = select i1 [[TOBOOL5]], i32 0, i32 1
+// CHECK11-NEXT: [[TMP38:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK11-NEXT: [[TOBOOL5:%.*]] = trunc i8 [[TMP38]] to i1
+// CHECK11-NEXT: [[TMP39:%.*]] = select i1 [[TOBOOL5]], i32 0, i32 1
+// CHECK11-NEXT: [[TMP40:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP39]], 0
// CHECK11-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP32]], align 4
-// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK11-NEXT: store i32 1, ptr [[TMP33]], align 4
-// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP27]], ptr [[TMP34]], align 8
-// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP28]], ptr [[TMP35]], align 8
-// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.17, ptr [[TMP36]], align 8
-// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.18, ptr [[TMP37]], align 8
-// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP38]], align 8
-// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK11-NEXT: store i64 100, ptr [[TMP40]], align 8
-// CHECK11-NEXT: [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP31]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l67.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK11-NEXT: [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
-// CHECK11-NEXT: br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP41]], align 4
+// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK11-NEXT: store i32 1, ptr [[TMP42]], align 4
+// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP35]], ptr [[TMP43]], align 8
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP36]], ptr [[TMP44]], align 8
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.17, ptr [[TMP45]], align 8
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.18, ptr [[TMP46]], align 8
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP47]], align 8
+// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP48]], align 8
+// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK11-NEXT: store i64 100, ptr [[TMP49]], align 8
+// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP50]], align 8
+// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP51]], align 4
+// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [[TMP40]], ptr [[TMP52]], align 4
+// CHECK11-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP53]], align 4
+// CHECK11-NEXT: [[TMP54:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP39]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l67.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK11-NEXT: [[TMP55:%.*]] = icmp ne i32 [[TMP54]], 0
+// CHECK11-NEXT: br i1 [[TMP55]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK11: omp_offload.failed8:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l67(i64 [[TMP23]]) #[[ATTR2]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l67(i64 [[TMP31]]) #[[ATTR2]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK11: omp_offload.cont9:
// CHECK11-NEXT: ret i32 0
diff --git a/clang/test/OpenMP/teams_distribute_parallel_for_simd_lastprivate_codegen.cpp b/clang/test/OpenMP/teams_distribute_parallel_for_simd_lastprivate_codegen.cpp
index c3743d5ec4fd9..2b5a3d788f2c0 100644
--- a/clang/test/OpenMP/teams_distribute_parallel_for_simd_lastprivate_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_parallel_for_simd_lastprivate_codegen.cpp
@@ -853,7 +853,7 @@ int main() {
// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP25]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP25]], align 4
// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 5, ptr [[TMP26]], align 4
// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -870,9 +870,17 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP32]], align 8
// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP33]], align 8
-// CHECK9-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l110.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
-// CHECK9-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK9-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l110.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK9-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l110(i64 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]], i64 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -880,18 +888,18 @@ int main() {
// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiET_v()
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP36]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP40]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP37:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP37]]
+// CHECK9-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP41]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -1318,7 +1326,7 @@ int main() {
// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1335,27 +1343,35 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP27]], align 8
// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK9-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK9-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK9-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK9-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49(i64 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP32]]
+// CHECK9-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP36]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
@@ -1858,7 +1874,7 @@ int main() {
// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP25]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP25]], align 4
// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 5, ptr [[TMP26]], align 4
// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1875,9 +1891,17 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP32]], align 4
// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP33]], align 8
-// CHECK11-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l110.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
-// CHECK11-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK11-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l110.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK11-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l110(i32 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]], i32 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1885,18 +1909,18 @@ int main() {
// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiET_v()
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP36]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP40]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP37:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP37]]
+// CHECK11-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP41]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -2317,7 +2341,7 @@ int main() {
// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2334,27 +2358,35 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP27]], align 4
// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK11-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK11-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK11-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK11-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49(i32 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP32]]
+// CHECK11-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP36]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
diff --git a/clang/test/OpenMP/teams_distribute_parallel_for_simd_num_threads_codegen.cpp b/clang/test/OpenMP/teams_distribute_parallel_for_simd_num_threads_codegen.cpp
index 189fe7049da80..4792f1279200f 100644
--- a/clang/test/OpenMP/teams_distribute_parallel_for_simd_num_threads_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_parallel_for_simd_num_threads_codegen.cpp
@@ -98,7 +98,7 @@ int main() {
// CHECK1-NEXT: store i8 [[CALL]], ptr [[A]], align 1
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -115,65 +115,82 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 2, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l50.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 2, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 2, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l50.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l50() #[[ATTR6:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: lpad:
-// CHECK1-NEXT: [[TMP11:%.*]] = landingpad { ptr, i32 }
+// CHECK1-NEXT: [[TMP15:%.*]] = landingpad { ptr, i32 }
// CHECK1-NEXT: cleanup
-// CHECK1-NEXT: [[TMP12:%.*]] = extractvalue { ptr, i32 } [[TMP11]], 0
-// CHECK1-NEXT: store ptr [[TMP12]], ptr [[EXN_SLOT]], align 8
-// CHECK1-NEXT: [[TMP13:%.*]] = extractvalue { ptr, i32 } [[TMP11]], 1
-// CHECK1-NEXT: store i32 [[TMP13]], ptr [[EHSELECTOR_SLOT]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = extractvalue { ptr, i32 } [[TMP15]], 0
+// CHECK1-NEXT: store ptr [[TMP16]], ptr [[EXN_SLOT]], align 8
+// CHECK1-NEXT: [[TMP17:%.*]] = extractvalue { ptr, i32 } [[TMP15]], 1
+// CHECK1-NEXT: store i32 [[TMP17]], ptr [[EHSELECTOR_SLOT]], align 4
// CHECK1-NEXT: call void @_ZN1SD1Ev(ptr noundef nonnull align 8 dereferenceable(24) [[S]]) #[[ATTR6]]
// CHECK1-NEXT: br label [[EH_RESUME:%.*]]
// CHECK1: omp_offload.cont:
-// CHECK1-NEXT: [[TMP14:%.*]] = load i8, ptr [[A]], align 1
-// CHECK1-NEXT: store i8 [[TMP14]], ptr [[A_CASTED]], align 1
-// CHECK1-NEXT: [[TMP15:%.*]] = load i64, ptr [[A_CASTED]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP15]], ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP15]], ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP21:%.*]] = load i8, ptr [[A]], align 1
-// CHECK1-NEXT: store i8 [[TMP21]], ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TMP22:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TMP23:%.*]] = zext i8 [[TMP22]] to i32
+// CHECK1-NEXT: [[TMP18:%.*]] = load i8, ptr [[A]], align 1
+// CHECK1-NEXT: store i8 [[TMP18]], ptr [[A_CASTED]], align 1
+// CHECK1-NEXT: [[TMP19:%.*]] = load i64, ptr [[A_CASTED]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP19]], ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP19]], ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP25:%.*]] = load i8, ptr [[A]], align 1
+// CHECK1-NEXT: store i8 [[TMP25]], ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK1-NEXT: [[TMP26:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK1-NEXT: [[TMP27:%.*]] = zext i8 [[TMP26]] to i32
+// CHECK1-NEXT: [[TMP28:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP27]], 0
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP24]], align 4
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP25]], align 4
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP19]], ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP20]], ptr [[TMP27]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP30]], align 8
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP31]], align 8
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP32]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP23]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l55.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
-// CHECK1-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP29]], align 4
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP30]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP23]], ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP24]], ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP35]], align 8
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP36]], align 8
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP37]], align 8
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP38]], align 8
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP39]], align 4
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP28]], ptr [[TMP40]], align 4
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP41]], align 4
+// CHECK1-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP27]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l55.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
+// CHECK1-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l55(i64 [[TMP15]]) #[[ATTR6]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l55(i64 [[TMP19]]) #[[ATTR6]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK1: omp_offload.cont4:
-// CHECK1-NEXT: [[TMP35:%.*]] = load i8, ptr [[A]], align 1
-// CHECK1-NEXT: [[CONV:%.*]] = sext i8 [[TMP35]] to i32
+// CHECK1-NEXT: [[TMP44:%.*]] = load i8, ptr [[A]], align 1
+// CHECK1-NEXT: [[CONV:%.*]] = sext i8 [[TMP44]] to i32
// CHECK1-NEXT: [[CALL6:%.*]] = invoke noundef signext i32 @_Z5tmainIcLi5EEiv()
// CHECK1-NEXT: to label [[INVOKE_CONT5:%.*]] unwind label [[LPAD]]
// CHECK1: invoke.cont5:
@@ -184,8 +201,8 @@ int main() {
// CHECK1-NEXT: [[ADD9:%.*]] = add nsw i32 [[ADD]], [[CALL8]]
// CHECK1-NEXT: store i32 [[ADD9]], ptr [[RETVAL]], align 4
// CHECK1-NEXT: call void @_ZN1SD1Ev(ptr noundef nonnull align 8 dereferenceable(24) [[S]]) #[[ATTR6]]
-// CHECK1-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK1-NEXT: ret i32 [[TMP36]]
+// CHECK1-NEXT: [[TMP45:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK1-NEXT: ret i32 [[TMP45]]
// CHECK1: eh.resume:
// CHECK1-NEXT: [[EXN:%.*]] = load ptr, ptr [[EXN_SLOT]], align 8
// CHECK1-NEXT: [[SEL:%.*]] = load i32, ptr [[EHSELECTOR_SLOT]], align 4
@@ -568,7 +585,7 @@ int main() {
// CHECK1-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -585,35 +602,51 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 5, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l36.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 5, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 5, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l36.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l36() #[[ATTR6]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 23, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l40.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 23, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 23, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l40.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK1-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l40() #[[ATTR6]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -630,7 +663,7 @@ int main() {
// CHECK1-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -647,9 +680,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l36.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l36.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l36() #[[ATTR6]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -660,40 +701,49 @@ int main() {
// CHECK1-NEXT: [[CALL:%.*]] = call noundef signext i8 @_ZN1ScvcEv(ptr noundef nonnull align 8 dereferenceable(24) [[REF_TMP]])
// CHECK1-NEXT: call void @_ZN1SD1Ev(ptr noundef nonnull align 8 dereferenceable(24) [[REF_TMP]]) #[[ATTR6]]
// CHECK1-NEXT: store i8 [[CALL]], ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TMP11:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK1-NEXT: [[TMP12:%.*]] = zext i8 [[TMP11]] to i32
+// CHECK1-NEXT: [[TMP15:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK1-NEXT: [[TMP16:%.*]] = zext i8 [[TMP15]] to i32
+// CHECK1-NEXT: [[TMP17:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP16]], 0
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP13]], align 4
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP14]], align 4
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP18]], align 4
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 100, ptr [[TMP21]], align 8
-// CHECK1-NEXT: [[TMP22:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP12]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l40.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0
-// CHECK1-NEXT: br i1 [[TMP23]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP25]], align 8
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 100, ptr [[TMP26]], align 8
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [[TMP17]], ptr [[TMP29]], align 4
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP16]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l40.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK1-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l40() #[[ATTR6]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK1: omp_offload.cont4:
// CHECK1-NEXT: ret i32 0
// CHECK1: terminate.lpad:
-// CHECK1-NEXT: [[TMP24:%.*]] = landingpad { ptr, i32 }
+// CHECK1-NEXT: [[TMP33:%.*]] = landingpad { ptr, i32 }
// CHECK1-NEXT: catch ptr null
-// CHECK1-NEXT: [[TMP25:%.*]] = extractvalue { ptr, i32 } [[TMP24]], 0
-// CHECK1-NEXT: call void @__clang_call_terminate(ptr [[TMP25]]) #[[ATTR9]]
+// CHECK1-NEXT: [[TMP34:%.*]] = extractvalue { ptr, i32 } [[TMP33]], 0
+// CHECK1-NEXT: call void @__clang_call_terminate(ptr [[TMP34]]) #[[ATTR9]]
// CHECK1-NEXT: unreachable
//
//
@@ -1775,7 +1825,7 @@ int main() {
// CHECK5-NEXT: store i8 [[CALL]], ptr [[A]], align 1
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK5-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK5-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1792,65 +1842,82 @@ int main() {
// CHECK5-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK5-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 2, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l50.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK5-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK5-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [i32 2, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK5-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK5-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 2, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l50.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK5-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l50() #[[ATTR6:[0-9]+]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK5: lpad:
-// CHECK5-NEXT: [[TMP11:%.*]] = landingpad { ptr, i32 }
+// CHECK5-NEXT: [[TMP15:%.*]] = landingpad { ptr, i32 }
// CHECK5-NEXT: cleanup
-// CHECK5-NEXT: [[TMP12:%.*]] = extractvalue { ptr, i32 } [[TMP11]], 0
-// CHECK5-NEXT: store ptr [[TMP12]], ptr [[EXN_SLOT]], align 8
-// CHECK5-NEXT: [[TMP13:%.*]] = extractvalue { ptr, i32 } [[TMP11]], 1
-// CHECK5-NEXT: store i32 [[TMP13]], ptr [[EHSELECTOR_SLOT]], align 4
+// CHECK5-NEXT: [[TMP16:%.*]] = extractvalue { ptr, i32 } [[TMP15]], 0
+// CHECK5-NEXT: store ptr [[TMP16]], ptr [[EXN_SLOT]], align 8
+// CHECK5-NEXT: [[TMP17:%.*]] = extractvalue { ptr, i32 } [[TMP15]], 1
+// CHECK5-NEXT: store i32 [[TMP17]], ptr [[EHSELECTOR_SLOT]], align 4
// CHECK5-NEXT: call void @_ZN1SD1Ev(ptr noundef nonnull align 8 dereferenceable(24) [[S]]) #[[ATTR6]]
// CHECK5-NEXT: br label [[EH_RESUME:%.*]]
// CHECK5: omp_offload.cont:
-// CHECK5-NEXT: [[TMP14:%.*]] = load i8, ptr [[A]], align 1
-// CHECK5-NEXT: store i8 [[TMP14]], ptr [[A_CASTED]], align 1
-// CHECK5-NEXT: [[TMP15:%.*]] = load i64, ptr [[A_CASTED]], align 8
-// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK5-NEXT: store i64 [[TMP15]], ptr [[TMP16]], align 8
-// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK5-NEXT: store i64 [[TMP15]], ptr [[TMP17]], align 8
-// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK5-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP21:%.*]] = load i8, ptr [[A]], align 1
-// CHECK5-NEXT: store i8 [[TMP21]], ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK5-NEXT: [[TMP22:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK5-NEXT: [[TMP23:%.*]] = zext i8 [[TMP22]] to i32
+// CHECK5-NEXT: [[TMP18:%.*]] = load i8, ptr [[A]], align 1
+// CHECK5-NEXT: store i8 [[TMP18]], ptr [[A_CASTED]], align 1
+// CHECK5-NEXT: [[TMP19:%.*]] = load i64, ptr [[A_CASTED]], align 8
+// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK5-NEXT: store i64 [[TMP19]], ptr [[TMP20]], align 8
+// CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK5-NEXT: store i64 [[TMP19]], ptr [[TMP21]], align 8
+// CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK5-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP25:%.*]] = load i8, ptr [[A]], align 1
+// CHECK5-NEXT: store i8 [[TMP25]], ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK5-NEXT: [[TMP26:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK5-NEXT: [[TMP27:%.*]] = zext i8 [[TMP26]] to i32
+// CHECK5-NEXT: [[TMP28:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP27]], 0
// CHECK5-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP24]], align 4
-// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK5-NEXT: store i32 1, ptr [[TMP25]], align 4
-// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK5-NEXT: store ptr [[TMP19]], ptr [[TMP26]], align 8
-// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[TMP20]], ptr [[TMP27]], align 8
-// CHECK5-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK5-NEXT: store ptr @.offload_sizes, ptr [[TMP28]], align 8
-// CHECK5-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK5-NEXT: store ptr @.offload_maptypes, ptr [[TMP29]], align 8
-// CHECK5-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP30]], align 8
-// CHECK5-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
-// CHECK5-NEXT: store ptr null, ptr [[TMP31]], align 8
-// CHECK5-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK5-NEXT: store i64 100, ptr [[TMP32]], align 8
-// CHECK5-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP23]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l55.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK5-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
-// CHECK5-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK5-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP29]], align 4
+// CHECK5-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK5-NEXT: store i32 1, ptr [[TMP30]], align 4
+// CHECK5-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
+// CHECK5-NEXT: store ptr [[TMP23]], ptr [[TMP31]], align 8
+// CHECK5-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[TMP24]], ptr [[TMP32]], align 8
+// CHECK5-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK5-NEXT: store ptr @.offload_sizes, ptr [[TMP33]], align 8
+// CHECK5-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK5-NEXT: store ptr @.offload_maptypes, ptr [[TMP34]], align 8
+// CHECK5-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP35]], align 8
+// CHECK5-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP36]], align 8
+// CHECK5-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK5-NEXT: store i64 100, ptr [[TMP37]], align 8
+// CHECK5-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP38]], align 8
+// CHECK5-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP39]], align 4
+// CHECK5-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [[TMP28]], ptr [[TMP40]], align 4
+// CHECK5-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP41]], align 4
+// CHECK5-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP27]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l55.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK5-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
+// CHECK5-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK5: omp_offload.failed3:
-// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l55(i64 [[TMP15]]) #[[ATTR6]]
+// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l55(i64 [[TMP19]]) #[[ATTR6]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK5: omp_offload.cont4:
-// CHECK5-NEXT: [[TMP35:%.*]] = load i8, ptr [[A]], align 1
-// CHECK5-NEXT: [[CONV:%.*]] = sext i8 [[TMP35]] to i32
+// CHECK5-NEXT: [[TMP44:%.*]] = load i8, ptr [[A]], align 1
+// CHECK5-NEXT: [[CONV:%.*]] = sext i8 [[TMP44]] to i32
// CHECK5-NEXT: [[CALL6:%.*]] = invoke noundef signext i32 @_Z5tmainIcLi5EEiv()
// CHECK5-NEXT: to label [[INVOKE_CONT5:%.*]] unwind label [[LPAD]]
// CHECK5: invoke.cont5:
@@ -1861,8 +1928,8 @@ int main() {
// CHECK5-NEXT: [[ADD9:%.*]] = add nsw i32 [[ADD]], [[CALL8]]
// CHECK5-NEXT: store i32 [[ADD9]], ptr [[RETVAL]], align 4
// CHECK5-NEXT: call void @_ZN1SD1Ev(ptr noundef nonnull align 8 dereferenceable(24) [[S]]) #[[ATTR6]]
-// CHECK5-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK5-NEXT: ret i32 [[TMP36]]
+// CHECK5-NEXT: [[TMP45:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK5-NEXT: ret i32 [[TMP45]]
// CHECK5: eh.resume:
// CHECK5-NEXT: [[EXN:%.*]] = load ptr, ptr [[EXN_SLOT]], align 8
// CHECK5-NEXT: [[SEL:%.*]] = load i32, ptr [[EHSELECTOR_SLOT]], align 4
@@ -2245,7 +2312,7 @@ int main() {
// CHECK5-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK5-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK5-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2262,35 +2329,51 @@ int main() {
// CHECK5-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK5-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 5, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l36.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK5-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK5-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [i32 5, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK5-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK5-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 5, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l36.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK5-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l36() #[[ATTR6]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK5: omp_offload.cont:
// CHECK5-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK5-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK5-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK5-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK5-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK5-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK5-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK5-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK5-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK5-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK5-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK5-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK5-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK5-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK5-NEXT: store i64 100, ptr [[TMP19]], align 8
-// CHECK5-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 23, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l40.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK5-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK5-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK5-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK5-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK5-NEXT: store i64 100, ptr [[TMP23]], align 8
+// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [i32 23, i32 0, i32 0], ptr [[TMP26]], align 4
+// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK5-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 23, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l40.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK5-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK5-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK5: omp_offload.failed3:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIcLi5EEiv_l40() #[[ATTR6]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -2307,7 +2390,7 @@ int main() {
// CHECK5-NEXT: [[_TMP1:%.*]] = alloca i32, align 4
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK5-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK5-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2324,9 +2407,17 @@ int main() {
// CHECK5-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 100, ptr [[TMP8]], align 8
-// CHECK5-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l36.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK5-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK5-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK5-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK5-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l36.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK5-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l36() #[[ATTR6]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -2337,40 +2428,49 @@ int main() {
// CHECK5-NEXT: [[CALL:%.*]] = call noundef signext i8 @_ZN1ScvcEv(ptr noundef nonnull align 8 dereferenceable(24) [[REF_TMP]])
// CHECK5-NEXT: call void @_ZN1SD1Ev(ptr noundef nonnull align 8 dereferenceable(24) [[REF_TMP]]) #[[ATTR6]]
// CHECK5-NEXT: store i8 [[CALL]], ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK5-NEXT: [[TMP11:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
-// CHECK5-NEXT: [[TMP12:%.*]] = zext i8 [[TMP11]] to i32
+// CHECK5-NEXT: [[TMP15:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
+// CHECK5-NEXT: [[TMP16:%.*]] = zext i8 [[TMP15]] to i32
+// CHECK5-NEXT: [[TMP17:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP16]], 0
// CHECK5-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP13]], align 4
-// CHECK5-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK5-NEXT: store i32 0, ptr [[TMP14]], align 4
-// CHECK5-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK5-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK5-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK5-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK5-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP19]], align 8
-// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP18]], align 4
+// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK5-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK5-NEXT: store ptr null, ptr [[TMP20]], align 8
-// CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK5-NEXT: store i64 100, ptr [[TMP21]], align 8
-// CHECK5-NEXT: [[TMP22:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP12]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l40.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK5-NEXT: [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0
-// CHECK5-NEXT: br i1 [[TMP23]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
+// CHECK5-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK5-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK5-NEXT: store ptr null, ptr [[TMP23]], align 8
+// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP24]], align 8
+// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP25]], align 8
+// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK5-NEXT: store i64 100, ptr [[TMP26]], align 8
+// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK5-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK5-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] [[TMP17]], ptr [[TMP29]], align 4
+// CHECK5-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK5-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP16]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l40.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK5-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK5-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK5: omp_offload.failed3:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainI1SLi1EEiv_l40() #[[ATTR6]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT4]]
// CHECK5: omp_offload.cont4:
// CHECK5-NEXT: ret i32 0
// CHECK5: terminate.lpad:
-// CHECK5-NEXT: [[TMP24:%.*]] = landingpad { ptr, i32 }
+// CHECK5-NEXT: [[TMP33:%.*]] = landingpad { ptr, i32 }
// CHECK5-NEXT: catch ptr null
-// CHECK5-NEXT: [[TMP25:%.*]] = extractvalue { ptr, i32 } [[TMP24]], 0
-// CHECK5-NEXT: call void @__clang_call_terminate(ptr [[TMP25]]) #[[ATTR9]]
+// CHECK5-NEXT: [[TMP34:%.*]] = extractvalue { ptr, i32 } [[TMP33]], 0
+// CHECK5-NEXT: call void @__clang_call_terminate(ptr [[TMP34]]) #[[ATTR9]]
// CHECK5-NEXT: unreachable
//
//
diff --git a/clang/test/OpenMP/teams_distribute_parallel_for_simd_private_codegen.cpp b/clang/test/OpenMP/teams_distribute_parallel_for_simd_private_codegen.cpp
index fa5849c680f26..35698f812ed88 100644
--- a/clang/test/OpenMP/teams_distribute_parallel_for_simd_private_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_parallel_for_simd_private_codegen.cpp
@@ -265,7 +265,7 @@ int main() {
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -282,9 +282,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l98.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l98.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l98() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -541,7 +549,7 @@ int main() {
// CHECK1-NEXT: store ptr undef, ptr [[_TMP1]], align 8
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -558,27 +566,35 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK1-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK1: arraydestroy.body:
-// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK1-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK1-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK1: arraydestroy.done2:
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK1-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK1-NEXT: ret i32 [[TMP12]]
+// CHECK1-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK1-NEXT: ret i32 [[TMP16]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN1SIiEC1Ev
@@ -1017,7 +1033,7 @@ int main() {
// CHECK3-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1034,9 +1050,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l98.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK3-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l98.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK3-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l98() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1287,7 +1311,7 @@ int main() {
// CHECK3-NEXT: store ptr undef, ptr [[_TMP1]], align 4
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1304,27 +1328,35 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK3-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK3-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK3-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK3-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK3: arraydestroy.body:
-// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK3-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK3-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK3: arraydestroy.done2:
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK3-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK3-NEXT: ret i32 [[TMP12]]
+// CHECK3-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK3-NEXT: ret i32 [[TMP16]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN1SIiEC1Ev
diff --git a/clang/test/OpenMP/teams_distribute_parallel_for_simd_proc_bind_codegen.cpp b/clang/test/OpenMP/teams_distribute_parallel_for_simd_proc_bind_codegen.cpp
index a34ddd6914e3f..590e65c5df49e 100644
--- a/clang/test/OpenMP/teams_distribute_parallel_for_simd_proc_bind_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_parallel_for_simd_proc_bind_codegen.cpp
@@ -61,7 +61,7 @@ int main() {
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -78,35 +78,51 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 1000, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l36.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l36.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l36() #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[KERNEL_ARGS2:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP11]], align 4
-// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
-// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
-// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
-// CHECK1-NEXT: store ptr null, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
-// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
-// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
-// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 1
+// CHECK1-NEXT: store i32 0, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 2
// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 3
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
-// CHECK1-NEXT: store i64 1000, ptr [[TMP19]], align 8
-// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l39.region_id, ptr [[KERNEL_ARGS2]])
-// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
-// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 4
+// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 5
+// CHECK1-NEXT: store ptr null, ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 8
+// CHECK1-NEXT: store i64 1000, ptr [[TMP23]], align 8
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS2]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP27]], align 4
+// CHECK1-NEXT: [[TMP28:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l39.region_id, ptr [[KERNEL_ARGS2]])
+// CHECK1-NEXT: [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
+// CHECK1-NEXT: br i1 [[TMP29]], label [[OMP_OFFLOAD_FAILED3:%.*]], label [[OMP_OFFLOAD_CONT4:%.*]]
// CHECK1: omp_offload.failed3:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l39() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT4]]
@@ -425,7 +441,7 @@ int main() {
// CHECK1-NEXT: [[TMP:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -442,9 +458,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 1000, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l29.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l29.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l29() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/teams_distribute_parallel_for_simd_reduction_codegen.cpp b/clang/test/OpenMP/teams_distribute_parallel_for_simd_reduction_codegen.cpp
index 8880ef3037a32..1df8bce9953a5 100644
--- a/clang/test/OpenMP/teams_distribute_parallel_for_simd_reduction_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_parallel_for_simd_reduction_codegen.cpp
@@ -120,7 +120,7 @@ int main() {
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -137,9 +137,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l70.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK1-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l70.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l70(i64 [[TMP1]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -420,7 +428,7 @@ int main() {
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -437,9 +445,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK1-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32(i64 [[TMP1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -724,7 +740,7 @@ int main() {
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -741,9 +757,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP14]], align 4
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP15]], align 8
-// CHECK3-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l70.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK3-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK3-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l70.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK3-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l70(i32 [[TMP1]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1020,7 +1044,7 @@ int main() {
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1037,9 +1061,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP14]], align 4
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP15]], align 8
-// CHECK3-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK3-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK3-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK3-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32(i32 [[TMP1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/teams_distribute_parallel_for_simd_schedule_codegen.cpp b/clang/test/OpenMP/teams_distribute_parallel_for_simd_schedule_codegen.cpp
index 390e6bc552f86..a1254b10fa10c 100644
--- a/clang/test/OpenMP/teams_distribute_parallel_for_simd_schedule_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_parallel_for_simd_schedule_codegen.cpp
@@ -263,7 +263,7 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -280,157 +280,197 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.4, ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l41.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.4, ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l41.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK1-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK1-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK1: omp_offload.failed8:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l41(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK1: omp_offload.cont9:
// CHECK1-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A10]], ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A10]], ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.8, ptr [[TMP41]], align 8
-// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP42]], align 8
-// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP43]], align 8
-// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP44]], align 8
-// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK1-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l46.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK1-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK1-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 8
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 8
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.8, ptr [[TMP49]], align 8
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP50]], align 8
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP51]], align 8
+// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP52]], align 8
+// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK1-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l46.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK1-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK1-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK1: omp_offload.failed16:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l46(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK1: omp_offload.cont17:
// CHECK1-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP48]], align 8
-// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A18]], ptr [[TMP49]], align 8
-// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP50]], align 8
-// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP60:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP60]], align 8
+// CHECK1-NEXT: [[TMP61:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A18]], ptr [[TMP61]], align 8
+// CHECK1-NEXT: [[TMP62:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP62]], align 8
+// CHECK1-NEXT: [[TMP63:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS23:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP53]], align 4
-// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP51]], ptr [[TMP55]], align 8
-// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP52]], ptr [[TMP56]], align 8
-// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.12, ptr [[TMP57]], align 8
-// CHECK1-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP58]], align 8
-// CHECK1-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP59]], align 8
-// CHECK1-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP60]], align 8
-// CHECK1-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP61]], align 8
-// CHECK1-NEXT: [[TMP62:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l52.region_id, ptr [[KERNEL_ARGS23]])
-// CHECK1-NEXT: [[TMP63:%.*]] = icmp ne i32 [[TMP62]], 0
-// CHECK1-NEXT: br i1 [[TMP63]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
+// CHECK1-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP65]], align 4
+// CHECK1-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP66]], align 4
+// CHECK1-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP63]], ptr [[TMP67]], align 8
+// CHECK1-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP64]], ptr [[TMP68]], align 8
+// CHECK1-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.12, ptr [[TMP69]], align 8
+// CHECK1-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP70]], align 8
+// CHECK1-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP71]], align 8
+// CHECK1-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP72]], align 8
+// CHECK1-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP73]], align 8
+// CHECK1-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP74]], align 8
+// CHECK1-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP75]], align 4
+// CHECK1-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP76]], align 4
+// CHECK1-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP77]], align 4
+// CHECK1-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l52.region_id, ptr [[KERNEL_ARGS23]])
+// CHECK1-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
+// CHECK1-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
// CHECK1: omp_offload.failed24:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l52(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT25]]
// CHECK1: omp_offload.cont25:
// CHECK1-NEXT: [[A26:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP64]], align 8
-// CHECK1-NEXT: [[TMP65:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A26]], ptr [[TMP65]], align 8
-// CHECK1-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP66]], align 8
-// CHECK1-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP80:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP80]], align 8
+// CHECK1-NEXT: [[TMP81:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A26]], ptr [[TMP81]], align 8
+// CHECK1-NEXT: [[TMP82:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP82]], align 8
+// CHECK1-NEXT: [[TMP83:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP84:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP69]], align 4
-// CHECK1-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP70]], align 4
-// CHECK1-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP67]], ptr [[TMP71]], align 8
-// CHECK1-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 8
-// CHECK1-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.16, ptr [[TMP73]], align 8
-// CHECK1-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP74]], align 8
-// CHECK1-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP75]], align 8
-// CHECK1-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP76]], align 8
-// CHECK1-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP77]], align 8
-// CHECK1-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l58.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK1-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
-// CHECK1-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK1-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP85]], align 4
+// CHECK1-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP86]], align 4
+// CHECK1-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP83]], ptr [[TMP87]], align 8
+// CHECK1-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP84]], ptr [[TMP88]], align 8
+// CHECK1-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.16, ptr [[TMP89]], align 8
+// CHECK1-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP90]], align 8
+// CHECK1-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP91]], align 8
+// CHECK1-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP92]], align 8
+// CHECK1-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP93]], align 8
+// CHECK1-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP94]], align 8
+// CHECK1-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP95]], align 4
+// CHECK1-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP96]], align 4
+// CHECK1-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP97]], align 4
+// CHECK1-NEXT: [[TMP98:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l58.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK1-NEXT: [[TMP99:%.*]] = icmp ne i32 [[TMP98]], 0
+// CHECK1-NEXT: br i1 [[TMP99]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK1: omp_offload.failed32:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l58(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK1: omp_offload.cont33:
// CHECK1-NEXT: [[A34:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A34]], i64 0, i64 0
-// CHECK1-NEXT: [[TMP80:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK1-NEXT: ret i32 [[TMP80]]
+// CHECK1-NEXT: [[TMP100:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK1-NEXT: ret i32 [[TMP100]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36
@@ -1327,7 +1367,7 @@ int main (int argc, char **argv) {
// CHECK2-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK2-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK2-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK2-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK2-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK2-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK2-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK2-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1344,157 +1384,197 @@ int main (int argc, char **argv) {
// CHECK2-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK2-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK2-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK2-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS]])
-// CHECK2-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK2-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK2-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK2-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK2-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK2-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK2-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK2-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK2-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK2-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK2-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS]])
+// CHECK2-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK2-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK2: omp_offload.failed:
// CHECK2-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK2-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK2: omp_offload.cont:
// CHECK2-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK2-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK2-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 8
-// CHECK2-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK2-NEXT: store ptr [[A2]], ptr [[TMP17]], align 8
-// CHECK2-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
-// CHECK2-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK2-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK2-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK2-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK2-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 8
+// CHECK2-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK2-NEXT: store ptr [[A2]], ptr [[TMP21]], align 8
+// CHECK2-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
+// CHECK2-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK2-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK2-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK2-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK2-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK2-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK2-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK2-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK2-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK2-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK2-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK2-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK2-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK2-NEXT: store ptr @.offload_sizes.4, ptr [[TMP25]], align 8
-// CHECK2-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK2-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP26]], align 8
-// CHECK2-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK2-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK2-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK2-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK2-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK2-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK2-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l41.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK2-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK2-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK2-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK2-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK2-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK2-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK2-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK2-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK2-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK2-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK2-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK2-NEXT: store ptr @.offload_sizes.4, ptr [[TMP29]], align 8
+// CHECK2-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK2-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP30]], align 8
+// CHECK2-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK2-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK2-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK2-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK2-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK2-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK2-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK2-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK2-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK2-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK2-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK2-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK2-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK2-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK2-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l41.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK2-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK2-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK2: omp_offload.failed8:
// CHECK2-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l41(ptr [[THIS1]]) #[[ATTR2]]
// CHECK2-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK2: omp_offload.cont9:
// CHECK2-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK2-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK2-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 8
-// CHECK2-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK2-NEXT: store ptr [[A10]], ptr [[TMP33]], align 8
-// CHECK2-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
-// CHECK2-NEXT: store ptr null, ptr [[TMP34]], align 8
-// CHECK2-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK2-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK2-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK2-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 8
+// CHECK2-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK2-NEXT: store ptr [[A10]], ptr [[TMP41]], align 8
+// CHECK2-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
+// CHECK2-NEXT: store ptr null, ptr [[TMP42]], align 8
+// CHECK2-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK2-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK2-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK2-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK2-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK2-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK2-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK2-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK2-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
-// CHECK2-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK2-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
-// CHECK2-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK2-NEXT: store ptr @.offload_sizes.8, ptr [[TMP41]], align 8
-// CHECK2-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK2-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP42]], align 8
-// CHECK2-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK2-NEXT: store ptr null, ptr [[TMP43]], align 8
-// CHECK2-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK2-NEXT: store ptr null, ptr [[TMP44]], align 8
-// CHECK2-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK2-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK2-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l46.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK2-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK2-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK2-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK2-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK2-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK2-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK2-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK2-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 8
+// CHECK2-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK2-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 8
+// CHECK2-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK2-NEXT: store ptr @.offload_sizes.8, ptr [[TMP49]], align 8
+// CHECK2-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK2-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP50]], align 8
+// CHECK2-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK2-NEXT: store ptr null, ptr [[TMP51]], align 8
+// CHECK2-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK2-NEXT: store ptr null, ptr [[TMP52]], align 8
+// CHECK2-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK2-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK2-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK2-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK2-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK2-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK2-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK2-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK2-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK2-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK2-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l46.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK2-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK2-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK2: omp_offload.failed16:
// CHECK2-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l46(ptr [[THIS1]]) #[[ATTR2]]
// CHECK2-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK2: omp_offload.cont17:
// CHECK2-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK2-NEXT: [[TMP48:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK2-NEXT: store ptr [[THIS1]], ptr [[TMP48]], align 8
-// CHECK2-NEXT: [[TMP49:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK2-NEXT: store ptr [[A18]], ptr [[TMP49]], align 8
-// CHECK2-NEXT: [[TMP50:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
-// CHECK2-NEXT: store ptr null, ptr [[TMP50]], align 8
-// CHECK2-NEXT: [[TMP51:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK2-NEXT: [[TMP52:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK2-NEXT: [[TMP60:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK2-NEXT: store ptr [[THIS1]], ptr [[TMP60]], align 8
+// CHECK2-NEXT: [[TMP61:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK2-NEXT: store ptr [[A18]], ptr [[TMP61]], align 8
+// CHECK2-NEXT: [[TMP62:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
+// CHECK2-NEXT: store ptr null, ptr [[TMP62]], align 8
+// CHECK2-NEXT: [[TMP63:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK2-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
// CHECK2-NEXT: [[KERNEL_ARGS23:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK2-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
-// CHECK2-NEXT: store i32 1, ptr [[TMP53]], align 4
-// CHECK2-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
-// CHECK2-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK2-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
-// CHECK2-NEXT: store ptr [[TMP51]], ptr [[TMP55]], align 8
-// CHECK2-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
-// CHECK2-NEXT: store ptr [[TMP52]], ptr [[TMP56]], align 8
-// CHECK2-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
-// CHECK2-NEXT: store ptr @.offload_sizes.12, ptr [[TMP57]], align 8
-// CHECK2-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
-// CHECK2-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP58]], align 8
-// CHECK2-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
-// CHECK2-NEXT: store ptr null, ptr [[TMP59]], align 8
-// CHECK2-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
-// CHECK2-NEXT: store ptr null, ptr [[TMP60]], align 8
-// CHECK2-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
-// CHECK2-NEXT: store i64 123, ptr [[TMP61]], align 8
-// CHECK2-NEXT: [[TMP62:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l52.region_id, ptr [[KERNEL_ARGS23]])
-// CHECK2-NEXT: [[TMP63:%.*]] = icmp ne i32 [[TMP62]], 0
-// CHECK2-NEXT: br i1 [[TMP63]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
+// CHECK2-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
+// CHECK2-NEXT: store i32 2, ptr [[TMP65]], align 4
+// CHECK2-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
+// CHECK2-NEXT: store i32 1, ptr [[TMP66]], align 4
+// CHECK2-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
+// CHECK2-NEXT: store ptr [[TMP63]], ptr [[TMP67]], align 8
+// CHECK2-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
+// CHECK2-NEXT: store ptr [[TMP64]], ptr [[TMP68]], align 8
+// CHECK2-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
+// CHECK2-NEXT: store ptr @.offload_sizes.12, ptr [[TMP69]], align 8
+// CHECK2-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
+// CHECK2-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP70]], align 8
+// CHECK2-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
+// CHECK2-NEXT: store ptr null, ptr [[TMP71]], align 8
+// CHECK2-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
+// CHECK2-NEXT: store ptr null, ptr [[TMP72]], align 8
+// CHECK2-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
+// CHECK2-NEXT: store i64 123, ptr [[TMP73]], align 8
+// CHECK2-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 9
+// CHECK2-NEXT: store i64 0, ptr [[TMP74]], align 8
+// CHECK2-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 10
+// CHECK2-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP75]], align 4
+// CHECK2-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 11
+// CHECK2-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP76]], align 4
+// CHECK2-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 12
+// CHECK2-NEXT: store i32 0, ptr [[TMP77]], align 4
+// CHECK2-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l52.region_id, ptr [[KERNEL_ARGS23]])
+// CHECK2-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
+// CHECK2-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
// CHECK2: omp_offload.failed24:
// CHECK2-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l52(ptr [[THIS1]]) #[[ATTR2]]
// CHECK2-NEXT: br label [[OMP_OFFLOAD_CONT25]]
// CHECK2: omp_offload.cont25:
// CHECK2-NEXT: [[A26:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK2-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK2-NEXT: store ptr [[THIS1]], ptr [[TMP64]], align 8
-// CHECK2-NEXT: [[TMP65:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
-// CHECK2-NEXT: store ptr [[A26]], ptr [[TMP65]], align 8
-// CHECK2-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i64 0, i64 0
-// CHECK2-NEXT: store ptr null, ptr [[TMP66]], align 8
-// CHECK2-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK2-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK2-NEXT: [[TMP80:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK2-NEXT: store ptr [[THIS1]], ptr [[TMP80]], align 8
+// CHECK2-NEXT: [[TMP81:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK2-NEXT: store ptr [[A26]], ptr [[TMP81]], align 8
+// CHECK2-NEXT: [[TMP82:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i64 0, i64 0
+// CHECK2-NEXT: store ptr null, ptr [[TMP82]], align 8
+// CHECK2-NEXT: [[TMP83:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK2-NEXT: [[TMP84:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
// CHECK2-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK2-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK2-NEXT: store i32 1, ptr [[TMP69]], align 4
-// CHECK2-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK2-NEXT: store i32 1, ptr [[TMP70]], align 4
-// CHECK2-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK2-NEXT: store ptr [[TMP67]], ptr [[TMP71]], align 8
-// CHECK2-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK2-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 8
-// CHECK2-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK2-NEXT: store ptr @.offload_sizes.16, ptr [[TMP73]], align 8
-// CHECK2-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK2-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP74]], align 8
-// CHECK2-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK2-NEXT: store ptr null, ptr [[TMP75]], align 8
-// CHECK2-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK2-NEXT: store ptr null, ptr [[TMP76]], align 8
-// CHECK2-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK2-NEXT: store i64 123, ptr [[TMP77]], align 8
-// CHECK2-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l58.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK2-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
-// CHECK2-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK2-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK2-NEXT: store i32 2, ptr [[TMP85]], align 4
+// CHECK2-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK2-NEXT: store i32 1, ptr [[TMP86]], align 4
+// CHECK2-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK2-NEXT: store ptr [[TMP83]], ptr [[TMP87]], align 8
+// CHECK2-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK2-NEXT: store ptr [[TMP84]], ptr [[TMP88]], align 8
+// CHECK2-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK2-NEXT: store ptr @.offload_sizes.16, ptr [[TMP89]], align 8
+// CHECK2-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK2-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP90]], align 8
+// CHECK2-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK2-NEXT: store ptr null, ptr [[TMP91]], align 8
+// CHECK2-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK2-NEXT: store ptr null, ptr [[TMP92]], align 8
+// CHECK2-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK2-NEXT: store i64 123, ptr [[TMP93]], align 8
+// CHECK2-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK2-NEXT: store i64 0, ptr [[TMP94]], align 8
+// CHECK2-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK2-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP95]], align 4
+// CHECK2-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK2-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP96]], align 4
+// CHECK2-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK2-NEXT: store i32 0, ptr [[TMP97]], align 4
+// CHECK2-NEXT: [[TMP98:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l58.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK2-NEXT: [[TMP99:%.*]] = icmp ne i32 [[TMP98]], 0
+// CHECK2-NEXT: br i1 [[TMP99]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK2: omp_offload.failed32:
// CHECK2-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l58(ptr [[THIS1]]) #[[ATTR2]]
// CHECK2-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK2: omp_offload.cont33:
// CHECK2-NEXT: [[A34:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK2-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A34]], i64 0, i64 0
-// CHECK2-NEXT: [[TMP80:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK2-NEXT: ret i32 [[TMP80]]
+// CHECK2-NEXT: [[TMP100:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK2-NEXT: ret i32 [[TMP100]]
//
//
// CHECK2-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36
@@ -2391,7 +2471,7 @@ int main (int argc, char **argv) {
// CHECK5-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK5-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK5-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK5-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK5-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK5-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2408,157 +2488,197 @@ int main (int argc, char **argv) {
// CHECK5-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK5-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK5-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK5-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS]])
-// CHECK5-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK5-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK5-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK5-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK5-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS]])
+// CHECK5-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK5-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK5: omp_offload.failed:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK5: omp_offload.cont:
// CHECK5-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 4
-// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[A2]], ptr [[TMP17]], align 4
-// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
-// CHECK5-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 4
+// CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[A2]], ptr [[TMP21]], align 4
+// CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
+// CHECK5-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK5-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK5-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK5-NEXT: store ptr @.offload_sizes.4, ptr [[TMP25]], align 4
-// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK5-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP26]], align 4
-// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK5-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK5-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK5-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK5-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK5-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l41.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK5-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK5-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK5-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK5-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK5-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK5-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK5-NEXT: store ptr @.offload_sizes.4, ptr [[TMP29]], align 4
+// CHECK5-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK5-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP30]], align 4
+// CHECK5-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK5-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK5-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK5-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK5-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK5-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK5-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK5-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK5-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l41.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK5-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK5-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK5: omp_offload.failed8:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l41(ptr [[THIS1]]) #[[ATTR2]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK5: omp_offload.cont9:
// CHECK5-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 4
-// CHECK5-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[A10]], ptr [[TMP33]], align 4
-// CHECK5-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
-// CHECK5-NEXT: store ptr null, ptr [[TMP34]], align 4
-// CHECK5-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 4
+// CHECK5-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[A10]], ptr [[TMP41]], align 4
+// CHECK5-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
+// CHECK5-NEXT: store ptr null, ptr [[TMP42]], align 4
+// CHECK5-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK5-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK5-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK5-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK5-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 4
-// CHECK5-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 4
-// CHECK5-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK5-NEXT: store ptr @.offload_sizes.8, ptr [[TMP41]], align 4
-// CHECK5-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK5-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP42]], align 4
-// CHECK5-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK5-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK5-NEXT: store ptr null, ptr [[TMP44]], align 4
-// CHECK5-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK5-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK5-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l46.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK5-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK5-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK5-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK5-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK5-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK5-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK5-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 4
+// CHECK5-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 4
+// CHECK5-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK5-NEXT: store ptr @.offload_sizes.8, ptr [[TMP49]], align 4
+// CHECK5-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK5-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP50]], align 4
+// CHECK5-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK5-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP52]], align 4
+// CHECK5-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK5-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK5-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK5-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK5-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK5-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK5-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l46.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK5-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK5-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK5: omp_offload.failed16:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l46(ptr [[THIS1]]) #[[ATTR2]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK5: omp_offload.cont17:
// CHECK5-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP48:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP48]], align 4
-// CHECK5-NEXT: [[TMP49:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[A18]], ptr [[TMP49]], align 4
-// CHECK5-NEXT: [[TMP50:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
-// CHECK5-NEXT: store ptr null, ptr [[TMP50]], align 4
-// CHECK5-NEXT: [[TMP51:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP52:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP60:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP60]], align 4
+// CHECK5-NEXT: [[TMP61:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[A18]], ptr [[TMP61]], align 4
+// CHECK5-NEXT: [[TMP62:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
+// CHECK5-NEXT: store ptr null, ptr [[TMP62]], align 4
+// CHECK5-NEXT: [[TMP63:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS23:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP53]], align 4
-// CHECK5-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
-// CHECK5-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK5-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
-// CHECK5-NEXT: store ptr [[TMP51]], ptr [[TMP55]], align 4
-// CHECK5-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[TMP52]], ptr [[TMP56]], align 4
-// CHECK5-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
-// CHECK5-NEXT: store ptr @.offload_sizes.12, ptr [[TMP57]], align 4
-// CHECK5-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
-// CHECK5-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP58]], align 4
-// CHECK5-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP59]], align 4
-// CHECK5-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
-// CHECK5-NEXT: store ptr null, ptr [[TMP60]], align 4
-// CHECK5-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
-// CHECK5-NEXT: store i64 123, ptr [[TMP61]], align 8
-// CHECK5-NEXT: [[TMP62:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l52.region_id, ptr [[KERNEL_ARGS23]])
-// CHECK5-NEXT: [[TMP63:%.*]] = icmp ne i32 [[TMP62]], 0
-// CHECK5-NEXT: br i1 [[TMP63]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
+// CHECK5-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP65]], align 4
+// CHECK5-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
+// CHECK5-NEXT: store i32 1, ptr [[TMP66]], align 4
+// CHECK5-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
+// CHECK5-NEXT: store ptr [[TMP63]], ptr [[TMP67]], align 4
+// CHECK5-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[TMP64]], ptr [[TMP68]], align 4
+// CHECK5-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
+// CHECK5-NEXT: store ptr @.offload_sizes.12, ptr [[TMP69]], align 4
+// CHECK5-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
+// CHECK5-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP70]], align 4
+// CHECK5-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP71]], align 4
+// CHECK5-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP72]], align 4
+// CHECK5-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
+// CHECK5-NEXT: store i64 123, ptr [[TMP73]], align 8
+// CHECK5-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP74]], align 8
+// CHECK5-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP75]], align 4
+// CHECK5-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP76]], align 4
+// CHECK5-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP77]], align 4
+// CHECK5-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l52.region_id, ptr [[KERNEL_ARGS23]])
+// CHECK5-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
+// CHECK5-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
// CHECK5: omp_offload.failed24:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l52(ptr [[THIS1]]) #[[ATTR2]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT25]]
// CHECK5: omp_offload.cont25:
// CHECK5-NEXT: [[A26:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP64]], align 4
-// CHECK5-NEXT: [[TMP65:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
-// CHECK5-NEXT: store ptr [[A26]], ptr [[TMP65]], align 4
-// CHECK5-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i32 0, i32 0
-// CHECK5-NEXT: store ptr null, ptr [[TMP66]], align 4
-// CHECK5-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP80:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP80]], align 4
+// CHECK5-NEXT: [[TMP81:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK5-NEXT: store ptr [[A26]], ptr [[TMP81]], align 4
+// CHECK5-NEXT: [[TMP82:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i32 0, i32 0
+// CHECK5-NEXT: store ptr null, ptr [[TMP82]], align 4
+// CHECK5-NEXT: [[TMP83:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK5-NEXT: [[TMP84:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
// CHECK5-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK5-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK5-NEXT: store i32 1, ptr [[TMP69]], align 4
-// CHECK5-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK5-NEXT: store i32 1, ptr [[TMP70]], align 4
-// CHECK5-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK5-NEXT: store ptr [[TMP67]], ptr [[TMP71]], align 4
-// CHECK5-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK5-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 4
-// CHECK5-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK5-NEXT: store ptr @.offload_sizes.16, ptr [[TMP73]], align 4
-// CHECK5-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK5-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP74]], align 4
-// CHECK5-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK5-NEXT: store ptr null, ptr [[TMP75]], align 4
-// CHECK5-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK5-NEXT: store ptr null, ptr [[TMP76]], align 4
-// CHECK5-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK5-NEXT: store i64 123, ptr [[TMP77]], align 8
-// CHECK5-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l58.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK5-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
-// CHECK5-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK5-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK5-NEXT: store i32 2, ptr [[TMP85]], align 4
+// CHECK5-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK5-NEXT: store i32 1, ptr [[TMP86]], align 4
+// CHECK5-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK5-NEXT: store ptr [[TMP83]], ptr [[TMP87]], align 4
+// CHECK5-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK5-NEXT: store ptr [[TMP84]], ptr [[TMP88]], align 4
+// CHECK5-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK5-NEXT: store ptr @.offload_sizes.16, ptr [[TMP89]], align 4
+// CHECK5-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK5-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP90]], align 4
+// CHECK5-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK5-NEXT: store ptr null, ptr [[TMP91]], align 4
+// CHECK5-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK5-NEXT: store ptr null, ptr [[TMP92]], align 4
+// CHECK5-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK5-NEXT: store i64 123, ptr [[TMP93]], align 8
+// CHECK5-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK5-NEXT: store i64 0, ptr [[TMP94]], align 8
+// CHECK5-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP95]], align 4
+// CHECK5-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP96]], align 4
+// CHECK5-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK5-NEXT: store i32 0, ptr [[TMP97]], align 4
+// CHECK5-NEXT: [[TMP98:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l58.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK5-NEXT: [[TMP99:%.*]] = icmp ne i32 [[TMP98]], 0
+// CHECK5-NEXT: br i1 [[TMP99]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK5: omp_offload.failed32:
// CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l58(ptr [[THIS1]]) #[[ATTR2]]
// CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK5: omp_offload.cont33:
// CHECK5-NEXT: [[A34:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK5-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A34]], i32 0, i32 0
-// CHECK5-NEXT: [[TMP80:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK5-NEXT: ret i32 [[TMP80]]
+// CHECK5-NEXT: [[TMP100:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK5-NEXT: ret i32 [[TMP100]]
//
//
// CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36
@@ -3428,7 +3548,7 @@ int main (int argc, char **argv) {
// CHECK6-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK6-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK6-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK6-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK6-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK6-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK6-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK6-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3445,157 +3565,197 @@ int main (int argc, char **argv) {
// CHECK6-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK6-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK6-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK6-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS]])
-// CHECK6-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK6-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK6-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK6-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK6-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK6-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK6-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK6-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK6-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK6-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK6-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS]])
+// CHECK6-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK6-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK6: omp_offload.failed:
// CHECK6-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK6-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK6: omp_offload.cont:
// CHECK6-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK6-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK6-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 4
-// CHECK6-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK6-NEXT: store ptr [[A2]], ptr [[TMP17]], align 4
-// CHECK6-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
-// CHECK6-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK6-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK6-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK6-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK6-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 4
+// CHECK6-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK6-NEXT: store ptr [[A2]], ptr [[TMP21]], align 4
+// CHECK6-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
+// CHECK6-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK6-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK6-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK6-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK6-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK6-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK6-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK6-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK6-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK6-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK6-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK6-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK6-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK6-NEXT: store ptr @.offload_sizes.4, ptr [[TMP25]], align 4
-// CHECK6-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK6-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP26]], align 4
-// CHECK6-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK6-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK6-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK6-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK6-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK6-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK6-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l41.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK6-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK6-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK6-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK6-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK6-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK6-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK6-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK6-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK6-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK6-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK6-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK6-NEXT: store ptr @.offload_sizes.4, ptr [[TMP29]], align 4
+// CHECK6-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK6-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP30]], align 4
+// CHECK6-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK6-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK6-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK6-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK6-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK6-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK6-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK6-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK6-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK6-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK6-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK6-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK6-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK6-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK6-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l41.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK6-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK6-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK6: omp_offload.failed8:
// CHECK6-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l41(ptr [[THIS1]]) #[[ATTR2]]
// CHECK6-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK6: omp_offload.cont9:
// CHECK6-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK6-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK6-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 4
-// CHECK6-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK6-NEXT: store ptr [[A10]], ptr [[TMP33]], align 4
-// CHECK6-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
-// CHECK6-NEXT: store ptr null, ptr [[TMP34]], align 4
-// CHECK6-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK6-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK6-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK6-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 4
+// CHECK6-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK6-NEXT: store ptr [[A10]], ptr [[TMP41]], align 4
+// CHECK6-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
+// CHECK6-NEXT: store ptr null, ptr [[TMP42]], align 4
+// CHECK6-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK6-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK6-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK6-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK6-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK6-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK6-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK6-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK6-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 4
-// CHECK6-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK6-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 4
-// CHECK6-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK6-NEXT: store ptr @.offload_sizes.8, ptr [[TMP41]], align 4
-// CHECK6-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK6-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP42]], align 4
-// CHECK6-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK6-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK6-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK6-NEXT: store ptr null, ptr [[TMP44]], align 4
-// CHECK6-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK6-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK6-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l46.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK6-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK6-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK6-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK6-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK6-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK6-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK6-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK6-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 4
+// CHECK6-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK6-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 4
+// CHECK6-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK6-NEXT: store ptr @.offload_sizes.8, ptr [[TMP49]], align 4
+// CHECK6-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK6-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP50]], align 4
+// CHECK6-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK6-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK6-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK6-NEXT: store ptr null, ptr [[TMP52]], align 4
+// CHECK6-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK6-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK6-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK6-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK6-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK6-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK6-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK6-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
+// CHECK6-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK6-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK6-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l46.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK6-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK6-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK6: omp_offload.failed16:
// CHECK6-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l46(ptr [[THIS1]]) #[[ATTR2]]
// CHECK6-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK6: omp_offload.cont17:
// CHECK6-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK6-NEXT: [[TMP48:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK6-NEXT: store ptr [[THIS1]], ptr [[TMP48]], align 4
-// CHECK6-NEXT: [[TMP49:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK6-NEXT: store ptr [[A18]], ptr [[TMP49]], align 4
-// CHECK6-NEXT: [[TMP50:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
-// CHECK6-NEXT: store ptr null, ptr [[TMP50]], align 4
-// CHECK6-NEXT: [[TMP51:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK6-NEXT: [[TMP52:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK6-NEXT: [[TMP60:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK6-NEXT: store ptr [[THIS1]], ptr [[TMP60]], align 4
+// CHECK6-NEXT: [[TMP61:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK6-NEXT: store ptr [[A18]], ptr [[TMP61]], align 4
+// CHECK6-NEXT: [[TMP62:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
+// CHECK6-NEXT: store ptr null, ptr [[TMP62]], align 4
+// CHECK6-NEXT: [[TMP63:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK6-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
// CHECK6-NEXT: [[KERNEL_ARGS23:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK6-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
-// CHECK6-NEXT: store i32 1, ptr [[TMP53]], align 4
-// CHECK6-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
-// CHECK6-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK6-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
-// CHECK6-NEXT: store ptr [[TMP51]], ptr [[TMP55]], align 4
-// CHECK6-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
-// CHECK6-NEXT: store ptr [[TMP52]], ptr [[TMP56]], align 4
-// CHECK6-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
-// CHECK6-NEXT: store ptr @.offload_sizes.12, ptr [[TMP57]], align 4
-// CHECK6-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
-// CHECK6-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP58]], align 4
-// CHECK6-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
-// CHECK6-NEXT: store ptr null, ptr [[TMP59]], align 4
-// CHECK6-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
-// CHECK6-NEXT: store ptr null, ptr [[TMP60]], align 4
-// CHECK6-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
-// CHECK6-NEXT: store i64 123, ptr [[TMP61]], align 8
-// CHECK6-NEXT: [[TMP62:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l52.region_id, ptr [[KERNEL_ARGS23]])
-// CHECK6-NEXT: [[TMP63:%.*]] = icmp ne i32 [[TMP62]], 0
-// CHECK6-NEXT: br i1 [[TMP63]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
+// CHECK6-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0
+// CHECK6-NEXT: store i32 2, ptr [[TMP65]], align 4
+// CHECK6-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1
+// CHECK6-NEXT: store i32 1, ptr [[TMP66]], align 4
+// CHECK6-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2
+// CHECK6-NEXT: store ptr [[TMP63]], ptr [[TMP67]], align 4
+// CHECK6-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3
+// CHECK6-NEXT: store ptr [[TMP64]], ptr [[TMP68]], align 4
+// CHECK6-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4
+// CHECK6-NEXT: store ptr @.offload_sizes.12, ptr [[TMP69]], align 4
+// CHECK6-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5
+// CHECK6-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP70]], align 4
+// CHECK6-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6
+// CHECK6-NEXT: store ptr null, ptr [[TMP71]], align 4
+// CHECK6-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7
+// CHECK6-NEXT: store ptr null, ptr [[TMP72]], align 4
+// CHECK6-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8
+// CHECK6-NEXT: store i64 123, ptr [[TMP73]], align 8
+// CHECK6-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 9
+// CHECK6-NEXT: store i64 0, ptr [[TMP74]], align 8
+// CHECK6-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 10
+// CHECK6-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP75]], align 4
+// CHECK6-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 11
+// CHECK6-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP76]], align 4
+// CHECK6-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 12
+// CHECK6-NEXT: store i32 0, ptr [[TMP77]], align 4
+// CHECK6-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l52.region_id, ptr [[KERNEL_ARGS23]])
+// CHECK6-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
+// CHECK6-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
// CHECK6: omp_offload.failed24:
// CHECK6-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l52(ptr [[THIS1]]) #[[ATTR2]]
// CHECK6-NEXT: br label [[OMP_OFFLOAD_CONT25]]
// CHECK6: omp_offload.cont25:
// CHECK6-NEXT: [[A26:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK6-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK6-NEXT: store ptr [[THIS1]], ptr [[TMP64]], align 4
-// CHECK6-NEXT: [[TMP65:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
-// CHECK6-NEXT: store ptr [[A26]], ptr [[TMP65]], align 4
-// CHECK6-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i32 0, i32 0
-// CHECK6-NEXT: store ptr null, ptr [[TMP66]], align 4
-// CHECK6-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
-// CHECK6-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK6-NEXT: [[TMP80:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK6-NEXT: store ptr [[THIS1]], ptr [[TMP80]], align 4
+// CHECK6-NEXT: [[TMP81:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
+// CHECK6-NEXT: store ptr [[A26]], ptr [[TMP81]], align 4
+// CHECK6-NEXT: [[TMP82:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i32 0, i32 0
+// CHECK6-NEXT: store ptr null, ptr [[TMP82]], align 4
+// CHECK6-NEXT: [[TMP83:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0
+// CHECK6-NEXT: [[TMP84:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0
// CHECK6-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK6-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
-// CHECK6-NEXT: store i32 1, ptr [[TMP69]], align 4
-// CHECK6-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
-// CHECK6-NEXT: store i32 1, ptr [[TMP70]], align 4
-// CHECK6-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
-// CHECK6-NEXT: store ptr [[TMP67]], ptr [[TMP71]], align 4
-// CHECK6-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
-// CHECK6-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 4
-// CHECK6-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
-// CHECK6-NEXT: store ptr @.offload_sizes.16, ptr [[TMP73]], align 4
-// CHECK6-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
-// CHECK6-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP74]], align 4
-// CHECK6-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
-// CHECK6-NEXT: store ptr null, ptr [[TMP75]], align 4
-// CHECK6-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
-// CHECK6-NEXT: store ptr null, ptr [[TMP76]], align 4
-// CHECK6-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
-// CHECK6-NEXT: store i64 123, ptr [[TMP77]], align 8
-// CHECK6-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l58.region_id, ptr [[KERNEL_ARGS31]])
-// CHECK6-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0
-// CHECK6-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
+// CHECK6-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0
+// CHECK6-NEXT: store i32 2, ptr [[TMP85]], align 4
+// CHECK6-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1
+// CHECK6-NEXT: store i32 1, ptr [[TMP86]], align 4
+// CHECK6-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2
+// CHECK6-NEXT: store ptr [[TMP83]], ptr [[TMP87]], align 4
+// CHECK6-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3
+// CHECK6-NEXT: store ptr [[TMP84]], ptr [[TMP88]], align 4
+// CHECK6-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4
+// CHECK6-NEXT: store ptr @.offload_sizes.16, ptr [[TMP89]], align 4
+// CHECK6-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5
+// CHECK6-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP90]], align 4
+// CHECK6-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6
+// CHECK6-NEXT: store ptr null, ptr [[TMP91]], align 4
+// CHECK6-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7
+// CHECK6-NEXT: store ptr null, ptr [[TMP92]], align 4
+// CHECK6-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8
+// CHECK6-NEXT: store i64 123, ptr [[TMP93]], align 8
+// CHECK6-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9
+// CHECK6-NEXT: store i64 0, ptr [[TMP94]], align 8
+// CHECK6-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10
+// CHECK6-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP95]], align 4
+// CHECK6-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11
+// CHECK6-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP96]], align 4
+// CHECK6-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12
+// CHECK6-NEXT: store i32 0, ptr [[TMP97]], align 4
+// CHECK6-NEXT: [[TMP98:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l58.region_id, ptr [[KERNEL_ARGS31]])
+// CHECK6-NEXT: [[TMP99:%.*]] = icmp ne i32 [[TMP98]], 0
+// CHECK6-NEXT: br i1 [[TMP99]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]]
// CHECK6: omp_offload.failed32:
// CHECK6-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l58(ptr [[THIS1]]) #[[ATTR2]]
// CHECK6-NEXT: br label [[OMP_OFFLOAD_CONT33]]
// CHECK6: omp_offload.cont33:
// CHECK6-NEXT: [[A34:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK6-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A34]], i32 0, i32 0
-// CHECK6-NEXT: [[TMP80:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK6-NEXT: ret i32 [[TMP80]]
+// CHECK6-NEXT: [[TMP100:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK6-NEXT: ret i32 [[TMP100]]
//
//
// CHECK6-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36
@@ -4908,7 +5068,7 @@ int main (int argc, char **argv) {
// CHECK13-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK13-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK13-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK13-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK13-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK13-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK13-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -4925,294 +5085,334 @@ int main (int argc, char **argv) {
// CHECK13-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK13-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK13-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK13-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l154.region_id, ptr [[KERNEL_ARGS]])
-// CHECK13-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK13-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK13-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK13-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK13-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK13-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK13-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l154.region_id, ptr [[KERNEL_ARGS]])
+// CHECK13-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK13-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK13: omp_offload.failed:
// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l154(i64 [[TMP4]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK13: omp_offload.cont:
-// CHECK13-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK13-NEXT: [[TMP35:%.*]] = load i64, ptr [[N_CASTED3]], align 8
-// CHECK13-NEXT: [[TMP36:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK13-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK13-NEXT: [[TMP39:%.*]] = load i64, ptr [[N_CASTED3]], align 8
+// CHECK13-NEXT: [[TMP40:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK13-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES7]], ptr align 8 @.offload_sizes.4, i64 24, i1 false)
-// CHECK13-NEXT: [[TMP37:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP35]], ptr [[TMP37]], align 8
-// CHECK13-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP35]], ptr [[TMP38]], align 8
-// CHECK13-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK13-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP40]], align 8
-// CHECK13-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP41]], align 8
-// CHECK13-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
-// CHECK13-NEXT: store ptr null, ptr [[TMP42]], align 8
-// CHECK13-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP43]], align 8
-// CHECK13-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 8
-// CHECK13-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK13-NEXT: store i64 [[TMP36]], ptr [[TMP45]], align 8
-// CHECK13-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK13-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP39]], ptr [[TMP41]], align 8
+// CHECK13-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP39]], ptr [[TMP42]], align 8
+// CHECK13-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK13-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP44]], align 8
+// CHECK13-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP45]], align 8
+// CHECK13-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
// CHECK13-NEXT: store ptr null, ptr [[TMP46]], align 8
-// CHECK13-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP50:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP50]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK13-NEXT: [[TMP51:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK13-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP51]], 0
+// CHECK13-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP47]], align 8
+// CHECK13-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 8
+// CHECK13-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK13-NEXT: store i64 [[TMP40]], ptr [[TMP49]], align 8
+// CHECK13-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK13-NEXT: store ptr null, ptr [[TMP50]], align 8
+// CHECK13-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP54:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP54]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK13-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK13-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP55]], 0
// CHECK13-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK13-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK13-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK13-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK13-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP52]], 1
-// CHECK13-NEXT: [[TMP53:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK13-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK13-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP56]], 1
+// CHECK13-NEXT: [[TMP57:%.*]] = zext i32 [[ADD14]] to i64
// CHECK13-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK13-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK13-NEXT: store i32 3, ptr [[TMP55]], align 4
-// CHECK13-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP47]], ptr [[TMP56]], align 8
-// CHECK13-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 8
-// CHECK13-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK13-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 8
-// CHECK13-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP59]], align 8
-// CHECK13-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP60]], align 8
-// CHECK13-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP61]], align 8
-// CHECK13-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK13-NEXT: store i64 [[TMP53]], ptr [[TMP62]], align 8
-// CHECK13-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l159.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK13-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
-// CHECK13-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK13-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP58]], align 4
+// CHECK13-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK13-NEXT: store i32 3, ptr [[TMP59]], align 4
+// CHECK13-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP51]], ptr [[TMP60]], align 8
+// CHECK13-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 8
+// CHECK13-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK13-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 8
+// CHECK13-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP63]], align 8
+// CHECK13-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP64]], align 8
+// CHECK13-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP65]], align 8
+// CHECK13-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK13-NEXT: store i64 [[TMP57]], ptr [[TMP66]], align 8
+// CHECK13-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP67]], align 8
+// CHECK13-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP68]], align 4
+// CHECK13-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK13-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP70]], align 4
+// CHECK13-NEXT: [[TMP71:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l159.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK13-NEXT: [[TMP72:%.*]] = icmp ne i32 [[TMP71]], 0
+// CHECK13-NEXT: br i1 [[TMP72]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK13: omp_offload.failed16:
-// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l159(i64 [[TMP35]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l159(i64 [[TMP39]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK13: omp_offload.cont17:
-// CHECK13-NEXT: [[TMP65:%.*]] = load i32, ptr [[M]], align 4
-// CHECK13-NEXT: store i32 [[TMP65]], ptr [[M_CASTED]], align 4
-// CHECK13-NEXT: [[TMP66:%.*]] = load i64, ptr [[M_CASTED]], align 8
-// CHECK13-NEXT: [[TMP67:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP67]], ptr [[N_CASTED18]], align 4
-// CHECK13-NEXT: [[TMP68:%.*]] = load i64, ptr [[N_CASTED18]], align 8
-// CHECK13-NEXT: [[TMP69:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK13-NEXT: [[TMP73:%.*]] = load i32, ptr [[M]], align 4
+// CHECK13-NEXT: store i32 [[TMP73]], ptr [[M_CASTED]], align 4
+// CHECK13-NEXT: [[TMP74:%.*]] = load i64, ptr [[M_CASTED]], align 8
+// CHECK13-NEXT: [[TMP75:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP75]], ptr [[N_CASTED18]], align 4
+// CHECK13-NEXT: [[TMP76:%.*]] = load i64, ptr [[N_CASTED18]], align 8
+// CHECK13-NEXT: [[TMP77:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK13-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES22]], ptr align 8 @.offload_sizes.8, i64 32, i1 false)
-// CHECK13-NEXT: [[TMP70:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP66]], ptr [[TMP70]], align 8
-// CHECK13-NEXT: [[TMP71:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP66]], ptr [[TMP71]], align 8
-// CHECK13-NEXT: [[TMP72:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP72]], align 8
-// CHECK13-NEXT: [[TMP73:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP68]], ptr [[TMP73]], align 8
-// CHECK13-NEXT: [[TMP74:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP68]], ptr [[TMP74]], align 8
-// CHECK13-NEXT: [[TMP75:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 1
-// CHECK13-NEXT: store ptr null, ptr [[TMP75]], align 8
-// CHECK13-NEXT: [[TMP76:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP76]], align 8
-// CHECK13-NEXT: [[TMP77:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP77]], align 8
-// CHECK13-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 2
-// CHECK13-NEXT: store ptr null, ptr [[TMP78]], align 8
-// CHECK13-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP79]], align 8
-// CHECK13-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP80]], align 8
-// CHECK13-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 3
-// CHECK13-NEXT: store i64 [[TMP69]], ptr [[TMP81]], align 8
-// CHECK13-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 3
-// CHECK13-NEXT: store ptr null, ptr [[TMP82]], align 8
-// CHECK13-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP86:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP86]], ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK13-NEXT: [[TMP87:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK13-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP87]], 0
+// CHECK13-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP74]], ptr [[TMP78]], align 8
+// CHECK13-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP74]], ptr [[TMP79]], align 8
+// CHECK13-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP80]], align 8
+// CHECK13-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP76]], ptr [[TMP81]], align 8
+// CHECK13-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP76]], ptr [[TMP82]], align 8
+// CHECK13-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 1
+// CHECK13-NEXT: store ptr null, ptr [[TMP83]], align 8
+// CHECK13-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP84]], align 8
+// CHECK13-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP85]], align 8
+// CHECK13-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 2
+// CHECK13-NEXT: store ptr null, ptr [[TMP86]], align 8
+// CHECK13-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP87]], align 8
+// CHECK13-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP88]], align 8
+// CHECK13-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 3
+// CHECK13-NEXT: store i64 [[TMP77]], ptr [[TMP89]], align 8
+// CHECK13-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 3
+// CHECK13-NEXT: store ptr null, ptr [[TMP90]], align 8
+// CHECK13-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP94:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP94]], ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK13-NEXT: [[TMP95:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK13-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP95]], 0
// CHECK13-NEXT: [[DIV27:%.*]] = sdiv i32 [[SUB26]], 1
// CHECK13-NEXT: [[SUB28:%.*]] = sub nsw i32 [[DIV27]], 1
// CHECK13-NEXT: store i32 [[SUB28]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK13-NEXT: [[TMP88:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK13-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP88]], 1
-// CHECK13-NEXT: [[TMP89:%.*]] = zext i32 [[ADD29]] to i64
+// CHECK13-NEXT: [[TMP96:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK13-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP96]], 1
+// CHECK13-NEXT: [[TMP97:%.*]] = zext i32 [[ADD29]] to i64
// CHECK13-NEXT: [[KERNEL_ARGS30:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP90]], align 4
-// CHECK13-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
-// CHECK13-NEXT: store i32 4, ptr [[TMP91]], align 4
-// CHECK13-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP83]], ptr [[TMP92]], align 8
-// CHECK13-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP84]], ptr [[TMP93]], align 8
-// CHECK13-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
-// CHECK13-NEXT: store ptr [[TMP85]], ptr [[TMP94]], align 8
-// CHECK13-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP95]], align 8
-// CHECK13-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP96]], align 8
-// CHECK13-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP97]], align 8
-// CHECK13-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
-// CHECK13-NEXT: store i64 [[TMP89]], ptr [[TMP98]], align 8
-// CHECK13-NEXT: [[TMP99:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l164.region_id, ptr [[KERNEL_ARGS30]])
-// CHECK13-NEXT: [[TMP100:%.*]] = icmp ne i32 [[TMP99]], 0
-// CHECK13-NEXT: br i1 [[TMP100]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
+// CHECK13-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP98]], align 4
+// CHECK13-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
+// CHECK13-NEXT: store i32 4, ptr [[TMP99]], align 4
+// CHECK13-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP91]], ptr [[TMP100]], align 8
+// CHECK13-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP92]], ptr [[TMP101]], align 8
+// CHECK13-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
+// CHECK13-NEXT: store ptr [[TMP93]], ptr [[TMP102]], align 8
+// CHECK13-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP103]], align 8
+// CHECK13-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP104]], align 8
+// CHECK13-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP105]], align 8
+// CHECK13-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
+// CHECK13-NEXT: store i64 [[TMP97]], ptr [[TMP106]], align 8
+// CHECK13-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP107]], align 8
+// CHECK13-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP108]], align 4
+// CHECK13-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP109]], align 4
+// CHECK13-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP110]], align 4
+// CHECK13-NEXT: [[TMP111:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l164.region_id, ptr [[KERNEL_ARGS30]])
+// CHECK13-NEXT: [[TMP112:%.*]] = icmp ne i32 [[TMP111]], 0
+// CHECK13-NEXT: br i1 [[TMP112]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
// CHECK13: omp_offload.failed31:
-// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l164(i64 [[TMP66]], i64 [[TMP68]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l164(i64 [[TMP74]], i64 [[TMP76]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT32]]
// CHECK13: omp_offload.cont32:
-// CHECK13-NEXT: [[TMP101:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP101]], ptr [[N_CASTED33]], align 4
-// CHECK13-NEXT: [[TMP102:%.*]] = load i64, ptr [[N_CASTED33]], align 8
-// CHECK13-NEXT: [[TMP103:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK13-NEXT: [[TMP113:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP113]], ptr [[N_CASTED33]], align 4
+// CHECK13-NEXT: [[TMP114:%.*]] = load i64, ptr [[N_CASTED33]], align 8
+// CHECK13-NEXT: [[TMP115:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK13-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES37]], ptr align 8 @.offload_sizes.12, i64 24, i1 false)
-// CHECK13-NEXT: [[TMP104:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP102]], ptr [[TMP104]], align 8
-// CHECK13-NEXT: [[TMP105:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP102]], ptr [[TMP105]], align 8
-// CHECK13-NEXT: [[TMP106:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP106]], align 8
-// CHECK13-NEXT: [[TMP107:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP107]], align 8
-// CHECK13-NEXT: [[TMP108:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP108]], align 8
-// CHECK13-NEXT: [[TMP109:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i64 0, i64 1
-// CHECK13-NEXT: store ptr null, ptr [[TMP109]], align 8
-// CHECK13-NEXT: [[TMP110:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP110]], align 8
-// CHECK13-NEXT: [[TMP111:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP111]], align 8
-// CHECK13-NEXT: [[TMP112:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 2
-// CHECK13-NEXT: store i64 [[TMP103]], ptr [[TMP112]], align 8
-// CHECK13-NEXT: [[TMP113:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i64 0, i64 2
-// CHECK13-NEXT: store ptr null, ptr [[TMP113]], align 8
-// CHECK13-NEXT: [[TMP114:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP115:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP116:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP117:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP117]], ptr [[DOTCAPTURE_EXPR_39]], align 4
-// CHECK13-NEXT: [[TMP118:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_39]], align 4
-// CHECK13-NEXT: [[SUB41:%.*]] = sub nsw i32 [[TMP118]], 0
+// CHECK13-NEXT: [[TMP116:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP114]], ptr [[TMP116]], align 8
+// CHECK13-NEXT: [[TMP117:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP114]], ptr [[TMP117]], align 8
+// CHECK13-NEXT: [[TMP118:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP118]], align 8
+// CHECK13-NEXT: [[TMP119:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP119]], align 8
+// CHECK13-NEXT: [[TMP120:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP120]], align 8
+// CHECK13-NEXT: [[TMP121:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i64 0, i64 1
+// CHECK13-NEXT: store ptr null, ptr [[TMP121]], align 8
+// CHECK13-NEXT: [[TMP122:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP122]], align 8
+// CHECK13-NEXT: [[TMP123:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP123]], align 8
+// CHECK13-NEXT: [[TMP124:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 2
+// CHECK13-NEXT: store i64 [[TMP115]], ptr [[TMP124]], align 8
+// CHECK13-NEXT: [[TMP125:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i64 0, i64 2
+// CHECK13-NEXT: store ptr null, ptr [[TMP125]], align 8
+// CHECK13-NEXT: [[TMP126:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP127:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP128:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP129:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP129]], ptr [[DOTCAPTURE_EXPR_39]], align 4
+// CHECK13-NEXT: [[TMP130:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_39]], align 4
+// CHECK13-NEXT: [[SUB41:%.*]] = sub nsw i32 [[TMP130]], 0
// CHECK13-NEXT: [[DIV42:%.*]] = sdiv i32 [[SUB41]], 1
// CHECK13-NEXT: [[SUB43:%.*]] = sub nsw i32 [[DIV42]], 1
// CHECK13-NEXT: store i32 [[SUB43]], ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK13-NEXT: [[TMP119:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK13-NEXT: [[ADD44:%.*]] = add nsw i32 [[TMP119]], 1
-// CHECK13-NEXT: [[TMP120:%.*]] = zext i32 [[ADD44]] to i64
+// CHECK13-NEXT: [[TMP131:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
+// CHECK13-NEXT: [[ADD44:%.*]] = add nsw i32 [[TMP131]], 1
+// CHECK13-NEXT: [[TMP132:%.*]] = zext i32 [[ADD44]] to i64
// CHECK13-NEXT: [[KERNEL_ARGS45:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP121:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP121]], align 4
-// CHECK13-NEXT: [[TMP122:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 1
-// CHECK13-NEXT: store i32 3, ptr [[TMP122]], align 4
-// CHECK13-NEXT: [[TMP123:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP114]], ptr [[TMP123]], align 8
-// CHECK13-NEXT: [[TMP124:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP115]], ptr [[TMP124]], align 8
-// CHECK13-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 4
-// CHECK13-NEXT: store ptr [[TMP116]], ptr [[TMP125]], align 8
-// CHECK13-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP126]], align 8
-// CHECK13-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP127]], align 8
-// CHECK13-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP128]], align 8
-// CHECK13-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 8
-// CHECK13-NEXT: store i64 [[TMP120]], ptr [[TMP129]], align 8
-// CHECK13-NEXT: [[TMP130:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l169.region_id, ptr [[KERNEL_ARGS45]])
-// CHECK13-NEXT: [[TMP131:%.*]] = icmp ne i32 [[TMP130]], 0
-// CHECK13-NEXT: br i1 [[TMP131]], label [[OMP_OFFLOAD_FAILED46:%.*]], label [[OMP_OFFLOAD_CONT47:%.*]]
+// CHECK13-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP133]], align 4
+// CHECK13-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 1
+// CHECK13-NEXT: store i32 3, ptr [[TMP134]], align 4
+// CHECK13-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP126]], ptr [[TMP135]], align 8
+// CHECK13-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP127]], ptr [[TMP136]], align 8
+// CHECK13-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 4
+// CHECK13-NEXT: store ptr [[TMP128]], ptr [[TMP137]], align 8
+// CHECK13-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP138]], align 8
+// CHECK13-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP139]], align 8
+// CHECK13-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP140]], align 8
+// CHECK13-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 8
+// CHECK13-NEXT: store i64 [[TMP132]], ptr [[TMP141]], align 8
+// CHECK13-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP142]], align 8
+// CHECK13-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP143]], align 4
+// CHECK13-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP144]], align 4
+// CHECK13-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP145]], align 4
+// CHECK13-NEXT: [[TMP146:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l169.region_id, ptr [[KERNEL_ARGS45]])
+// CHECK13-NEXT: [[TMP147:%.*]] = icmp ne i32 [[TMP146]], 0
+// CHECK13-NEXT: br i1 [[TMP147]], label [[OMP_OFFLOAD_FAILED46:%.*]], label [[OMP_OFFLOAD_CONT47:%.*]]
// CHECK13: omp_offload.failed46:
-// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l169(i64 [[TMP102]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l169(i64 [[TMP114]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT47]]
// CHECK13: omp_offload.cont47:
-// CHECK13-NEXT: [[TMP132:%.*]] = load i32, ptr [[M]], align 4
-// CHECK13-NEXT: store i32 [[TMP132]], ptr [[M_CASTED48]], align 4
-// CHECK13-NEXT: [[TMP133:%.*]] = load i64, ptr [[M_CASTED48]], align 8
-// CHECK13-NEXT: [[TMP134:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP134]], ptr [[N_CASTED49]], align 4
-// CHECK13-NEXT: [[TMP135:%.*]] = load i64, ptr [[N_CASTED49]], align 8
-// CHECK13-NEXT: [[TMP136:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK13-NEXT: [[TMP148:%.*]] = load i32, ptr [[M]], align 4
+// CHECK13-NEXT: store i32 [[TMP148]], ptr [[M_CASTED48]], align 4
+// CHECK13-NEXT: [[TMP149:%.*]] = load i64, ptr [[M_CASTED48]], align 8
+// CHECK13-NEXT: [[TMP150:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP150]], ptr [[N_CASTED49]], align 4
+// CHECK13-NEXT: [[TMP151:%.*]] = load i64, ptr [[N_CASTED49]], align 8
+// CHECK13-NEXT: [[TMP152:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK13-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES53]], ptr align 8 @.offload_sizes.16, i64 32, i1 false)
-// CHECK13-NEXT: [[TMP137:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP133]], ptr [[TMP137]], align 8
-// CHECK13-NEXT: [[TMP138:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP133]], ptr [[TMP138]], align 8
-// CHECK13-NEXT: [[TMP139:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP139]], align 8
-// CHECK13-NEXT: [[TMP140:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP135]], ptr [[TMP140]], align 8
-// CHECK13-NEXT: [[TMP141:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 1
-// CHECK13-NEXT: store i64 [[TMP135]], ptr [[TMP141]], align 8
-// CHECK13-NEXT: [[TMP142:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 1
-// CHECK13-NEXT: store ptr null, ptr [[TMP142]], align 8
-// CHECK13-NEXT: [[TMP143:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 2
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP143]], align 8
-// CHECK13-NEXT: [[TMP144:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 2
-// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP144]], align 8
-// CHECK13-NEXT: [[TMP145:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 2
-// CHECK13-NEXT: store ptr null, ptr [[TMP145]], align 8
-// CHECK13-NEXT: [[TMP146:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP146]], align 8
-// CHECK13-NEXT: [[TMP147:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP147]], align 8
-// CHECK13-NEXT: [[TMP148:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 3
-// CHECK13-NEXT: store i64 [[TMP136]], ptr [[TMP148]], align 8
-// CHECK13-NEXT: [[TMP149:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 3
-// CHECK13-NEXT: store ptr null, ptr [[TMP149]], align 8
-// CHECK13-NEXT: [[TMP150:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP151:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP152:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP153:%.*]] = load i32, ptr [[N]], align 4
-// CHECK13-NEXT: store i32 [[TMP153]], ptr [[DOTCAPTURE_EXPR_55]], align 4
-// CHECK13-NEXT: [[TMP154:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_55]], align 4
-// CHECK13-NEXT: [[SUB57:%.*]] = sub nsw i32 [[TMP154]], 0
+// CHECK13-NEXT: [[TMP153:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP149]], ptr [[TMP153]], align 8
+// CHECK13-NEXT: [[TMP154:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP149]], ptr [[TMP154]], align 8
+// CHECK13-NEXT: [[TMP155:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP155]], align 8
+// CHECK13-NEXT: [[TMP156:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP151]], ptr [[TMP156]], align 8
+// CHECK13-NEXT: [[TMP157:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 1
+// CHECK13-NEXT: store i64 [[TMP151]], ptr [[TMP157]], align 8
+// CHECK13-NEXT: [[TMP158:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 1
+// CHECK13-NEXT: store ptr null, ptr [[TMP158]], align 8
+// CHECK13-NEXT: [[TMP159:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 2
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP159]], align 8
+// CHECK13-NEXT: [[TMP160:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 2
+// CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP160]], align 8
+// CHECK13-NEXT: [[TMP161:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 2
+// CHECK13-NEXT: store ptr null, ptr [[TMP161]], align 8
+// CHECK13-NEXT: [[TMP162:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP162]], align 8
+// CHECK13-NEXT: [[TMP163:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP163]], align 8
+// CHECK13-NEXT: [[TMP164:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 3
+// CHECK13-NEXT: store i64 [[TMP152]], ptr [[TMP164]], align 8
+// CHECK13-NEXT: [[TMP165:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 3
+// CHECK13-NEXT: store ptr null, ptr [[TMP165]], align 8
+// CHECK13-NEXT: [[TMP166:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP167:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP168:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP169:%.*]] = load i32, ptr [[N]], align 4
+// CHECK13-NEXT: store i32 [[TMP169]], ptr [[DOTCAPTURE_EXPR_55]], align 4
+// CHECK13-NEXT: [[TMP170:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_55]], align 4
+// CHECK13-NEXT: [[SUB57:%.*]] = sub nsw i32 [[TMP170]], 0
// CHECK13-NEXT: [[DIV58:%.*]] = sdiv i32 [[SUB57]], 1
// CHECK13-NEXT: [[SUB59:%.*]] = sub nsw i32 [[DIV58]], 1
// CHECK13-NEXT: store i32 [[SUB59]], ptr [[DOTCAPTURE_EXPR_56]], align 4
-// CHECK13-NEXT: [[TMP155:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_56]], align 4
-// CHECK13-NEXT: [[ADD60:%.*]] = add nsw i32 [[TMP155]], 1
-// CHECK13-NEXT: [[TMP156:%.*]] = zext i32 [[ADD60]] to i64
+// CHECK13-NEXT: [[TMP171:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_56]], align 4
+// CHECK13-NEXT: [[ADD60:%.*]] = add nsw i32 [[TMP171]], 1
+// CHECK13-NEXT: [[TMP172:%.*]] = zext i32 [[ADD60]] to i64
// CHECK13-NEXT: [[KERNEL_ARGS61:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP157:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP157]], align 4
-// CHECK13-NEXT: [[TMP158:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 1
-// CHECK13-NEXT: store i32 4, ptr [[TMP158]], align 4
-// CHECK13-NEXT: [[TMP159:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP150]], ptr [[TMP159]], align 8
-// CHECK13-NEXT: [[TMP160:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP151]], ptr [[TMP160]], align 8
-// CHECK13-NEXT: [[TMP161:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 4
-// CHECK13-NEXT: store ptr [[TMP152]], ptr [[TMP161]], align 8
-// CHECK13-NEXT: [[TMP162:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP162]], align 8
-// CHECK13-NEXT: [[TMP163:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP163]], align 8
-// CHECK13-NEXT: [[TMP164:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP164]], align 8
-// CHECK13-NEXT: [[TMP165:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 8
-// CHECK13-NEXT: store i64 [[TMP156]], ptr [[TMP165]], align 8
-// CHECK13-NEXT: [[TMP166:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l174.region_id, ptr [[KERNEL_ARGS61]])
-// CHECK13-NEXT: [[TMP167:%.*]] = icmp ne i32 [[TMP166]], 0
-// CHECK13-NEXT: br i1 [[TMP167]], label [[OMP_OFFLOAD_FAILED62:%.*]], label [[OMP_OFFLOAD_CONT63:%.*]]
+// CHECK13-NEXT: [[TMP173:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP173]], align 4
+// CHECK13-NEXT: [[TMP174:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 1
+// CHECK13-NEXT: store i32 4, ptr [[TMP174]], align 4
+// CHECK13-NEXT: [[TMP175:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP166]], ptr [[TMP175]], align 8
+// CHECK13-NEXT: [[TMP176:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP167]], ptr [[TMP176]], align 8
+// CHECK13-NEXT: [[TMP177:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 4
+// CHECK13-NEXT: store ptr [[TMP168]], ptr [[TMP177]], align 8
+// CHECK13-NEXT: [[TMP178:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP178]], align 8
+// CHECK13-NEXT: [[TMP179:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP179]], align 8
+// CHECK13-NEXT: [[TMP180:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP180]], align 8
+// CHECK13-NEXT: [[TMP181:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 8
+// CHECK13-NEXT: store i64 [[TMP172]], ptr [[TMP181]], align 8
+// CHECK13-NEXT: [[TMP182:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP182]], align 8
+// CHECK13-NEXT: [[TMP183:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP183]], align 4
+// CHECK13-NEXT: [[TMP184:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP184]], align 4
+// CHECK13-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP185]], align 4
+// CHECK13-NEXT: [[TMP186:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l174.region_id, ptr [[KERNEL_ARGS61]])
+// CHECK13-NEXT: [[TMP187:%.*]] = icmp ne i32 [[TMP186]], 0
+// CHECK13-NEXT: br i1 [[TMP187]], label [[OMP_OFFLOAD_FAILED62:%.*]], label [[OMP_OFFLOAD_CONT63:%.*]]
// CHECK13: omp_offload.failed62:
-// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l174(i64 [[TMP133]], i64 [[TMP135]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l174(i64 [[TMP149]], i64 [[TMP151]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT63]]
// CHECK13: omp_offload.cont63:
-// CHECK13-NEXT: [[TMP168:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK13-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP168]])
+// CHECK13-NEXT: [[TMP188:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK13-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP188]])
// CHECK13-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK13-NEXT: [[TMP169:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK13-NEXT: call void @llvm.stackrestore(ptr [[TMP169]])
-// CHECK13-NEXT: [[TMP170:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK13-NEXT: ret i32 [[TMP170]]
+// CHECK13-NEXT: [[TMP189:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK13-NEXT: call void @llvm.stackrestore(ptr [[TMP189]])
+// CHECK13-NEXT: [[TMP190:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK13-NEXT: ret i32 [[TMP190]]
//
//
// CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l154
@@ -6482,7 +6682,7 @@ int main (int argc, char **argv) {
// CHECK13-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK13-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK13-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK13-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK13-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK13-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK13-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -6499,165 +6699,205 @@ int main (int argc, char **argv) {
// CHECK13-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK13-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK13-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK13-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l122.region_id, ptr [[KERNEL_ARGS]])
-// CHECK13-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK13-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK13-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK13-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK13-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK13-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK13-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l122.region_id, ptr [[KERNEL_ARGS]])
+// CHECK13-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK13-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK13: omp_offload.failed:
// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l122(ptr [[A]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK13: omp_offload.cont:
-// CHECK13-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP16]], align 8
-// CHECK13-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP17]], align 8
-// CHECK13-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK13-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP20]], align 8
+// CHECK13-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP21]], align 8
+// CHECK13-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK13-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK13-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK13-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK13-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK13-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK13-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK13-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK13-NEXT: store ptr @.offload_sizes.24, ptr [[TMP25]], align 8
-// CHECK13-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP26]], align 8
-// CHECK13-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK13-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK13-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK13-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK13-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l127.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK13-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK13-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK13-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK13-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK13-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK13-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK13-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK13-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK13-NEXT: store ptr @.offload_sizes.24, ptr [[TMP29]], align 8
+// CHECK13-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP30]], align 8
+// CHECK13-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK13-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK13-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK13-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK13-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK13-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK13-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK13-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK13-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l127.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK13-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK13-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK13: omp_offload.failed6:
// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l127(ptr [[A]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK13: omp_offload.cont7:
-// CHECK13-NEXT: [[TMP32:%.*]] = load i32, ptr [[M]], align 4
-// CHECK13-NEXT: store i32 [[TMP32]], ptr [[M_CASTED]], align 4
-// CHECK13-NEXT: [[TMP33:%.*]] = load i64, ptr [[M_CASTED]], align 8
-// CHECK13-NEXT: [[TMP34:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP33]], ptr [[TMP34]], align 8
-// CHECK13-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP33]], ptr [[TMP35]], align 8
-// CHECK13-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP36]], align 8
-// CHECK13-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP37]], align 8
-// CHECK13-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP38]], align 8
-// CHECK13-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
-// CHECK13-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK13-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP40:%.*]] = load i32, ptr [[M]], align 4
+// CHECK13-NEXT: store i32 [[TMP40]], ptr [[M_CASTED]], align 4
+// CHECK13-NEXT: [[TMP41:%.*]] = load i64, ptr [[M_CASTED]], align 8
+// CHECK13-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP41]], ptr [[TMP42]], align 8
+// CHECK13-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP41]], ptr [[TMP43]], align 8
+// CHECK13-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP44]], align 8
+// CHECK13-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP45]], align 8
+// CHECK13-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP46]], align 8
+// CHECK13-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
+// CHECK13-NEXT: store ptr null, ptr [[TMP47]], align 8
+// CHECK13-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK13-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP42]], align 4
-// CHECK13-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK13-NEXT: store i32 2, ptr [[TMP43]], align 4
-// CHECK13-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP40]], ptr [[TMP44]], align 8
-// CHECK13-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP41]], ptr [[TMP45]], align 8
-// CHECK13-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK13-NEXT: store ptr @.offload_sizes.28, ptr [[TMP46]], align 8
-// CHECK13-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP47]], align 8
-// CHECK13-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP48]], align 8
-// CHECK13-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP49]], align 8
-// CHECK13-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK13-NEXT: store i64 10, ptr [[TMP50]], align 8
-// CHECK13-NEXT: [[TMP51:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l132.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK13-NEXT: [[TMP52:%.*]] = icmp ne i32 [[TMP51]], 0
-// CHECK13-NEXT: br i1 [[TMP52]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK13-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP50]], align 4
+// CHECK13-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK13-NEXT: store i32 2, ptr [[TMP51]], align 4
+// CHECK13-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP48]], ptr [[TMP52]], align 8
+// CHECK13-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 8
+// CHECK13-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK13-NEXT: store ptr @.offload_sizes.28, ptr [[TMP54]], align 8
+// CHECK13-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP55]], align 8
+// CHECK13-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP56]], align 8
+// CHECK13-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP57]], align 8
+// CHECK13-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK13-NEXT: store i64 10, ptr [[TMP58]], align 8
+// CHECK13-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP59]], align 8
+// CHECK13-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP60]], align 4
+// CHECK13-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4
+// CHECK13-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP62]], align 4
+// CHECK13-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l132.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK13-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
+// CHECK13-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK13: omp_offload.failed13:
-// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l132(i64 [[TMP33]], ptr [[A]]) #[[ATTR3]]
+// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l132(i64 [[TMP41]], ptr [[A]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT14]]
// CHECK13: omp_offload.cont14:
-// CHECK13-NEXT: [[TMP53:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP53]], align 8
-// CHECK13-NEXT: [[TMP54:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP54]], align 8
-// CHECK13-NEXT: [[TMP55:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP55]], align 8
-// CHECK13-NEXT: [[TMP56:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP57:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP65:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP65]], align 8
+// CHECK13-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP66]], align 8
+// CHECK13-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP67]], align 8
+// CHECK13-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP69:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
// CHECK13-NEXT: [[KERNEL_ARGS19:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP58]], align 4
-// CHECK13-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
-// CHECK13-NEXT: store i32 1, ptr [[TMP59]], align 4
-// CHECK13-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP56]], ptr [[TMP60]], align 8
-// CHECK13-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP57]], ptr [[TMP61]], align 8
-// CHECK13-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
-// CHECK13-NEXT: store ptr @.offload_sizes.32, ptr [[TMP62]], align 8
-// CHECK13-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP63]], align 8
-// CHECK13-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP64]], align 8
-// CHECK13-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP65]], align 8
-// CHECK13-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
-// CHECK13-NEXT: store i64 10, ptr [[TMP66]], align 8
-// CHECK13-NEXT: [[TMP67:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l137.region_id, ptr [[KERNEL_ARGS19]])
-// CHECK13-NEXT: [[TMP68:%.*]] = icmp ne i32 [[TMP67]], 0
-// CHECK13-NEXT: br i1 [[TMP68]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
+// CHECK13-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP70]], align 4
+// CHECK13-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
+// CHECK13-NEXT: store i32 1, ptr [[TMP71]], align 4
+// CHECK13-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 8
+// CHECK13-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP69]], ptr [[TMP73]], align 8
+// CHECK13-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
+// CHECK13-NEXT: store ptr @.offload_sizes.32, ptr [[TMP74]], align 8
+// CHECK13-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP75]], align 8
+// CHECK13-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP76]], align 8
+// CHECK13-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP77]], align 8
+// CHECK13-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
+// CHECK13-NEXT: store i64 10, ptr [[TMP78]], align 8
+// CHECK13-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP79]], align 8
+// CHECK13-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP80]], align 4
+// CHECK13-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP81]], align 4
+// CHECK13-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP82]], align 4
+// CHECK13-NEXT: [[TMP83:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l137.region_id, ptr [[KERNEL_ARGS19]])
+// CHECK13-NEXT: [[TMP84:%.*]] = icmp ne i32 [[TMP83]], 0
+// CHECK13-NEXT: br i1 [[TMP84]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
// CHECK13: omp_offload.failed20:
// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l137(ptr [[A]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT21]]
// CHECK13: omp_offload.cont21:
-// CHECK13-NEXT: [[TMP69:%.*]] = load i32, ptr [[M]], align 4
-// CHECK13-NEXT: store i32 [[TMP69]], ptr [[M_CASTED22]], align 4
-// CHECK13-NEXT: [[TMP70:%.*]] = load i64, ptr [[M_CASTED22]], align 8
-// CHECK13-NEXT: [[TMP71:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP70]], ptr [[TMP71]], align 8
-// CHECK13-NEXT: [[TMP72:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
-// CHECK13-NEXT: store i64 [[TMP70]], ptr [[TMP72]], align 8
-// CHECK13-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i64 0, i64 0
-// CHECK13-NEXT: store ptr null, ptr [[TMP73]], align 8
-// CHECK13-NEXT: [[TMP74:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 1
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP74]], align 8
-// CHECK13-NEXT: [[TMP75:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 1
-// CHECK13-NEXT: store ptr [[A]], ptr [[TMP75]], align 8
-// CHECK13-NEXT: [[TMP76:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i64 0, i64 1
-// CHECK13-NEXT: store ptr null, ptr [[TMP76]], align 8
-// CHECK13-NEXT: [[TMP77:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
-// CHECK13-NEXT: [[TMP78:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP85:%.*]] = load i32, ptr [[M]], align 4
+// CHECK13-NEXT: store i32 [[TMP85]], ptr [[M_CASTED22]], align 4
+// CHECK13-NEXT: [[TMP86:%.*]] = load i64, ptr [[M_CASTED22]], align 8
+// CHECK13-NEXT: [[TMP87:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP86]], ptr [[TMP87]], align 8
+// CHECK13-NEXT: [[TMP88:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
+// CHECK13-NEXT: store i64 [[TMP86]], ptr [[TMP88]], align 8
+// CHECK13-NEXT: [[TMP89:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i64 0, i64 0
+// CHECK13-NEXT: store ptr null, ptr [[TMP89]], align 8
+// CHECK13-NEXT: [[TMP90:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 1
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP90]], align 8
+// CHECK13-NEXT: [[TMP91:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 1
+// CHECK13-NEXT: store ptr [[A]], ptr [[TMP91]], align 8
+// CHECK13-NEXT: [[TMP92:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i64 0, i64 1
+// CHECK13-NEXT: store ptr null, ptr [[TMP92]], align 8
+// CHECK13-NEXT: [[TMP93:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
+// CHECK13-NEXT: [[TMP94:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
// CHECK13-NEXT: [[KERNEL_ARGS27:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK13-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 0
-// CHECK13-NEXT: store i32 1, ptr [[TMP79]], align 4
-// CHECK13-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 1
-// CHECK13-NEXT: store i32 2, ptr [[TMP80]], align 4
-// CHECK13-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 2
-// CHECK13-NEXT: store ptr [[TMP77]], ptr [[TMP81]], align 8
-// CHECK13-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 3
-// CHECK13-NEXT: store ptr [[TMP78]], ptr [[TMP82]], align 8
-// CHECK13-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 4
-// CHECK13-NEXT: store ptr @.offload_sizes.36, ptr [[TMP83]], align 8
-// CHECK13-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 5
-// CHECK13-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP84]], align 8
-// CHECK13-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 6
-// CHECK13-NEXT: store ptr null, ptr [[TMP85]], align 8
-// CHECK13-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 7
-// CHECK13-NEXT: store ptr null, ptr [[TMP86]], align 8
-// CHECK13-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 8
-// CHECK13-NEXT: store i64 10, ptr [[TMP87]], align 8
-// CHECK13-NEXT: [[TMP88:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l142.region_id, ptr [[KERNEL_ARGS27]])
-// CHECK13-NEXT: [[TMP89:%.*]] = icmp ne i32 [[TMP88]], 0
-// CHECK13-NEXT: br i1 [[TMP89]], label [[OMP_OFFLOAD_FAILED28:%.*]], label [[OMP_OFFLOAD_CONT29:%.*]]
+// CHECK13-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 0
+// CHECK13-NEXT: store i32 2, ptr [[TMP95]], align 4
+// CHECK13-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 1
+// CHECK13-NEXT: store i32 2, ptr [[TMP96]], align 4
+// CHECK13-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 2
+// CHECK13-NEXT: store ptr [[TMP93]], ptr [[TMP97]], align 8
+// CHECK13-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 3
+// CHECK13-NEXT: store ptr [[TMP94]], ptr [[TMP98]], align 8
+// CHECK13-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 4
+// CHECK13-NEXT: store ptr @.offload_sizes.36, ptr [[TMP99]], align 8
+// CHECK13-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 5
+// CHECK13-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP100]], align 8
+// CHECK13-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 6
+// CHECK13-NEXT: store ptr null, ptr [[TMP101]], align 8
+// CHECK13-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 7
+// CHECK13-NEXT: store ptr null, ptr [[TMP102]], align 8
+// CHECK13-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 8
+// CHECK13-NEXT: store i64 10, ptr [[TMP103]], align 8
+// CHECK13-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 9
+// CHECK13-NEXT: store i64 0, ptr [[TMP104]], align 8
+// CHECK13-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 10
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP105]], align 4
+// CHECK13-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 11
+// CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP106]], align 4
+// CHECK13-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 12
+// CHECK13-NEXT: store i32 0, ptr [[TMP107]], align 4
+// CHECK13-NEXT: [[TMP108:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l142.region_id, ptr [[KERNEL_ARGS27]])
+// CHECK13-NEXT: [[TMP109:%.*]] = icmp ne i32 [[TMP108]], 0
+// CHECK13-NEXT: br i1 [[TMP109]], label [[OMP_OFFLOAD_FAILED28:%.*]], label [[OMP_OFFLOAD_CONT29:%.*]]
// CHECK13: omp_offload.failed28:
-// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l142(i64 [[TMP70]], ptr [[A]]) #[[ATTR3]]
+// CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l142(i64 [[TMP86]], ptr [[A]]) #[[ATTR3]]
// CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT29]]
// CHECK13: omp_offload.cont29:
// CHECK13-NEXT: ret i32 0
@@ -7646,7 +7886,7 @@ int main (int argc, char **argv) {
// CHECK14-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK14-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK14-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK14-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK14-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK14-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK14-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK14-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -7663,294 +7903,334 @@ int main (int argc, char **argv) {
// CHECK14-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK14-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK14-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK14-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l154.region_id, ptr [[KERNEL_ARGS]])
-// CHECK14-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK14-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK14-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK14-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK14-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK14-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK14-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK14-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK14-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK14-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK14-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l154.region_id, ptr [[KERNEL_ARGS]])
+// CHECK14-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK14-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK14: omp_offload.failed:
// CHECK14-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l154(i64 [[TMP4]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK14-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK14: omp_offload.cont:
-// CHECK14-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK14-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK14-NEXT: [[TMP35:%.*]] = load i64, ptr [[N_CASTED3]], align 8
-// CHECK14-NEXT: [[TMP36:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK14-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK14-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK14-NEXT: [[TMP39:%.*]] = load i64, ptr [[N_CASTED3]], align 8
+// CHECK14-NEXT: [[TMP40:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK14-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES7]], ptr align 8 @.offload_sizes.4, i64 24, i1 false)
-// CHECK14-NEXT: [[TMP37:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK14-NEXT: store i64 [[TMP35]], ptr [[TMP37]], align 8
-// CHECK14-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK14-NEXT: store i64 [[TMP35]], ptr [[TMP38]], align 8
-// CHECK14-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
-// CHECK14-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK14-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK14-NEXT: store i64 [[TMP1]], ptr [[TMP40]], align 8
-// CHECK14-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK14-NEXT: store i64 [[TMP1]], ptr [[TMP41]], align 8
-// CHECK14-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
-// CHECK14-NEXT: store ptr null, ptr [[TMP42]], align 8
-// CHECK14-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK14-NEXT: store ptr [[VLA]], ptr [[TMP43]], align 8
-// CHECK14-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK14-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 8
-// CHECK14-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK14-NEXT: store i64 [[TMP36]], ptr [[TMP45]], align 8
-// CHECK14-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK14-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK14-NEXT: store i64 [[TMP39]], ptr [[TMP41]], align 8
+// CHECK14-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK14-NEXT: store i64 [[TMP39]], ptr [[TMP42]], align 8
+// CHECK14-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
+// CHECK14-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK14-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK14-NEXT: store i64 [[TMP1]], ptr [[TMP44]], align 8
+// CHECK14-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK14-NEXT: store i64 [[TMP1]], ptr [[TMP45]], align 8
+// CHECK14-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
// CHECK14-NEXT: store ptr null, ptr [[TMP46]], align 8
-// CHECK14-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK14-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK14-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK14-NEXT: [[TMP50:%.*]] = load i32, ptr [[N]], align 4
-// CHECK14-NEXT: store i32 [[TMP50]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK14-NEXT: [[TMP51:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK14-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP51]], 0
+// CHECK14-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK14-NEXT: store ptr [[VLA]], ptr [[TMP47]], align 8
+// CHECK14-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK14-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 8
+// CHECK14-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK14-NEXT: store i64 [[TMP40]], ptr [[TMP49]], align 8
+// CHECK14-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK14-NEXT: store ptr null, ptr [[TMP50]], align 8
+// CHECK14-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK14-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK14-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK14-NEXT: [[TMP54:%.*]] = load i32, ptr [[N]], align 4
+// CHECK14-NEXT: store i32 [[TMP54]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK14-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK14-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP55]], 0
// CHECK14-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK14-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK14-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK14-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK14-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP52]], 1
-// CHECK14-NEXT: [[TMP53:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK14-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK14-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP56]], 1
+// CHECK14-NEXT: [[TMP57:%.*]] = zext i32 [[ADD14]] to i64
// CHECK14-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK14-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK14-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK14-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK14-NEXT: store i32 3, ptr [[TMP55]], align 4
-// CHECK14-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK14-NEXT: store ptr [[TMP47]], ptr [[TMP56]], align 8
-// CHECK14-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK14-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 8
-// CHECK14-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK14-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 8
-// CHECK14-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK14-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP59]], align 8
-// CHECK14-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK14-NEXT: store ptr null, ptr [[TMP60]], align 8
-// CHECK14-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK14-NEXT: store ptr null, ptr [[TMP61]], align 8
-// CHECK14-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK14-NEXT: store i64 [[TMP53]], ptr [[TMP62]], align 8
-// CHECK14-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l159.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK14-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
-// CHECK14-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK14-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK14-NEXT: store i32 2, ptr [[TMP58]], align 4
+// CHECK14-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK14-NEXT: store i32 3, ptr [[TMP59]], align 4
+// CHECK14-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK14-NEXT: store ptr [[TMP51]], ptr [[TMP60]], align 8
+// CHECK14-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK14-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 8
+// CHECK14-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK14-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 8
+// CHECK14-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK14-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP63]], align 8
+// CHECK14-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK14-NEXT: store ptr null, ptr [[TMP64]], align 8
+// CHECK14-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK14-NEXT: store ptr null, ptr [[TMP65]], align 8
+// CHECK14-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK14-NEXT: store i64 [[TMP57]], ptr [[TMP66]], align 8
+// CHECK14-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK14-NEXT: store i64 0, ptr [[TMP67]], align 8
+// CHECK14-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK14-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP68]], align 4
+// CHECK14-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK14-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK14-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK14-NEXT: store i32 0, ptr [[TMP70]], align 4
+// CHECK14-NEXT: [[TMP71:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l159.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK14-NEXT: [[TMP72:%.*]] = icmp ne i32 [[TMP71]], 0
+// CHECK14-NEXT: br i1 [[TMP72]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK14: omp_offload.failed16:
-// CHECK14-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l159(i64 [[TMP35]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK14-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l159(i64 [[TMP39]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK14-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK14: omp_offload.cont17:
-// CHECK14-NEXT: [[TMP65:%.*]] = load i32, ptr [[M]], align 4
-// CHECK14-NEXT: store i32 [[TMP65]], ptr [[M_CASTED]], align 4
-// CHECK14-NEXT: [[TMP66:%.*]] = load i64, ptr [[M_CASTED]], align 8
-// CHECK14-NEXT: [[TMP67:%.*]] = load i32, ptr [[N]], align 4
-// CHECK14-NEXT: store i32 [[TMP67]], ptr [[N_CASTED18]], align 4
-// CHECK14-NEXT: [[TMP68:%.*]] = load i64, ptr [[N_CASTED18]], align 8
-// CHECK14-NEXT: [[TMP69:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK14-NEXT: [[TMP73:%.*]] = load i32, ptr [[M]], align 4
+// CHECK14-NEXT: store i32 [[TMP73]], ptr [[M_CASTED]], align 4
+// CHECK14-NEXT: [[TMP74:%.*]] = load i64, ptr [[M_CASTED]], align 8
+// CHECK14-NEXT: [[TMP75:%.*]] = load i32, ptr [[N]], align 4
+// CHECK14-NEXT: store i32 [[TMP75]], ptr [[N_CASTED18]], align 4
+// CHECK14-NEXT: [[TMP76:%.*]] = load i64, ptr [[N_CASTED18]], align 8
+// CHECK14-NEXT: [[TMP77:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK14-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES22]], ptr align 8 @.offload_sizes.8, i64 32, i1 false)
-// CHECK14-NEXT: [[TMP70:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK14-NEXT: store i64 [[TMP66]], ptr [[TMP70]], align 8
-// CHECK14-NEXT: [[TMP71:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK14-NEXT: store i64 [[TMP66]], ptr [[TMP71]], align 8
-// CHECK14-NEXT: [[TMP72:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
-// CHECK14-NEXT: store ptr null, ptr [[TMP72]], align 8
-// CHECK14-NEXT: [[TMP73:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
-// CHECK14-NEXT: store i64 [[TMP68]], ptr [[TMP73]], align 8
-// CHECK14-NEXT: [[TMP74:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
-// CHECK14-NEXT: store i64 [[TMP68]], ptr [[TMP74]], align 8
-// CHECK14-NEXT: [[TMP75:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 1
-// CHECK14-NEXT: store ptr null, ptr [[TMP75]], align 8
-// CHECK14-NEXT: [[TMP76:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
-// CHECK14-NEXT: store i64 [[TMP1]], ptr [[TMP76]], align 8
-// CHECK14-NEXT: [[TMP77:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
-// CHECK14-NEXT: store i64 [[TMP1]], ptr [[TMP77]], align 8
-// CHECK14-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 2
-// CHECK14-NEXT: store ptr null, ptr [[TMP78]], align 8
-// CHECK14-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 3
-// CHECK14-NEXT: store ptr [[VLA]], ptr [[TMP79]], align 8
-// CHECK14-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 3
-// CHECK14-NEXT: store ptr [[VLA]], ptr [[TMP80]], align 8
-// CHECK14-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 3
-// CHECK14-NEXT: store i64 [[TMP69]], ptr [[TMP81]], align 8
-// CHECK14-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 3
-// CHECK14-NEXT: store ptr null, ptr [[TMP82]], align 8
-// CHECK14-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK14-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK14-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
-// CHECK14-NEXT: [[TMP86:%.*]] = load i32, ptr [[N]], align 4
-// CHECK14-NEXT: store i32 [[TMP86]], ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK14-NEXT: [[TMP87:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK14-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP87]], 0
+// CHECK14-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK14-NEXT: store i64 [[TMP74]], ptr [[TMP78]], align 8
+// CHECK14-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK14-NEXT: store i64 [[TMP74]], ptr [[TMP79]], align 8
+// CHECK14-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
+// CHECK14-NEXT: store ptr null, ptr [[TMP80]], align 8
+// CHECK14-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
+// CHECK14-NEXT: store i64 [[TMP76]], ptr [[TMP81]], align 8
+// CHECK14-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
+// CHECK14-NEXT: store i64 [[TMP76]], ptr [[TMP82]], align 8
+// CHECK14-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 1
+// CHECK14-NEXT: store ptr null, ptr [[TMP83]], align 8
+// CHECK14-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
+// CHECK14-NEXT: store i64 [[TMP1]], ptr [[TMP84]], align 8
+// CHECK14-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
+// CHECK14-NEXT: store i64 [[TMP1]], ptr [[TMP85]], align 8
+// CHECK14-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 2
+// CHECK14-NEXT: store ptr null, ptr [[TMP86]], align 8
+// CHECK14-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 3
+// CHECK14-NEXT: store ptr [[VLA]], ptr [[TMP87]], align 8
+// CHECK14-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 3
+// CHECK14-NEXT: store ptr [[VLA]], ptr [[TMP88]], align 8
+// CHECK14-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 3
+// CHECK14-NEXT: store i64 [[TMP77]], ptr [[TMP89]], align 8
+// CHECK14-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 3
+// CHECK14-NEXT: store ptr null, ptr [[TMP90]], align 8
+// CHECK14-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK14-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK14-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
+// CHECK14-NEXT: [[TMP94:%.*]] = load i32, ptr [[N]], align 4
+// CHECK14-NEXT: store i32 [[TMP94]], ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK14-NEXT: [[TMP95:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK14-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP95]], 0
// CHECK14-NEXT: [[DIV27:%.*]] = sdiv i32 [[SUB26]], 1
// CHECK14-NEXT: [[SUB28:%.*]] = sub nsw i32 [[DIV27]], 1
// CHECK14-NEXT: store i32 [[SUB28]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK14-NEXT: [[TMP88:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK14-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP88]], 1
-// CHECK14-NEXT: [[TMP89:%.*]] = zext i32 [[ADD29]] to i64
+// CHECK14-NEXT: [[TMP96:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK14-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP96]], 1
+// CHECK14-NEXT: [[TMP97:%.*]] = zext i32 [[ADD29]] to i64
// CHECK14-NEXT: [[KERNEL_ARGS30:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK14-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
-// CHECK14-NEXT: store i32 1, ptr [[TMP90]], align 4
-// CHECK14-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
-// CHECK14-NEXT: store i32 4, ptr [[TMP91]], align 4
-// CHECK14-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
-// CHECK14-NEXT: store ptr [[TMP83]], ptr [[TMP92]], align 8
-// CHECK14-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
-// CHECK14-NEXT: store ptr [[TMP84]], ptr [[TMP93]], align 8
-// CHECK14-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
-// CHECK14-NEXT: store ptr [[TMP85]], ptr [[TMP94]], align 8
-// CHECK14-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
-// CHECK14-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP95]], align 8
-// CHECK14-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
-// CHECK14-NEXT: store ptr null, ptr [[TMP96]], align 8
-// CHECK14-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
-// CHECK14-NEXT: store ptr null, ptr [[TMP97]], align 8
-// CHECK14-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
-// CHECK14-NEXT: store i64 [[TMP89]], ptr [[TMP98]], align 8
-// CHECK14-NEXT: [[TMP99:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l164.region_id, ptr [[KERNEL_ARGS30]])
-// CHECK14-NEXT: [[TMP100:%.*]] = icmp ne i32 [[TMP99]], 0
-// CHECK14-NEXT: br i1 [[TMP100]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
+// CHECK14-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
+// CHECK14-NEXT: store i32 2, ptr [[TMP98]], align 4
+// CHECK14-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
+// CHECK14-NEXT: store i32 4, ptr [[TMP99]], align 4
+// CHECK14-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
+// CHECK14-NEXT: store ptr [[TMP91]], ptr [[TMP100]], align 8
+// CHECK14-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
+// CHECK14-NEXT: store ptr [[TMP92]], ptr [[TMP101]], align 8
+// CHECK14-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
+// CHECK14-NEXT: store ptr [[TMP93]], ptr [[TMP102]], align 8
+// CHECK14-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
+// CHECK14-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP103]], align 8
+// CHECK14-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
+// CHECK14-NEXT: store ptr null, ptr [[TMP104]], align 8
+// CHECK14-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
+// CHECK14-NEXT: store ptr null, ptr [[TMP105]], align 8
+// CHECK14-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
+// CHECK14-NEXT: store i64 [[TMP97]], ptr [[TMP106]], align 8
+// CHECK14-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 9
+// CHECK14-NEXT: store i64 0, ptr [[TMP107]], align 8
+// CHECK14-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 10
+// CHECK14-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP108]], align 4
+// CHECK14-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 11
+// CHECK14-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP109]], align 4
+// CHECK14-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 12
+// CHECK14-NEXT: store i32 0, ptr [[TMP110]], align 4
+// CHECK14-NEXT: [[TMP111:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l164.region_id, ptr [[KERNEL_ARGS30]])
+// CHECK14-NEXT: [[TMP112:%.*]] = icmp ne i32 [[TMP111]], 0
+// CHECK14-NEXT: br i1 [[TMP112]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
// CHECK14: omp_offload.failed31:
-// CHECK14-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l164(i64 [[TMP66]], i64 [[TMP68]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK14-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l164(i64 [[TMP74]], i64 [[TMP76]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK14-NEXT: br label [[OMP_OFFLOAD_CONT32]]
// CHECK14: omp_offload.cont32:
-// CHECK14-NEXT: [[TMP101:%.*]] = load i32, ptr [[N]], align 4
-// CHECK14-NEXT: store i32 [[TMP101]], ptr [[N_CASTED33]], align 4
-// CHECK14-NEXT: [[TMP102:%.*]] = load i64, ptr [[N_CASTED33]], align 8
-// CHECK14-NEXT: [[TMP103:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK14-NEXT: [[TMP113:%.*]] = load i32, ptr [[N]], align 4
+// CHECK14-NEXT: store i32 [[TMP113]], ptr [[N_CASTED33]], align 4
+// CHECK14-NEXT: [[TMP114:%.*]] = load i64, ptr [[N_CASTED33]], align 8
+// CHECK14-NEXT: [[TMP115:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK14-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES37]], ptr align 8 @.offload_sizes.12, i64 24, i1 false)
-// CHECK14-NEXT: [[TMP104:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
-// CHECK14-NEXT: store i64 [[TMP102]], ptr [[TMP104]], align 8
-// CHECK14-NEXT: [[TMP105:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
-// CHECK14-NEXT: store i64 [[TMP102]], ptr [[TMP105]], align 8
-// CHECK14-NEXT: [[TMP106:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i64 0, i64 0
-// CHECK14-NEXT: store ptr null, ptr [[TMP106]], align 8
-// CHECK14-NEXT: [[TMP107:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 1
-// CHECK14-NEXT: store i64 [[TMP1]], ptr [[TMP107]], align 8
-// CHECK14-NEXT: [[TMP108:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 1
-// CHECK14-NEXT: store i64 [[TMP1]], ptr [[TMP108]], align 8
-// CHECK14-NEXT: [[TMP109:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i64 0, i64 1
-// CHECK14-NEXT: store ptr null, ptr [[TMP109]], align 8
-// CHECK14-NEXT: [[TMP110:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 2
-// CHECK14-NEXT: store ptr [[VLA]], ptr [[TMP110]], align 8
-// CHECK14-NEXT: [[TMP111:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 2
-// CHECK14-NEXT: store ptr [[VLA]], ptr [[TMP111]], align 8
-// CHECK14-NEXT: [[TMP112:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 2
-// CHECK14-NEXT: store i64 [[TMP103]], ptr [[TMP112]], align 8
-// CHECK14-NEXT: [[TMP113:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i64 0, i64 2
-// CHECK14-NEXT: store ptr null, ptr [[TMP113]], align 8
-// CHECK14-NEXT: [[TMP114:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
-// CHECK14-NEXT: [[TMP115:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
-// CHECK14-NEXT: [[TMP116:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 0
-// CHECK14-NEXT: [[TMP117:%.*]] = load i32, ptr [[N]], align 4
-// CHECK14-NEXT: store i32 [[TMP117]], ptr [[DOTCAPTURE_EXPR_39]], align 4
-// CHECK14-NEXT: [[TMP118:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_39]], align 4
-// CHECK14-NEXT: [[SUB41:%.*]] = sub nsw i32 [[TMP118]], 0
+// CHECK14-NEXT: [[TMP116:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
+// CHECK14-NEXT: store i64 [[TMP114]], ptr [[TMP116]], align 8
+// CHECK14-NEXT: [[TMP117:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
+// CHECK14-NEXT: store i64 [[TMP114]], ptr [[TMP117]], align 8
+// CHECK14-NEXT: [[TMP118:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i64 0, i64 0
+// CHECK14-NEXT: store ptr null, ptr [[TMP118]], align 8
+// CHECK14-NEXT: [[TMP119:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 1
+// CHECK14-NEXT: store i64 [[TMP1]], ptr [[TMP119]], align 8
+// CHECK14-NEXT: [[TMP120:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 1
+// CHECK14-NEXT: store i64 [[TMP1]], ptr [[TMP120]], align 8
+// CHECK14-NEXT: [[TMP121:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i64 0, i64 1
+// CHECK14-NEXT: store ptr null, ptr [[TMP121]], align 8
+// CHECK14-NEXT: [[TMP122:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 2
+// CHECK14-NEXT: store ptr [[VLA]], ptr [[TMP122]], align 8
+// CHECK14-NEXT: [[TMP123:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 2
+// CHECK14-NEXT: store ptr [[VLA]], ptr [[TMP123]], align 8
+// CHECK14-NEXT: [[TMP124:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 2
+// CHECK14-NEXT: store i64 [[TMP115]], ptr [[TMP124]], align 8
+// CHECK14-NEXT: [[TMP125:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i64 0, i64 2
+// CHECK14-NEXT: store ptr null, ptr [[TMP125]], align 8
+// CHECK14-NEXT: [[TMP126:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
+// CHECK14-NEXT: [[TMP127:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
+// CHECK14-NEXT: [[TMP128:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 0
+// CHECK14-NEXT: [[TMP129:%.*]] = load i32, ptr [[N]], align 4
+// CHECK14-NEXT: store i32 [[TMP129]], ptr [[DOTCAPTURE_EXPR_39]], align 4
+// CHECK14-NEXT: [[TMP130:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_39]], align 4
+// CHECK14-NEXT: [[SUB41:%.*]] = sub nsw i32 [[TMP130]], 0
// CHECK14-NEXT: [[DIV42:%.*]] = sdiv i32 [[SUB41]], 1
// CHECK14-NEXT: [[SUB43:%.*]] = sub nsw i32 [[DIV42]], 1
// CHECK14-NEXT: store i32 [[SUB43]], ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK14-NEXT: [[TMP119:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK14-NEXT: [[ADD44:%.*]] = add nsw i32 [[TMP119]], 1
-// CHECK14-NEXT: [[TMP120:%.*]] = zext i32 [[ADD44]] to i64
+// CHECK14-NEXT: [[TMP131:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
+// CHECK14-NEXT: [[ADD44:%.*]] = add nsw i32 [[TMP131]], 1
+// CHECK14-NEXT: [[TMP132:%.*]] = zext i32 [[ADD44]] to i64
// CHECK14-NEXT: [[KERNEL_ARGS45:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK14-NEXT: [[TMP121:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 0
-// CHECK14-NEXT: store i32 1, ptr [[TMP121]], align 4
-// CHECK14-NEXT: [[TMP122:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 1
-// CHECK14-NEXT: store i32 3, ptr [[TMP122]], align 4
-// CHECK14-NEXT: [[TMP123:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 2
-// CHECK14-NEXT: store ptr [[TMP114]], ptr [[TMP123]], align 8
-// CHECK14-NEXT: [[TMP124:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 3
-// CHECK14-NEXT: store ptr [[TMP115]], ptr [[TMP124]], align 8
-// CHECK14-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 4
-// CHECK14-NEXT: store ptr [[TMP116]], ptr [[TMP125]], align 8
-// CHECK14-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 5
-// CHECK14-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP126]], align 8
-// CHECK14-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 6
-// CHECK14-NEXT: store ptr null, ptr [[TMP127]], align 8
-// CHECK14-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 7
-// CHECK14-NEXT: store ptr null, ptr [[TMP128]], align 8
-// CHECK14-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 8
-// CHECK14-NEXT: store i64 [[TMP120]], ptr [[TMP129]], align 8
-// CHECK14-NEXT: [[TMP130:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l169.region_id, ptr [[KERNEL_ARGS45]])
-// CHECK14-NEXT: [[TMP131:%.*]] = icmp ne i32 [[TMP130]], 0
-// CHECK14-NEXT: br i1 [[TMP131]], label [[OMP_OFFLOAD_FAILED46:%.*]], label [[OMP_OFFLOAD_CONT47:%.*]]
+// CHECK14-NEXT: [[TMP133:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 0
+// CHECK14-NEXT: store i32 2, ptr [[TMP133]], align 4
+// CHECK14-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 1
+// CHECK14-NEXT: store i32 3, ptr [[TMP134]], align 4
+// CHECK14-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 2
+// CHECK14-NEXT: store ptr [[TMP126]], ptr [[TMP135]], align 8
+// CHECK14-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 3
+// CHECK14-NEXT: store ptr [[TMP127]], ptr [[TMP136]], align 8
+// CHECK14-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 4
+// CHECK14-NEXT: store ptr [[TMP128]], ptr [[TMP137]], align 8
+// CHECK14-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 5
+// CHECK14-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP138]], align 8
+// CHECK14-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 6
+// CHECK14-NEXT: store ptr null, ptr [[TMP139]], align 8
+// CHECK14-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 7
+// CHECK14-NEXT: store ptr null, ptr [[TMP140]], align 8
+// CHECK14-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 8
+// CHECK14-NEXT: store i64 [[TMP132]], ptr [[TMP141]], align 8
+// CHECK14-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 9
+// CHECK14-NEXT: store i64 0, ptr [[TMP142]], align 8
+// CHECK14-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 10
+// CHECK14-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP143]], align 4
+// CHECK14-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 11
+// CHECK14-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP144]], align 4
+// CHECK14-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 12
+// CHECK14-NEXT: store i32 0, ptr [[TMP145]], align 4
+// CHECK14-NEXT: [[TMP146:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l169.region_id, ptr [[KERNEL_ARGS45]])
+// CHECK14-NEXT: [[TMP147:%.*]] = icmp ne i32 [[TMP146]], 0
+// CHECK14-NEXT: br i1 [[TMP147]], label [[OMP_OFFLOAD_FAILED46:%.*]], label [[OMP_OFFLOAD_CONT47:%.*]]
// CHECK14: omp_offload.failed46:
-// CHECK14-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l169(i64 [[TMP102]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK14-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l169(i64 [[TMP114]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK14-NEXT: br label [[OMP_OFFLOAD_CONT47]]
// CHECK14: omp_offload.cont47:
-// CHECK14-NEXT: [[TMP132:%.*]] = load i32, ptr [[M]], align 4
-// CHECK14-NEXT: store i32 [[TMP132]], ptr [[M_CASTED48]], align 4
-// CHECK14-NEXT: [[TMP133:%.*]] = load i64, ptr [[M_CASTED48]], align 8
-// CHECK14-NEXT: [[TMP134:%.*]] = load i32, ptr [[N]], align 4
-// CHECK14-NEXT: store i32 [[TMP134]], ptr [[N_CASTED49]], align 4
-// CHECK14-NEXT: [[TMP135:%.*]] = load i64, ptr [[N_CASTED49]], align 8
-// CHECK14-NEXT: [[TMP136:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK14-NEXT: [[TMP148:%.*]] = load i32, ptr [[M]], align 4
+// CHECK14-NEXT: store i32 [[TMP148]], ptr [[M_CASTED48]], align 4
+// CHECK14-NEXT: [[TMP149:%.*]] = load i64, ptr [[M_CASTED48]], align 8
+// CHECK14-NEXT: [[TMP150:%.*]] = load i32, ptr [[N]], align 4
+// CHECK14-NEXT: store i32 [[TMP150]], ptr [[N_CASTED49]], align 4
+// CHECK14-NEXT: [[TMP151:%.*]] = load i64, ptr [[N_CASTED49]], align 8
+// CHECK14-NEXT: [[TMP152:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK14-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES53]], ptr align 8 @.offload_sizes.16, i64 32, i1 false)
-// CHECK14-NEXT: [[TMP137:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
-// CHECK14-NEXT: store i64 [[TMP133]], ptr [[TMP137]], align 8
-// CHECK14-NEXT: [[TMP138:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
-// CHECK14-NEXT: store i64 [[TMP133]], ptr [[TMP138]], align 8
-// CHECK14-NEXT: [[TMP139:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 0
-// CHECK14-NEXT: store ptr null, ptr [[TMP139]], align 8
-// CHECK14-NEXT: [[TMP140:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 1
-// CHECK14-NEXT: store i64 [[TMP135]], ptr [[TMP140]], align 8
-// CHECK14-NEXT: [[TMP141:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 1
-// CHECK14-NEXT: store i64 [[TMP135]], ptr [[TMP141]], align 8
-// CHECK14-NEXT: [[TMP142:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 1
-// CHECK14-NEXT: store ptr null, ptr [[TMP142]], align 8
-// CHECK14-NEXT: [[TMP143:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 2
-// CHECK14-NEXT: store i64 [[TMP1]], ptr [[TMP143]], align 8
-// CHECK14-NEXT: [[TMP144:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 2
-// CHECK14-NEXT: store i64 [[TMP1]], ptr [[TMP144]], align 8
-// CHECK14-NEXT: [[TMP145:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 2
-// CHECK14-NEXT: store ptr null, ptr [[TMP145]], align 8
-// CHECK14-NEXT: [[TMP146:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 3
-// CHECK14-NEXT: store ptr [[VLA]], ptr [[TMP146]], align 8
-// CHECK14-NEXT: [[TMP147:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 3
-// CHECK14-NEXT: store ptr [[VLA]], ptr [[TMP147]], align 8
-// CHECK14-NEXT: [[TMP148:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 3
-// CHECK14-NEXT: store i64 [[TMP136]], ptr [[TMP148]], align 8
-// CHECK14-NEXT: [[TMP149:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 3
-// CHECK14-NEXT: store ptr null, ptr [[TMP149]], align 8
-// CHECK14-NEXT: [[TMP150:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
-// CHECK14-NEXT: [[TMP151:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
-// CHECK14-NEXT: [[TMP152:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 0
-// CHECK14-NEXT: [[TMP153:%.*]] = load i32, ptr [[N]], align 4
-// CHECK14-NEXT: store i32 [[TMP153]], ptr [[DOTCAPTURE_EXPR_55]], align 4
-// CHECK14-NEXT: [[TMP154:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_55]], align 4
-// CHECK14-NEXT: [[SUB57:%.*]] = sub nsw i32 [[TMP154]], 0
+// CHECK14-NEXT: [[TMP153:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
+// CHECK14-NEXT: store i64 [[TMP149]], ptr [[TMP153]], align 8
+// CHECK14-NEXT: [[TMP154:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
+// CHECK14-NEXT: store i64 [[TMP149]], ptr [[TMP154]], align 8
+// CHECK14-NEXT: [[TMP155:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 0
+// CHECK14-NEXT: store ptr null, ptr [[TMP155]], align 8
+// CHECK14-NEXT: [[TMP156:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 1
+// CHECK14-NEXT: store i64 [[TMP151]], ptr [[TMP156]], align 8
+// CHECK14-NEXT: [[TMP157:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 1
+// CHECK14-NEXT: store i64 [[TMP151]], ptr [[TMP157]], align 8
+// CHECK14-NEXT: [[TMP158:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 1
+// CHECK14-NEXT: store ptr null, ptr [[TMP158]], align 8
+// CHECK14-NEXT: [[TMP159:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 2
+// CHECK14-NEXT: store i64 [[TMP1]], ptr [[TMP159]], align 8
+// CHECK14-NEXT: [[TMP160:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 2
+// CHECK14-NEXT: store i64 [[TMP1]], ptr [[TMP160]], align 8
+// CHECK14-NEXT: [[TMP161:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 2
+// CHECK14-NEXT: store ptr null, ptr [[TMP161]], align 8
+// CHECK14-NEXT: [[TMP162:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 3
+// CHECK14-NEXT: store ptr [[VLA]], ptr [[TMP162]], align 8
+// CHECK14-NEXT: [[TMP163:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 3
+// CHECK14-NEXT: store ptr [[VLA]], ptr [[TMP163]], align 8
+// CHECK14-NEXT: [[TMP164:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 3
+// CHECK14-NEXT: store i64 [[TMP152]], ptr [[TMP164]], align 8
+// CHECK14-NEXT: [[TMP165:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i64 0, i64 3
+// CHECK14-NEXT: store ptr null, ptr [[TMP165]], align 8
+// CHECK14-NEXT: [[TMP166:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
+// CHECK14-NEXT: [[TMP167:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
+// CHECK14-NEXT: [[TMP168:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 0
+// CHECK14-NEXT: [[TMP169:%.*]] = load i32, ptr [[N]], align 4
+// CHECK14-NEXT: store i32 [[TMP169]], ptr [[DOTCAPTURE_EXPR_55]], align 4
+// CHECK14-NEXT: [[TMP170:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_55]], align 4
+// CHECK14-NEXT: [[SUB57:%.*]] = sub nsw i32 [[TMP170]], 0
// CHECK14-NEXT: [[DIV58:%.*]] = sdiv i32 [[SUB57]], 1
// CHECK14-NEXT: [[SUB59:%.*]] = sub nsw i32 [[DIV58]], 1
// CHECK14-NEXT: store i32 [[SUB59]], ptr [[DOTCAPTURE_EXPR_56]], align 4
-// CHECK14-NEXT: [[TMP155:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_56]], align 4
-// CHECK14-NEXT: [[ADD60:%.*]] = add nsw i32 [[TMP155]], 1
-// CHECK14-NEXT: [[TMP156:%.*]] = zext i32 [[ADD60]] to i64
+// CHECK14-NEXT: [[TMP171:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_56]], align 4
+// CHECK14-NEXT: [[ADD60:%.*]] = add nsw i32 [[TMP171]], 1
+// CHECK14-NEXT: [[TMP172:%.*]] = zext i32 [[ADD60]] to i64
// CHECK14-NEXT: [[KERNEL_ARGS61:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK14-NEXT: [[TMP157:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 0
-// CHECK14-NEXT: store i32 1, ptr [[TMP157]], align 4
-// CHECK14-NEXT: [[TMP158:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 1
-// CHECK14-NEXT: store i32 4, ptr [[TMP158]], align 4
-// CHECK14-NEXT: [[TMP159:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 2
-// CHECK14-NEXT: store ptr [[TMP150]], ptr [[TMP159]], align 8
-// CHECK14-NEXT: [[TMP160:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 3
-// CHECK14-NEXT: store ptr [[TMP151]], ptr [[TMP160]], align 8
-// CHECK14-NEXT: [[TMP161:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 4
-// CHECK14-NEXT: store ptr [[TMP152]], ptr [[TMP161]], align 8
-// CHECK14-NEXT: [[TMP162:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 5
-// CHECK14-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP162]], align 8
-// CHECK14-NEXT: [[TMP163:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 6
-// CHECK14-NEXT: store ptr null, ptr [[TMP163]], align 8
-// CHECK14-NEXT: [[TMP164:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 7
-// CHECK14-NEXT: store ptr null, ptr [[TMP164]], align 8
-// CHECK14-NEXT: [[TMP165:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 8
-// CHECK14-NEXT: store i64 [[TMP156]], ptr [[TMP165]], align 8
-// CHECK14-NEXT: [[TMP166:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l174.region_id, ptr [[KERNEL_ARGS61]])
-// CHECK14-NEXT: [[TMP167:%.*]] = icmp ne i32 [[TMP166]], 0
-// CHECK14-NEXT: br i1 [[TMP167]], label [[OMP_OFFLOAD_FAILED62:%.*]], label [[OMP_OFFLOAD_CONT63:%.*]]
+// CHECK14-NEXT: [[TMP173:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 0
+// CHECK14-NEXT: store i32 2, ptr [[TMP173]], align 4
+// CHECK14-NEXT: [[TMP174:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 1
+// CHECK14-NEXT: store i32 4, ptr [[TMP174]], align 4
+// CHECK14-NEXT: [[TMP175:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 2
+// CHECK14-NEXT: store ptr [[TMP166]], ptr [[TMP175]], align 8
+// CHECK14-NEXT: [[TMP176:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 3
+// CHECK14-NEXT: store ptr [[TMP167]], ptr [[TMP176]], align 8
+// CHECK14-NEXT: [[TMP177:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 4
+// CHECK14-NEXT: store ptr [[TMP168]], ptr [[TMP177]], align 8
+// CHECK14-NEXT: [[TMP178:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 5
+// CHECK14-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP178]], align 8
+// CHECK14-NEXT: [[TMP179:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 6
+// CHECK14-NEXT: store ptr null, ptr [[TMP179]], align 8
+// CHECK14-NEXT: [[TMP180:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 7
+// CHECK14-NEXT: store ptr null, ptr [[TMP180]], align 8
+// CHECK14-NEXT: [[TMP181:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 8
+// CHECK14-NEXT: store i64 [[TMP172]], ptr [[TMP181]], align 8
+// CHECK14-NEXT: [[TMP182:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 9
+// CHECK14-NEXT: store i64 0, ptr [[TMP182]], align 8
+// CHECK14-NEXT: [[TMP183:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 10
+// CHECK14-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP183]], align 4
+// CHECK14-NEXT: [[TMP184:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 11
+// CHECK14-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP184]], align 4
+// CHECK14-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 12
+// CHECK14-NEXT: store i32 0, ptr [[TMP185]], align 4
+// CHECK14-NEXT: [[TMP186:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l174.region_id, ptr [[KERNEL_ARGS61]])
+// CHECK14-NEXT: [[TMP187:%.*]] = icmp ne i32 [[TMP186]], 0
+// CHECK14-NEXT: br i1 [[TMP187]], label [[OMP_OFFLOAD_FAILED62:%.*]], label [[OMP_OFFLOAD_CONT63:%.*]]
// CHECK14: omp_offload.failed62:
-// CHECK14-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l174(i64 [[TMP133]], i64 [[TMP135]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK14-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l174(i64 [[TMP149]], i64 [[TMP151]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK14-NEXT: br label [[OMP_OFFLOAD_CONT63]]
// CHECK14: omp_offload.cont63:
-// CHECK14-NEXT: [[TMP168:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK14-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP168]])
+// CHECK14-NEXT: [[TMP188:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK14-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP188]])
// CHECK14-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK14-NEXT: [[TMP169:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK14-NEXT: call void @llvm.stackrestore(ptr [[TMP169]])
-// CHECK14-NEXT: [[TMP170:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK14-NEXT: ret i32 [[TMP170]]
+// CHECK14-NEXT: [[TMP189:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK14-NEXT: call void @llvm.stackrestore(ptr [[TMP189]])
+// CHECK14-NEXT: [[TMP190:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK14-NEXT: ret i32 [[TMP190]]
//
//
// CHECK14-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l154
@@ -9220,7 +9500,7 @@ int main (int argc, char **argv) {
// CHECK14-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK14-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK14-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK14-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK14-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK14-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK14-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK14-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -9237,165 +9517,205 @@ int main (int argc, char **argv) {
// CHECK14-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK14-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK14-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK14-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l122.region_id, ptr [[KERNEL_ARGS]])
-// CHECK14-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK14-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK14-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK14-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK14-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK14-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK14-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK14-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK14-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK14-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK14-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l122.region_id, ptr [[KERNEL_ARGS]])
+// CHECK14-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK14-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK14: omp_offload.failed:
// CHECK14-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l122(ptr [[A]]) #[[ATTR3]]
// CHECK14-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK14: omp_offload.cont:
-// CHECK14-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK14-NEXT: store ptr [[A]], ptr [[TMP16]], align 8
-// CHECK14-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK14-NEXT: store ptr [[A]], ptr [[TMP17]], align 8
-// CHECK14-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
-// CHECK14-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK14-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK14-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK14-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK14-NEXT: store ptr [[A]], ptr [[TMP20]], align 8
+// CHECK14-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK14-NEXT: store ptr [[A]], ptr [[TMP21]], align 8
+// CHECK14-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
+// CHECK14-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK14-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK14-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK14-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK14-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK14-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK14-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK14-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK14-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK14-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK14-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK14-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK14-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK14-NEXT: store ptr @.offload_sizes.24, ptr [[TMP25]], align 8
-// CHECK14-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK14-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP26]], align 8
-// CHECK14-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK14-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK14-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK14-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK14-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK14-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK14-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l127.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK14-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK14-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK14-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK14-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK14-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK14-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK14-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK14-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK14-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK14-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK14-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK14-NEXT: store ptr @.offload_sizes.24, ptr [[TMP29]], align 8
+// CHECK14-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK14-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP30]], align 8
+// CHECK14-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK14-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK14-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK14-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK14-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK14-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK14-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK14-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK14-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK14-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK14-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK14-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK14-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK14-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK14-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l127.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK14-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK14-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK14: omp_offload.failed6:
// CHECK14-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l127(ptr [[A]]) #[[ATTR3]]
// CHECK14-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK14: omp_offload.cont7:
-// CHECK14-NEXT: [[TMP32:%.*]] = load i32, ptr [[M]], align 4
-// CHECK14-NEXT: store i32 [[TMP32]], ptr [[M_CASTED]], align 4
-// CHECK14-NEXT: [[TMP33:%.*]] = load i64, ptr [[M_CASTED]], align 8
-// CHECK14-NEXT: [[TMP34:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK14-NEXT: store i64 [[TMP33]], ptr [[TMP34]], align 8
-// CHECK14-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK14-NEXT: store i64 [[TMP33]], ptr [[TMP35]], align 8
-// CHECK14-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
-// CHECK14-NEXT: store ptr null, ptr [[TMP36]], align 8
-// CHECK14-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
-// CHECK14-NEXT: store ptr [[A]], ptr [[TMP37]], align 8
-// CHECK14-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
-// CHECK14-NEXT: store ptr [[A]], ptr [[TMP38]], align 8
-// CHECK14-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
-// CHECK14-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK14-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK14-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK14-NEXT: [[TMP40:%.*]] = load i32, ptr [[M]], align 4
+// CHECK14-NEXT: store i32 [[TMP40]], ptr [[M_CASTED]], align 4
+// CHECK14-NEXT: [[TMP41:%.*]] = load i64, ptr [[M_CASTED]], align 8
+// CHECK14-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK14-NEXT: store i64 [[TMP41]], ptr [[TMP42]], align 8
+// CHECK14-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK14-NEXT: store i64 [[TMP41]], ptr [[TMP43]], align 8
+// CHECK14-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
+// CHECK14-NEXT: store ptr null, ptr [[TMP44]], align 8
+// CHECK14-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
+// CHECK14-NEXT: store ptr [[A]], ptr [[TMP45]], align 8
+// CHECK14-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
+// CHECK14-NEXT: store ptr [[A]], ptr [[TMP46]], align 8
+// CHECK14-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
+// CHECK14-NEXT: store ptr null, ptr [[TMP47]], align 8
+// CHECK14-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK14-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK14-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK14-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK14-NEXT: store i32 1, ptr [[TMP42]], align 4
-// CHECK14-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK14-NEXT: store i32 2, ptr [[TMP43]], align 4
-// CHECK14-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK14-NEXT: store ptr [[TMP40]], ptr [[TMP44]], align 8
-// CHECK14-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK14-NEXT: store ptr [[TMP41]], ptr [[TMP45]], align 8
-// CHECK14-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK14-NEXT: store ptr @.offload_sizes.28, ptr [[TMP46]], align 8
-// CHECK14-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK14-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP47]], align 8
-// CHECK14-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK14-NEXT: store ptr null, ptr [[TMP48]], align 8
-// CHECK14-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK14-NEXT: store ptr null, ptr [[TMP49]], align 8
-// CHECK14-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK14-NEXT: store i64 10, ptr [[TMP50]], align 8
-// CHECK14-NEXT: [[TMP51:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l132.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK14-NEXT: [[TMP52:%.*]] = icmp ne i32 [[TMP51]], 0
-// CHECK14-NEXT: br i1 [[TMP52]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK14-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK14-NEXT: store i32 2, ptr [[TMP50]], align 4
+// CHECK14-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK14-NEXT: store i32 2, ptr [[TMP51]], align 4
+// CHECK14-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK14-NEXT: store ptr [[TMP48]], ptr [[TMP52]], align 8
+// CHECK14-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK14-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 8
+// CHECK14-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK14-NEXT: store ptr @.offload_sizes.28, ptr [[TMP54]], align 8
+// CHECK14-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK14-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP55]], align 8
+// CHECK14-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK14-NEXT: store ptr null, ptr [[TMP56]], align 8
+// CHECK14-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK14-NEXT: store ptr null, ptr [[TMP57]], align 8
+// CHECK14-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK14-NEXT: store i64 10, ptr [[TMP58]], align 8
+// CHECK14-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK14-NEXT: store i64 0, ptr [[TMP59]], align 8
+// CHECK14-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK14-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP60]], align 4
+// CHECK14-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK14-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4
+// CHECK14-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK14-NEXT: store i32 0, ptr [[TMP62]], align 4
+// CHECK14-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l132.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK14-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
+// CHECK14-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK14: omp_offload.failed13:
-// CHECK14-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l132(i64 [[TMP33]], ptr [[A]]) #[[ATTR3]]
+// CHECK14-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l132(i64 [[TMP41]], ptr [[A]]) #[[ATTR3]]
// CHECK14-NEXT: br label [[OMP_OFFLOAD_CONT14]]
// CHECK14: omp_offload.cont14:
-// CHECK14-NEXT: [[TMP53:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK14-NEXT: store ptr [[A]], ptr [[TMP53]], align 8
-// CHECK14-NEXT: [[TMP54:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
-// CHECK14-NEXT: store ptr [[A]], ptr [[TMP54]], align 8
-// CHECK14-NEXT: [[TMP55:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i64 0, i64 0
-// CHECK14-NEXT: store ptr null, ptr [[TMP55]], align 8
-// CHECK14-NEXT: [[TMP56:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK14-NEXT: [[TMP57:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK14-NEXT: [[TMP65:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK14-NEXT: store ptr [[A]], ptr [[TMP65]], align 8
+// CHECK14-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK14-NEXT: store ptr [[A]], ptr [[TMP66]], align 8
+// CHECK14-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i64 0, i64 0
+// CHECK14-NEXT: store ptr null, ptr [[TMP67]], align 8
+// CHECK14-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK14-NEXT: [[TMP69:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
// CHECK14-NEXT: [[KERNEL_ARGS19:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK14-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
-// CHECK14-NEXT: store i32 1, ptr [[TMP58]], align 4
-// CHECK14-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
-// CHECK14-NEXT: store i32 1, ptr [[TMP59]], align 4
-// CHECK14-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
-// CHECK14-NEXT: store ptr [[TMP56]], ptr [[TMP60]], align 8
-// CHECK14-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
-// CHECK14-NEXT: store ptr [[TMP57]], ptr [[TMP61]], align 8
-// CHECK14-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
-// CHECK14-NEXT: store ptr @.offload_sizes.32, ptr [[TMP62]], align 8
-// CHECK14-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
-// CHECK14-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP63]], align 8
-// CHECK14-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
-// CHECK14-NEXT: store ptr null, ptr [[TMP64]], align 8
-// CHECK14-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
-// CHECK14-NEXT: store ptr null, ptr [[TMP65]], align 8
-// CHECK14-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
-// CHECK14-NEXT: store i64 10, ptr [[TMP66]], align 8
-// CHECK14-NEXT: [[TMP67:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l137.region_id, ptr [[KERNEL_ARGS19]])
-// CHECK14-NEXT: [[TMP68:%.*]] = icmp ne i32 [[TMP67]], 0
-// CHECK14-NEXT: br i1 [[TMP68]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
+// CHECK14-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
+// CHECK14-NEXT: store i32 2, ptr [[TMP70]], align 4
+// CHECK14-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
+// CHECK14-NEXT: store i32 1, ptr [[TMP71]], align 4
+// CHECK14-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
+// CHECK14-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 8
+// CHECK14-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
+// CHECK14-NEXT: store ptr [[TMP69]], ptr [[TMP73]], align 8
+// CHECK14-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
+// CHECK14-NEXT: store ptr @.offload_sizes.32, ptr [[TMP74]], align 8
+// CHECK14-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
+// CHECK14-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP75]], align 8
+// CHECK14-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
+// CHECK14-NEXT: store ptr null, ptr [[TMP76]], align 8
+// CHECK14-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
+// CHECK14-NEXT: store ptr null, ptr [[TMP77]], align 8
+// CHECK14-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
+// CHECK14-NEXT: store i64 10, ptr [[TMP78]], align 8
+// CHECK14-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 9
+// CHECK14-NEXT: store i64 0, ptr [[TMP79]], align 8
+// CHECK14-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 10
+// CHECK14-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP80]], align 4
+// CHECK14-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 11
+// CHECK14-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP81]], align 4
+// CHECK14-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 12
+// CHECK14-NEXT: store i32 0, ptr [[TMP82]], align 4
+// CHECK14-NEXT: [[TMP83:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l137.region_id, ptr [[KERNEL_ARGS19]])
+// CHECK14-NEXT: [[TMP84:%.*]] = icmp ne i32 [[TMP83]], 0
+// CHECK14-NEXT: br i1 [[TMP84]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
// CHECK14: omp_offload.failed20:
// CHECK14-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l137(ptr [[A]]) #[[ATTR3]]
// CHECK14-NEXT: br label [[OMP_OFFLOAD_CONT21]]
// CHECK14: omp_offload.cont21:
-// CHECK14-NEXT: [[TMP69:%.*]] = load i32, ptr [[M]], align 4
-// CHECK14-NEXT: store i32 [[TMP69]], ptr [[M_CASTED22]], align 4
-// CHECK14-NEXT: [[TMP70:%.*]] = load i64, ptr [[M_CASTED22]], align 8
-// CHECK14-NEXT: [[TMP71:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
-// CHECK14-NEXT: store i64 [[TMP70]], ptr [[TMP71]], align 8
-// CHECK14-NEXT: [[TMP72:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
-// CHECK14-NEXT: store i64 [[TMP70]], ptr [[TMP72]], align 8
-// CHECK14-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i64 0, i64 0
-// CHECK14-NEXT: store ptr null, ptr [[TMP73]], align 8
-// CHECK14-NEXT: [[TMP74:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 1
-// CHECK14-NEXT: store ptr [[A]], ptr [[TMP74]], align 8
-// CHECK14-NEXT: [[TMP75:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 1
-// CHECK14-NEXT: store ptr [[A]], ptr [[TMP75]], align 8
-// CHECK14-NEXT: [[TMP76:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i64 0, i64 1
-// CHECK14-NEXT: store ptr null, ptr [[TMP76]], align 8
-// CHECK14-NEXT: [[TMP77:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
-// CHECK14-NEXT: [[TMP78:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
+// CHECK14-NEXT: [[TMP85:%.*]] = load i32, ptr [[M]], align 4
+// CHECK14-NEXT: store i32 [[TMP85]], ptr [[M_CASTED22]], align 4
+// CHECK14-NEXT: [[TMP86:%.*]] = load i64, ptr [[M_CASTED22]], align 8
+// CHECK14-NEXT: [[TMP87:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
+// CHECK14-NEXT: store i64 [[TMP86]], ptr [[TMP87]], align 8
+// CHECK14-NEXT: [[TMP88:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
+// CHECK14-NEXT: store i64 [[TMP86]], ptr [[TMP88]], align 8
+// CHECK14-NEXT: [[TMP89:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i64 0, i64 0
+// CHECK14-NEXT: store ptr null, ptr [[TMP89]], align 8
+// CHECK14-NEXT: [[TMP90:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 1
+// CHECK14-NEXT: store ptr [[A]], ptr [[TMP90]], align 8
+// CHECK14-NEXT: [[TMP91:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 1
+// CHECK14-NEXT: store ptr [[A]], ptr [[TMP91]], align 8
+// CHECK14-NEXT: [[TMP92:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i64 0, i64 1
+// CHECK14-NEXT: store ptr null, ptr [[TMP92]], align 8
+// CHECK14-NEXT: [[TMP93:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
+// CHECK14-NEXT: [[TMP94:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
// CHECK14-NEXT: [[KERNEL_ARGS27:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK14-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 0
-// CHECK14-NEXT: store i32 1, ptr [[TMP79]], align 4
-// CHECK14-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 1
-// CHECK14-NEXT: store i32 2, ptr [[TMP80]], align 4
-// CHECK14-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 2
-// CHECK14-NEXT: store ptr [[TMP77]], ptr [[TMP81]], align 8
-// CHECK14-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 3
-// CHECK14-NEXT: store ptr [[TMP78]], ptr [[TMP82]], align 8
-// CHECK14-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 4
-// CHECK14-NEXT: store ptr @.offload_sizes.36, ptr [[TMP83]], align 8
-// CHECK14-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 5
-// CHECK14-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP84]], align 8
-// CHECK14-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 6
-// CHECK14-NEXT: store ptr null, ptr [[TMP85]], align 8
-// CHECK14-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 7
-// CHECK14-NEXT: store ptr null, ptr [[TMP86]], align 8
-// CHECK14-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 8
-// CHECK14-NEXT: store i64 10, ptr [[TMP87]], align 8
-// CHECK14-NEXT: [[TMP88:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l142.region_id, ptr [[KERNEL_ARGS27]])
-// CHECK14-NEXT: [[TMP89:%.*]] = icmp ne i32 [[TMP88]], 0
-// CHECK14-NEXT: br i1 [[TMP89]], label [[OMP_OFFLOAD_FAILED28:%.*]], label [[OMP_OFFLOAD_CONT29:%.*]]
+// CHECK14-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 0
+// CHECK14-NEXT: store i32 2, ptr [[TMP95]], align 4
+// CHECK14-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 1
+// CHECK14-NEXT: store i32 2, ptr [[TMP96]], align 4
+// CHECK14-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 2
+// CHECK14-NEXT: store ptr [[TMP93]], ptr [[TMP97]], align 8
+// CHECK14-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 3
+// CHECK14-NEXT: store ptr [[TMP94]], ptr [[TMP98]], align 8
+// CHECK14-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 4
+// CHECK14-NEXT: store ptr @.offload_sizes.36, ptr [[TMP99]], align 8
+// CHECK14-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 5
+// CHECK14-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP100]], align 8
+// CHECK14-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 6
+// CHECK14-NEXT: store ptr null, ptr [[TMP101]], align 8
+// CHECK14-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 7
+// CHECK14-NEXT: store ptr null, ptr [[TMP102]], align 8
+// CHECK14-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 8
+// CHECK14-NEXT: store i64 10, ptr [[TMP103]], align 8
+// CHECK14-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 9
+// CHECK14-NEXT: store i64 0, ptr [[TMP104]], align 8
+// CHECK14-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 10
+// CHECK14-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP105]], align 4
+// CHECK14-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 11
+// CHECK14-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP106]], align 4
+// CHECK14-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 12
+// CHECK14-NEXT: store i32 0, ptr [[TMP107]], align 4
+// CHECK14-NEXT: [[TMP108:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l142.region_id, ptr [[KERNEL_ARGS27]])
+// CHECK14-NEXT: [[TMP109:%.*]] = icmp ne i32 [[TMP108]], 0
+// CHECK14-NEXT: br i1 [[TMP109]], label [[OMP_OFFLOAD_FAILED28:%.*]], label [[OMP_OFFLOAD_CONT29:%.*]]
// CHECK14: omp_offload.failed28:
-// CHECK14-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l142(i64 [[TMP70]], ptr [[A]]) #[[ATTR3]]
+// CHECK14-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l142(i64 [[TMP86]], ptr [[A]]) #[[ATTR3]]
// CHECK14-NEXT: br label [[OMP_OFFLOAD_CONT29]]
// CHECK14: omp_offload.cont29:
// CHECK14-NEXT: ret i32 0
@@ -10384,7 +10704,7 @@ int main (int argc, char **argv) {
// CHECK17-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK17-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK17-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK17-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK17-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK17-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK17-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -10401,298 +10721,338 @@ int main (int argc, char **argv) {
// CHECK17-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK17-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK17-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK17-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l154.region_id, ptr [[KERNEL_ARGS]])
-// CHECK17-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK17-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK17-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK17-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK17-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK17-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK17-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l154.region_id, ptr [[KERNEL_ARGS]])
+// CHECK17-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK17-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK17: omp_offload.failed:
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l154(i32 [[TMP3]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK17: omp_offload.cont:
-// CHECK17-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK17-NEXT: [[TMP35:%.*]] = load i32, ptr [[N_CASTED3]], align 4
-// CHECK17-NEXT: [[TMP36:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK17-NEXT: [[TMP37:%.*]] = sext i32 [[TMP36]] to i64
+// CHECK17-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK17-NEXT: [[TMP39:%.*]] = load i32, ptr [[N_CASTED3]], align 4
+// CHECK17-NEXT: [[TMP40:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK17-NEXT: [[TMP41:%.*]] = sext i32 [[TMP40]] to i64
// CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES7]], ptr align 4 @.offload_sizes.4, i32 24, i1 false)
-// CHECK17-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK17-NEXT: store i32 [[TMP35]], ptr [[TMP38]], align 4
-// CHECK17-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK17-NEXT: store i32 [[TMP35]], ptr [[TMP39]], align 4
-// CHECK17-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP40]], align 4
-// CHECK17-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK17-NEXT: store i32 [[TMP0]], ptr [[TMP41]], align 4
-// CHECK17-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK17-NEXT: store i32 [[TMP0]], ptr [[TMP42]], align 4
-// CHECK17-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
-// CHECK17-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK17-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 4
-// CHECK17-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP45]], align 4
-// CHECK17-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK17-NEXT: store i64 [[TMP37]], ptr [[TMP46]], align 4
-// CHECK17-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK17-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK17-NEXT: store i32 [[TMP39]], ptr [[TMP42]], align 4
+// CHECK17-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK17-NEXT: store i32 [[TMP39]], ptr [[TMP43]], align 4
+// CHECK17-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP44]], align 4
+// CHECK17-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK17-NEXT: store i32 [[TMP0]], ptr [[TMP45]], align 4
+// CHECK17-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK17-NEXT: store i32 [[TMP0]], ptr [[TMP46]], align 4
+// CHECK17-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
// CHECK17-NEXT: store ptr null, ptr [[TMP47]], align 4
-// CHECK17-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP51:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP51]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK17-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK17-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP52]], 0
+// CHECK17-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 4
+// CHECK17-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP49]], align 4
+// CHECK17-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK17-NEXT: store i64 [[TMP41]], ptr [[TMP50]], align 4
+// CHECK17-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK17-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK17-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP54:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP55:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP55]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK17-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK17-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP56]], 0
// CHECK17-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK17-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK17-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK17-NEXT: [[TMP53:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK17-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP53]], 1
-// CHECK17-NEXT: [[TMP54:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK17-NEXT: [[TMP57:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK17-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP57]], 1
+// CHECK17-NEXT: [[TMP58:%.*]] = zext i32 [[ADD14]] to i64
// CHECK17-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP55]], align 4
-// CHECK17-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK17-NEXT: store i32 3, ptr [[TMP56]], align 4
-// CHECK17-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 4
-// CHECK17-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 4
-// CHECK17-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK17-NEXT: store ptr [[TMP50]], ptr [[TMP59]], align 4
-// CHECK17-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP60]], align 4
-// CHECK17-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP61]], align 4
-// CHECK17-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP62]], align 4
-// CHECK17-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK17-NEXT: store i64 [[TMP54]], ptr [[TMP63]], align 8
-// CHECK17-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l159.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK17-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
-// CHECK17-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK17-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP59]], align 4
+// CHECK17-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK17-NEXT: store i32 3, ptr [[TMP60]], align 4
+// CHECK17-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 4
+// CHECK17-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 4
+// CHECK17-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK17-NEXT: store ptr [[TMP54]], ptr [[TMP63]], align 4
+// CHECK17-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP64]], align 4
+// CHECK17-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP65]], align 4
+// CHECK17-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP66]], align 4
+// CHECK17-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK17-NEXT: store i64 [[TMP58]], ptr [[TMP67]], align 8
+// CHECK17-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP68]], align 8
+// CHECK17-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK17-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP70]], align 4
+// CHECK17-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP71]], align 4
+// CHECK17-NEXT: [[TMP72:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l159.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK17-NEXT: [[TMP73:%.*]] = icmp ne i32 [[TMP72]], 0
+// CHECK17-NEXT: br i1 [[TMP73]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK17: omp_offload.failed16:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l159(i32 [[TMP35]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l159(i32 [[TMP39]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK17: omp_offload.cont17:
-// CHECK17-NEXT: [[TMP66:%.*]] = load i32, ptr [[M]], align 4
-// CHECK17-NEXT: store i32 [[TMP66]], ptr [[M_CASTED]], align 4
-// CHECK17-NEXT: [[TMP67:%.*]] = load i32, ptr [[M_CASTED]], align 4
-// CHECK17-NEXT: [[TMP68:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP68]], ptr [[N_CASTED18]], align 4
-// CHECK17-NEXT: [[TMP69:%.*]] = load i32, ptr [[N_CASTED18]], align 4
-// CHECK17-NEXT: [[TMP70:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK17-NEXT: [[TMP71:%.*]] = sext i32 [[TMP70]] to i64
+// CHECK17-NEXT: [[TMP74:%.*]] = load i32, ptr [[M]], align 4
+// CHECK17-NEXT: store i32 [[TMP74]], ptr [[M_CASTED]], align 4
+// CHECK17-NEXT: [[TMP75:%.*]] = load i32, ptr [[M_CASTED]], align 4
+// CHECK17-NEXT: [[TMP76:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP76]], ptr [[N_CASTED18]], align 4
+// CHECK17-NEXT: [[TMP77:%.*]] = load i32, ptr [[N_CASTED18]], align 4
+// CHECK17-NEXT: [[TMP78:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK17-NEXT: [[TMP79:%.*]] = sext i32 [[TMP78]] to i64
// CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES22]], ptr align 4 @.offload_sizes.8, i32 32, i1 false)
-// CHECK17-NEXT: [[TMP72:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK17-NEXT: store i32 [[TMP67]], ptr [[TMP72]], align 4
-// CHECK17-NEXT: [[TMP73:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK17-NEXT: store i32 [[TMP67]], ptr [[TMP73]], align 4
-// CHECK17-NEXT: [[TMP74:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP74]], align 4
-// CHECK17-NEXT: [[TMP75:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
-// CHECK17-NEXT: store i32 [[TMP69]], ptr [[TMP75]], align 4
-// CHECK17-NEXT: [[TMP76:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
-// CHECK17-NEXT: store i32 [[TMP69]], ptr [[TMP76]], align 4
-// CHECK17-NEXT: [[TMP77:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 1
-// CHECK17-NEXT: store ptr null, ptr [[TMP77]], align 4
-// CHECK17-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
-// CHECK17-NEXT: store i32 [[TMP0]], ptr [[TMP78]], align 4
-// CHECK17-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
-// CHECK17-NEXT: store i32 [[TMP0]], ptr [[TMP79]], align 4
-// CHECK17-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 2
-// CHECK17-NEXT: store ptr null, ptr [[TMP80]], align 4
-// CHECK17-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP81]], align 4
-// CHECK17-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP82]], align 4
-// CHECK17-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 3
-// CHECK17-NEXT: store i64 [[TMP71]], ptr [[TMP83]], align 4
-// CHECK17-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 3
-// CHECK17-NEXT: store ptr null, ptr [[TMP84]], align 4
-// CHECK17-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP88:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP88]], ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK17-NEXT: [[TMP89:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK17-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP89]], 0
+// CHECK17-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK17-NEXT: store i32 [[TMP75]], ptr [[TMP80]], align 4
+// CHECK17-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK17-NEXT: store i32 [[TMP75]], ptr [[TMP81]], align 4
+// CHECK17-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP82]], align 4
+// CHECK17-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
+// CHECK17-NEXT: store i32 [[TMP77]], ptr [[TMP83]], align 4
+// CHECK17-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
+// CHECK17-NEXT: store i32 [[TMP77]], ptr [[TMP84]], align 4
+// CHECK17-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 1
+// CHECK17-NEXT: store ptr null, ptr [[TMP85]], align 4
+// CHECK17-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
+// CHECK17-NEXT: store i32 [[TMP0]], ptr [[TMP86]], align 4
+// CHECK17-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
+// CHECK17-NEXT: store i32 [[TMP0]], ptr [[TMP87]], align 4
+// CHECK17-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 2
+// CHECK17-NEXT: store ptr null, ptr [[TMP88]], align 4
+// CHECK17-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP89]], align 4
+// CHECK17-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP90]], align 4
+// CHECK17-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 3
+// CHECK17-NEXT: store i64 [[TMP79]], ptr [[TMP91]], align 4
+// CHECK17-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 3
+// CHECK17-NEXT: store ptr null, ptr [[TMP92]], align 4
+// CHECK17-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP94:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP95:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP96:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP96]], ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK17-NEXT: [[TMP97:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK17-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP97]], 0
// CHECK17-NEXT: [[DIV27:%.*]] = sdiv i32 [[SUB26]], 1
// CHECK17-NEXT: [[SUB28:%.*]] = sub nsw i32 [[DIV27]], 1
// CHECK17-NEXT: store i32 [[SUB28]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK17-NEXT: [[TMP90:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK17-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP90]], 1
-// CHECK17-NEXT: [[TMP91:%.*]] = zext i32 [[ADD29]] to i64
+// CHECK17-NEXT: [[TMP98:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK17-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP98]], 1
+// CHECK17-NEXT: [[TMP99:%.*]] = zext i32 [[ADD29]] to i64
// CHECK17-NEXT: [[KERNEL_ARGS30:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP92]], align 4
-// CHECK17-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
-// CHECK17-NEXT: store i32 4, ptr [[TMP93]], align 4
-// CHECK17-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP85]], ptr [[TMP94]], align 4
-// CHECK17-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP86]], ptr [[TMP95]], align 4
-// CHECK17-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
-// CHECK17-NEXT: store ptr [[TMP87]], ptr [[TMP96]], align 4
-// CHECK17-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP97]], align 4
-// CHECK17-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP98]], align 4
-// CHECK17-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP99]], align 4
-// CHECK17-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
-// CHECK17-NEXT: store i64 [[TMP91]], ptr [[TMP100]], align 8
-// CHECK17-NEXT: [[TMP101:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l164.region_id, ptr [[KERNEL_ARGS30]])
-// CHECK17-NEXT: [[TMP102:%.*]] = icmp ne i32 [[TMP101]], 0
-// CHECK17-NEXT: br i1 [[TMP102]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
+// CHECK17-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP100]], align 4
+// CHECK17-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
+// CHECK17-NEXT: store i32 4, ptr [[TMP101]], align 4
+// CHECK17-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP93]], ptr [[TMP102]], align 4
+// CHECK17-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP94]], ptr [[TMP103]], align 4
+// CHECK17-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
+// CHECK17-NEXT: store ptr [[TMP95]], ptr [[TMP104]], align 4
+// CHECK17-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP105]], align 4
+// CHECK17-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP106]], align 4
+// CHECK17-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP107]], align 4
+// CHECK17-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
+// CHECK17-NEXT: store i64 [[TMP99]], ptr [[TMP108]], align 8
+// CHECK17-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP109]], align 8
+// CHECK17-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP110]], align 4
+// CHECK17-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP111]], align 4
+// CHECK17-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP112]], align 4
+// CHECK17-NEXT: [[TMP113:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l164.region_id, ptr [[KERNEL_ARGS30]])
+// CHECK17-NEXT: [[TMP114:%.*]] = icmp ne i32 [[TMP113]], 0
+// CHECK17-NEXT: br i1 [[TMP114]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
// CHECK17: omp_offload.failed31:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l164(i32 [[TMP67]], i32 [[TMP69]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l164(i32 [[TMP75]], i32 [[TMP77]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT32]]
// CHECK17: omp_offload.cont32:
-// CHECK17-NEXT: [[TMP103:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP103]], ptr [[N_CASTED33]], align 4
-// CHECK17-NEXT: [[TMP104:%.*]] = load i32, ptr [[N_CASTED33]], align 4
-// CHECK17-NEXT: [[TMP105:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK17-NEXT: [[TMP106:%.*]] = sext i32 [[TMP105]] to i64
+// CHECK17-NEXT: [[TMP115:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP115]], ptr [[N_CASTED33]], align 4
+// CHECK17-NEXT: [[TMP116:%.*]] = load i32, ptr [[N_CASTED33]], align 4
+// CHECK17-NEXT: [[TMP117:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK17-NEXT: [[TMP118:%.*]] = sext i32 [[TMP117]] to i64
// CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES37]], ptr align 4 @.offload_sizes.12, i32 24, i1 false)
-// CHECK17-NEXT: [[TMP107:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
-// CHECK17-NEXT: store i32 [[TMP104]], ptr [[TMP107]], align 4
-// CHECK17-NEXT: [[TMP108:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
-// CHECK17-NEXT: store i32 [[TMP104]], ptr [[TMP108]], align 4
-// CHECK17-NEXT: [[TMP109:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i32 0, i32 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP109]], align 4
-// CHECK17-NEXT: [[TMP110:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 1
-// CHECK17-NEXT: store i32 [[TMP0]], ptr [[TMP110]], align 4
-// CHECK17-NEXT: [[TMP111:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 1
-// CHECK17-NEXT: store i32 [[TMP0]], ptr [[TMP111]], align 4
-// CHECK17-NEXT: [[TMP112:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i32 0, i32 1
-// CHECK17-NEXT: store ptr null, ptr [[TMP112]], align 4
-// CHECK17-NEXT: [[TMP113:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP113]], align 4
-// CHECK17-NEXT: [[TMP114:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP114]], align 4
-// CHECK17-NEXT: [[TMP115:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 2
-// CHECK17-NEXT: store i64 [[TMP106]], ptr [[TMP115]], align 4
-// CHECK17-NEXT: [[TMP116:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i32 0, i32 2
-// CHECK17-NEXT: store ptr null, ptr [[TMP116]], align 4
-// CHECK17-NEXT: [[TMP117:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP118:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP119:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP120:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP120]], ptr [[DOTCAPTURE_EXPR_39]], align 4
-// CHECK17-NEXT: [[TMP121:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_39]], align 4
-// CHECK17-NEXT: [[SUB41:%.*]] = sub nsw i32 [[TMP121]], 0
+// CHECK17-NEXT: [[TMP119:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
+// CHECK17-NEXT: store i32 [[TMP116]], ptr [[TMP119]], align 4
+// CHECK17-NEXT: [[TMP120:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
+// CHECK17-NEXT: store i32 [[TMP116]], ptr [[TMP120]], align 4
+// CHECK17-NEXT: [[TMP121:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i32 0, i32 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP121]], align 4
+// CHECK17-NEXT: [[TMP122:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 1
+// CHECK17-NEXT: store i32 [[TMP0]], ptr [[TMP122]], align 4
+// CHECK17-NEXT: [[TMP123:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 1
+// CHECK17-NEXT: store i32 [[TMP0]], ptr [[TMP123]], align 4
+// CHECK17-NEXT: [[TMP124:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i32 0, i32 1
+// CHECK17-NEXT: store ptr null, ptr [[TMP124]], align 4
+// CHECK17-NEXT: [[TMP125:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP125]], align 4
+// CHECK17-NEXT: [[TMP126:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP126]], align 4
+// CHECK17-NEXT: [[TMP127:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 2
+// CHECK17-NEXT: store i64 [[TMP118]], ptr [[TMP127]], align 4
+// CHECK17-NEXT: [[TMP128:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i32 0, i32 2
+// CHECK17-NEXT: store ptr null, ptr [[TMP128]], align 4
+// CHECK17-NEXT: [[TMP129:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP130:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP131:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP132:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP132]], ptr [[DOTCAPTURE_EXPR_39]], align 4
+// CHECK17-NEXT: [[TMP133:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_39]], align 4
+// CHECK17-NEXT: [[SUB41:%.*]] = sub nsw i32 [[TMP133]], 0
// CHECK17-NEXT: [[DIV42:%.*]] = sdiv i32 [[SUB41]], 1
// CHECK17-NEXT: [[SUB43:%.*]] = sub nsw i32 [[DIV42]], 1
// CHECK17-NEXT: store i32 [[SUB43]], ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK17-NEXT: [[TMP122:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK17-NEXT: [[ADD44:%.*]] = add nsw i32 [[TMP122]], 1
-// CHECK17-NEXT: [[TMP123:%.*]] = zext i32 [[ADD44]] to i64
+// CHECK17-NEXT: [[TMP134:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
+// CHECK17-NEXT: [[ADD44:%.*]] = add nsw i32 [[TMP134]], 1
+// CHECK17-NEXT: [[TMP135:%.*]] = zext i32 [[ADD44]] to i64
// CHECK17-NEXT: [[KERNEL_ARGS45:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP124:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP124]], align 4
-// CHECK17-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 1
-// CHECK17-NEXT: store i32 3, ptr [[TMP125]], align 4
-// CHECK17-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP117]], ptr [[TMP126]], align 4
-// CHECK17-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP118]], ptr [[TMP127]], align 4
-// CHECK17-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 4
-// CHECK17-NEXT: store ptr [[TMP119]], ptr [[TMP128]], align 4
-// CHECK17-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP129]], align 4
-// CHECK17-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP130]], align 4
-// CHECK17-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP131]], align 4
-// CHECK17-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 8
-// CHECK17-NEXT: store i64 [[TMP123]], ptr [[TMP132]], align 8
-// CHECK17-NEXT: [[TMP133:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l169.region_id, ptr [[KERNEL_ARGS45]])
-// CHECK17-NEXT: [[TMP134:%.*]] = icmp ne i32 [[TMP133]], 0
-// CHECK17-NEXT: br i1 [[TMP134]], label [[OMP_OFFLOAD_FAILED46:%.*]], label [[OMP_OFFLOAD_CONT47:%.*]]
+// CHECK17-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP136]], align 4
+// CHECK17-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 1
+// CHECK17-NEXT: store i32 3, ptr [[TMP137]], align 4
+// CHECK17-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP129]], ptr [[TMP138]], align 4
+// CHECK17-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP130]], ptr [[TMP139]], align 4
+// CHECK17-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 4
+// CHECK17-NEXT: store ptr [[TMP131]], ptr [[TMP140]], align 4
+// CHECK17-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP141]], align 4
+// CHECK17-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP142]], align 4
+// CHECK17-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP143]], align 4
+// CHECK17-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 8
+// CHECK17-NEXT: store i64 [[TMP135]], ptr [[TMP144]], align 8
+// CHECK17-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP145]], align 8
+// CHECK17-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP146]], align 4
+// CHECK17-NEXT: [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP147]], align 4
+// CHECK17-NEXT: [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP148]], align 4
+// CHECK17-NEXT: [[TMP149:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l169.region_id, ptr [[KERNEL_ARGS45]])
+// CHECK17-NEXT: [[TMP150:%.*]] = icmp ne i32 [[TMP149]], 0
+// CHECK17-NEXT: br i1 [[TMP150]], label [[OMP_OFFLOAD_FAILED46:%.*]], label [[OMP_OFFLOAD_CONT47:%.*]]
// CHECK17: omp_offload.failed46:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l169(i32 [[TMP104]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l169(i32 [[TMP116]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT47]]
// CHECK17: omp_offload.cont47:
-// CHECK17-NEXT: [[TMP135:%.*]] = load i32, ptr [[M]], align 4
-// CHECK17-NEXT: store i32 [[TMP135]], ptr [[M_CASTED48]], align 4
-// CHECK17-NEXT: [[TMP136:%.*]] = load i32, ptr [[M_CASTED48]], align 4
-// CHECK17-NEXT: [[TMP137:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP137]], ptr [[N_CASTED49]], align 4
-// CHECK17-NEXT: [[TMP138:%.*]] = load i32, ptr [[N_CASTED49]], align 4
-// CHECK17-NEXT: [[TMP139:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK17-NEXT: [[TMP140:%.*]] = sext i32 [[TMP139]] to i64
+// CHECK17-NEXT: [[TMP151:%.*]] = load i32, ptr [[M]], align 4
+// CHECK17-NEXT: store i32 [[TMP151]], ptr [[M_CASTED48]], align 4
+// CHECK17-NEXT: [[TMP152:%.*]] = load i32, ptr [[M_CASTED48]], align 4
+// CHECK17-NEXT: [[TMP153:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP153]], ptr [[N_CASTED49]], align 4
+// CHECK17-NEXT: [[TMP154:%.*]] = load i32, ptr [[N_CASTED49]], align 4
+// CHECK17-NEXT: [[TMP155:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK17-NEXT: [[TMP156:%.*]] = sext i32 [[TMP155]] to i64
// CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES53]], ptr align 4 @.offload_sizes.16, i32 32, i1 false)
-// CHECK17-NEXT: [[TMP141:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
-// CHECK17-NEXT: store i32 [[TMP136]], ptr [[TMP141]], align 4
-// CHECK17-NEXT: [[TMP142:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
-// CHECK17-NEXT: store i32 [[TMP136]], ptr [[TMP142]], align 4
-// CHECK17-NEXT: [[TMP143:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP143]], align 4
-// CHECK17-NEXT: [[TMP144:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 1
-// CHECK17-NEXT: store i32 [[TMP138]], ptr [[TMP144]], align 4
-// CHECK17-NEXT: [[TMP145:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 1
-// CHECK17-NEXT: store i32 [[TMP138]], ptr [[TMP145]], align 4
-// CHECK17-NEXT: [[TMP146:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 1
-// CHECK17-NEXT: store ptr null, ptr [[TMP146]], align 4
-// CHECK17-NEXT: [[TMP147:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 2
-// CHECK17-NEXT: store i32 [[TMP0]], ptr [[TMP147]], align 4
-// CHECK17-NEXT: [[TMP148:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 2
-// CHECK17-NEXT: store i32 [[TMP0]], ptr [[TMP148]], align 4
-// CHECK17-NEXT: [[TMP149:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 2
-// CHECK17-NEXT: store ptr null, ptr [[TMP149]], align 4
-// CHECK17-NEXT: [[TMP150:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP150]], align 4
-// CHECK17-NEXT: [[TMP151:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP151]], align 4
-// CHECK17-NEXT: [[TMP152:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 3
-// CHECK17-NEXT: store i64 [[TMP140]], ptr [[TMP152]], align 4
-// CHECK17-NEXT: [[TMP153:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 3
-// CHECK17-NEXT: store ptr null, ptr [[TMP153]], align 4
-// CHECK17-NEXT: [[TMP154:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP155:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP156:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP157:%.*]] = load i32, ptr [[N]], align 4
-// CHECK17-NEXT: store i32 [[TMP157]], ptr [[DOTCAPTURE_EXPR_55]], align 4
-// CHECK17-NEXT: [[TMP158:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_55]], align 4
-// CHECK17-NEXT: [[SUB57:%.*]] = sub nsw i32 [[TMP158]], 0
+// CHECK17-NEXT: [[TMP157:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
+// CHECK17-NEXT: store i32 [[TMP152]], ptr [[TMP157]], align 4
+// CHECK17-NEXT: [[TMP158:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
+// CHECK17-NEXT: store i32 [[TMP152]], ptr [[TMP158]], align 4
+// CHECK17-NEXT: [[TMP159:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP159]], align 4
+// CHECK17-NEXT: [[TMP160:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 1
+// CHECK17-NEXT: store i32 [[TMP154]], ptr [[TMP160]], align 4
+// CHECK17-NEXT: [[TMP161:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 1
+// CHECK17-NEXT: store i32 [[TMP154]], ptr [[TMP161]], align 4
+// CHECK17-NEXT: [[TMP162:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 1
+// CHECK17-NEXT: store ptr null, ptr [[TMP162]], align 4
+// CHECK17-NEXT: [[TMP163:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 2
+// CHECK17-NEXT: store i32 [[TMP0]], ptr [[TMP163]], align 4
+// CHECK17-NEXT: [[TMP164:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 2
+// CHECK17-NEXT: store i32 [[TMP0]], ptr [[TMP164]], align 4
+// CHECK17-NEXT: [[TMP165:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 2
+// CHECK17-NEXT: store ptr null, ptr [[TMP165]], align 4
+// CHECK17-NEXT: [[TMP166:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP166]], align 4
+// CHECK17-NEXT: [[TMP167:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP167]], align 4
+// CHECK17-NEXT: [[TMP168:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 3
+// CHECK17-NEXT: store i64 [[TMP156]], ptr [[TMP168]], align 4
+// CHECK17-NEXT: [[TMP169:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 3
+// CHECK17-NEXT: store ptr null, ptr [[TMP169]], align 4
+// CHECK17-NEXT: [[TMP170:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP171:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP172:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP173:%.*]] = load i32, ptr [[N]], align 4
+// CHECK17-NEXT: store i32 [[TMP173]], ptr [[DOTCAPTURE_EXPR_55]], align 4
+// CHECK17-NEXT: [[TMP174:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_55]], align 4
+// CHECK17-NEXT: [[SUB57:%.*]] = sub nsw i32 [[TMP174]], 0
// CHECK17-NEXT: [[DIV58:%.*]] = sdiv i32 [[SUB57]], 1
// CHECK17-NEXT: [[SUB59:%.*]] = sub nsw i32 [[DIV58]], 1
// CHECK17-NEXT: store i32 [[SUB59]], ptr [[DOTCAPTURE_EXPR_56]], align 4
-// CHECK17-NEXT: [[TMP159:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_56]], align 4
-// CHECK17-NEXT: [[ADD60:%.*]] = add nsw i32 [[TMP159]], 1
-// CHECK17-NEXT: [[TMP160:%.*]] = zext i32 [[ADD60]] to i64
+// CHECK17-NEXT: [[TMP175:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_56]], align 4
+// CHECK17-NEXT: [[ADD60:%.*]] = add nsw i32 [[TMP175]], 1
+// CHECK17-NEXT: [[TMP176:%.*]] = zext i32 [[ADD60]] to i64
// CHECK17-NEXT: [[KERNEL_ARGS61:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP161:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP161]], align 4
-// CHECK17-NEXT: [[TMP162:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 1
-// CHECK17-NEXT: store i32 4, ptr [[TMP162]], align 4
-// CHECK17-NEXT: [[TMP163:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP154]], ptr [[TMP163]], align 4
-// CHECK17-NEXT: [[TMP164:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP155]], ptr [[TMP164]], align 4
-// CHECK17-NEXT: [[TMP165:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 4
-// CHECK17-NEXT: store ptr [[TMP156]], ptr [[TMP165]], align 4
-// CHECK17-NEXT: [[TMP166:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP166]], align 4
-// CHECK17-NEXT: [[TMP167:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP167]], align 4
-// CHECK17-NEXT: [[TMP168:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP168]], align 4
-// CHECK17-NEXT: [[TMP169:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 8
-// CHECK17-NEXT: store i64 [[TMP160]], ptr [[TMP169]], align 8
-// CHECK17-NEXT: [[TMP170:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l174.region_id, ptr [[KERNEL_ARGS61]])
-// CHECK17-NEXT: [[TMP171:%.*]] = icmp ne i32 [[TMP170]], 0
-// CHECK17-NEXT: br i1 [[TMP171]], label [[OMP_OFFLOAD_FAILED62:%.*]], label [[OMP_OFFLOAD_CONT63:%.*]]
+// CHECK17-NEXT: [[TMP177:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP177]], align 4
+// CHECK17-NEXT: [[TMP178:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 1
+// CHECK17-NEXT: store i32 4, ptr [[TMP178]], align 4
+// CHECK17-NEXT: [[TMP179:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP170]], ptr [[TMP179]], align 4
+// CHECK17-NEXT: [[TMP180:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP171]], ptr [[TMP180]], align 4
+// CHECK17-NEXT: [[TMP181:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 4
+// CHECK17-NEXT: store ptr [[TMP172]], ptr [[TMP181]], align 4
+// CHECK17-NEXT: [[TMP182:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP182]], align 4
+// CHECK17-NEXT: [[TMP183:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP183]], align 4
+// CHECK17-NEXT: [[TMP184:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP184]], align 4
+// CHECK17-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 8
+// CHECK17-NEXT: store i64 [[TMP176]], ptr [[TMP185]], align 8
+// CHECK17-NEXT: [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP186]], align 8
+// CHECK17-NEXT: [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP187]], align 4
+// CHECK17-NEXT: [[TMP188:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP188]], align 4
+// CHECK17-NEXT: [[TMP189:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP189]], align 4
+// CHECK17-NEXT: [[TMP190:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l174.region_id, ptr [[KERNEL_ARGS61]])
+// CHECK17-NEXT: [[TMP191:%.*]] = icmp ne i32 [[TMP190]], 0
+// CHECK17-NEXT: br i1 [[TMP191]], label [[OMP_OFFLOAD_FAILED62:%.*]], label [[OMP_OFFLOAD_CONT63:%.*]]
// CHECK17: omp_offload.failed62:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l174(i32 [[TMP136]], i32 [[TMP138]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l174(i32 [[TMP152]], i32 [[TMP154]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT63]]
// CHECK17: omp_offload.cont63:
-// CHECK17-NEXT: [[TMP172:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK17-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP172]])
+// CHECK17-NEXT: [[TMP192:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK17-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP192]])
// CHECK17-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK17-NEXT: [[TMP173:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK17-NEXT: call void @llvm.stackrestore(ptr [[TMP173]])
-// CHECK17-NEXT: [[TMP174:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK17-NEXT: ret i32 [[TMP174]]
+// CHECK17-NEXT: [[TMP193:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK17-NEXT: call void @llvm.stackrestore(ptr [[TMP193]])
+// CHECK17-NEXT: [[TMP194:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK17-NEXT: ret i32 [[TMP194]]
//
//
// CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l154
@@ -11937,7 +12297,7 @@ int main (int argc, char **argv) {
// CHECK17-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK17-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK17-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK17-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK17-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK17-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -11954,165 +12314,205 @@ int main (int argc, char **argv) {
// CHECK17-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK17-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK17-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK17-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l122.region_id, ptr [[KERNEL_ARGS]])
-// CHECK17-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK17-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK17-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK17-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK17-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK17-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK17-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l122.region_id, ptr [[KERNEL_ARGS]])
+// CHECK17-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK17-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK17: omp_offload.failed:
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l122(ptr [[A]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK17: omp_offload.cont:
-// CHECK17-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP16]], align 4
-// CHECK17-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP17]], align 4
-// CHECK17-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK17-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP20]], align 4
+// CHECK17-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP21]], align 4
+// CHECK17-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK17-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK17-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK17-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK17-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK17-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK17-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK17-NEXT: store ptr @.offload_sizes.24, ptr [[TMP25]], align 4
-// CHECK17-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP26]], align 4
-// CHECK17-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK17-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK17-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK17-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK17-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l127.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK17-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK17-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK17-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK17-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK17-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK17-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK17-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK17-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK17-NEXT: store ptr @.offload_sizes.24, ptr [[TMP29]], align 4
+// CHECK17-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP30]], align 4
+// CHECK17-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK17-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK17-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK17-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK17-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK17-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK17-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK17-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK17-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l127.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK17-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK17-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK17: omp_offload.failed6:
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l127(ptr [[A]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK17: omp_offload.cont7:
-// CHECK17-NEXT: [[TMP32:%.*]] = load i32, ptr [[M]], align 4
-// CHECK17-NEXT: store i32 [[TMP32]], ptr [[M_CASTED]], align 4
-// CHECK17-NEXT: [[TMP33:%.*]] = load i32, ptr [[M_CASTED]], align 4
-// CHECK17-NEXT: [[TMP34:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK17-NEXT: store i32 [[TMP33]], ptr [[TMP34]], align 4
-// CHECK17-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK17-NEXT: store i32 [[TMP33]], ptr [[TMP35]], align 4
-// CHECK17-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP36]], align 4
-// CHECK17-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP37]], align 4
-// CHECK17-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP38]], align 4
-// CHECK17-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1
-// CHECK17-NEXT: store ptr null, ptr [[TMP39]], align 4
-// CHECK17-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP40:%.*]] = load i32, ptr [[M]], align 4
+// CHECK17-NEXT: store i32 [[TMP40]], ptr [[M_CASTED]], align 4
+// CHECK17-NEXT: [[TMP41:%.*]] = load i32, ptr [[M_CASTED]], align 4
+// CHECK17-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK17-NEXT: store i32 [[TMP41]], ptr [[TMP42]], align 4
+// CHECK17-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK17-NEXT: store i32 [[TMP41]], ptr [[TMP43]], align 4
+// CHECK17-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP44]], align 4
+// CHECK17-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP45]], align 4
+// CHECK17-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP46]], align 4
+// CHECK17-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1
+// CHECK17-NEXT: store ptr null, ptr [[TMP47]], align 4
+// CHECK17-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP42]], align 4
-// CHECK17-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK17-NEXT: store i32 2, ptr [[TMP43]], align 4
-// CHECK17-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP40]], ptr [[TMP44]], align 4
-// CHECK17-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP41]], ptr [[TMP45]], align 4
-// CHECK17-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK17-NEXT: store ptr @.offload_sizes.28, ptr [[TMP46]], align 4
-// CHECK17-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP47]], align 4
-// CHECK17-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP48]], align 4
-// CHECK17-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP49]], align 4
-// CHECK17-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK17-NEXT: store i64 10, ptr [[TMP50]], align 8
-// CHECK17-NEXT: [[TMP51:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l132.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK17-NEXT: [[TMP52:%.*]] = icmp ne i32 [[TMP51]], 0
-// CHECK17-NEXT: br i1 [[TMP52]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK17-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP50]], align 4
+// CHECK17-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK17-NEXT: store i32 2, ptr [[TMP51]], align 4
+// CHECK17-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP48]], ptr [[TMP52]], align 4
+// CHECK17-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 4
+// CHECK17-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK17-NEXT: store ptr @.offload_sizes.28, ptr [[TMP54]], align 4
+// CHECK17-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP55]], align 4
+// CHECK17-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP56]], align 4
+// CHECK17-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP57]], align 4
+// CHECK17-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK17-NEXT: store i64 10, ptr [[TMP58]], align 8
+// CHECK17-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP59]], align 8
+// CHECK17-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP60]], align 4
+// CHECK17-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4
+// CHECK17-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP62]], align 4
+// CHECK17-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l132.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK17-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
+// CHECK17-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK17: omp_offload.failed13:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l132(i32 [[TMP33]], ptr [[A]]) #[[ATTR3]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l132(i32 [[TMP41]], ptr [[A]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT14]]
// CHECK17: omp_offload.cont14:
-// CHECK17-NEXT: [[TMP53:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP53]], align 4
-// CHECK17-NEXT: [[TMP54:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP54]], align 4
-// CHECK17-NEXT: [[TMP55:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i32 0, i32 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP55]], align 4
-// CHECK17-NEXT: [[TMP56:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP57:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP65:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP65]], align 4
+// CHECK17-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP66]], align 4
+// CHECK17-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i32 0, i32 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP67]], align 4
+// CHECK17-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP69:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS19:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP58]], align 4
-// CHECK17-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
-// CHECK17-NEXT: store i32 1, ptr [[TMP59]], align 4
-// CHECK17-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP56]], ptr [[TMP60]], align 4
-// CHECK17-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP57]], ptr [[TMP61]], align 4
-// CHECK17-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
-// CHECK17-NEXT: store ptr @.offload_sizes.32, ptr [[TMP62]], align 4
-// CHECK17-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP63]], align 4
-// CHECK17-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP64]], align 4
-// CHECK17-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP65]], align 4
-// CHECK17-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
-// CHECK17-NEXT: store i64 10, ptr [[TMP66]], align 8
-// CHECK17-NEXT: [[TMP67:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l137.region_id, ptr [[KERNEL_ARGS19]])
-// CHECK17-NEXT: [[TMP68:%.*]] = icmp ne i32 [[TMP67]], 0
-// CHECK17-NEXT: br i1 [[TMP68]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
+// CHECK17-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP70]], align 4
+// CHECK17-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
+// CHECK17-NEXT: store i32 1, ptr [[TMP71]], align 4
+// CHECK17-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 4
+// CHECK17-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP69]], ptr [[TMP73]], align 4
+// CHECK17-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
+// CHECK17-NEXT: store ptr @.offload_sizes.32, ptr [[TMP74]], align 4
+// CHECK17-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP75]], align 4
+// CHECK17-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP76]], align 4
+// CHECK17-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP77]], align 4
+// CHECK17-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
+// CHECK17-NEXT: store i64 10, ptr [[TMP78]], align 8
+// CHECK17-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP79]], align 8
+// CHECK17-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP80]], align 4
+// CHECK17-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP81]], align 4
+// CHECK17-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP82]], align 4
+// CHECK17-NEXT: [[TMP83:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l137.region_id, ptr [[KERNEL_ARGS19]])
+// CHECK17-NEXT: [[TMP84:%.*]] = icmp ne i32 [[TMP83]], 0
+// CHECK17-NEXT: br i1 [[TMP84]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
// CHECK17: omp_offload.failed20:
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l137(ptr [[A]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT21]]
// CHECK17: omp_offload.cont21:
-// CHECK17-NEXT: [[TMP69:%.*]] = load i32, ptr [[M]], align 4
-// CHECK17-NEXT: store i32 [[TMP69]], ptr [[M_CASTED22]], align 4
-// CHECK17-NEXT: [[TMP70:%.*]] = load i32, ptr [[M_CASTED22]], align 4
-// CHECK17-NEXT: [[TMP71:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
-// CHECK17-NEXT: store i32 [[TMP70]], ptr [[TMP71]], align 4
-// CHECK17-NEXT: [[TMP72:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
-// CHECK17-NEXT: store i32 [[TMP70]], ptr [[TMP72]], align 4
-// CHECK17-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i32 0, i32 0
-// CHECK17-NEXT: store ptr null, ptr [[TMP73]], align 4
-// CHECK17-NEXT: [[TMP74:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 1
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP74]], align 4
-// CHECK17-NEXT: [[TMP75:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 1
-// CHECK17-NEXT: store ptr [[A]], ptr [[TMP75]], align 4
-// CHECK17-NEXT: [[TMP76:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i32 0, i32 1
-// CHECK17-NEXT: store ptr null, ptr [[TMP76]], align 4
-// CHECK17-NEXT: [[TMP77:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
-// CHECK17-NEXT: [[TMP78:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP85:%.*]] = load i32, ptr [[M]], align 4
+// CHECK17-NEXT: store i32 [[TMP85]], ptr [[M_CASTED22]], align 4
+// CHECK17-NEXT: [[TMP86:%.*]] = load i32, ptr [[M_CASTED22]], align 4
+// CHECK17-NEXT: [[TMP87:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
+// CHECK17-NEXT: store i32 [[TMP86]], ptr [[TMP87]], align 4
+// CHECK17-NEXT: [[TMP88:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
+// CHECK17-NEXT: store i32 [[TMP86]], ptr [[TMP88]], align 4
+// CHECK17-NEXT: [[TMP89:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i32 0, i32 0
+// CHECK17-NEXT: store ptr null, ptr [[TMP89]], align 4
+// CHECK17-NEXT: [[TMP90:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 1
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP90]], align 4
+// CHECK17-NEXT: [[TMP91:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 1
+// CHECK17-NEXT: store ptr [[A]], ptr [[TMP91]], align 4
+// CHECK17-NEXT: [[TMP92:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i32 0, i32 1
+// CHECK17-NEXT: store ptr null, ptr [[TMP92]], align 4
+// CHECK17-NEXT: [[TMP93:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
+// CHECK17-NEXT: [[TMP94:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS27:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK17-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP79]], align 4
-// CHECK17-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 1
-// CHECK17-NEXT: store i32 2, ptr [[TMP80]], align 4
-// CHECK17-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 2
-// CHECK17-NEXT: store ptr [[TMP77]], ptr [[TMP81]], align 4
-// CHECK17-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 3
-// CHECK17-NEXT: store ptr [[TMP78]], ptr [[TMP82]], align 4
-// CHECK17-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 4
-// CHECK17-NEXT: store ptr @.offload_sizes.36, ptr [[TMP83]], align 4
-// CHECK17-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 5
-// CHECK17-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP84]], align 4
-// CHECK17-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 6
-// CHECK17-NEXT: store ptr null, ptr [[TMP85]], align 4
-// CHECK17-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 7
-// CHECK17-NEXT: store ptr null, ptr [[TMP86]], align 4
-// CHECK17-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 8
-// CHECK17-NEXT: store i64 10, ptr [[TMP87]], align 8
-// CHECK17-NEXT: [[TMP88:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l142.region_id, ptr [[KERNEL_ARGS27]])
-// CHECK17-NEXT: [[TMP89:%.*]] = icmp ne i32 [[TMP88]], 0
-// CHECK17-NEXT: br i1 [[TMP89]], label [[OMP_OFFLOAD_FAILED28:%.*]], label [[OMP_OFFLOAD_CONT29:%.*]]
+// CHECK17-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 0
+// CHECK17-NEXT: store i32 2, ptr [[TMP95]], align 4
+// CHECK17-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 1
+// CHECK17-NEXT: store i32 2, ptr [[TMP96]], align 4
+// CHECK17-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 2
+// CHECK17-NEXT: store ptr [[TMP93]], ptr [[TMP97]], align 4
+// CHECK17-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 3
+// CHECK17-NEXT: store ptr [[TMP94]], ptr [[TMP98]], align 4
+// CHECK17-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 4
+// CHECK17-NEXT: store ptr @.offload_sizes.36, ptr [[TMP99]], align 4
+// CHECK17-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 5
+// CHECK17-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP100]], align 4
+// CHECK17-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 6
+// CHECK17-NEXT: store ptr null, ptr [[TMP101]], align 4
+// CHECK17-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 7
+// CHECK17-NEXT: store ptr null, ptr [[TMP102]], align 4
+// CHECK17-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 8
+// CHECK17-NEXT: store i64 10, ptr [[TMP103]], align 8
+// CHECK17-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP104]], align 8
+// CHECK17-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP105]], align 4
+// CHECK17-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP106]], align 4
+// CHECK17-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP107]], align 4
+// CHECK17-NEXT: [[TMP108:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l142.region_id, ptr [[KERNEL_ARGS27]])
+// CHECK17-NEXT: [[TMP109:%.*]] = icmp ne i32 [[TMP108]], 0
+// CHECK17-NEXT: br i1 [[TMP109]], label [[OMP_OFFLOAD_FAILED28:%.*]], label [[OMP_OFFLOAD_CONT29:%.*]]
// CHECK17: omp_offload.failed28:
-// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l142(i32 [[TMP70]], ptr [[A]]) #[[ATTR3]]
+// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l142(i32 [[TMP86]], ptr [[A]]) #[[ATTR3]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT29]]
// CHECK17: omp_offload.cont29:
// CHECK17-NEXT: ret i32 0
@@ -13074,7 +13474,7 @@ int main (int argc, char **argv) {
// CHECK19-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK19-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK19-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK19-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK19-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK19-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK19-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -13091,298 +13491,338 @@ int main (int argc, char **argv) {
// CHECK19-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK19-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK19-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK19-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l154.region_id, ptr [[KERNEL_ARGS]])
-// CHECK19-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK19-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK19-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK19-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK19-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
+// CHECK19-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK19-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l154.region_id, ptr [[KERNEL_ARGS]])
+// CHECK19-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK19-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK19: omp_offload.failed:
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l154(i32 [[TMP3]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK19: omp_offload.cont:
-// CHECK19-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK19-NEXT: [[TMP35:%.*]] = load i32, ptr [[N_CASTED3]], align 4
-// CHECK19-NEXT: [[TMP36:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK19-NEXT: [[TMP37:%.*]] = sext i32 [[TMP36]] to i64
+// CHECK19-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK19-NEXT: [[TMP39:%.*]] = load i32, ptr [[N_CASTED3]], align 4
+// CHECK19-NEXT: [[TMP40:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK19-NEXT: [[TMP41:%.*]] = sext i32 [[TMP40]] to i64
// CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES7]], ptr align 4 @.offload_sizes.4, i32 24, i1 false)
-// CHECK19-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP35]], ptr [[TMP38]], align 4
-// CHECK19-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP35]], ptr [[TMP39]], align 4
-// CHECK19-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP40]], align 4
-// CHECK19-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP41]], align 4
-// CHECK19-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP42]], align 4
-// CHECK19-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
-// CHECK19-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK19-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 4
-// CHECK19-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP45]], align 4
-// CHECK19-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK19-NEXT: store i64 [[TMP37]], ptr [[TMP46]], align 4
-// CHECK19-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK19-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP39]], ptr [[TMP42]], align 4
+// CHECK19-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP39]], ptr [[TMP43]], align 4
+// CHECK19-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP44]], align 4
+// CHECK19-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP45]], align 4
+// CHECK19-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP46]], align 4
+// CHECK19-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
// CHECK19-NEXT: store ptr null, ptr [[TMP47]], align 4
-// CHECK19-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP51:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP51]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK19-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK19-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP52]], 0
+// CHECK19-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 4
+// CHECK19-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP49]], align 4
+// CHECK19-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK19-NEXT: store i64 [[TMP41]], ptr [[TMP50]], align 4
+// CHECK19-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK19-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK19-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP54:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP55:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP55]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK19-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK19-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP56]], 0
// CHECK19-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK19-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK19-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK19-NEXT: [[TMP53:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK19-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP53]], 1
-// CHECK19-NEXT: [[TMP54:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK19-NEXT: [[TMP57:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK19-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP57]], 1
+// CHECK19-NEXT: [[TMP58:%.*]] = zext i32 [[ADD14]] to i64
// CHECK19-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP55]], align 4
-// CHECK19-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK19-NEXT: store i32 3, ptr [[TMP56]], align 4
-// CHECK19-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 4
-// CHECK19-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 4
-// CHECK19-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK19-NEXT: store ptr [[TMP50]], ptr [[TMP59]], align 4
-// CHECK19-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP60]], align 4
-// CHECK19-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP61]], align 4
-// CHECK19-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP62]], align 4
-// CHECK19-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK19-NEXT: store i64 [[TMP54]], ptr [[TMP63]], align 8
-// CHECK19-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l159.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK19-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
-// CHECK19-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK19-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP59]], align 4
+// CHECK19-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK19-NEXT: store i32 3, ptr [[TMP60]], align 4
+// CHECK19-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 4
+// CHECK19-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 4
+// CHECK19-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK19-NEXT: store ptr [[TMP54]], ptr [[TMP63]], align 4
+// CHECK19-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.5, ptr [[TMP64]], align 4
+// CHECK19-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP65]], align 4
+// CHECK19-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP66]], align 4
+// CHECK19-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK19-NEXT: store i64 [[TMP58]], ptr [[TMP67]], align 8
+// CHECK19-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP68]], align 8
+// CHECK19-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK19-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP70]], align 4
+// CHECK19-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP71]], align 4
+// CHECK19-NEXT: [[TMP72:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l159.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK19-NEXT: [[TMP73:%.*]] = icmp ne i32 [[TMP72]], 0
+// CHECK19-NEXT: br i1 [[TMP73]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK19: omp_offload.failed16:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l159(i32 [[TMP35]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l159(i32 [[TMP39]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK19: omp_offload.cont17:
-// CHECK19-NEXT: [[TMP66:%.*]] = load i32, ptr [[M]], align 4
-// CHECK19-NEXT: store i32 [[TMP66]], ptr [[M_CASTED]], align 4
-// CHECK19-NEXT: [[TMP67:%.*]] = load i32, ptr [[M_CASTED]], align 4
-// CHECK19-NEXT: [[TMP68:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP68]], ptr [[N_CASTED18]], align 4
-// CHECK19-NEXT: [[TMP69:%.*]] = load i32, ptr [[N_CASTED18]], align 4
-// CHECK19-NEXT: [[TMP70:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK19-NEXT: [[TMP71:%.*]] = sext i32 [[TMP70]] to i64
+// CHECK19-NEXT: [[TMP74:%.*]] = load i32, ptr [[M]], align 4
+// CHECK19-NEXT: store i32 [[TMP74]], ptr [[M_CASTED]], align 4
+// CHECK19-NEXT: [[TMP75:%.*]] = load i32, ptr [[M_CASTED]], align 4
+// CHECK19-NEXT: [[TMP76:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP76]], ptr [[N_CASTED18]], align 4
+// CHECK19-NEXT: [[TMP77:%.*]] = load i32, ptr [[N_CASTED18]], align 4
+// CHECK19-NEXT: [[TMP78:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK19-NEXT: [[TMP79:%.*]] = sext i32 [[TMP78]] to i64
// CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES22]], ptr align 4 @.offload_sizes.8, i32 32, i1 false)
-// CHECK19-NEXT: [[TMP72:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP67]], ptr [[TMP72]], align 4
-// CHECK19-NEXT: [[TMP73:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP67]], ptr [[TMP73]], align 4
-// CHECK19-NEXT: [[TMP74:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP74]], align 4
-// CHECK19-NEXT: [[TMP75:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP69]], ptr [[TMP75]], align 4
-// CHECK19-NEXT: [[TMP76:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP69]], ptr [[TMP76]], align 4
-// CHECK19-NEXT: [[TMP77:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 1
-// CHECK19-NEXT: store ptr null, ptr [[TMP77]], align 4
-// CHECK19-NEXT: [[TMP78:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP78]], align 4
-// CHECK19-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP79]], align 4
-// CHECK19-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 2
-// CHECK19-NEXT: store ptr null, ptr [[TMP80]], align 4
-// CHECK19-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP81]], align 4
-// CHECK19-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP82]], align 4
-// CHECK19-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 3
-// CHECK19-NEXT: store i64 [[TMP71]], ptr [[TMP83]], align 4
-// CHECK19-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 3
-// CHECK19-NEXT: store ptr null, ptr [[TMP84]], align 4
-// CHECK19-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP88:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP88]], ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK19-NEXT: [[TMP89:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK19-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP89]], 0
+// CHECK19-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP75]], ptr [[TMP80]], align 4
+// CHECK19-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP75]], ptr [[TMP81]], align 4
+// CHECK19-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP82]], align 4
+// CHECK19-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP77]], ptr [[TMP83]], align 4
+// CHECK19-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP77]], ptr [[TMP84]], align 4
+// CHECK19-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 1
+// CHECK19-NEXT: store ptr null, ptr [[TMP85]], align 4
+// CHECK19-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP86]], align 4
+// CHECK19-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP87]], align 4
+// CHECK19-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 2
+// CHECK19-NEXT: store ptr null, ptr [[TMP88]], align 4
+// CHECK19-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP89]], align 4
+// CHECK19-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP90]], align 4
+// CHECK19-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 3
+// CHECK19-NEXT: store i64 [[TMP79]], ptr [[TMP91]], align 4
+// CHECK19-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 3
+// CHECK19-NEXT: store ptr null, ptr [[TMP92]], align 4
+// CHECK19-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP94:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP95:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP96:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP96]], ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK19-NEXT: [[TMP97:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK19-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP97]], 0
// CHECK19-NEXT: [[DIV27:%.*]] = sdiv i32 [[SUB26]], 1
// CHECK19-NEXT: [[SUB28:%.*]] = sub nsw i32 [[DIV27]], 1
// CHECK19-NEXT: store i32 [[SUB28]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK19-NEXT: [[TMP90:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK19-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP90]], 1
-// CHECK19-NEXT: [[TMP91:%.*]] = zext i32 [[ADD29]] to i64
+// CHECK19-NEXT: [[TMP98:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK19-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP98]], 1
+// CHECK19-NEXT: [[TMP99:%.*]] = zext i32 [[ADD29]] to i64
// CHECK19-NEXT: [[KERNEL_ARGS30:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP92]], align 4
-// CHECK19-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
-// CHECK19-NEXT: store i32 4, ptr [[TMP93]], align 4
-// CHECK19-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP85]], ptr [[TMP94]], align 4
-// CHECK19-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP86]], ptr [[TMP95]], align 4
-// CHECK19-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
-// CHECK19-NEXT: store ptr [[TMP87]], ptr [[TMP96]], align 4
-// CHECK19-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP97]], align 4
-// CHECK19-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP98]], align 4
-// CHECK19-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP99]], align 4
-// CHECK19-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
-// CHECK19-NEXT: store i64 [[TMP91]], ptr [[TMP100]], align 8
-// CHECK19-NEXT: [[TMP101:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l164.region_id, ptr [[KERNEL_ARGS30]])
-// CHECK19-NEXT: [[TMP102:%.*]] = icmp ne i32 [[TMP101]], 0
-// CHECK19-NEXT: br i1 [[TMP102]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
+// CHECK19-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP100]], align 4
+// CHECK19-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
+// CHECK19-NEXT: store i32 4, ptr [[TMP101]], align 4
+// CHECK19-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP93]], ptr [[TMP102]], align 4
+// CHECK19-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP94]], ptr [[TMP103]], align 4
+// CHECK19-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
+// CHECK19-NEXT: store ptr [[TMP95]], ptr [[TMP104]], align 4
+// CHECK19-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP105]], align 4
+// CHECK19-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP106]], align 4
+// CHECK19-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP107]], align 4
+// CHECK19-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
+// CHECK19-NEXT: store i64 [[TMP99]], ptr [[TMP108]], align 8
+// CHECK19-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP109]], align 8
+// CHECK19-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP110]], align 4
+// CHECK19-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP111]], align 4
+// CHECK19-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP112]], align 4
+// CHECK19-NEXT: [[TMP113:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l164.region_id, ptr [[KERNEL_ARGS30]])
+// CHECK19-NEXT: [[TMP114:%.*]] = icmp ne i32 [[TMP113]], 0
+// CHECK19-NEXT: br i1 [[TMP114]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
// CHECK19: omp_offload.failed31:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l164(i32 [[TMP67]], i32 [[TMP69]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l164(i32 [[TMP75]], i32 [[TMP77]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT32]]
// CHECK19: omp_offload.cont32:
-// CHECK19-NEXT: [[TMP103:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP103]], ptr [[N_CASTED33]], align 4
-// CHECK19-NEXT: [[TMP104:%.*]] = load i32, ptr [[N_CASTED33]], align 4
-// CHECK19-NEXT: [[TMP105:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK19-NEXT: [[TMP106:%.*]] = sext i32 [[TMP105]] to i64
+// CHECK19-NEXT: [[TMP115:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP115]], ptr [[N_CASTED33]], align 4
+// CHECK19-NEXT: [[TMP116:%.*]] = load i32, ptr [[N_CASTED33]], align 4
+// CHECK19-NEXT: [[TMP117:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK19-NEXT: [[TMP118:%.*]] = sext i32 [[TMP117]] to i64
// CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES37]], ptr align 4 @.offload_sizes.12, i32 24, i1 false)
-// CHECK19-NEXT: [[TMP107:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP104]], ptr [[TMP107]], align 4
-// CHECK19-NEXT: [[TMP108:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP104]], ptr [[TMP108]], align 4
-// CHECK19-NEXT: [[TMP109:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP109]], align 4
-// CHECK19-NEXT: [[TMP110:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP110]], align 4
-// CHECK19-NEXT: [[TMP111:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP111]], align 4
-// CHECK19-NEXT: [[TMP112:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i32 0, i32 1
-// CHECK19-NEXT: store ptr null, ptr [[TMP112]], align 4
-// CHECK19-NEXT: [[TMP113:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP113]], align 4
-// CHECK19-NEXT: [[TMP114:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP114]], align 4
-// CHECK19-NEXT: [[TMP115:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 2
-// CHECK19-NEXT: store i64 [[TMP106]], ptr [[TMP115]], align 4
-// CHECK19-NEXT: [[TMP116:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i32 0, i32 2
-// CHECK19-NEXT: store ptr null, ptr [[TMP116]], align 4
-// CHECK19-NEXT: [[TMP117:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP118:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP119:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP120:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP120]], ptr [[DOTCAPTURE_EXPR_39]], align 4
-// CHECK19-NEXT: [[TMP121:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_39]], align 4
-// CHECK19-NEXT: [[SUB41:%.*]] = sub nsw i32 [[TMP121]], 0
+// CHECK19-NEXT: [[TMP119:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP116]], ptr [[TMP119]], align 4
+// CHECK19-NEXT: [[TMP120:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP116]], ptr [[TMP120]], align 4
+// CHECK19-NEXT: [[TMP121:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP121]], align 4
+// CHECK19-NEXT: [[TMP122:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP122]], align 4
+// CHECK19-NEXT: [[TMP123:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP123]], align 4
+// CHECK19-NEXT: [[TMP124:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i32 0, i32 1
+// CHECK19-NEXT: store ptr null, ptr [[TMP124]], align 4
+// CHECK19-NEXT: [[TMP125:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP125]], align 4
+// CHECK19-NEXT: [[TMP126:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP126]], align 4
+// CHECK19-NEXT: [[TMP127:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 2
+// CHECK19-NEXT: store i64 [[TMP118]], ptr [[TMP127]], align 4
+// CHECK19-NEXT: [[TMP128:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS36]], i32 0, i32 2
+// CHECK19-NEXT: store ptr null, ptr [[TMP128]], align 4
+// CHECK19-NEXT: [[TMP129:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS34]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP130:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS35]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP131:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES37]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP132:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP132]], ptr [[DOTCAPTURE_EXPR_39]], align 4
+// CHECK19-NEXT: [[TMP133:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_39]], align 4
+// CHECK19-NEXT: [[SUB41:%.*]] = sub nsw i32 [[TMP133]], 0
// CHECK19-NEXT: [[DIV42:%.*]] = sdiv i32 [[SUB41]], 1
// CHECK19-NEXT: [[SUB43:%.*]] = sub nsw i32 [[DIV42]], 1
// CHECK19-NEXT: store i32 [[SUB43]], ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK19-NEXT: [[TMP122:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
-// CHECK19-NEXT: [[ADD44:%.*]] = add nsw i32 [[TMP122]], 1
-// CHECK19-NEXT: [[TMP123:%.*]] = zext i32 [[ADD44]] to i64
+// CHECK19-NEXT: [[TMP134:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4
+// CHECK19-NEXT: [[ADD44:%.*]] = add nsw i32 [[TMP134]], 1
+// CHECK19-NEXT: [[TMP135:%.*]] = zext i32 [[ADD44]] to i64
// CHECK19-NEXT: [[KERNEL_ARGS45:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP124:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP124]], align 4
-// CHECK19-NEXT: [[TMP125:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 1
-// CHECK19-NEXT: store i32 3, ptr [[TMP125]], align 4
-// CHECK19-NEXT: [[TMP126:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP117]], ptr [[TMP126]], align 4
-// CHECK19-NEXT: [[TMP127:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP118]], ptr [[TMP127]], align 4
-// CHECK19-NEXT: [[TMP128:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 4
-// CHECK19-NEXT: store ptr [[TMP119]], ptr [[TMP128]], align 4
-// CHECK19-NEXT: [[TMP129:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP129]], align 4
-// CHECK19-NEXT: [[TMP130:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP130]], align 4
-// CHECK19-NEXT: [[TMP131:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP131]], align 4
-// CHECK19-NEXT: [[TMP132:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 8
-// CHECK19-NEXT: store i64 [[TMP123]], ptr [[TMP132]], align 8
-// CHECK19-NEXT: [[TMP133:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l169.region_id, ptr [[KERNEL_ARGS45]])
-// CHECK19-NEXT: [[TMP134:%.*]] = icmp ne i32 [[TMP133]], 0
-// CHECK19-NEXT: br i1 [[TMP134]], label [[OMP_OFFLOAD_FAILED46:%.*]], label [[OMP_OFFLOAD_CONT47:%.*]]
+// CHECK19-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP136]], align 4
+// CHECK19-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 1
+// CHECK19-NEXT: store i32 3, ptr [[TMP137]], align 4
+// CHECK19-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP129]], ptr [[TMP138]], align 4
+// CHECK19-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP130]], ptr [[TMP139]], align 4
+// CHECK19-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 4
+// CHECK19-NEXT: store ptr [[TMP131]], ptr [[TMP140]], align 4
+// CHECK19-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.13, ptr [[TMP141]], align 4
+// CHECK19-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP142]], align 4
+// CHECK19-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP143]], align 4
+// CHECK19-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 8
+// CHECK19-NEXT: store i64 [[TMP135]], ptr [[TMP144]], align 8
+// CHECK19-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP145]], align 8
+// CHECK19-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP146]], align 4
+// CHECK19-NEXT: [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP147]], align 4
+// CHECK19-NEXT: [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS45]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP148]], align 4
+// CHECK19-NEXT: [[TMP149:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l169.region_id, ptr [[KERNEL_ARGS45]])
+// CHECK19-NEXT: [[TMP150:%.*]] = icmp ne i32 [[TMP149]], 0
+// CHECK19-NEXT: br i1 [[TMP150]], label [[OMP_OFFLOAD_FAILED46:%.*]], label [[OMP_OFFLOAD_CONT47:%.*]]
// CHECK19: omp_offload.failed46:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l169(i32 [[TMP104]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l169(i32 [[TMP116]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT47]]
// CHECK19: omp_offload.cont47:
-// CHECK19-NEXT: [[TMP135:%.*]] = load i32, ptr [[M]], align 4
-// CHECK19-NEXT: store i32 [[TMP135]], ptr [[M_CASTED48]], align 4
-// CHECK19-NEXT: [[TMP136:%.*]] = load i32, ptr [[M_CASTED48]], align 4
-// CHECK19-NEXT: [[TMP137:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP137]], ptr [[N_CASTED49]], align 4
-// CHECK19-NEXT: [[TMP138:%.*]] = load i32, ptr [[N_CASTED49]], align 4
-// CHECK19-NEXT: [[TMP139:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK19-NEXT: [[TMP140:%.*]] = sext i32 [[TMP139]] to i64
+// CHECK19-NEXT: [[TMP151:%.*]] = load i32, ptr [[M]], align 4
+// CHECK19-NEXT: store i32 [[TMP151]], ptr [[M_CASTED48]], align 4
+// CHECK19-NEXT: [[TMP152:%.*]] = load i32, ptr [[M_CASTED48]], align 4
+// CHECK19-NEXT: [[TMP153:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP153]], ptr [[N_CASTED49]], align 4
+// CHECK19-NEXT: [[TMP154:%.*]] = load i32, ptr [[N_CASTED49]], align 4
+// CHECK19-NEXT: [[TMP155:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK19-NEXT: [[TMP156:%.*]] = sext i32 [[TMP155]] to i64
// CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES53]], ptr align 4 @.offload_sizes.16, i32 32, i1 false)
-// CHECK19-NEXT: [[TMP141:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP136]], ptr [[TMP141]], align 4
-// CHECK19-NEXT: [[TMP142:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP136]], ptr [[TMP142]], align 4
-// CHECK19-NEXT: [[TMP143:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP143]], align 4
-// CHECK19-NEXT: [[TMP144:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP138]], ptr [[TMP144]], align 4
-// CHECK19-NEXT: [[TMP145:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 1
-// CHECK19-NEXT: store i32 [[TMP138]], ptr [[TMP145]], align 4
-// CHECK19-NEXT: [[TMP146:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 1
-// CHECK19-NEXT: store ptr null, ptr [[TMP146]], align 4
-// CHECK19-NEXT: [[TMP147:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 2
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP147]], align 4
-// CHECK19-NEXT: [[TMP148:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 2
-// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP148]], align 4
-// CHECK19-NEXT: [[TMP149:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 2
-// CHECK19-NEXT: store ptr null, ptr [[TMP149]], align 4
-// CHECK19-NEXT: [[TMP150:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP150]], align 4
-// CHECK19-NEXT: [[TMP151:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP151]], align 4
-// CHECK19-NEXT: [[TMP152:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 3
-// CHECK19-NEXT: store i64 [[TMP140]], ptr [[TMP152]], align 4
-// CHECK19-NEXT: [[TMP153:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 3
-// CHECK19-NEXT: store ptr null, ptr [[TMP153]], align 4
-// CHECK19-NEXT: [[TMP154:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP155:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP156:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP157:%.*]] = load i32, ptr [[N]], align 4
-// CHECK19-NEXT: store i32 [[TMP157]], ptr [[DOTCAPTURE_EXPR_55]], align 4
-// CHECK19-NEXT: [[TMP158:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_55]], align 4
-// CHECK19-NEXT: [[SUB57:%.*]] = sub nsw i32 [[TMP158]], 0
+// CHECK19-NEXT: [[TMP157:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP152]], ptr [[TMP157]], align 4
+// CHECK19-NEXT: [[TMP158:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP152]], ptr [[TMP158]], align 4
+// CHECK19-NEXT: [[TMP159:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP159]], align 4
+// CHECK19-NEXT: [[TMP160:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP154]], ptr [[TMP160]], align 4
+// CHECK19-NEXT: [[TMP161:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 1
+// CHECK19-NEXT: store i32 [[TMP154]], ptr [[TMP161]], align 4
+// CHECK19-NEXT: [[TMP162:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 1
+// CHECK19-NEXT: store ptr null, ptr [[TMP162]], align 4
+// CHECK19-NEXT: [[TMP163:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 2
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP163]], align 4
+// CHECK19-NEXT: [[TMP164:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 2
+// CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP164]], align 4
+// CHECK19-NEXT: [[TMP165:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 2
+// CHECK19-NEXT: store ptr null, ptr [[TMP165]], align 4
+// CHECK19-NEXT: [[TMP166:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP166]], align 4
+// CHECK19-NEXT: [[TMP167:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP167]], align 4
+// CHECK19-NEXT: [[TMP168:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 3
+// CHECK19-NEXT: store i64 [[TMP156]], ptr [[TMP168]], align 4
+// CHECK19-NEXT: [[TMP169:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS52]], i32 0, i32 3
+// CHECK19-NEXT: store ptr null, ptr [[TMP169]], align 4
+// CHECK19-NEXT: [[TMP170:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS50]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP171:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS51]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP172:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES53]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP173:%.*]] = load i32, ptr [[N]], align 4
+// CHECK19-NEXT: store i32 [[TMP173]], ptr [[DOTCAPTURE_EXPR_55]], align 4
+// CHECK19-NEXT: [[TMP174:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_55]], align 4
+// CHECK19-NEXT: [[SUB57:%.*]] = sub nsw i32 [[TMP174]], 0
// CHECK19-NEXT: [[DIV58:%.*]] = sdiv i32 [[SUB57]], 1
// CHECK19-NEXT: [[SUB59:%.*]] = sub nsw i32 [[DIV58]], 1
// CHECK19-NEXT: store i32 [[SUB59]], ptr [[DOTCAPTURE_EXPR_56]], align 4
-// CHECK19-NEXT: [[TMP159:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_56]], align 4
-// CHECK19-NEXT: [[ADD60:%.*]] = add nsw i32 [[TMP159]], 1
-// CHECK19-NEXT: [[TMP160:%.*]] = zext i32 [[ADD60]] to i64
+// CHECK19-NEXT: [[TMP175:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_56]], align 4
+// CHECK19-NEXT: [[ADD60:%.*]] = add nsw i32 [[TMP175]], 1
+// CHECK19-NEXT: [[TMP176:%.*]] = zext i32 [[ADD60]] to i64
// CHECK19-NEXT: [[KERNEL_ARGS61:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP161:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP161]], align 4
-// CHECK19-NEXT: [[TMP162:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 1
-// CHECK19-NEXT: store i32 4, ptr [[TMP162]], align 4
-// CHECK19-NEXT: [[TMP163:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP154]], ptr [[TMP163]], align 4
-// CHECK19-NEXT: [[TMP164:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP155]], ptr [[TMP164]], align 4
-// CHECK19-NEXT: [[TMP165:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 4
-// CHECK19-NEXT: store ptr [[TMP156]], ptr [[TMP165]], align 4
-// CHECK19-NEXT: [[TMP166:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP166]], align 4
-// CHECK19-NEXT: [[TMP167:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP167]], align 4
-// CHECK19-NEXT: [[TMP168:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP168]], align 4
-// CHECK19-NEXT: [[TMP169:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 8
-// CHECK19-NEXT: store i64 [[TMP160]], ptr [[TMP169]], align 8
-// CHECK19-NEXT: [[TMP170:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l174.region_id, ptr [[KERNEL_ARGS61]])
-// CHECK19-NEXT: [[TMP171:%.*]] = icmp ne i32 [[TMP170]], 0
-// CHECK19-NEXT: br i1 [[TMP171]], label [[OMP_OFFLOAD_FAILED62:%.*]], label [[OMP_OFFLOAD_CONT63:%.*]]
+// CHECK19-NEXT: [[TMP177:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP177]], align 4
+// CHECK19-NEXT: [[TMP178:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 1
+// CHECK19-NEXT: store i32 4, ptr [[TMP178]], align 4
+// CHECK19-NEXT: [[TMP179:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP170]], ptr [[TMP179]], align 4
+// CHECK19-NEXT: [[TMP180:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP171]], ptr [[TMP180]], align 4
+// CHECK19-NEXT: [[TMP181:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 4
+// CHECK19-NEXT: store ptr [[TMP172]], ptr [[TMP181]], align 4
+// CHECK19-NEXT: [[TMP182:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP182]], align 4
+// CHECK19-NEXT: [[TMP183:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP183]], align 4
+// CHECK19-NEXT: [[TMP184:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP184]], align 4
+// CHECK19-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 8
+// CHECK19-NEXT: store i64 [[TMP176]], ptr [[TMP185]], align 8
+// CHECK19-NEXT: [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP186]], align 8
+// CHECK19-NEXT: [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP187]], align 4
+// CHECK19-NEXT: [[TMP188:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP188]], align 4
+// CHECK19-NEXT: [[TMP189:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS61]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP189]], align 4
+// CHECK19-NEXT: [[TMP190:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l174.region_id, ptr [[KERNEL_ARGS61]])
+// CHECK19-NEXT: [[TMP191:%.*]] = icmp ne i32 [[TMP190]], 0
+// CHECK19-NEXT: br i1 [[TMP191]], label [[OMP_OFFLOAD_FAILED62:%.*]], label [[OMP_OFFLOAD_CONT63:%.*]]
// CHECK19: omp_offload.failed62:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l174(i32 [[TMP136]], i32 [[TMP138]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l174(i32 [[TMP152]], i32 [[TMP154]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT63]]
// CHECK19: omp_offload.cont63:
-// CHECK19-NEXT: [[TMP172:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK19-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP172]])
+// CHECK19-NEXT: [[TMP192:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK19-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP192]])
// CHECK19-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK19-NEXT: [[TMP173:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK19-NEXT: call void @llvm.stackrestore(ptr [[TMP173]])
-// CHECK19-NEXT: [[TMP174:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK19-NEXT: ret i32 [[TMP174]]
+// CHECK19-NEXT: [[TMP193:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK19-NEXT: call void @llvm.stackrestore(ptr [[TMP193]])
+// CHECK19-NEXT: [[TMP194:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK19-NEXT: ret i32 [[TMP194]]
//
//
// CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l154
@@ -14627,7 +15067,7 @@ int main (int argc, char **argv) {
// CHECK19-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK19-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK19-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK19-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK19-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK19-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -14644,165 +15084,205 @@ int main (int argc, char **argv) {
// CHECK19-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK19-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK19-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK19-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l122.region_id, ptr [[KERNEL_ARGS]])
-// CHECK19-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK19-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK19-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK19-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK19-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK19-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK19-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l122.region_id, ptr [[KERNEL_ARGS]])
+// CHECK19-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK19-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK19: omp_offload.failed:
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l122(ptr [[A]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK19: omp_offload.cont:
-// CHECK19-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP16]], align 4
-// CHECK19-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP17]], align 4
-// CHECK19-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK19-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP20]], align 4
+// CHECK19-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP21]], align 4
+// CHECK19-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK19-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK19-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK19-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK19-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK19-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK19-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK19-NEXT: store ptr @.offload_sizes.24, ptr [[TMP25]], align 4
-// CHECK19-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP26]], align 4
-// CHECK19-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK19-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK19-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK19-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK19-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l127.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK19-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK19-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK19-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK19-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK19-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK19-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK19-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK19-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK19-NEXT: store ptr @.offload_sizes.24, ptr [[TMP29]], align 4
+// CHECK19-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.25, ptr [[TMP30]], align 4
+// CHECK19-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK19-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK19-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK19-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK19-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK19-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK19-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4
+// CHECK19-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK19-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l127.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK19-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK19-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK19: omp_offload.failed6:
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l127(ptr [[A]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK19: omp_offload.cont7:
-// CHECK19-NEXT: [[TMP32:%.*]] = load i32, ptr [[M]], align 4
-// CHECK19-NEXT: store i32 [[TMP32]], ptr [[M_CASTED]], align 4
-// CHECK19-NEXT: [[TMP33:%.*]] = load i32, ptr [[M_CASTED]], align 4
-// CHECK19-NEXT: [[TMP34:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP33]], ptr [[TMP34]], align 4
-// CHECK19-NEXT: [[TMP35:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP33]], ptr [[TMP35]], align 4
-// CHECK19-NEXT: [[TMP36:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP36]], align 4
-// CHECK19-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP37]], align 4
-// CHECK19-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP38]], align 4
-// CHECK19-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1
-// CHECK19-NEXT: store ptr null, ptr [[TMP39]], align 4
-// CHECK19-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP40:%.*]] = load i32, ptr [[M]], align 4
+// CHECK19-NEXT: store i32 [[TMP40]], ptr [[M_CASTED]], align 4
+// CHECK19-NEXT: [[TMP41:%.*]] = load i32, ptr [[M_CASTED]], align 4
+// CHECK19-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP41]], ptr [[TMP42]], align 4
+// CHECK19-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP41]], ptr [[TMP43]], align 4
+// CHECK19-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP44]], align 4
+// CHECK19-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP45]], align 4
+// CHECK19-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP46]], align 4
+// CHECK19-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1
+// CHECK19-NEXT: store ptr null, ptr [[TMP47]], align 4
+// CHECK19-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP42]], align 4
-// CHECK19-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK19-NEXT: store i32 2, ptr [[TMP43]], align 4
-// CHECK19-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP40]], ptr [[TMP44]], align 4
-// CHECK19-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP41]], ptr [[TMP45]], align 4
-// CHECK19-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK19-NEXT: store ptr @.offload_sizes.28, ptr [[TMP46]], align 4
-// CHECK19-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP47]], align 4
-// CHECK19-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP48]], align 4
-// CHECK19-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP49]], align 4
-// CHECK19-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK19-NEXT: store i64 10, ptr [[TMP50]], align 8
-// CHECK19-NEXT: [[TMP51:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l132.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK19-NEXT: [[TMP52:%.*]] = icmp ne i32 [[TMP51]], 0
-// CHECK19-NEXT: br i1 [[TMP52]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK19-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP50]], align 4
+// CHECK19-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK19-NEXT: store i32 2, ptr [[TMP51]], align 4
+// CHECK19-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP48]], ptr [[TMP52]], align 4
+// CHECK19-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 4
+// CHECK19-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK19-NEXT: store ptr @.offload_sizes.28, ptr [[TMP54]], align 4
+// CHECK19-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.29, ptr [[TMP55]], align 4
+// CHECK19-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP56]], align 4
+// CHECK19-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP57]], align 4
+// CHECK19-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK19-NEXT: store i64 10, ptr [[TMP58]], align 8
+// CHECK19-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP59]], align 8
+// CHECK19-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP60]], align 4
+// CHECK19-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4
+// CHECK19-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP62]], align 4
+// CHECK19-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l132.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK19-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
+// CHECK19-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK19: omp_offload.failed13:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l132(i32 [[TMP33]], ptr [[A]]) #[[ATTR3]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l132(i32 [[TMP41]], ptr [[A]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT14]]
// CHECK19: omp_offload.cont14:
-// CHECK19-NEXT: [[TMP53:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP53]], align 4
-// CHECK19-NEXT: [[TMP54:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP54]], align 4
-// CHECK19-NEXT: [[TMP55:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP55]], align 4
-// CHECK19-NEXT: [[TMP56:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP57:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP65:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP65]], align 4
+// CHECK19-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP66]], align 4
+// CHECK19-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP67]], align 4
+// CHECK19-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP69:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS19:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP58]], align 4
-// CHECK19-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
-// CHECK19-NEXT: store i32 1, ptr [[TMP59]], align 4
-// CHECK19-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP56]], ptr [[TMP60]], align 4
-// CHECK19-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP57]], ptr [[TMP61]], align 4
-// CHECK19-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
-// CHECK19-NEXT: store ptr @.offload_sizes.32, ptr [[TMP62]], align 4
-// CHECK19-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP63]], align 4
-// CHECK19-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP64]], align 4
-// CHECK19-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP65]], align 4
-// CHECK19-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
-// CHECK19-NEXT: store i64 10, ptr [[TMP66]], align 8
-// CHECK19-NEXT: [[TMP67:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l137.region_id, ptr [[KERNEL_ARGS19]])
-// CHECK19-NEXT: [[TMP68:%.*]] = icmp ne i32 [[TMP67]], 0
-// CHECK19-NEXT: br i1 [[TMP68]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
+// CHECK19-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP70]], align 4
+// CHECK19-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1
+// CHECK19-NEXT: store i32 1, ptr [[TMP71]], align 4
+// CHECK19-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP68]], ptr [[TMP72]], align 4
+// CHECK19-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP69]], ptr [[TMP73]], align 4
+// CHECK19-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4
+// CHECK19-NEXT: store ptr @.offload_sizes.32, ptr [[TMP74]], align 4
+// CHECK19-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.33, ptr [[TMP75]], align 4
+// CHECK19-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP76]], align 4
+// CHECK19-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP77]], align 4
+// CHECK19-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8
+// CHECK19-NEXT: store i64 10, ptr [[TMP78]], align 8
+// CHECK19-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP79]], align 8
+// CHECK19-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP80]], align 4
+// CHECK19-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP81]], align 4
+// CHECK19-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP82]], align 4
+// CHECK19-NEXT: [[TMP83:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l137.region_id, ptr [[KERNEL_ARGS19]])
+// CHECK19-NEXT: [[TMP84:%.*]] = icmp ne i32 [[TMP83]], 0
+// CHECK19-NEXT: br i1 [[TMP84]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]]
// CHECK19: omp_offload.failed20:
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l137(ptr [[A]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT21]]
// CHECK19: omp_offload.cont21:
-// CHECK19-NEXT: [[TMP69:%.*]] = load i32, ptr [[M]], align 4
-// CHECK19-NEXT: store i32 [[TMP69]], ptr [[M_CASTED22]], align 4
-// CHECK19-NEXT: [[TMP70:%.*]] = load i32, ptr [[M_CASTED22]], align 4
-// CHECK19-NEXT: [[TMP71:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP70]], ptr [[TMP71]], align 4
-// CHECK19-NEXT: [[TMP72:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
-// CHECK19-NEXT: store i32 [[TMP70]], ptr [[TMP72]], align 4
-// CHECK19-NEXT: [[TMP73:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i32 0, i32 0
-// CHECK19-NEXT: store ptr null, ptr [[TMP73]], align 4
-// CHECK19-NEXT: [[TMP74:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 1
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP74]], align 4
-// CHECK19-NEXT: [[TMP75:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 1
-// CHECK19-NEXT: store ptr [[A]], ptr [[TMP75]], align 4
-// CHECK19-NEXT: [[TMP76:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i32 0, i32 1
-// CHECK19-NEXT: store ptr null, ptr [[TMP76]], align 4
-// CHECK19-NEXT: [[TMP77:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP78:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP85:%.*]] = load i32, ptr [[M]], align 4
+// CHECK19-NEXT: store i32 [[TMP85]], ptr [[M_CASTED22]], align 4
+// CHECK19-NEXT: [[TMP86:%.*]] = load i32, ptr [[M_CASTED22]], align 4
+// CHECK19-NEXT: [[TMP87:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP86]], ptr [[TMP87]], align 4
+// CHECK19-NEXT: [[TMP88:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
+// CHECK19-NEXT: store i32 [[TMP86]], ptr [[TMP88]], align 4
+// CHECK19-NEXT: [[TMP89:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i32 0, i32 0
+// CHECK19-NEXT: store ptr null, ptr [[TMP89]], align 4
+// CHECK19-NEXT: [[TMP90:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 1
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP90]], align 4
+// CHECK19-NEXT: [[TMP91:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 1
+// CHECK19-NEXT: store ptr [[A]], ptr [[TMP91]], align 4
+// CHECK19-NEXT: [[TMP92:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS25]], i32 0, i32 1
+// CHECK19-NEXT: store ptr null, ptr [[TMP92]], align 4
+// CHECK19-NEXT: [[TMP93:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS23]], i32 0, i32 0
+// CHECK19-NEXT: [[TMP94:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS24]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS27:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK19-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP79]], align 4
-// CHECK19-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 1
-// CHECK19-NEXT: store i32 2, ptr [[TMP80]], align 4
-// CHECK19-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 2
-// CHECK19-NEXT: store ptr [[TMP77]], ptr [[TMP81]], align 4
-// CHECK19-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 3
-// CHECK19-NEXT: store ptr [[TMP78]], ptr [[TMP82]], align 4
-// CHECK19-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 4
-// CHECK19-NEXT: store ptr @.offload_sizes.36, ptr [[TMP83]], align 4
-// CHECK19-NEXT: [[TMP84:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 5
-// CHECK19-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP84]], align 4
-// CHECK19-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 6
-// CHECK19-NEXT: store ptr null, ptr [[TMP85]], align 4
-// CHECK19-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 7
-// CHECK19-NEXT: store ptr null, ptr [[TMP86]], align 4
-// CHECK19-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 8
-// CHECK19-NEXT: store i64 10, ptr [[TMP87]], align 8
-// CHECK19-NEXT: [[TMP88:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l142.region_id, ptr [[KERNEL_ARGS27]])
-// CHECK19-NEXT: [[TMP89:%.*]] = icmp ne i32 [[TMP88]], 0
-// CHECK19-NEXT: br i1 [[TMP89]], label [[OMP_OFFLOAD_FAILED28:%.*]], label [[OMP_OFFLOAD_CONT29:%.*]]
+// CHECK19-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 0
+// CHECK19-NEXT: store i32 2, ptr [[TMP95]], align 4
+// CHECK19-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 1
+// CHECK19-NEXT: store i32 2, ptr [[TMP96]], align 4
+// CHECK19-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 2
+// CHECK19-NEXT: store ptr [[TMP93]], ptr [[TMP97]], align 4
+// CHECK19-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 3
+// CHECK19-NEXT: store ptr [[TMP94]], ptr [[TMP98]], align 4
+// CHECK19-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 4
+// CHECK19-NEXT: store ptr @.offload_sizes.36, ptr [[TMP99]], align 4
+// CHECK19-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 5
+// CHECK19-NEXT: store ptr @.offload_maptypes.37, ptr [[TMP100]], align 4
+// CHECK19-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 6
+// CHECK19-NEXT: store ptr null, ptr [[TMP101]], align 4
+// CHECK19-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 7
+// CHECK19-NEXT: store ptr null, ptr [[TMP102]], align 4
+// CHECK19-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 8
+// CHECK19-NEXT: store i64 10, ptr [[TMP103]], align 8
+// CHECK19-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP104]], align 8
+// CHECK19-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP105]], align 4
+// CHECK19-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP106]], align 4
+// CHECK19-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS27]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP107]], align 4
+// CHECK19-NEXT: [[TMP108:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l142.region_id, ptr [[KERNEL_ARGS27]])
+// CHECK19-NEXT: [[TMP109:%.*]] = icmp ne i32 [[TMP108]], 0
+// CHECK19-NEXT: br i1 [[TMP109]], label [[OMP_OFFLOAD_FAILED28:%.*]], label [[OMP_OFFLOAD_CONT29:%.*]]
// CHECK19: omp_offload.failed28:
-// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l142(i32 [[TMP70]], ptr [[A]]) #[[ATTR3]]
+// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l142(i32 [[TMP86]], ptr [[A]]) #[[ATTR3]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT29]]
// CHECK19: omp_offload.cont29:
// CHECK19-NEXT: ret i32 0
diff --git a/clang/test/OpenMP/teams_distribute_private_codegen.cpp b/clang/test/OpenMP/teams_distribute_private_codegen.cpp
index 28a91e7df7f16..e01cf84087b9f 100644
--- a/clang/test/OpenMP/teams_distribute_private_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_private_codegen.cpp
@@ -243,7 +243,7 @@ int main() {
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -260,9 +260,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l93.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l93.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l93() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -404,7 +412,7 @@ int main() {
// CHECK1-NEXT: store ptr undef, ptr [[_TMP1]], align 8
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -421,27 +429,35 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK1-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK1: arraydestroy.body:
-// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK1-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK1-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK1: arraydestroy.done2:
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK1-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK1-NEXT: ret i32 [[TMP12]]
+// CHECK1-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK1-NEXT: ret i32 [[TMP16]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN1SIiEC1Ev
@@ -762,7 +778,7 @@ int main() {
// CHECK3-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -779,9 +795,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l93.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK3-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l93.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK3-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l93() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -921,7 +945,7 @@ int main() {
// CHECK3-NEXT: store ptr undef, ptr [[_TMP1]], align 4
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -938,27 +962,35 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK3-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK3-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK3-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK3-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK3: arraydestroy.body:
-// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK3-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK3-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK3: arraydestroy.done2:
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK3-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK3-NEXT: ret i32 [[TMP12]]
+// CHECK3-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK3-NEXT: ret i32 [[TMP16]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN1SIiEC1Ev
diff --git a/clang/test/OpenMP/teams_distribute_reduction_codegen.cpp b/clang/test/OpenMP/teams_distribute_reduction_codegen.cpp
index 017f9ab23a556..9659ee44f480b 100644
--- a/clang/test/OpenMP/teams_distribute_reduction_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_reduction_codegen.cpp
@@ -105,7 +105,7 @@ int main() {
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -122,9 +122,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l63.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK1-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l63.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l63(i64 [[TMP1]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -277,7 +285,7 @@ int main() {
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -294,9 +302,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK1-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32(i64 [[TMP1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -453,7 +469,7 @@ int main() {
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -470,9 +486,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP14]], align 4
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP15]], align 8
-// CHECK3-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l63.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK3-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK3-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l63.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK3-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l63(i32 [[TMP1]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -625,7 +649,7 @@ int main() {
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -642,9 +666,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP14]], align 4
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP15]], align 8
-// CHECK3-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK3-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP18]], align 4
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK3-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK3-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32(i32 [[TMP1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/teams_distribute_simd_codegen.cpp b/clang/test/OpenMP/teams_distribute_simd_codegen.cpp
index 15e7ed9788a42..249b8922165e3 100644
--- a/clang/test/OpenMP/teams_distribute_simd_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_simd_codegen.cpp
@@ -295,87 +295,104 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[TMP29:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP29]], 1
// CHECK1-NEXT: [[TMP30:%.*]] = zext i32 [[ADD]] to i64
+// CHECK1-NEXT: [[TMP31:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP26]], 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP31]], align 4
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK1-NEXT: store i32 5, ptr [[TMP32]], align 4
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP24]], ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP25]], ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP35]], align 8
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP36]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP37]], align 8
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP32]], align 4
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK1-NEXT: store i32 5, ptr [[TMP33]], align 4
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP24]], ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP25]], ptr [[TMP35]], align 8
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP36]], align 8
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP37]], align 8
+// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK1-NEXT: store ptr null, ptr [[TMP38]], align 8
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK1-NEXT: store i64 [[TMP30]], ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 [[TMP26]], i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l30.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP41:%.*]] = icmp ne i32 [[TMP40]], 0
-// CHECK1-NEXT: br i1 [[TMP41]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP39]], align 8
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK1-NEXT: store i64 [[TMP30]], ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] [[TMP31]], ptr [[TMP42]], align 4
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP43]], align 4
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP44]], align 4
+// CHECK1-NEXT: [[TMP45:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 [[TMP26]], i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l30.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP46:%.*]] = icmp ne i32 [[TMP45]], 0
+// CHECK1-NEXT: br i1 [[TMP46]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l30(i64 [[TMP2]], i64 [[TMP4]], ptr @a, i64 [[TMP6]], i64 [[TMP8]]) #[[ATTR3:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
-// CHECK1-NEXT: [[TMP42:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: store i32 [[TMP42]], ptr [[N_CASTED4]], align 4
-// CHECK1-NEXT: [[TMP43:%.*]] = load i64, ptr [[N_CASTED4]], align 8
-// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP43]], ptr [[TMP44]], align 8
-// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK1-NEXT: store i64 [[TMP43]], ptr [[TMP45]], align 8
-// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP46]], align 8
-// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
-// CHECK1-NEXT: store ptr @a, ptr [[TMP47]], align 8
-// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
-// CHECK1-NEXT: store ptr @a, ptr [[TMP48]], align 8
-// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 1
-// CHECK1-NEXT: store ptr null, ptr [[TMP49]], align 8
-// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP52:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK1-NEXT: store i32 [[TMP52]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK1-NEXT: [[TMP53:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK1-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP53]], 0
+// CHECK1-NEXT: [[TMP47:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: store i32 [[TMP47]], ptr [[N_CASTED4]], align 4
+// CHECK1-NEXT: [[TMP48:%.*]] = load i64, ptr [[N_CASTED4]], align 8
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP48]], ptr [[TMP49]], align 8
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK1-NEXT: store i64 [[TMP48]], ptr [[TMP50]], align 8
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP51]], align 8
+// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
+// CHECK1-NEXT: store ptr @a, ptr [[TMP52]], align 8
+// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
+// CHECK1-NEXT: store ptr @a, ptr [[TMP53]], align 8
+// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 1
+// CHECK1-NEXT: store ptr null, ptr [[TMP54]], align 8
+// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP57:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK1-NEXT: store i32 [[TMP57]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK1-NEXT: [[TMP58:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK1-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP58]], 0
// CHECK1-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK1-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK1-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK1-NEXT: [[TMP54:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK1-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP54]], 1
-// CHECK1-NEXT: [[TMP55:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK1-NEXT: [[TMP59:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK1-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP59]], 1
+// CHECK1-NEXT: [[TMP60:%.*]] = zext i32 [[ADD14]] to i64
// CHECK1-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP56]], align 4
-// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK1-NEXT: store i32 2, ptr [[TMP57]], align 4
-// CHECK1-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP50]], ptr [[TMP58]], align 8
-// CHECK1-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP51]], ptr [[TMP59]], align 8
-// CHECK1-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.2, ptr [[TMP60]], align 8
-// CHECK1-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP61]], align 8
-// CHECK1-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP62]], align 8
-// CHECK1-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP63]], align 8
-// CHECK1-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK1-NEXT: store i64 [[TMP55]], ptr [[TMP64]], align 8
-// CHECK1-NEXT: [[TMP65:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l36.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK1-NEXT: [[TMP66:%.*]] = icmp ne i32 [[TMP65]], 0
-// CHECK1-NEXT: br i1 [[TMP66]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK1-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP61]], align 4
+// CHECK1-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK1-NEXT: store i32 2, ptr [[TMP62]], align 4
+// CHECK1-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP55]], ptr [[TMP63]], align 8
+// CHECK1-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP56]], ptr [[TMP64]], align 8
+// CHECK1-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.2, ptr [[TMP65]], align 8
+// CHECK1-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP66]], align 8
+// CHECK1-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP67]], align 8
+// CHECK1-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP68]], align 8
+// CHECK1-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK1-NEXT: store i64 [[TMP60]], ptr [[TMP69]], align 8
+// CHECK1-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP70]], align 8
+// CHECK1-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP71]], align 4
+// CHECK1-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP72]], align 4
+// CHECK1-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP73]], align 4
+// CHECK1-NEXT: [[TMP74:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l36.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK1-NEXT: [[TMP75:%.*]] = icmp ne i32 [[TMP74]], 0
+// CHECK1-NEXT: br i1 [[TMP75]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK1: omp_offload.failed16:
-// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l36(i64 [[TMP43]], ptr @a) #[[ATTR3]]
+// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l36(i64 [[TMP48]], ptr @a) #[[ATTR3]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK1: omp_offload.cont17:
-// CHECK1-NEXT: [[TMP67:%.*]] = load i32, ptr @a, align 4
-// CHECK1-NEXT: ret i32 [[TMP67]]
+// CHECK1-NEXT: [[TMP76:%.*]] = load i32, ptr @a, align 4
+// CHECK1-NEXT: ret i32 [[TMP76]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l30
@@ -716,87 +733,104 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[TMP29:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP29]], 1
// CHECK3-NEXT: [[TMP30:%.*]] = zext i32 [[ADD]] to i64
+// CHECK3-NEXT: [[TMP31:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP26]], 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP31]], align 4
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK3-NEXT: store i32 5, ptr [[TMP32]], align 4
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP24]], ptr [[TMP33]], align 4
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP25]], ptr [[TMP34]], align 4
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes, ptr [[TMP35]], align 4
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP36]], align 4
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP37]], align 4
-// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK3-NEXT: store i32 5, ptr [[TMP33]], align 4
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP24]], ptr [[TMP34]], align 4
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP25]], ptr [[TMP35]], align 4
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes, ptr [[TMP36]], align 4
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK3-NEXT: store ptr null, ptr [[TMP38]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK3-NEXT: store i64 [[TMP30]], ptr [[TMP39]], align 8
-// CHECK3-NEXT: [[TMP40:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 [[TMP26]], i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l30.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP41:%.*]] = icmp ne i32 [[TMP40]], 0
-// CHECK3-NEXT: br i1 [[TMP41]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP39]], align 4
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK3-NEXT: store i64 [[TMP30]], ptr [[TMP40]], align 8
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP41]], align 8
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] [[TMP31]], ptr [[TMP42]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP43]], align 4
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP44]], align 4
+// CHECK3-NEXT: [[TMP45:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 [[TMP26]], i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l30.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP46:%.*]] = icmp ne i32 [[TMP45]], 0
+// CHECK3-NEXT: br i1 [[TMP46]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l30(i32 [[TMP2]], i32 [[TMP4]], ptr @a, i32 [[TMP6]], i32 [[TMP8]]) #[[ATTR3:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
-// CHECK3-NEXT: [[TMP42:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: store i32 [[TMP42]], ptr [[N_CASTED4]], align 4
-// CHECK3-NEXT: [[TMP43:%.*]] = load i32, ptr [[N_CASTED4]], align 4
-// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP43]], ptr [[TMP44]], align 4
-// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK3-NEXT: store i32 [[TMP43]], ptr [[TMP45]], align 4
-// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP46]], align 4
-// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
-// CHECK3-NEXT: store ptr @a, ptr [[TMP47]], align 4
-// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
-// CHECK3-NEXT: store ptr @a, ptr [[TMP48]], align 4
-// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 1
-// CHECK3-NEXT: store ptr null, ptr [[TMP49]], align 4
-// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP52:%.*]] = load i32, ptr [[N_ADDR]], align 4
-// CHECK3-NEXT: store i32 [[TMP52]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK3-NEXT: [[TMP53:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK3-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP53]], 0
+// CHECK3-NEXT: [[TMP47:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: store i32 [[TMP47]], ptr [[N_CASTED4]], align 4
+// CHECK3-NEXT: [[TMP48:%.*]] = load i32, ptr [[N_CASTED4]], align 4
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP48]], ptr [[TMP49]], align 4
+// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK3-NEXT: store i32 [[TMP48]], ptr [[TMP50]], align 4
+// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
+// CHECK3-NEXT: store ptr @a, ptr [[TMP52]], align 4
+// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
+// CHECK3-NEXT: store ptr @a, ptr [[TMP53]], align 4
+// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 1
+// CHECK3-NEXT: store ptr null, ptr [[TMP54]], align 4
+// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP57:%.*]] = load i32, ptr [[N_ADDR]], align 4
+// CHECK3-NEXT: store i32 [[TMP57]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK3-NEXT: [[TMP58:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK3-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP58]], 0
// CHECK3-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK3-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK3-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK3-NEXT: [[TMP54:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK3-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP54]], 1
-// CHECK3-NEXT: [[TMP55:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK3-NEXT: [[TMP59:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK3-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP59]], 1
+// CHECK3-NEXT: [[TMP60:%.*]] = zext i32 [[ADD14]] to i64
// CHECK3-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP56]], align 4
-// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK3-NEXT: store i32 2, ptr [[TMP57]], align 4
-// CHECK3-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP50]], ptr [[TMP58]], align 4
-// CHECK3-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP51]], ptr [[TMP59]], align 4
-// CHECK3-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.2, ptr [[TMP60]], align 4
-// CHECK3-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP61]], align 4
-// CHECK3-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP62]], align 4
-// CHECK3-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP63]], align 4
-// CHECK3-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK3-NEXT: store i64 [[TMP55]], ptr [[TMP64]], align 8
-// CHECK3-NEXT: [[TMP65:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l36.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK3-NEXT: [[TMP66:%.*]] = icmp ne i32 [[TMP65]], 0
-// CHECK3-NEXT: br i1 [[TMP66]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK3-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP61]], align 4
+// CHECK3-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK3-NEXT: store i32 2, ptr [[TMP62]], align 4
+// CHECK3-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP55]], ptr [[TMP63]], align 4
+// CHECK3-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP56]], ptr [[TMP64]], align 4
+// CHECK3-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.2, ptr [[TMP65]], align 4
+// CHECK3-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP66]], align 4
+// CHECK3-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP67]], align 4
+// CHECK3-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP68]], align 4
+// CHECK3-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK3-NEXT: store i64 [[TMP60]], ptr [[TMP69]], align 8
+// CHECK3-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP70]], align 8
+// CHECK3-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP71]], align 4
+// CHECK3-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP72]], align 4
+// CHECK3-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP73]], align 4
+// CHECK3-NEXT: [[TMP74:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l36.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK3-NEXT: [[TMP75:%.*]] = icmp ne i32 [[TMP74]], 0
+// CHECK3-NEXT: br i1 [[TMP75]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK3: omp_offload.failed16:
-// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l36(i32 [[TMP43]], ptr @a) #[[ATTR3]]
+// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l36(i32 [[TMP48]], ptr @a) #[[ATTR3]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK3: omp_offload.cont17:
-// CHECK3-NEXT: [[TMP67:%.*]] = load i32, ptr @a, align 4
-// CHECK3-NEXT: ret i32 [[TMP67]]
+// CHECK3-NEXT: [[TMP76:%.*]] = load i32, ptr @a, align 4
+// CHECK3-NEXT: ret i32 [[TMP76]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z21teams_argument_globali_l30
@@ -1380,7 +1414,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1397,18 +1431,26 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK9-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l75.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK9-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP34]], align 4
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l75.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK9-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l75(i64 [[TMP4]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[VLA]], i64 0
-// CHECK9-NEXT: [[TMP34:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK9-NEXT: [[TMP35:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP35]])
-// CHECK9-NEXT: ret i32 [[TMP34]]
+// CHECK9-NEXT: [[TMP38:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK9-NEXT: [[TMP39:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP39]])
+// CHECK9-NEXT: ret i32 [[TMP38]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l75
@@ -1601,7 +1643,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1618,18 +1660,26 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK11-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l75.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK11-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP34]], align 4
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l75.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK11-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l75(i32 [[TMP3]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[VLA]], i32 0
-// CHECK11-NEXT: [[TMP34:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK11-NEXT: [[TMP35:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP35]])
-// CHECK11-NEXT: ret i32 [[TMP34]]
+// CHECK11-NEXT: [[TMP38:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK11-NEXT: [[TMP39:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP39]])
+// CHECK11-NEXT: ret i32 [[TMP38]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z15teams_local_argv_l75
@@ -1964,7 +2014,7 @@ int main (int argc, char **argv) {
// CHECK17-NEXT: [[TMP17:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK17-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK17-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK17-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK17-NEXT: store i32 3, ptr [[TMP19]], align 4
// CHECK17-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1981,17 +2031,25 @@ int main (int argc, char **argv) {
// CHECK17-NEXT: store ptr null, ptr [[TMP25]], align 8
// CHECK17-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK17-NEXT: store i64 123, ptr [[TMP26]], align 8
-// CHECK17-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l123.region_id, ptr [[KERNEL_ARGS]])
-// CHECK17-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK17-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK17-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK17-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK17-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP29]], align 4
+// CHECK17-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK17-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l123.region_id, ptr [[KERNEL_ARGS]])
+// CHECK17-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK17-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK17: omp_offload.failed:
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l123(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK17: omp_offload.cont:
// CHECK17-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK17-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A2]], i64 0, i64 0
-// CHECK17-NEXT: [[TMP29:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK17-NEXT: ret i32 [[TMP29]]
+// CHECK17-NEXT: [[TMP33:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK17-NEXT: ret i32 [[TMP33]]
//
//
// CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l123
@@ -2141,7 +2199,7 @@ int main (int argc, char **argv) {
// CHECK19-NEXT: [[TMP17:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK19-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK19-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK19-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK19-NEXT: store i32 3, ptr [[TMP19]], align 4
// CHECK19-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2158,17 +2216,25 @@ int main (int argc, char **argv) {
// CHECK19-NEXT: store ptr null, ptr [[TMP25]], align 4
// CHECK19-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK19-NEXT: store i64 123, ptr [[TMP26]], align 8
-// CHECK19-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l123.region_id, ptr [[KERNEL_ARGS]])
-// CHECK19-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK19-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK19-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK19-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK19-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP29]], align 4
+// CHECK19-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK19-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l123.region_id, ptr [[KERNEL_ARGS]])
+// CHECK19-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK19-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK19: omp_offload.failed:
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l123(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK19: omp_offload.cont:
// CHECK19-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK19-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A2]], i32 0, i32 0
-// CHECK19-NEXT: [[TMP29:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK19-NEXT: ret i32 [[TMP29]]
+// CHECK19-NEXT: [[TMP33:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK19-NEXT: ret i32 [[TMP33]]
//
//
// CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l123
@@ -2317,7 +2383,7 @@ int main (int argc, char **argv) {
// CHECK21-NEXT: [[TMP17:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK21-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK21-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK21-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK21-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK21-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK21-NEXT: store i32 3, ptr [[TMP19]], align 4
// CHECK21-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2334,17 +2400,25 @@ int main (int argc, char **argv) {
// CHECK21-NEXT: store ptr null, ptr [[TMP25]], align 8
// CHECK21-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK21-NEXT: store i64 123, ptr [[TMP26]], align 8
-// CHECK21-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l123.region_id, ptr [[KERNEL_ARGS]])
-// CHECK21-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK21-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK21-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK21-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK21-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK21-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK21-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK21-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP29]], align 4
+// CHECK21-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK21-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK21-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l123.region_id, ptr [[KERNEL_ARGS]])
+// CHECK21-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK21-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK21: omp_offload.failed:
// CHECK21-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l123(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK21-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK21: omp_offload.cont:
// CHECK21-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK21-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A2]], i64 0, i64 0
-// CHECK21-NEXT: [[TMP29:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK21-NEXT: ret i32 [[TMP29]]
+// CHECK21-NEXT: [[TMP33:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK21-NEXT: ret i32 [[TMP33]]
//
//
// CHECK21-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l123
@@ -2544,7 +2618,7 @@ int main (int argc, char **argv) {
// CHECK23-NEXT: [[TMP17:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK23-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK23-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK23-NEXT: store i32 1, ptr [[TMP18]], align 4
+// CHECK23-NEXT: store i32 2, ptr [[TMP18]], align 4
// CHECK23-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK23-NEXT: store i32 3, ptr [[TMP19]], align 4
// CHECK23-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2561,17 +2635,25 @@ int main (int argc, char **argv) {
// CHECK23-NEXT: store ptr null, ptr [[TMP25]], align 4
// CHECK23-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK23-NEXT: store i64 123, ptr [[TMP26]], align 8
-// CHECK23-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l123.region_id, ptr [[KERNEL_ARGS]])
-// CHECK23-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
-// CHECK23-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK23-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK23-NEXT: store i64 0, ptr [[TMP27]], align 8
+// CHECK23-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK23-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
+// CHECK23-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK23-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP29]], align 4
+// CHECK23-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK23-NEXT: store i32 0, ptr [[TMP30]], align 4
+// CHECK23-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l123.region_id, ptr [[KERNEL_ARGS]])
+// CHECK23-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
+// CHECK23-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK23: omp_offload.failed:
// CHECK23-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l123(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK23-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK23: omp_offload.cont:
// CHECK23-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK23-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A2]], i32 0, i32 0
-// CHECK23-NEXT: [[TMP29:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK23-NEXT: ret i32 [[TMP29]]
+// CHECK23-NEXT: [[TMP33:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK23-NEXT: ret i32 [[TMP33]]
//
//
// CHECK23-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l123
@@ -3097,7 +3179,7 @@ int main (int argc, char **argv) {
// CHECK33-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK33-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK33-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK33-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK33-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK33-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK33-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK33-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3114,20 +3196,28 @@ int main (int argc, char **argv) {
// CHECK33-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK33-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK33-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK33-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l192.region_id, ptr [[KERNEL_ARGS]])
-// CHECK33-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK33-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK33-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK33-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK33-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK33-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK33-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK33-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP34]], align 4
+// CHECK33-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK33-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK33-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l192.region_id, ptr [[KERNEL_ARGS]])
+// CHECK33-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK33-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK33: omp_offload.failed:
// CHECK33-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l192(i64 [[TMP4]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK33-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK33: omp_offload.cont:
-// CHECK33-NEXT: [[TMP34:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK33-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP34]])
+// CHECK33-NEXT: [[TMP38:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK33-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP38]])
// CHECK33-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK33-NEXT: [[TMP35:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK33-NEXT: call void @llvm.stackrestore(ptr [[TMP35]])
-// CHECK33-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK33-NEXT: ret i32 [[TMP36]]
+// CHECK33-NEXT: [[TMP39:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK33-NEXT: call void @llvm.stackrestore(ptr [[TMP39]])
+// CHECK33-NEXT: [[TMP40:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK33-NEXT: ret i32 [[TMP40]]
//
//
// CHECK33-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l192
@@ -3295,28 +3385,37 @@ int main (int argc, char **argv) {
// CHECK33-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK33-NEXT: [[TMP14:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK33-NEXT: [[TMP15:%.*]] = load i32, ptr [[TE]], align 4
+// CHECK33-NEXT: [[TMP16:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP15]], 0
// CHECK33-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK33-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK33-NEXT: store i32 1, ptr [[TMP16]], align 4
-// CHECK33-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK33-NEXT: store i32 3, ptr [[TMP17]], align 4
-// CHECK33-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK33-NEXT: store ptr [[TMP13]], ptr [[TMP18]], align 8
-// CHECK33-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK33-NEXT: store ptr [[TMP14]], ptr [[TMP19]], align 8
-// CHECK33-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK33-NEXT: store ptr @.offload_sizes.2, ptr [[TMP20]], align 8
-// CHECK33-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK33-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP21]], align 8
-// CHECK33-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK33-NEXT: store ptr null, ptr [[TMP22]], align 8
-// CHECK33-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK33-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK33-NEXT: store i32 2, ptr [[TMP17]], align 4
+// CHECK33-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK33-NEXT: store i32 3, ptr [[TMP18]], align 4
+// CHECK33-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK33-NEXT: store ptr [[TMP13]], ptr [[TMP19]], align 8
+// CHECK33-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK33-NEXT: store ptr [[TMP14]], ptr [[TMP20]], align 8
+// CHECK33-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK33-NEXT: store ptr @.offload_sizes.2, ptr [[TMP21]], align 8
+// CHECK33-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK33-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP22]], align 8
+// CHECK33-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK33-NEXT: store ptr null, ptr [[TMP23]], align 8
-// CHECK33-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK33-NEXT: store i64 10, ptr [[TMP24]], align 8
-// CHECK33-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 [[TMP15]], i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l181.region_id, ptr [[KERNEL_ARGS]])
-// CHECK33-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK33-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK33-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK33-NEXT: store ptr null, ptr [[TMP24]], align 8
+// CHECK33-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK33-NEXT: store i64 10, ptr [[TMP25]], align 8
+// CHECK33-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK33-NEXT: store i64 0, ptr [[TMP26]], align 8
+// CHECK33-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK33-NEXT: store [3 x i32] [[TMP16]], ptr [[TMP27]], align 4
+// CHECK33-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK33-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP28]], align 4
+// CHECK33-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK33-NEXT: store i32 0, ptr [[TMP29]], align 4
+// CHECK33-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 [[TMP15]], i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l181.region_id, ptr [[KERNEL_ARGS]])
+// CHECK33-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
+// CHECK33-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK33: omp_offload.failed:
// CHECK33-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l181(i64 [[TMP1]], i64 [[TMP3]], ptr [[A]]) #[[ATTR3]]
// CHECK33-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -3490,7 +3589,7 @@ int main (int argc, char **argv) {
// CHECK35-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK35-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK35-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK35-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK35-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK35-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK35-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK35-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3507,20 +3606,28 @@ int main (int argc, char **argv) {
// CHECK35-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK35-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK35-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK35-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l192.region_id, ptr [[KERNEL_ARGS]])
-// CHECK35-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK35-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK35-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK35-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK35-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK35-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK35-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK35-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP34]], align 4
+// CHECK35-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK35-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK35-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l192.region_id, ptr [[KERNEL_ARGS]])
+// CHECK35-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK35-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK35: omp_offload.failed:
// CHECK35-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l192(i32 [[TMP3]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK35-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK35: omp_offload.cont:
-// CHECK35-NEXT: [[TMP34:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK35-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP34]])
+// CHECK35-NEXT: [[TMP38:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK35-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP38]])
// CHECK35-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK35-NEXT: [[TMP35:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK35-NEXT: call void @llvm.stackrestore(ptr [[TMP35]])
-// CHECK35-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK35-NEXT: ret i32 [[TMP36]]
+// CHECK35-NEXT: [[TMP39:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK35-NEXT: call void @llvm.stackrestore(ptr [[TMP39]])
+// CHECK35-NEXT: [[TMP40:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK35-NEXT: ret i32 [[TMP40]]
//
//
// CHECK35-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l192
@@ -3687,28 +3794,37 @@ int main (int argc, char **argv) {
// CHECK35-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK35-NEXT: [[TMP14:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK35-NEXT: [[TMP15:%.*]] = load i32, ptr [[TE]], align 4
+// CHECK35-NEXT: [[TMP16:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP15]], 0
// CHECK35-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK35-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK35-NEXT: store i32 1, ptr [[TMP16]], align 4
-// CHECK35-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK35-NEXT: store i32 3, ptr [[TMP17]], align 4
-// CHECK35-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK35-NEXT: store ptr [[TMP13]], ptr [[TMP18]], align 4
-// CHECK35-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK35-NEXT: store ptr [[TMP14]], ptr [[TMP19]], align 4
-// CHECK35-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK35-NEXT: store ptr @.offload_sizes.2, ptr [[TMP20]], align 4
-// CHECK35-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK35-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP21]], align 4
-// CHECK35-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK35-NEXT: store ptr null, ptr [[TMP22]], align 4
-// CHECK35-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK35-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK35-NEXT: store i32 2, ptr [[TMP17]], align 4
+// CHECK35-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK35-NEXT: store i32 3, ptr [[TMP18]], align 4
+// CHECK35-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK35-NEXT: store ptr [[TMP13]], ptr [[TMP19]], align 4
+// CHECK35-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK35-NEXT: store ptr [[TMP14]], ptr [[TMP20]], align 4
+// CHECK35-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK35-NEXT: store ptr @.offload_sizes.2, ptr [[TMP21]], align 4
+// CHECK35-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK35-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP22]], align 4
+// CHECK35-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK35-NEXT: store ptr null, ptr [[TMP23]], align 4
-// CHECK35-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK35-NEXT: store i64 10, ptr [[TMP24]], align 8
-// CHECK35-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 [[TMP15]], i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l181.region_id, ptr [[KERNEL_ARGS]])
-// CHECK35-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK35-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK35-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK35-NEXT: store ptr null, ptr [[TMP24]], align 4
+// CHECK35-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK35-NEXT: store i64 10, ptr [[TMP25]], align 8
+// CHECK35-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK35-NEXT: store i64 0, ptr [[TMP26]], align 8
+// CHECK35-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK35-NEXT: store [3 x i32] [[TMP16]], ptr [[TMP27]], align 4
+// CHECK35-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK35-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP28]], align 4
+// CHECK35-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK35-NEXT: store i32 0, ptr [[TMP29]], align 4
+// CHECK35-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 [[TMP15]], i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l181.region_id, ptr [[KERNEL_ARGS]])
+// CHECK35-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
+// CHECK35-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK35: omp_offload.failed:
// CHECK35-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l181(i32 [[TMP1]], i32 [[TMP3]], ptr [[A]]) #[[ATTR3]]
// CHECK35-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -3891,7 +4007,7 @@ int main (int argc, char **argv) {
// CHECK37-NEXT: [[TMP27:%.*]] = zext i32 [[ADD]] to i64
// CHECK37-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK37-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK37-NEXT: store i32 1, ptr [[TMP28]], align 4
+// CHECK37-NEXT: store i32 2, ptr [[TMP28]], align 4
// CHECK37-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK37-NEXT: store i32 4, ptr [[TMP29]], align 4
// CHECK37-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -3908,20 +4024,28 @@ int main (int argc, char **argv) {
// CHECK37-NEXT: store ptr null, ptr [[TMP35]], align 8
// CHECK37-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK37-NEXT: store i64 [[TMP27]], ptr [[TMP36]], align 8
-// CHECK37-NEXT: [[TMP37:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l192.region_id, ptr [[KERNEL_ARGS]])
-// CHECK37-NEXT: [[TMP38:%.*]] = icmp ne i32 [[TMP37]], 0
-// CHECK37-NEXT: br i1 [[TMP38]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK37-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK37-NEXT: store i64 0, ptr [[TMP37]], align 8
+// CHECK37-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK37-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP38]], align 4
+// CHECK37-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK37-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP39]], align 4
+// CHECK37-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK37-NEXT: store i32 0, ptr [[TMP40]], align 4
+// CHECK37-NEXT: [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l192.region_id, ptr [[KERNEL_ARGS]])
+// CHECK37-NEXT: [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
+// CHECK37-NEXT: br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK37: omp_offload.failed:
// CHECK37-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l192(i64 [[TMP4]], i64 [[TMP6]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK37-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK37: omp_offload.cont:
-// CHECK37-NEXT: [[TMP39:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK37-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP39]])
+// CHECK37-NEXT: [[TMP43:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK37-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP43]])
// CHECK37-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK37-NEXT: [[TMP40:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK37-NEXT: call void @llvm.stackrestore(ptr [[TMP40]])
-// CHECK37-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK37-NEXT: ret i32 [[TMP41]]
+// CHECK37-NEXT: [[TMP44:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK37-NEXT: call void @llvm.stackrestore(ptr [[TMP44]])
+// CHECK37-NEXT: [[TMP45:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK37-NEXT: ret i32 [[TMP45]]
//
//
// CHECK37-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l192
@@ -4136,28 +4260,37 @@ int main (int argc, char **argv) {
// CHECK37-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK37-NEXT: [[TMP14:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK37-NEXT: [[TMP15:%.*]] = load i32, ptr [[TE]], align 4
+// CHECK37-NEXT: [[TMP16:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP15]], 0
// CHECK37-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK37-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK37-NEXT: store i32 1, ptr [[TMP16]], align 4
-// CHECK37-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK37-NEXT: store i32 3, ptr [[TMP17]], align 4
-// CHECK37-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK37-NEXT: store ptr [[TMP13]], ptr [[TMP18]], align 8
-// CHECK37-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK37-NEXT: store ptr [[TMP14]], ptr [[TMP19]], align 8
-// CHECK37-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK37-NEXT: store ptr @.offload_sizes.2, ptr [[TMP20]], align 8
-// CHECK37-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK37-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP21]], align 8
-// CHECK37-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK37-NEXT: store ptr null, ptr [[TMP22]], align 8
-// CHECK37-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK37-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK37-NEXT: store i32 2, ptr [[TMP17]], align 4
+// CHECK37-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK37-NEXT: store i32 3, ptr [[TMP18]], align 4
+// CHECK37-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK37-NEXT: store ptr [[TMP13]], ptr [[TMP19]], align 8
+// CHECK37-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK37-NEXT: store ptr [[TMP14]], ptr [[TMP20]], align 8
+// CHECK37-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK37-NEXT: store ptr @.offload_sizes.2, ptr [[TMP21]], align 8
+// CHECK37-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK37-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP22]], align 8
+// CHECK37-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK37-NEXT: store ptr null, ptr [[TMP23]], align 8
-// CHECK37-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK37-NEXT: store i64 10, ptr [[TMP24]], align 8
-// CHECK37-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 [[TMP15]], i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l181.region_id, ptr [[KERNEL_ARGS]])
-// CHECK37-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK37-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK37-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK37-NEXT: store ptr null, ptr [[TMP24]], align 8
+// CHECK37-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK37-NEXT: store i64 10, ptr [[TMP25]], align 8
+// CHECK37-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK37-NEXT: store i64 0, ptr [[TMP26]], align 8
+// CHECK37-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK37-NEXT: store [3 x i32] [[TMP16]], ptr [[TMP27]], align 4
+// CHECK37-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK37-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP28]], align 4
+// CHECK37-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK37-NEXT: store i32 0, ptr [[TMP29]], align 4
+// CHECK37-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 [[TMP15]], i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l181.region_id, ptr [[KERNEL_ARGS]])
+// CHECK37-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
+// CHECK37-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK37: omp_offload.failed:
// CHECK37-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l181(i64 [[TMP1]], i64 [[TMP3]], ptr [[A]]) #[[ATTR3]]
// CHECK37-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -4341,7 +4474,7 @@ int main (int argc, char **argv) {
// CHECK39-NEXT: [[TMP27:%.*]] = zext i32 [[ADD]] to i64
// CHECK39-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK39-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK39-NEXT: store i32 1, ptr [[TMP28]], align 4
+// CHECK39-NEXT: store i32 2, ptr [[TMP28]], align 4
// CHECK39-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK39-NEXT: store i32 4, ptr [[TMP29]], align 4
// CHECK39-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -4358,20 +4491,28 @@ int main (int argc, char **argv) {
// CHECK39-NEXT: store ptr null, ptr [[TMP35]], align 4
// CHECK39-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK39-NEXT: store i64 [[TMP27]], ptr [[TMP36]], align 8
-// CHECK39-NEXT: [[TMP37:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l192.region_id, ptr [[KERNEL_ARGS]])
-// CHECK39-NEXT: [[TMP38:%.*]] = icmp ne i32 [[TMP37]], 0
-// CHECK39-NEXT: br i1 [[TMP38]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK39-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK39-NEXT: store i64 0, ptr [[TMP37]], align 8
+// CHECK39-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK39-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP38]], align 4
+// CHECK39-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK39-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP39]], align 4
+// CHECK39-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK39-NEXT: store i32 0, ptr [[TMP40]], align 4
+// CHECK39-NEXT: [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l192.region_id, ptr [[KERNEL_ARGS]])
+// CHECK39-NEXT: [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
+// CHECK39-NEXT: br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK39: omp_offload.failed:
// CHECK39-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l192(i32 [[TMP3]], i32 [[TMP5]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK39-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK39: omp_offload.cont:
-// CHECK39-NEXT: [[TMP39:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK39-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP39]])
+// CHECK39-NEXT: [[TMP43:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK39-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP43]])
// CHECK39-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK39-NEXT: [[TMP40:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK39-NEXT: call void @llvm.stackrestore(ptr [[TMP40]])
-// CHECK39-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK39-NEXT: ret i32 [[TMP41]]
+// CHECK39-NEXT: [[TMP44:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK39-NEXT: call void @llvm.stackrestore(ptr [[TMP44]])
+// CHECK39-NEXT: [[TMP45:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK39-NEXT: ret i32 [[TMP45]]
//
//
// CHECK39-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l192
@@ -4584,28 +4725,37 @@ int main (int argc, char **argv) {
// CHECK39-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK39-NEXT: [[TMP14:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK39-NEXT: [[TMP15:%.*]] = load i32, ptr [[TE]], align 4
+// CHECK39-NEXT: [[TMP16:%.*]] = insertvalue [3 x i32] zeroinitializer, i32 [[TMP15]], 0
// CHECK39-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK39-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK39-NEXT: store i32 1, ptr [[TMP16]], align 4
-// CHECK39-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK39-NEXT: store i32 3, ptr [[TMP17]], align 4
-// CHECK39-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK39-NEXT: store ptr [[TMP13]], ptr [[TMP18]], align 4
-// CHECK39-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK39-NEXT: store ptr [[TMP14]], ptr [[TMP19]], align 4
-// CHECK39-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK39-NEXT: store ptr @.offload_sizes.2, ptr [[TMP20]], align 4
-// CHECK39-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK39-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP21]], align 4
-// CHECK39-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK39-NEXT: store ptr null, ptr [[TMP22]], align 4
-// CHECK39-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK39-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK39-NEXT: store i32 2, ptr [[TMP17]], align 4
+// CHECK39-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK39-NEXT: store i32 3, ptr [[TMP18]], align 4
+// CHECK39-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK39-NEXT: store ptr [[TMP13]], ptr [[TMP19]], align 4
+// CHECK39-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK39-NEXT: store ptr [[TMP14]], ptr [[TMP20]], align 4
+// CHECK39-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK39-NEXT: store ptr @.offload_sizes.2, ptr [[TMP21]], align 4
+// CHECK39-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK39-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP22]], align 4
+// CHECK39-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// CHECK39-NEXT: store ptr null, ptr [[TMP23]], align 4
-// CHECK39-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK39-NEXT: store i64 10, ptr [[TMP24]], align 8
-// CHECK39-NEXT: [[TMP25:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 [[TMP15]], i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l181.region_id, ptr [[KERNEL_ARGS]])
-// CHECK39-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0
-// CHECK39-NEXT: br i1 [[TMP26]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK39-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK39-NEXT: store ptr null, ptr [[TMP24]], align 4
+// CHECK39-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK39-NEXT: store i64 10, ptr [[TMP25]], align 8
+// CHECK39-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK39-NEXT: store i64 0, ptr [[TMP26]], align 8
+// CHECK39-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK39-NEXT: store [3 x i32] [[TMP16]], ptr [[TMP27]], align 4
+// CHECK39-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK39-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP28]], align 4
+// CHECK39-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK39-NEXT: store i32 0, ptr [[TMP29]], align 4
+// CHECK39-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 [[TMP15]], i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l181.region_id, ptr [[KERNEL_ARGS]])
+// CHECK39-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
+// CHECK39-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK39: omp_offload.failed:
// CHECK39-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l181(i32 [[TMP1]], i32 [[TMP3]], ptr [[A]]) #[[ATTR3]]
// CHECK39-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/teams_distribute_simd_collapse_codegen.cpp b/clang/test/OpenMP/teams_distribute_simd_collapse_codegen.cpp
index 347b25bf5fffe..efde7b7425f65 100644
--- a/clang/test/OpenMP/teams_distribute_simd_collapse_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_simd_collapse_codegen.cpp
@@ -128,7 +128,7 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -145,9 +145,17 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 56088, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -155,8 +163,8 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[A3:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x [456 x i32]], ptr [[A3]], i64 0, i64 0
// CHECK1-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds [456 x i32], ptr [[ARRAYIDX]], i64 0, i64 0
-// CHECK1-NEXT: [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
-// CHECK1-NEXT: ret i32 [[TMP16]]
+// CHECK1-NEXT: [[TMP20:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
+// CHECK1-NEXT: ret i32 [[TMP20]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -296,7 +304,7 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -313,9 +321,17 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 56088, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -323,8 +339,8 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[A3:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x [456 x i32]], ptr [[A3]], i32 0, i32 0
// CHECK3-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds [456 x i32], ptr [[ARRAYIDX]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
-// CHECK3-NEXT: ret i32 [[TMP16]]
+// CHECK3-NEXT: [[TMP20:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
+// CHECK3-NEXT: ret i32 [[TMP20]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -672,7 +688,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[ADD:%.*]] = add nsw i64 [[TMP35]], 1
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP36]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP36]], align 4
// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 5, ptr [[TMP37]], align 4
// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -689,20 +705,28 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP43]], align 8
// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 [[ADD]], ptr [[TMP44]], align 8
-// CHECK9-NEXT: [[TMP45:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l83.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP46:%.*]] = icmp ne i32 [[TMP45]], 0
-// CHECK9-NEXT: br i1 [[TMP46]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP46]], align 4
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP47]], align 4
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP48]], align 4
+// CHECK9-NEXT: [[TMP49:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l83.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP50:%.*]] = icmp ne i32 [[TMP49]], 0
+// CHECK9-NEXT: br i1 [[TMP50]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l83(i64 [[TMP7]], i64 [[TMP9]], i64 [[TMP1]], i64 [[TMP3]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP47:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef signext [[TMP47]])
+// CHECK9-NEXT: [[TMP51:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef signext [[TMP51]])
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK9-NEXT: [[TMP48:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP48]])
-// CHECK9-NEXT: [[TMP49:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP49]]
+// CHECK9-NEXT: [[TMP52:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP52]])
+// CHECK9-NEXT: [[TMP53:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP53]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l83
@@ -912,7 +936,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK9-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK9-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -929,9 +953,17 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 20, ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l69.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK9-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK9-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l69.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK9-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l69(ptr [[A]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1144,7 +1176,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[ADD:%.*]] = add nsw i64 [[TMP34]], 1
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP35]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP35]], align 4
// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 5, ptr [[TMP36]], align 4
// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1161,20 +1193,28 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP42]], align 4
// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 [[ADD]], ptr [[TMP43]], align 8
-// CHECK11-NEXT: [[TMP44:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l83.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP45:%.*]] = icmp ne i32 [[TMP44]], 0
-// CHECK11-NEXT: br i1 [[TMP45]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP44]], align 8
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP47]], align 4
+// CHECK11-NEXT: [[TMP48:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l83.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP49:%.*]] = icmp ne i32 [[TMP48]], 0
+// CHECK11-NEXT: br i1 [[TMP49]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l83(i32 [[TMP5]], i32 [[TMP7]], i32 [[TMP0]], i32 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP46:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef [[TMP46]])
+// CHECK11-NEXT: [[TMP50:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10ELi2EEiT_(i32 noundef [[TMP50]])
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK11-NEXT: [[TMP47:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP47]])
-// CHECK11-NEXT: [[TMP48:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP48]]
+// CHECK11-NEXT: [[TMP51:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP51]])
+// CHECK11-NEXT: [[TMP52:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP52]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l83
@@ -1382,7 +1422,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK11-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK11-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1399,9 +1439,17 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK11-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 20, ptr [[TMP13]], align 8
-// CHECK11-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l69.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK11-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP16]], align 4
+// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK11-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l69.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK11-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10ELi2EEiT__l69(ptr [[A]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/teams_distribute_simd_dist_schedule_codegen.cpp b/clang/test/OpenMP/teams_distribute_simd_dist_schedule_codegen.cpp
index 51a61e29f6cce..a90b9ebf04ffb 100644
--- a/clang/test/OpenMP/teams_distribute_simd_dist_schedule_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_simd_dist_schedule_codegen.cpp
@@ -168,7 +168,7 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -185,87 +185,111 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK1-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 8
-// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP17]], align 8
-// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 8
+// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A2]], ptr [[TMP21]], align 8
+// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.2, ptr [[TMP25]], align 8
-// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP26]], align 8
-// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l33.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.2, ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP36]], align 4
+// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK1-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l33.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK1-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK1-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK1: omp_offload.failed8:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l33(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK1: omp_offload.cont9:
// CHECK1-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 8
-// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK1-NEXT: store ptr [[A10]], ptr [[TMP33]], align 8
-// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
-// CHECK1-NEXT: store ptr null, ptr [[TMP34]], align 8
-// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 8
+// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK1-NEXT: store ptr [[A10]], ptr [[TMP41]], align 8
+// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0
+// CHECK1-NEXT: store ptr null, ptr [[TMP42]], align 8
+// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK1-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK1-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
-// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK1-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
-// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK1-NEXT: store ptr @.offload_sizes.5, ptr [[TMP41]], align 8
-// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK1-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP42]], align 8
-// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK1-NEXT: store ptr null, ptr [[TMP43]], align 8
-// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK1-NEXT: store ptr null, ptr [[TMP44]], align 8
-// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK1-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK1-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l38.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK1-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK1-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK1-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK1-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK1-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 8
+// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK1-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 8
+// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK1-NEXT: store ptr @.offload_sizes.5, ptr [[TMP49]], align 8
+// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK1-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP50]], align 8
+// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK1-NEXT: store ptr null, ptr [[TMP51]], align 8
+// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK1-NEXT: store ptr null, ptr [[TMP52]], align 8
+// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK1-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP56]], align 4
+// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK1-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l38.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK1-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK1-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK1: omp_offload.failed16:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l38(ptr [[THIS1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK1: omp_offload.cont17:
// CHECK1-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A18]], i64 0, i64 0
-// CHECK1-NEXT: [[TMP48:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK1-NEXT: ret i32 [[TMP48]]
+// CHECK1-NEXT: [[TMP60:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK1-NEXT: ret i32 [[TMP60]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -584,7 +608,7 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -601,87 +625,111 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 123, ptr [[TMP13]], align 8
-// CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK3-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28(ptr [[THIS1]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP16]], align 4
-// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A2]], ptr [[TMP17]], align 4
-// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 4
+// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A2]], ptr [[TMP21]], align 4
+// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.2, ptr [[TMP25]], align 4
-// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP26]], align 4
-// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
-// CHECK3-NEXT: store i64 123, ptr [[TMP29]], align 8
-// CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l33.region_id, ptr [[KERNEL_ARGS7]])
-// CHECK3-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK3-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
+// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.2, ptr [[TMP29]], align 4
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
+// CHECK3-NEXT: store i64 123, ptr [[TMP33]], align 8
+// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP36]], align 4
+// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK3-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l33.region_id, ptr [[KERNEL_ARGS7]])
+// CHECK3-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK3-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
// CHECK3: omp_offload.failed8:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l33(ptr [[THIS1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK3: omp_offload.cont9:
// CHECK3-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP32]], align 4
-// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
-// CHECK3-NEXT: store ptr [[A10]], ptr [[TMP33]], align 4
-// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
-// CHECK3-NEXT: store ptr null, ptr [[TMP34]], align 4
-// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 4
+// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
+// CHECK3-NEXT: store ptr [[A10]], ptr [[TMP41]], align 4
+// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0
+// CHECK3-NEXT: store ptr null, ptr [[TMP42]], align 4
+// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0
+// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK3-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK3-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 4
-// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK3-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 4
-// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK3-NEXT: store ptr @.offload_sizes.5, ptr [[TMP41]], align 4
-// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK3-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP42]], align 4
-// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK3-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK3-NEXT: store ptr null, ptr [[TMP44]], align 4
-// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK3-NEXT: store i64 123, ptr [[TMP45]], align 8
-// CHECK3-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l38.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK3-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK3-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK3-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK3-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK3-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 4
+// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK3-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 4
+// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK3-NEXT: store ptr @.offload_sizes.5, ptr [[TMP49]], align 4
+// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK3-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP50]], align 4
+// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK3-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK3-NEXT: store ptr null, ptr [[TMP52]], align 4
+// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK3-NEXT: store i64 123, ptr [[TMP53]], align 8
+// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP56]], align 4
+// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK3-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l38.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK3-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK3-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK3: omp_offload.failed16:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l38(ptr [[THIS1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK3: omp_offload.cont17:
// CHECK3-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0
// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A18]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP48:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK3-NEXT: ret i32 [[TMP48]]
+// CHECK3-NEXT: [[TMP60:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+// CHECK3-NEXT: ret i32 [[TMP60]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l28
@@ -1290,7 +1338,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1307,148 +1355,172 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK9-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK9-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP34]], align 4
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK9-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100(i64 [[TMP4]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK9-NEXT: [[TMP35:%.*]] = load i64, ptr [[N_CASTED3]], align 8
-// CHECK9-NEXT: [[TMP36:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK9-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK9-NEXT: [[TMP39:%.*]] = load i64, ptr [[N_CASTED3]], align 8
+// CHECK9-NEXT: [[TMP40:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK9-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES7]], ptr align 8 @.offload_sizes.2, i64 24, i1 false)
-// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP35]], ptr [[TMP37]], align 8
-// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP35]], ptr [[TMP38]], align 8
-// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP39]], align 8
-// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP40]], align 8
-// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP41]], align 8
-// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP42]], align 8
-// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP43]], align 8
-// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 8
-// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK9-NEXT: store i64 [[TMP36]], ptr [[TMP45]], align 8
-// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP39]], ptr [[TMP41]], align 8
+// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP39]], ptr [[TMP42]], align 8
+// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP43]], align 8
+// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP44]], align 8
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
// CHECK9-NEXT: store ptr null, ptr [[TMP46]], align 8
-// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP50:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP50]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK9-NEXT: [[TMP51:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK9-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP51]], 0
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP47]], align 8
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 8
+// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK9-NEXT: store i64 [[TMP40]], ptr [[TMP49]], align 8
+// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP50]], align 8
+// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP54:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP54]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK9-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK9-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP55]], 0
// CHECK9-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK9-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK9-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK9-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK9-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP52]], 1
-// CHECK9-NEXT: [[TMP53:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK9-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK9-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP56]], 1
+// CHECK9-NEXT: [[TMP57:%.*]] = zext i32 [[ADD14]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP54]], align 4
-// CHECK9-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK9-NEXT: store i32 3, ptr [[TMP55]], align 4
-// CHECK9-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP47]], ptr [[TMP56]], align 8
-// CHECK9-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 8
-// CHECK9-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 8
-// CHECK9-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP59]], align 8
-// CHECK9-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP60]], align 8
-// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP61]], align 8
-// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP53]], ptr [[TMP62]], align 8
-// CHECK9-NEXT: [[TMP63:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l105.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK9-NEXT: [[TMP64:%.*]] = icmp ne i32 [[TMP63]], 0
-// CHECK9-NEXT: br i1 [[TMP64]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK9-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP58]], align 4
+// CHECK9-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK9-NEXT: store i32 3, ptr [[TMP59]], align 4
+// CHECK9-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP51]], ptr [[TMP60]], align 8
+// CHECK9-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 8
+// CHECK9-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 8
+// CHECK9-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP63]], align 8
+// CHECK9-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP64]], align 8
+// CHECK9-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP65]], align 8
+// CHECK9-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP57]], ptr [[TMP66]], align 8
+// CHECK9-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP67]], align 8
+// CHECK9-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP68]], align 4
+// CHECK9-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP69]], align 4
+// CHECK9-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP70]], align 4
+// CHECK9-NEXT: [[TMP71:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l105.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK9-NEXT: [[TMP72:%.*]] = icmp ne i32 [[TMP71]], 0
+// CHECK9-NEXT: br i1 [[TMP72]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK9: omp_offload.failed16:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l105(i64 [[TMP35]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l105(i64 [[TMP39]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK9: omp_offload.cont17:
-// CHECK9-NEXT: [[TMP65:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP65]], ptr [[N_CASTED18]], align 4
-// CHECK9-NEXT: [[TMP66:%.*]] = load i64, ptr [[N_CASTED18]], align 8
-// CHECK9-NEXT: [[TMP67:%.*]] = mul nuw i64 [[TMP1]], 4
+// CHECK9-NEXT: [[TMP73:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP73]], ptr [[N_CASTED18]], align 4
+// CHECK9-NEXT: [[TMP74:%.*]] = load i64, ptr [[N_CASTED18]], align 8
+// CHECK9-NEXT: [[TMP75:%.*]] = mul nuw i64 [[TMP1]], 4
// CHECK9-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES22]], ptr align 8 @.offload_sizes.5, i64 24, i1 false)
-// CHECK9-NEXT: [[TMP68:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP66]], ptr [[TMP68]], align 8
-// CHECK9-NEXT: [[TMP69:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP66]], ptr [[TMP69]], align 8
-// CHECK9-NEXT: [[TMP70:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP70]], align 8
-// CHECK9-NEXT: [[TMP71:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP71]], align 8
-// CHECK9-NEXT: [[TMP72:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
-// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP72]], align 8
-// CHECK9-NEXT: [[TMP73:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 1
-// CHECK9-NEXT: store ptr null, ptr [[TMP73]], align 8
-// CHECK9-NEXT: [[TMP74:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP74]], align 8
-// CHECK9-NEXT: [[TMP75:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP75]], align 8
-// CHECK9-NEXT: [[TMP76:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 2
-// CHECK9-NEXT: store i64 [[TMP67]], ptr [[TMP76]], align 8
-// CHECK9-NEXT: [[TMP77:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 2
-// CHECK9-NEXT: store ptr null, ptr [[TMP77]], align 8
-// CHECK9-NEXT: [[TMP78:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP79:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP80:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP81:%.*]] = load i32, ptr [[N]], align 4
-// CHECK9-NEXT: store i32 [[TMP81]], ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK9-NEXT: [[TMP82:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK9-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP82]], 0
+// CHECK9-NEXT: [[TMP76:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP74]], ptr [[TMP76]], align 8
+// CHECK9-NEXT: [[TMP77:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP74]], ptr [[TMP77]], align 8
+// CHECK9-NEXT: [[TMP78:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP78]], align 8
+// CHECK9-NEXT: [[TMP79:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP79]], align 8
+// CHECK9-NEXT: [[TMP80:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
+// CHECK9-NEXT: store i64 [[TMP1]], ptr [[TMP80]], align 8
+// CHECK9-NEXT: [[TMP81:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 1
+// CHECK9-NEXT: store ptr null, ptr [[TMP81]], align 8
+// CHECK9-NEXT: [[TMP82:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP82]], align 8
+// CHECK9-NEXT: [[TMP83:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[VLA]], ptr [[TMP83]], align 8
+// CHECK9-NEXT: [[TMP84:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 2
+// CHECK9-NEXT: store i64 [[TMP75]], ptr [[TMP84]], align 8
+// CHECK9-NEXT: [[TMP85:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 2
+// CHECK9-NEXT: store ptr null, ptr [[TMP85]], align 8
+// CHECK9-NEXT: [[TMP86:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP87:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP88:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP89:%.*]] = load i32, ptr [[N]], align 4
+// CHECK9-NEXT: store i32 [[TMP89]], ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK9-NEXT: [[TMP90:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK9-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP90]], 0
// CHECK9-NEXT: [[DIV27:%.*]] = sdiv i32 [[SUB26]], 1
// CHECK9-NEXT: [[SUB28:%.*]] = sub nsw i32 [[DIV27]], 1
// CHECK9-NEXT: store i32 [[SUB28]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK9-NEXT: [[TMP83:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK9-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP83]], 1
-// CHECK9-NEXT: [[TMP84:%.*]] = zext i32 [[ADD29]] to i64
+// CHECK9-NEXT: [[TMP91:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK9-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP91]], 1
+// CHECK9-NEXT: [[TMP92:%.*]] = zext i32 [[ADD29]] to i64
// CHECK9-NEXT: [[KERNEL_ARGS30:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP85]], align 4
-// CHECK9-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
-// CHECK9-NEXT: store i32 3, ptr [[TMP86]], align 4
-// CHECK9-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP78]], ptr [[TMP87]], align 8
-// CHECK9-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP79]], ptr [[TMP88]], align 8
-// CHECK9-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
-// CHECK9-NEXT: store ptr [[TMP80]], ptr [[TMP89]], align 8
-// CHECK9-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP90]], align 8
-// CHECK9-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP91]], align 8
-// CHECK9-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP92]], align 8
-// CHECK9-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
-// CHECK9-NEXT: store i64 [[TMP84]], ptr [[TMP93]], align 8
-// CHECK9-NEXT: [[TMP94:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l110.region_id, ptr [[KERNEL_ARGS30]])
-// CHECK9-NEXT: [[TMP95:%.*]] = icmp ne i32 [[TMP94]], 0
-// CHECK9-NEXT: br i1 [[TMP95]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
+// CHECK9-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP93]], align 4
+// CHECK9-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
+// CHECK9-NEXT: store i32 3, ptr [[TMP94]], align 4
+// CHECK9-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP86]], ptr [[TMP95]], align 8
+// CHECK9-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP87]], ptr [[TMP96]], align 8
+// CHECK9-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
+// CHECK9-NEXT: store ptr [[TMP88]], ptr [[TMP97]], align 8
+// CHECK9-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP98]], align 8
+// CHECK9-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP99]], align 8
+// CHECK9-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP100]], align 8
+// CHECK9-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
+// CHECK9-NEXT: store i64 [[TMP92]], ptr [[TMP101]], align 8
+// CHECK9-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP102]], align 8
+// CHECK9-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP103]], align 4
+// CHECK9-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP104]], align 4
+// CHECK9-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP105]], align 4
+// CHECK9-NEXT: [[TMP106:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l110.region_id, ptr [[KERNEL_ARGS30]])
+// CHECK9-NEXT: [[TMP107:%.*]] = icmp ne i32 [[TMP106]], 0
+// CHECK9-NEXT: br i1 [[TMP107]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
// CHECK9: omp_offload.failed31:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l110(i64 [[TMP66]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l110(i64 [[TMP74]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT32]]
// CHECK9: omp_offload.cont32:
-// CHECK9-NEXT: [[TMP96:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP96]])
+// CHECK9-NEXT: [[TMP108:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP108]])
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK9-NEXT: [[TMP97:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP97]])
-// CHECK9-NEXT: [[TMP98:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP98]]
+// CHECK9-NEXT: [[TMP109:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK9-NEXT: call void @llvm.stackrestore(ptr [[TMP109]])
+// CHECK9-NEXT: [[TMP110:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP110]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100
@@ -1872,7 +1944,7 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK9-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK9-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1889,77 +1961,101 @@ int main (int argc, char **argv) {
// CHECK9-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l79.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK9-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK9-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l79.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK9-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l79(ptr [[A]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP16]], align 8
-// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP17]], align 8
-// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP18]], align 8
-// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP20]], align 8
+// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP21]], align 8
+// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP22]], align 8
+// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK9-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
-// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
-// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.11, ptr [[TMP25]], align 8
-// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP26]], align 8
-// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP28]], align 8
-// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK9-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK9-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK9-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK9-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK9-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8
+// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.11, ptr [[TMP29]], align 8
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP30]], align 8
+// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP32]], align 8
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK9-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP36]], align 4
+// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK9-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK9-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK9-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK9: omp_offload.failed6:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84(ptr [[A]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK9: omp_offload.cont7:
-// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP32]], align 8
-// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[A]], ptr [[TMP33]], align 8
-// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP34]], align 8
-// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP40]], align 8
+// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[A]], ptr [[TMP41]], align 8
+// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP42]], align 8
+// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK9-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 8
-// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 8
-// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.14, ptr [[TMP41]], align 8
-// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.15, ptr [[TMP42]], align 8
-// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP43]], align 8
-// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP44]], align 8
-// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK9-NEXT: store i64 10, ptr [[TMP45]], align 8
-// CHECK9-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK9-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK9-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK9-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 8
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 8
+// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.14, ptr [[TMP49]], align 8
+// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.15, ptr [[TMP50]], align 8
+// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP51]], align 8
+// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP52]], align 8
+// CHECK9-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK9-NEXT: store i64 10, ptr [[TMP53]], align 8
+// CHECK9-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK9-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK9-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP56]], align 4
+// CHECK9-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK9-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK9-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK9-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK9: omp_offload.failed13:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89(ptr [[A]]) #[[ATTR3]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT14]]
@@ -2326,7 +2422,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP23]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP23]], align 4
// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 3, ptr [[TMP24]], align 4
// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2343,150 +2439,174 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP30]], align 4
// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8
-// CHECK11-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
-// CHECK11-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP32]], align 8
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP34]], align 4
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
+// CHECK11-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100(i32 [[TMP3]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP34:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP34]], ptr [[N_CASTED3]], align 4
-// CHECK11-NEXT: [[TMP35:%.*]] = load i32, ptr [[N_CASTED3]], align 4
-// CHECK11-NEXT: [[TMP36:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK11-NEXT: [[TMP37:%.*]] = sext i32 [[TMP36]] to i64
+// CHECK11-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
+// CHECK11-NEXT: [[TMP39:%.*]] = load i32, ptr [[N_CASTED3]], align 4
+// CHECK11-NEXT: [[TMP40:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK11-NEXT: [[TMP41:%.*]] = sext i32 [[TMP40]] to i64
// CHECK11-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES7]], ptr align 4 @.offload_sizes.2, i32 24, i1 false)
-// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP35]], ptr [[TMP38]], align 4
-// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP35]], ptr [[TMP39]], align 4
-// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP40]], align 4
-// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP41]], align 4
-// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP42]], align 4
-// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP44]], align 4
-// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP45]], align 4
-// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
-// CHECK11-NEXT: store i64 [[TMP37]], ptr [[TMP46]], align 4
-// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP39]], ptr [[TMP42]], align 4
+// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP39]], ptr [[TMP43]], align 4
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP44]], align 4
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
// CHECK11-NEXT: store ptr null, ptr [[TMP47]], align 4
-// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP51:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP51]], ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK11-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
-// CHECK11-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP52]], 0
+// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 4
+// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP49]], align 4
+// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2
+// CHECK11-NEXT: store i64 [[TMP41]], ptr [[TMP50]], align 4
+// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP54:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP55:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP55]], ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK11-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
+// CHECK11-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP56]], 0
// CHECK11-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK11-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1
// CHECK11-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK11-NEXT: [[TMP53:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
-// CHECK11-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP53]], 1
-// CHECK11-NEXT: [[TMP54:%.*]] = zext i32 [[ADD14]] to i64
+// CHECK11-NEXT: [[TMP57:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4
+// CHECK11-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP57]], 1
+// CHECK11-NEXT: [[TMP58:%.*]] = zext i32 [[ADD14]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP55]], align 4
-// CHECK11-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
-// CHECK11-NEXT: store i32 3, ptr [[TMP56]], align 4
-// CHECK11-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP48]], ptr [[TMP57]], align 4
-// CHECK11-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP49]], ptr [[TMP58]], align 4
-// CHECK11-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP50]], ptr [[TMP59]], align 4
-// CHECK11-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP60]], align 4
-// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP61]], align 4
-// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP62]], align 4
-// CHECK11-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP54]], ptr [[TMP63]], align 8
-// CHECK11-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l105.region_id, ptr [[KERNEL_ARGS15]])
-// CHECK11-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
-// CHECK11-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
+// CHECK11-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP59]], align 4
+// CHECK11-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1
+// CHECK11-NEXT: store i32 3, ptr [[TMP60]], align 4
+// CHECK11-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 4
+// CHECK11-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 4
+// CHECK11-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP54]], ptr [[TMP63]], align 4
+// CHECK11-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP64]], align 4
+// CHECK11-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP65]], align 4
+// CHECK11-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP66]], align 4
+// CHECK11-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP58]], ptr [[TMP67]], align 8
+// CHECK11-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP68]], align 8
+// CHECK11-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4
+// CHECK11-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP70]], align 4
+// CHECK11-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP71]], align 4
+// CHECK11-NEXT: [[TMP72:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l105.region_id, ptr [[KERNEL_ARGS15]])
+// CHECK11-NEXT: [[TMP73:%.*]] = icmp ne i32 [[TMP72]], 0
+// CHECK11-NEXT: br i1 [[TMP73]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]]
// CHECK11: omp_offload.failed16:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l105(i32 [[TMP35]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l105(i32 [[TMP39]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT17]]
// CHECK11: omp_offload.cont17:
-// CHECK11-NEXT: [[TMP66:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP66]], ptr [[N_CASTED18]], align 4
-// CHECK11-NEXT: [[TMP67:%.*]] = load i32, ptr [[N_CASTED18]], align 4
-// CHECK11-NEXT: [[TMP68:%.*]] = mul nuw i32 [[TMP0]], 4
-// CHECK11-NEXT: [[TMP69:%.*]] = sext i32 [[TMP68]] to i64
+// CHECK11-NEXT: [[TMP74:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP74]], ptr [[N_CASTED18]], align 4
+// CHECK11-NEXT: [[TMP75:%.*]] = load i32, ptr [[N_CASTED18]], align 4
+// CHECK11-NEXT: [[TMP76:%.*]] = mul nuw i32 [[TMP0]], 4
+// CHECK11-NEXT: [[TMP77:%.*]] = sext i32 [[TMP76]] to i64
// CHECK11-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES22]], ptr align 4 @.offload_sizes.5, i32 24, i1 false)
-// CHECK11-NEXT: [[TMP70:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP67]], ptr [[TMP70]], align 4
-// CHECK11-NEXT: [[TMP71:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP67]], ptr [[TMP71]], align 4
-// CHECK11-NEXT: [[TMP72:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP72]], align 4
-// CHECK11-NEXT: [[TMP73:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP73]], align 4
-// CHECK11-NEXT: [[TMP74:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
-// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP74]], align 4
-// CHECK11-NEXT: [[TMP75:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 1
-// CHECK11-NEXT: store ptr null, ptr [[TMP75]], align 4
-// CHECK11-NEXT: [[TMP76:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP76]], align 4
-// CHECK11-NEXT: [[TMP77:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP77]], align 4
-// CHECK11-NEXT: [[TMP78:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 2
-// CHECK11-NEXT: store i64 [[TMP69]], ptr [[TMP78]], align 4
-// CHECK11-NEXT: [[TMP79:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 2
-// CHECK11-NEXT: store ptr null, ptr [[TMP79]], align 4
-// CHECK11-NEXT: [[TMP80:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP81:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP82:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP83:%.*]] = load i32, ptr [[N]], align 4
-// CHECK11-NEXT: store i32 [[TMP83]], ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK11-NEXT: [[TMP84:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
-// CHECK11-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP84]], 0
+// CHECK11-NEXT: [[TMP78:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP75]], ptr [[TMP78]], align 4
+// CHECK11-NEXT: [[TMP79:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP75]], ptr [[TMP79]], align 4
+// CHECK11-NEXT: [[TMP80:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP80]], align 4
+// CHECK11-NEXT: [[TMP81:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP81]], align 4
+// CHECK11-NEXT: [[TMP82:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 1
+// CHECK11-NEXT: store i32 [[TMP0]], ptr [[TMP82]], align 4
+// CHECK11-NEXT: [[TMP83:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 1
+// CHECK11-NEXT: store ptr null, ptr [[TMP83]], align 4
+// CHECK11-NEXT: [[TMP84:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP84]], align 4
+// CHECK11-NEXT: [[TMP85:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[VLA]], ptr [[TMP85]], align 4
+// CHECK11-NEXT: [[TMP86:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 2
+// CHECK11-NEXT: store i64 [[TMP77]], ptr [[TMP86]], align 4
+// CHECK11-NEXT: [[TMP87:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 2
+// CHECK11-NEXT: store ptr null, ptr [[TMP87]], align 4
+// CHECK11-NEXT: [[TMP88:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP89:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP90:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES22]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP91:%.*]] = load i32, ptr [[N]], align 4
+// CHECK11-NEXT: store i32 [[TMP91]], ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK11-NEXT: [[TMP92:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_24]], align 4
+// CHECK11-NEXT: [[SUB26:%.*]] = sub nsw i32 [[TMP92]], 0
// CHECK11-NEXT: [[DIV27:%.*]] = sdiv i32 [[SUB26]], 1
// CHECK11-NEXT: [[SUB28:%.*]] = sub nsw i32 [[DIV27]], 1
// CHECK11-NEXT: store i32 [[SUB28]], ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK11-NEXT: [[TMP85:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
-// CHECK11-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP85]], 1
-// CHECK11-NEXT: [[TMP86:%.*]] = zext i32 [[ADD29]] to i64
+// CHECK11-NEXT: [[TMP93:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4
+// CHECK11-NEXT: [[ADD29:%.*]] = add nsw i32 [[TMP93]], 1
+// CHECK11-NEXT: [[TMP94:%.*]] = zext i32 [[ADD29]] to i64
// CHECK11-NEXT: [[KERNEL_ARGS30:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP87]], align 4
-// CHECK11-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
-// CHECK11-NEXT: store i32 3, ptr [[TMP88]], align 4
-// CHECK11-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP80]], ptr [[TMP89]], align 4
-// CHECK11-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP81]], ptr [[TMP90]], align 4
-// CHECK11-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
-// CHECK11-NEXT: store ptr [[TMP82]], ptr [[TMP91]], align 4
-// CHECK11-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP92]], align 4
-// CHECK11-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP93]], align 4
-// CHECK11-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP94]], align 4
-// CHECK11-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
-// CHECK11-NEXT: store i64 [[TMP86]], ptr [[TMP95]], align 8
-// CHECK11-NEXT: [[TMP96:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l110.region_id, ptr [[KERNEL_ARGS30]])
-// CHECK11-NEXT: [[TMP97:%.*]] = icmp ne i32 [[TMP96]], 0
-// CHECK11-NEXT: br i1 [[TMP97]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
+// CHECK11-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP95]], align 4
+// CHECK11-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 1
+// CHECK11-NEXT: store i32 3, ptr [[TMP96]], align 4
+// CHECK11-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP88]], ptr [[TMP97]], align 4
+// CHECK11-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP89]], ptr [[TMP98]], align 4
+// CHECK11-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 4
+// CHECK11-NEXT: store ptr [[TMP90]], ptr [[TMP99]], align 4
+// CHECK11-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP100]], align 4
+// CHECK11-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP101]], align 4
+// CHECK11-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP102]], align 4
+// CHECK11-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 8
+// CHECK11-NEXT: store i64 [[TMP94]], ptr [[TMP103]], align 8
+// CHECK11-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP104]], align 8
+// CHECK11-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP105]], align 4
+// CHECK11-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP106]], align 4
+// CHECK11-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS30]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP107]], align 4
+// CHECK11-NEXT: [[TMP108:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l110.region_id, ptr [[KERNEL_ARGS30]])
+// CHECK11-NEXT: [[TMP109:%.*]] = icmp ne i32 [[TMP108]], 0
+// CHECK11-NEXT: br i1 [[TMP109]], label [[OMP_OFFLOAD_FAILED31:%.*]], label [[OMP_OFFLOAD_CONT32:%.*]]
// CHECK11: omp_offload.failed31:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l110(i32 [[TMP67]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l110(i32 [[TMP75]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT32]]
// CHECK11: omp_offload.cont32:
-// CHECK11-NEXT: [[TMP98:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
-// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP98]])
+// CHECK11-NEXT: [[TMP110:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
+// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP110]])
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
-// CHECK11-NEXT: [[TMP99:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP99]])
-// CHECK11-NEXT: [[TMP100:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP100]]
+// CHECK11-NEXT: [[TMP111:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK11-NEXT: call void @llvm.stackrestore(ptr [[TMP111]])
+// CHECK11-NEXT: [[TMP112:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP112]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l100
@@ -2907,7 +3027,7 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK11-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK11-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2924,77 +3044,101 @@ int main (int argc, char **argv) {
// CHECK11-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK11-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 10, ptr [[TMP13]], align 8
-// CHECK11-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l79.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK11-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP16]], align 4
+// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK11-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l79.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK11-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l79(ptr [[A]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP16]], align 4
-// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP17]], align 4
-// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP18]], align 4
-// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP20]], align 4
+// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP21]], align 4
+// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP22]], align 4
+// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP21]], align 4
-// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK11-NEXT: store i32 1, ptr [[TMP22]], align 4
-// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
-// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
-// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.11, ptr [[TMP25]], align 4
-// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP26]], align 4
-// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP28]], align 4
-// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK11-NEXT: store i64 10, ptr [[TMP29]], align 8
-// CHECK11-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK11-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK11-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP25]], align 4
+// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK11-NEXT: store i32 1, ptr [[TMP26]], align 4
+// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4
+// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4
+// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.11, ptr [[TMP29]], align 4
+// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP30]], align 4
+// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP32]], align 4
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK11-NEXT: store i64 10, ptr [[TMP33]], align 8
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP36]], align 4
+// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK11-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK11-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK11-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK11: omp_offload.failed6:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l84(ptr [[A]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK11: omp_offload.cont7:
-// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP32]], align 4
-// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[A]], ptr [[TMP33]], align 4
-// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP34]], align 4
-// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP40]], align 4
+// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[A]], ptr [[TMP41]], align 4
+// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP42]], align 4
+// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP37]], align 4
-// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
-// CHECK11-NEXT: store i32 1, ptr [[TMP38]], align 4
-// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP35]], ptr [[TMP39]], align 4
-// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP36]], ptr [[TMP40]], align 4
-// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.14, ptr [[TMP41]], align 4
-// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.15, ptr [[TMP42]], align 4
-// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP43]], align 4
-// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP44]], align 4
-// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
-// CHECK11-NEXT: store i64 10, ptr [[TMP45]], align 8
-// CHECK11-NEXT: [[TMP46:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89.region_id, ptr [[KERNEL_ARGS12]])
-// CHECK11-NEXT: [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
-// CHECK11-NEXT: br i1 [[TMP47]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1
+// CHECK11-NEXT: store i32 1, ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 4
+// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 4
+// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.14, ptr [[TMP49]], align 4
+// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.15, ptr [[TMP50]], align 4
+// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP51]], align 4
+// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP52]], align 4
+// CHECK11-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8
+// CHECK11-NEXT: store i64 10, ptr [[TMP53]], align 8
+// CHECK11-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP54]], align 8
+// CHECK11-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4
+// CHECK11-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP56]], align 4
+// CHECK11-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP57]], align 4
+// CHECK11-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89.region_id, ptr [[KERNEL_ARGS12]])
+// CHECK11-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
+// CHECK11-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]]
// CHECK11: omp_offload.failed13:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l89(ptr [[A]]) #[[ATTR3]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT14]]
diff --git a/clang/test/OpenMP/teams_distribute_simd_firstprivate_codegen.cpp b/clang/test/OpenMP/teams_distribute_simd_firstprivate_codegen.cpp
index 66aee147e585a..38609ccbfecbe 100644
--- a/clang/test/OpenMP/teams_distribute_simd_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_simd_firstprivate_codegen.cpp
@@ -306,7 +306,7 @@ int main() {
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 5, ptr [[TMP22]], align 4
// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -323,9 +323,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP29]], align 8
-// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP32]], align 4
+// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK1-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK1-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94(i64 [[TMP1]], ptr @vec, ptr @s_arr, ptr @var, i64 [[TMP3]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -586,7 +594,7 @@ int main() {
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -603,27 +611,35 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK1-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK1-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP31]], align 4
+// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK1-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK1-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56(i64 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK1-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK1: arraydestroy.body:
-// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK1-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK1-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK1: arraydestroy.done2:
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK1-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK1-NEXT: ret i32 [[TMP32]]
+// CHECK1-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK1-NEXT: ret i32 [[TMP36]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN2StC2Ev
@@ -1107,7 +1123,7 @@ int main() {
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 5, ptr [[TMP22]], align 4
// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1124,9 +1140,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4
// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP29]], align 8
-// CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK3-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK3-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK3-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94(i32 [[TMP1]], ptr @vec, ptr @s_arr, ptr @var, i32 [[TMP3]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1385,7 +1409,7 @@ int main() {
// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1402,27 +1426,35 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP27]], align 4
// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK3-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK3-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP31]], align 4
+// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK3-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK3-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56(i32 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK3-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK3-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK3: arraydestroy.body:
-// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK3-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK3-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK3: arraydestroy.done2:
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK3-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK3-NEXT: ret i32 [[TMP32]]
+// CHECK3-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK3-NEXT: ret i32 [[TMP36]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN2StC2Ev
diff --git a/clang/test/OpenMP/teams_distribute_simd_lastprivate_codegen.cpp b/clang/test/OpenMP/teams_distribute_simd_lastprivate_codegen.cpp
index cf2a7d3238b8f..80a7ccad9bba1 100644
--- a/clang/test/OpenMP/teams_distribute_simd_lastprivate_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_simd_lastprivate_codegen.cpp
@@ -582,7 +582,7 @@ int main() {
// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP25]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP25]], align 4
// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 5, ptr [[TMP26]], align 4
// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -599,9 +599,17 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP32]], align 8
// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP33]], align 8
-// CHECK9-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l96.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
-// CHECK9-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP36]], align 4
+// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK9-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l96.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK9-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l96(i64 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]], i64 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -609,18 +617,18 @@ int main() {
// CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiET_v()
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP36]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP40]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP37:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP37]]
+// CHECK9-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP41]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -892,7 +900,7 @@ int main() {
// CHECK9-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -909,27 +917,35 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP27]], align 8
// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK9-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK9-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP31]], align 4
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK9-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK9-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49(i64 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
// CHECK9-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done2:
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP32]]
+// CHECK9-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP36]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
@@ -1283,7 +1299,7 @@ int main() {
// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP25]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP25]], align 4
// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 5, ptr [[TMP26]], align 4
// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1300,9 +1316,17 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP32]], align 4
// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP33]], align 8
-// CHECK11-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l96.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
-// CHECK11-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP34]], align 8
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP36]], align 4
+// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP37]], align 4
+// CHECK11-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l96.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0
+// CHECK11-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l96(i32 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]], i32 [[TMP5]]) #[[ATTR4:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1310,18 +1334,18 @@ int main() {
// CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiET_v()
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP36]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP40]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP37:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP37]]
+// CHECK11-NEXT: [[TMP41:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP41]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -1591,7 +1615,7 @@ int main() {
// CHECK11-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP20]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP20]], align 4
// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 4, ptr [[TMP21]], align 4
// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1608,27 +1632,35 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP27]], align 4
// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 2, ptr [[TMP28]], align 8
-// CHECK11-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
-// CHECK11-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP29]], align 8
+// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
+// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP31]], align 4
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP32]], align 4
+// CHECK11-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
+// CHECK11-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l49(i32 [[TMP2]], ptr [[VEC]], ptr [[S_ARR]], ptr [[TMP3]]) #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
// CHECK11-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP31]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP35]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done2:
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP32:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP32]]
+// CHECK11-NEXT: [[TMP36:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP36]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
diff --git a/clang/test/OpenMP/teams_distribute_simd_private_codegen.cpp b/clang/test/OpenMP/teams_distribute_simd_private_codegen.cpp
index 199aac6efa275..e4fb4beba475d 100644
--- a/clang/test/OpenMP/teams_distribute_simd_private_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_simd_private_codegen.cpp
@@ -244,7 +244,7 @@ int main() {
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -261,9 +261,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -412,7 +420,7 @@ int main() {
// CHECK1-NEXT: store ptr undef, ptr [[_TMP1]], align 8
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -429,27 +437,35 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK1-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK1-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56() #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK1-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK1-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK1: arraydestroy.body:
-// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK1-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK1-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK1: arraydestroy.done2:
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK1-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK1-NEXT: ret i32 [[TMP12]]
+// CHECK1-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK1-NEXT: ret i32 [[TMP16]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN1SIiEC1Ev
@@ -777,7 +793,7 @@ int main() {
// CHECK3-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -794,9 +810,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK3-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK3-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l94() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -943,7 +967,7 @@ int main() {
// CHECK3-NEXT: store ptr undef, ptr [[_TMP1]], align 4
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -960,27 +984,35 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP8]], align 8
-// CHECK3-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK3-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP11]], align 4
+// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK3-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK3-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l56() #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK3-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK3-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK3: arraydestroy.body:
-// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
// CHECK3-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK3-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE2:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK3: arraydestroy.done2:
// CHECK3-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
-// CHECK3-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK3-NEXT: ret i32 [[TMP12]]
+// CHECK3-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK3-NEXT: ret i32 [[TMP16]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN1SIiEC1Ev
diff --git a/clang/test/OpenMP/teams_distribute_simd_reduction_codegen.cpp b/clang/test/OpenMP/teams_distribute_simd_reduction_codegen.cpp
index 1845b3f7e5378..22912243f2015 100644
--- a/clang/test/OpenMP/teams_distribute_simd_reduction_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_simd_reduction_codegen.cpp
@@ -105,7 +105,7 @@ int main() {
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -122,9 +122,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l63.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK1-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP18]], align 4
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l63.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l63(i64 [[TMP1]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -284,7 +292,7 @@ int main() {
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -301,9 +309,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 2, ptr [[TMP15]], align 8
-// CHECK1-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK1-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP18]], align 4
+// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK1-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK1-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32(i64 [[TMP1]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -467,7 +483,7 @@ int main() {
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -484,9 +500,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP14]], align 4
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP15]], align 8
-// CHECK3-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l63.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK3-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP18]], align 4
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK3-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l63.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK3-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l63(i32 [[TMP1]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -646,7 +670,7 @@ int main() {
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP7]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP8]], align 4
// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -663,9 +687,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP14]], align 4
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 2, ptr [[TMP15]], align 8
-// CHECK3-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
-// CHECK3-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP16]], align 8
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP18]], align 4
+// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP19]], align 4
+// CHECK3-NEXT: [[TMP20:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 1, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+// CHECK3-NEXT: br i1 [[TMP21]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32(i32 [[TMP1]]) #[[ATTR2]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/clang/test/OpenMP/teams_firstprivate_codegen.cpp b/clang/test/OpenMP/teams_firstprivate_codegen.cpp
index 7eaab2e002872..4dc09e0fec5f0 100644
--- a/clang/test/OpenMP/teams_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/teams_firstprivate_codegen.cpp
@@ -348,7 +348,7 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
// CHECK9-NEXT: [[TMP20:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 5, ptr [[TMP22]], align 4
// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -365,66 +365,82 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
// CHECK9-NEXT: store ptr null, ptr [[TMP28]], align 8
// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 0, ptr [[TMP29]], align 8
-// CHECK9-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l109.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK9-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK9-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l109.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK9-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l109(i64 [[TMP1]], ptr [[VEC]], ptr [[S_ARR]], ptr [[VAR]], i64 [[TMP3]]) #[[ATTR4:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP32:%.*]] = load i32, ptr [[T_VAR]], align 4
-// CHECK9-NEXT: store i32 [[TMP32]], ptr [[T_VAR_CASTED1]], align 4
-// CHECK9-NEXT: [[TMP33:%.*]] = load i64, ptr [[T_VAR_CASTED1]], align 8
-// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP33]], ptr [[TMP34]], align 8
-// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
-// CHECK9-NEXT: store i64 [[TMP33]], ptr [[TMP35]], align 8
-// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS4]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP36]], align 8
-// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP36:%.*]] = load i32, ptr [[T_VAR]], align 4
+// CHECK9-NEXT: store i32 [[TMP36]], ptr [[T_VAR_CASTED1]], align 4
+// CHECK9-NEXT: [[TMP37:%.*]] = load i64, ptr [[T_VAR_CASTED1]], align 8
+// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP37]], ptr [[TMP38]], align 8
+// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
+// CHECK9-NEXT: store i64 [[TMP37]], ptr [[TMP39]], align 8
+// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS4]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP40]], align 8
+// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP39]], align 4
-// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK9-NEXT: store i32 1, ptr [[TMP40]], align 4
-// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP37]], ptr [[TMP41]], align 8
-// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP38]], ptr [[TMP42]], align 8
-// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.2, ptr [[TMP43]], align 8
-// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP44]], align 8
-// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP45]], align 8
-// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP46]], align 8
-// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK9-NEXT: store i64 0, ptr [[TMP47]], align 8
-// CHECK9-NEXT: [[TMP48:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l116.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK9-NEXT: [[TMP49:%.*]] = icmp ne i32 [[TMP48]], 0
-// CHECK9-NEXT: br i1 [[TMP49]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP43]], align 4
+// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK9-NEXT: store i32 1, ptr [[TMP44]], align 4
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP41]], ptr [[TMP45]], align 8
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP42]], ptr [[TMP46]], align 8
+// CHECK9-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.2, ptr [[TMP47]], align 8
+// CHECK9-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP48]], align 8
+// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP49]], align 8
+// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP50]], align 8
+// CHECK9-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK9-NEXT: store i64 0, ptr [[TMP51]], align 8
+// CHECK9-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP52]], align 8
+// CHECK9-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP53]], align 4
+// CHECK9-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP54]], align 4
+// CHECK9-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP55]], align 4
+// CHECK9-NEXT: [[TMP56:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l116.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK9-NEXT: [[TMP57:%.*]] = icmp ne i32 [[TMP56]], 0
+// CHECK9-NEXT: br i1 [[TMP57]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK9: omp_offload.failed6:
-// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l116(i64 [[TMP33]]) #[[ATTR4]]
+// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l116(i64 [[TMP37]]) #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK9: omp_offload.cont7:
// CHECK9-NEXT: [[CALL:%.*]] = call signext i32 @_Z5tmainIiET_v()
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr nonnull align 4 dereferenceable(4) [[VAR]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP50]], [[OMP_OFFLOAD_CONT7]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP58]], [[OMP_OFFLOAD_CONT7]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE8:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done8:
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP51:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP51]]
+// CHECK9-NEXT: [[TMP59:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP59]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -661,7 +677,7 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP14]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP14]], align 4
// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 4, ptr [[TMP15]], align 4
// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -678,43 +694,59 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
// CHECK9-NEXT: store ptr null, ptr [[TMP21]], align 8
// CHECK9-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 0, ptr [[TMP22]], align 8
-// CHECK9-NEXT: [[TMP23:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l75.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP24:%.*]] = icmp ne i32 [[TMP23]], 0
-// CHECK9-NEXT: br i1 [[TMP24]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP23]], align 8
+// CHECK9-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP24]], align 4
+// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP26]], align 4
+// CHECK9-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l75.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
+// CHECK9-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l75(ptr [[T_VAR]], ptr [[VEC]], ptr [[S_ARR]], ptr [[VAR]]) #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK9: omp_offload.cont:
-// CHECK9-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[T_VAR]], ptr [[TMP25]], align 8
-// CHECK9-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK9-NEXT: store ptr [[T_VAR]], ptr [[TMP26]], align 8
-// CHECK9-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
-// CHECK9-NEXT: store ptr null, ptr [[TMP27]], align 8
-// CHECK9-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP29:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[T_VAR]], ptr [[TMP29]], align 8
+// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK9-NEXT: store ptr [[T_VAR]], ptr [[TMP30]], align 8
+// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
+// CHECK9-NEXT: store ptr null, ptr [[TMP31]], align 8
+// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS4:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK9-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP30]], align 4
-// CHECK9-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 1
-// CHECK9-NEXT: store i32 1, ptr [[TMP31]], align 4
-// CHECK9-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 2
-// CHECK9-NEXT: store ptr [[TMP28]], ptr [[TMP32]], align 8
-// CHECK9-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 3
-// CHECK9-NEXT: store ptr [[TMP29]], ptr [[TMP33]], align 8
-// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 4
-// CHECK9-NEXT: store ptr @.offload_sizes.8, ptr [[TMP34]], align 8
-// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 5
-// CHECK9-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP35]], align 8
-// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 6
-// CHECK9-NEXT: store ptr null, ptr [[TMP36]], align 8
-// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 7
-// CHECK9-NEXT: store ptr null, ptr [[TMP37]], align 8
-// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 8
-// CHECK9-NEXT: store i64 0, ptr [[TMP38]], align 8
-// CHECK9-NEXT: [[TMP39:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l81.region_id, ptr [[KERNEL_ARGS4]])
-// CHECK9-NEXT: [[TMP40:%.*]] = icmp ne i32 [[TMP39]], 0
-// CHECK9-NEXT: br i1 [[TMP40]], label [[OMP_OFFLOAD_FAILED5:%.*]], label [[OMP_OFFLOAD_CONT6:%.*]]
+// CHECK9-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 0
+// CHECK9-NEXT: store i32 2, ptr [[TMP34]], align 4
+// CHECK9-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 1
+// CHECK9-NEXT: store i32 1, ptr [[TMP35]], align 4
+// CHECK9-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 2
+// CHECK9-NEXT: store ptr [[TMP32]], ptr [[TMP36]], align 8
+// CHECK9-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 3
+// CHECK9-NEXT: store ptr [[TMP33]], ptr [[TMP37]], align 8
+// CHECK9-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 4
+// CHECK9-NEXT: store ptr @.offload_sizes.8, ptr [[TMP38]], align 8
+// CHECK9-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 5
+// CHECK9-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP39]], align 8
+// CHECK9-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 6
+// CHECK9-NEXT: store ptr null, ptr [[TMP40]], align 8
+// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 7
+// CHECK9-NEXT: store ptr null, ptr [[TMP41]], align 8
+// CHECK9-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 8
+// CHECK9-NEXT: store i64 0, ptr [[TMP42]], align 8
+// CHECK9-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP43]], align 8
+// CHECK9-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP44]], align 4
+// CHECK9-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP45]], align 4
+// CHECK9-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP46]], align 4
+// CHECK9-NEXT: [[TMP47:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l81.region_id, ptr [[KERNEL_ARGS4]])
+// CHECK9-NEXT: [[TMP48:%.*]] = icmp ne i32 [[TMP47]], 0
+// CHECK9-NEXT: br i1 [[TMP48]], label [[OMP_OFFLOAD_FAILED5:%.*]], label [[OMP_OFFLOAD_CONT6:%.*]]
// CHECK9: omp_offload.failed5:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l81(ptr [[T_VAR]]) #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT6]]
@@ -722,18 +754,18 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
// CHECK9-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr nonnull align 4 dereferenceable(4) [[VAR]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP41]], [[OMP_OFFLOAD_CONT6]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP49]], [[OMP_OFFLOAD_CONT6]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE7:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done7:
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP42:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP42]]
+// CHECK9-NEXT: [[TMP50:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP50]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
@@ -1109,7 +1141,7 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
// CHECK11-NEXT: [[TMP20:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP21]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP21]], align 4
// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 5, ptr [[TMP22]], align 4
// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1126,66 +1158,82 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
// CHECK11-NEXT: store ptr null, ptr [[TMP28]], align 4
// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 0, ptr [[TMP29]], align 8
-// CHECK11-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l109.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
-// CHECK11-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP30]], align 8
+// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP31]], align 4
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP33]], align 4
+// CHECK11-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l109.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK11-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l109(i32 [[TMP1]], ptr [[VEC]], ptr [[S_ARR]], ptr [[VAR]], i32 [[TMP3]]) #[[ATTR4:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP32:%.*]] = load i32, ptr [[T_VAR]], align 4
-// CHECK11-NEXT: store i32 [[TMP32]], ptr [[T_VAR_CASTED1]], align 4
-// CHECK11-NEXT: [[TMP33:%.*]] = load i32, ptr [[T_VAR_CASTED1]], align 4
-// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP33]], ptr [[TMP34]], align 4
-// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
-// CHECK11-NEXT: store i32 [[TMP33]], ptr [[TMP35]], align 4
-// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS4]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP36]], align 4
-// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP36:%.*]] = load i32, ptr [[T_VAR]], align 4
+// CHECK11-NEXT: store i32 [[TMP36]], ptr [[T_VAR_CASTED1]], align 4
+// CHECK11-NEXT: [[TMP37:%.*]] = load i32, ptr [[T_VAR_CASTED1]], align 4
+// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP37]], ptr [[TMP38]], align 4
+// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
+// CHECK11-NEXT: store i32 [[TMP37]], ptr [[TMP39]], align 4
+// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS4]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP40]], align 4
+// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP39]], align 4
-// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
-// CHECK11-NEXT: store i32 1, ptr [[TMP40]], align 4
-// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP37]], ptr [[TMP41]], align 4
-// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP38]], ptr [[TMP42]], align 4
-// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.2, ptr [[TMP43]], align 4
-// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP44]], align 4
-// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP45]], align 4
-// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP46]], align 4
-// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
-// CHECK11-NEXT: store i64 0, ptr [[TMP47]], align 8
-// CHECK11-NEXT: [[TMP48:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l116.region_id, ptr [[KERNEL_ARGS5]])
-// CHECK11-NEXT: [[TMP49:%.*]] = icmp ne i32 [[TMP48]], 0
-// CHECK11-NEXT: br i1 [[TMP49]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
+// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP43]], align 4
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
+// CHECK11-NEXT: store i32 1, ptr [[TMP44]], align 4
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP41]], ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP42]], ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.2, ptr [[TMP47]], align 4
+// CHECK11-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.3, ptr [[TMP48]], align 4
+// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP49]], align 4
+// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP50]], align 4
+// CHECK11-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
+// CHECK11-NEXT: store i64 0, ptr [[TMP51]], align 8
+// CHECK11-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP52]], align 8
+// CHECK11-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP53]], align 4
+// CHECK11-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP54]], align 4
+// CHECK11-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP55]], align 4
+// CHECK11-NEXT: [[TMP56:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l116.region_id, ptr [[KERNEL_ARGS5]])
+// CHECK11-NEXT: [[TMP57:%.*]] = icmp ne i32 [[TMP56]], 0
+// CHECK11-NEXT: br i1 [[TMP57]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK11: omp_offload.failed6:
-// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l116(i32 [[TMP33]]) #[[ATTR4]]
+// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l116(i32 [[TMP37]]) #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT7]]
// CHECK11: omp_offload.cont7:
// CHECK11-NEXT: [[CALL:%.*]] = call i32 @_Z5tmainIiET_v()
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr nonnull align 4 dereferenceable(4) [[VAR]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP50]], [[OMP_OFFLOAD_CONT7]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP58]], [[OMP_OFFLOAD_CONT7]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE8:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done8:
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP51:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP51]]
+// CHECK11-NEXT: [[TMP59:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP59]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
@@ -1422,7 +1470,7 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
// CHECK11-NEXT: [[TMP13:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP14]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP14]], align 4
// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 4, ptr [[TMP15]], align 4
// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1439,43 +1487,59 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
// CHECK11-NEXT: store ptr null, ptr [[TMP21]], align 4
// CHECK11-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 0, ptr [[TMP22]], align 8
-// CHECK11-NEXT: [[TMP23:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l75.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP24:%.*]] = icmp ne i32 [[TMP23]], 0
-// CHECK11-NEXT: br i1 [[TMP24]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP23]], align 8
+// CHECK11-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP24]], align 4
+// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP25]], align 4
+// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP26]], align 4
+// CHECK11-NEXT: [[TMP27:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l75.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0
+// CHECK11-NEXT: br i1 [[TMP28]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l75(ptr [[T_VAR]], ptr [[VEC]], ptr [[S_ARR]], ptr [[VAR]]) #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK11: omp_offload.cont:
-// CHECK11-NEXT: [[TMP25:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[T_VAR]], ptr [[TMP25]], align 4
-// CHECK11-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK11-NEXT: store ptr [[T_VAR]], ptr [[TMP26]], align 4
-// CHECK11-NEXT: [[TMP27:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
-// CHECK11-NEXT: store ptr null, ptr [[TMP27]], align 4
-// CHECK11-NEXT: [[TMP28:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP29:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[T_VAR]], ptr [[TMP29]], align 4
+// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
+// CHECK11-NEXT: store ptr [[T_VAR]], ptr [[TMP30]], align 4
+// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
+// CHECK11-NEXT: store ptr null, ptr [[TMP31]], align 4
+// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
+// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS4:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
-// CHECK11-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP30]], align 4
-// CHECK11-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 1
-// CHECK11-NEXT: store i32 1, ptr [[TMP31]], align 4
-// CHECK11-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 2
-// CHECK11-NEXT: store ptr [[TMP28]], ptr [[TMP32]], align 4
-// CHECK11-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 3
-// CHECK11-NEXT: store ptr [[TMP29]], ptr [[TMP33]], align 4
-// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 4
-// CHECK11-NEXT: store ptr @.offload_sizes.8, ptr [[TMP34]], align 4
-// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 5
-// CHECK11-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP35]], align 4
-// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 6
-// CHECK11-NEXT: store ptr null, ptr [[TMP36]], align 4
-// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 7
-// CHECK11-NEXT: store ptr null, ptr [[TMP37]], align 4
-// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 8
-// CHECK11-NEXT: store i64 0, ptr [[TMP38]], align 8
-// CHECK11-NEXT: [[TMP39:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l81.region_id, ptr [[KERNEL_ARGS4]])
-// CHECK11-NEXT: [[TMP40:%.*]] = icmp ne i32 [[TMP39]], 0
-// CHECK11-NEXT: br i1 [[TMP40]], label [[OMP_OFFLOAD_FAILED5:%.*]], label [[OMP_OFFLOAD_CONT6:%.*]]
+// CHECK11-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 0
+// CHECK11-NEXT: store i32 2, ptr [[TMP34]], align 4
+// CHECK11-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 1
+// CHECK11-NEXT: store i32 1, ptr [[TMP35]], align 4
+// CHECK11-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 2
+// CHECK11-NEXT: store ptr [[TMP32]], ptr [[TMP36]], align 4
+// CHECK11-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 3
+// CHECK11-NEXT: store ptr [[TMP33]], ptr [[TMP37]], align 4
+// CHECK11-NEXT: [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 4
+// CHECK11-NEXT: store ptr @.offload_sizes.8, ptr [[TMP38]], align 4
+// CHECK11-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 5
+// CHECK11-NEXT: store ptr @.offload_maptypes.9, ptr [[TMP39]], align 4
+// CHECK11-NEXT: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 6
+// CHECK11-NEXT: store ptr null, ptr [[TMP40]], align 4
+// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 7
+// CHECK11-NEXT: store ptr null, ptr [[TMP41]], align 4
+// CHECK11-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 8
+// CHECK11-NEXT: store i64 0, ptr [[TMP42]], align 8
+// CHECK11-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP43]], align 8
+// CHECK11-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP44]], align 4
+// CHECK11-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP45]], align 4
+// CHECK11-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP46]], align 4
+// CHECK11-NEXT: [[TMP47:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l81.region_id, ptr [[KERNEL_ARGS4]])
+// CHECK11-NEXT: [[TMP48:%.*]] = icmp ne i32 [[TMP47]], 0
+// CHECK11-NEXT: br i1 [[TMP48]], label [[OMP_OFFLOAD_FAILED5:%.*]], label [[OMP_OFFLOAD_CONT6:%.*]]
// CHECK11: omp_offload.failed5:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l81(ptr [[T_VAR]]) #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT6]]
@@ -1483,18 +1547,18 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
// CHECK11-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr nonnull align 4 dereferenceable(4) [[VAR]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP41]], [[OMP_OFFLOAD_CONT6]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP49]], [[OMP_OFFLOAD_CONT6]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE7:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done7:
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP42:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP42]]
+// CHECK11-NEXT: [[TMP50:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP50]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
@@ -1898,7 +1962,7 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
// CHECK17-NEXT: [[TMP42:%.*]] = getelementptr inbounds [8 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK17-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP43]], align 4
+// CHECK17-NEXT: store i32 2, ptr [[TMP43]], align 4
// CHECK17-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK17-NEXT: store i32 8, ptr [[TMP44]], align 4
// CHECK17-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1915,15 +1979,23 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
// CHECK17-NEXT: store ptr null, ptr [[TMP50]], align 8
// CHECK17-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK17-NEXT: store i64 0, ptr [[TMP51]], align 8
-// CHECK17-NEXT: [[TMP52:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z10array_funcPfP2StiPg_l152.region_id, ptr [[KERNEL_ARGS]])
-// CHECK17-NEXT: [[TMP53:%.*]] = icmp ne i32 [[TMP52]], 0
-// CHECK17-NEXT: br i1 [[TMP53]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK17-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP52]], align 8
+// CHECK17-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP53]], align 4
+// CHECK17-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP54]], align 4
+// CHECK17-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP55]], align 4
+// CHECK17-NEXT: [[TMP56:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z10array_funcPfP2StiPg_l152.region_id, ptr [[KERNEL_ARGS]])
+// CHECK17-NEXT: [[TMP57:%.*]] = icmp ne i32 [[TMP56]], 0
+// CHECK17-NEXT: br i1 [[TMP57]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK17: omp_offload.failed:
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z10array_funcPfP2StiPg_l152(ptr [[TMP8]], ptr [[TMP9]], i64 [[TMP1]], ptr [[TMP10]], i64 [[TMP3]], i64 [[TMP5]], ptr [[VLA]], i64 [[TMP12]]) #[[ATTR4:[0-9]+]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK17: omp_offload.cont:
-// CHECK17-NEXT: [[TMP54:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK17-NEXT: call void @llvm.stackrestore(ptr [[TMP54]])
+// CHECK17-NEXT: [[TMP58:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK17-NEXT: call void @llvm.stackrestore(ptr [[TMP58]])
// CHECK17-NEXT: ret void
//
//
@@ -2128,7 +2200,7 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
// CHECK17-NEXT: [[TMP54:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK17-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK17-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK17-NEXT: store i32 1, ptr [[TMP55]], align 4
+// CHECK17-NEXT: store i32 2, ptr [[TMP55]], align 4
// CHECK17-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK17-NEXT: store i32 10, ptr [[TMP56]], align 4
// CHECK17-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2145,15 +2217,23 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
// CHECK17-NEXT: store ptr null, ptr [[TMP62]], align 8
// CHECK17-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK17-NEXT: store i64 0, ptr [[TMP63]], align 8
-// CHECK17-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2St7St_funcEPS_iPg_l144.region_id, ptr [[KERNEL_ARGS]])
-// CHECK17-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0
-// CHECK17-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK17-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK17-NEXT: store i64 0, ptr [[TMP64]], align 8
+// CHECK17-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP65]], align 4
+// CHECK17-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP66]], align 4
+// CHECK17-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK17-NEXT: store i32 0, ptr [[TMP67]], align 4
+// CHECK17-NEXT: [[TMP68:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2St7St_funcEPS_iPg_l144.region_id, ptr [[KERNEL_ARGS]])
+// CHECK17-NEXT: [[TMP69:%.*]] = icmp ne i32 [[TMP68]], 0
+// CHECK17-NEXT: br i1 [[TMP69]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK17: omp_offload.failed:
// CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2St7St_funcEPS_iPg_l144(ptr [[TMP9]], i64 [[TMP1]], ptr [[TMP10]], i64 [[TMP3]], i64 [[TMP5]], ptr [[VLA]], ptr [[THIS1]], i64 [[TMP12]]) #[[ATTR4]]
// CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK17: omp_offload.cont:
-// CHECK17-NEXT: [[TMP66:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK17-NEXT: call void @llvm.stackrestore(ptr [[TMP66]])
+// CHECK17-NEXT: [[TMP70:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK17-NEXT: call void @llvm.stackrestore(ptr [[TMP70]])
// CHECK17-NEXT: ret void
//
//
@@ -2352,7 +2432,7 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
// CHECK19-NEXT: [[TMP40:%.*]] = getelementptr inbounds [8 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK19-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP41]], align 4
+// CHECK19-NEXT: store i32 2, ptr [[TMP41]], align 4
// CHECK19-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK19-NEXT: store i32 8, ptr [[TMP42]], align 4
// CHECK19-NEXT: [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2369,15 +2449,23 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
// CHECK19-NEXT: store ptr null, ptr [[TMP48]], align 4
// CHECK19-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK19-NEXT: store i64 0, ptr [[TMP49]], align 8
-// CHECK19-NEXT: [[TMP50:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z10array_funcPfP2StiPe_l152.region_id, ptr [[KERNEL_ARGS]])
-// CHECK19-NEXT: [[TMP51:%.*]] = icmp ne i32 [[TMP50]], 0
-// CHECK19-NEXT: br i1 [[TMP51]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK19-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP50]], align 8
+// CHECK19-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP51]], align 4
+// CHECK19-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP52]], align 4
+// CHECK19-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP53]], align 4
+// CHECK19-NEXT: [[TMP54:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z10array_funcPfP2StiPe_l152.region_id, ptr [[KERNEL_ARGS]])
+// CHECK19-NEXT: [[TMP55:%.*]] = icmp ne i32 [[TMP54]], 0
+// CHECK19-NEXT: br i1 [[TMP55]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK19: omp_offload.failed:
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z10array_funcPfP2StiPe_l152(ptr [[TMP5]], ptr [[TMP6]], i32 [[TMP0]], ptr [[TMP7]], i32 [[TMP1]], i32 [[TMP2]], ptr [[VLA]], i32 [[TMP9]]) #[[ATTR4:[0-9]+]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK19: omp_offload.cont:
-// CHECK19-NEXT: [[TMP52:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK19-NEXT: call void @llvm.stackrestore(ptr [[TMP52]])
+// CHECK19-NEXT: [[TMP56:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK19-NEXT: call void @llvm.stackrestore(ptr [[TMP56]])
// CHECK19-NEXT: ret void
//
//
@@ -2580,7 +2668,7 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
// CHECK19-NEXT: [[TMP52:%.*]] = getelementptr inbounds [10 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
// CHECK19-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK19-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK19-NEXT: store i32 1, ptr [[TMP53]], align 4
+// CHECK19-NEXT: store i32 2, ptr [[TMP53]], align 4
// CHECK19-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK19-NEXT: store i32 10, ptr [[TMP54]], align 4
// CHECK19-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -2597,15 +2685,23 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
// CHECK19-NEXT: store ptr null, ptr [[TMP60]], align 4
// CHECK19-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK19-NEXT: store i64 0, ptr [[TMP61]], align 8
-// CHECK19-NEXT: [[TMP62:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2St7St_funcEPS_iPe_l144.region_id, ptr [[KERNEL_ARGS]])
-// CHECK19-NEXT: [[TMP63:%.*]] = icmp ne i32 [[TMP62]], 0
-// CHECK19-NEXT: br i1 [[TMP63]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK19-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK19-NEXT: store i64 0, ptr [[TMP62]], align 8
+// CHECK19-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP63]], align 4
+// CHECK19-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP64]], align 4
+// CHECK19-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK19-NEXT: store i32 0, ptr [[TMP65]], align 4
+// CHECK19-NEXT: [[TMP66:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2St7St_funcEPS_iPe_l144.region_id, ptr [[KERNEL_ARGS]])
+// CHECK19-NEXT: [[TMP67:%.*]] = icmp ne i32 [[TMP66]], 0
+// CHECK19-NEXT: br i1 [[TMP67]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK19: omp_offload.failed:
// CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2St7St_funcEPS_iPe_l144(ptr [[TMP6]], i32 [[TMP0]], ptr [[TMP7]], i32 [[TMP1]], i32 [[TMP2]], ptr [[VLA]], ptr [[THIS1]], i32 [[TMP9]]) #[[ATTR4]]
// CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK19: omp_offload.cont:
-// CHECK19-NEXT: [[TMP64:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
-// CHECK19-NEXT: call void @llvm.stackrestore(ptr [[TMP64]])
+// CHECK19-NEXT: [[TMP68:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
+// CHECK19-NEXT: call void @llvm.stackrestore(ptr [[TMP68]])
// CHECK19-NEXT: ret void
//
//
diff --git a/clang/test/OpenMP/teams_private_codegen.cpp b/clang/test/OpenMP/teams_private_codegen.cpp
index 423b2ad84293f..b8bb9ed3e7484 100644
--- a/clang/test/OpenMP/teams_private_codegen.cpp
+++ b/clang/test/OpenMP/teams_private_codegen.cpp
@@ -219,7 +219,7 @@ int main() {
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4
+// CHECK1-NEXT: store i32 2, ptr [[TMP6]], align 4
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP7]], align 4
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -236,9 +236,17 @@ int main() {
// CHECK1-NEXT: store ptr null, ptr [[TMP13]], align 8
// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8
-// CHECK1-NEXT: [[TMP15:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSC1ERi_l48.region_id, ptr [[KERNEL_ARGS]])
-// CHECK1-NEXT: [[TMP16:%.*]] = icmp ne i32 [[TMP15]], 0
-// CHECK1-NEXT: br i1 [[TMP16]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK1-NEXT: store i64 0, ptr [[TMP15]], align 8
+// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK1-NEXT: store i32 0, ptr [[TMP18]], align 4
+// CHECK1-NEXT: [[TMP19:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSC1ERi_l48.region_id, ptr [[KERNEL_ARGS]])
+// CHECK1-NEXT: [[TMP20:%.*]] = icmp ne i32 [[TMP19]], 0
+// CHECK1-NEXT: br i1 [[TMP20]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK1: omp_offload.failed:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSC1ERi_l48(ptr [[THIS1]]) #[[ATTR4:[0-9]+]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -404,7 +412,7 @@ int main() {
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4
+// CHECK3-NEXT: store i32 2, ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK3-NEXT: store i32 1, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -421,9 +429,17 @@ int main() {
// CHECK3-NEXT: store ptr null, ptr [[TMP13]], align 4
// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8
-// CHECK3-NEXT: [[TMP15:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSC1ERi_l48.region_id, ptr [[KERNEL_ARGS]])
-// CHECK3-NEXT: [[TMP16:%.*]] = icmp ne i32 [[TMP15]], 0
-// CHECK3-NEXT: br i1 [[TMP16]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK3-NEXT: store i64 0, ptr [[TMP15]], align 8
+// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK3-NEXT: store i32 0, ptr [[TMP18]], align 4
+// CHECK3-NEXT: [[TMP19:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSC1ERi_l48.region_id, ptr [[KERNEL_ARGS]])
+// CHECK3-NEXT: [[TMP20:%.*]] = icmp ne i32 [[TMP19]], 0
+// CHECK3-NEXT: br i1 [[TMP20]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK3: omp_offload.failed:
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSC1ERi_l48(ptr [[THIS1]]) #[[ATTR4:[0-9]+]]
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -555,7 +571,7 @@ int main() {
// CHECK9-NEXT: call void @_ZN1SIfEC1Ef(ptr noundef nonnull align 4 dereferenceable(4) [[VAR]], float noundef 3.000000e+00)
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK9-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK9-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -572,9 +588,17 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK9-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 0, ptr [[TMP8]], align 8
-// CHECK9-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l136.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK9-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK9-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK9-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l136.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK9-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l136() #[[ATTR4:[0-9]+]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -583,18 +607,18 @@ int main() {
// CHECK9-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[VAR]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done1:
// CHECK9-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP12]]
+// CHECK9-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP16]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN2SSC1ERi
@@ -714,7 +738,7 @@ int main() {
// CHECK9-NEXT: call void @_ZN1SIiEC1Ei(ptr noundef nonnull align 4 dereferenceable(4) [[VAR]], i32 noundef signext 3)
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK9-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK9-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -731,9 +755,17 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK9-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 0, ptr [[TMP8]], align 8
-// CHECK9-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l86.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK9-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK9-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK9-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK9-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l86.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK9-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l86() #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -741,18 +773,18 @@ int main() {
// CHECK9-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[VAR]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK9-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i64 2
// CHECK9-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK9: arraydestroy.body:
-// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK9-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK9-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK9-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK9: arraydestroy.done1:
// CHECK9-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK9-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK9-NEXT: ret i32 [[TMP12]]
+// CHECK9-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK9-NEXT: ret i32 [[TMP16]]
//
//
// CHECK9-LABEL: define {{[^@]+}}@_ZN2SSC2ERi
@@ -786,7 +818,7 @@ int main() {
// CHECK9-NEXT: [[TMP5:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP6]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP6]], align 4
// CHECK9-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 1, ptr [[TMP7]], align 4
// CHECK9-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -803,9 +835,17 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP13]], align 8
// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 0, ptr [[TMP14]], align 8
-// CHECK9-NEXT: [[TMP15:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSC1ERi_l48.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP16:%.*]] = icmp ne i32 [[TMP15]], 0
-// CHECK9-NEXT: br i1 [[TMP16]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP15]], align 8
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK9-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP18]], align 4
+// CHECK9-NEXT: [[TMP19:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSC1ERi_l48.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP20:%.*]] = icmp ne i32 [[TMP19]], 0
+// CHECK9-NEXT: br i1 [[TMP20]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSC1ERi_l48(ptr [[THIS1]]) #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1011,7 +1051,7 @@ int main() {
// CHECK9-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK9-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK9-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK9-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK9-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK9-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK9-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK9-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1028,9 +1068,17 @@ int main() {
// CHECK9-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK9-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK9-NEXT: store i64 0, ptr [[TMP13]], align 8
-// CHECK9-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN3SSTIiEC1Ev_l64.region_id, ptr [[KERNEL_ARGS]])
-// CHECK9-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK9-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK9-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK9-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK9-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK9-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK9-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK9-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK9-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK9-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN3SSTIiEC1Ev_l64.region_id, ptr [[KERNEL_ARGS]])
+// CHECK9-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK9-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK9: omp_offload.failed:
// CHECK9-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN3SSTIiEC1Ev_l64(ptr [[THIS1]]) #[[ATTR4]]
// CHECK9-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1120,7 +1168,7 @@ int main() {
// CHECK11-NEXT: call void @_ZN1SIfEC1Ef(ptr noundef nonnull align 4 dereferenceable(4) [[VAR]], float noundef 3.000000e+00)
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK11-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK11-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1137,9 +1185,17 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK11-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 0, ptr [[TMP8]], align 8
-// CHECK11-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l136.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK11-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK11-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK11-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK11-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l136.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK11-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l136() #[[ATTR4:[0-9]+]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1148,18 +1204,18 @@ int main() {
// CHECK11-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[VAR]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done1:
// CHECK11-NEXT: call void @_ZN1SIfED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP12]]
+// CHECK11-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP16]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN2SSC1ERi
@@ -1279,7 +1335,7 @@ int main() {
// CHECK11-NEXT: call void @_ZN1SIiEC1Ei(ptr noundef nonnull align 4 dereferenceable(4) [[VAR]], i32 noundef 3)
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP0]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP0]], align 4
// CHECK11-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 0, ptr [[TMP1]], align 4
// CHECK11-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1296,9 +1352,17 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK11-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 0, ptr [[TMP8]], align 8
-// CHECK11-NEXT: [[TMP9:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l86.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP10:%.*]] = icmp ne i32 [[TMP9]], 0
-// CHECK11-NEXT: br i1 [[TMP10]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP9]], align 8
+// CHECK11-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP10]], align 4
+// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
+// CHECK11-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP12]], align 4
+// CHECK11-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l86.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
+// CHECK11-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l86() #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1306,18 +1370,18 @@ int main() {
// CHECK11-NEXT: store i32 0, ptr [[RETVAL]], align 4
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[VAR]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], ptr [[S_ARR]], i32 0, i32 0
-// CHECK11-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAY_BEGIN]], i32 2
// CHECK11-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK11: arraydestroy.body:
-// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP11]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP15]], [[OMP_OFFLOAD_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK11-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
// CHECK11-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK11-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK11: arraydestroy.done1:
// CHECK11-NEXT: call void @_ZN1SIiED1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR4]]
-// CHECK11-NEXT: [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK11-NEXT: ret i32 [[TMP12]]
+// CHECK11-NEXT: [[TMP16:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK11-NEXT: ret i32 [[TMP16]]
//
//
// CHECK11-LABEL: define {{[^@]+}}@_ZN2SSC2ERi
@@ -1351,7 +1415,7 @@ int main() {
// CHECK11-NEXT: [[TMP5:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP6]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP6]], align 4
// CHECK11-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 1, ptr [[TMP7]], align 4
// CHECK11-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1368,9 +1432,17 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP13]], align 4
// CHECK11-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 0, ptr [[TMP14]], align 8
-// CHECK11-NEXT: [[TMP15:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSC1ERi_l48.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP16:%.*]] = icmp ne i32 [[TMP15]], 0
-// CHECK11-NEXT: br i1 [[TMP16]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP15]], align 8
+// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP17]], align 4
+// CHECK11-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP18]], align 4
+// CHECK11-NEXT: [[TMP19:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSC1ERi_l48.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP20:%.*]] = icmp ne i32 [[TMP19]], 0
+// CHECK11-NEXT: br i1 [[TMP20]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSC1ERi_l48(ptr [[THIS1]]) #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
@@ -1576,7 +1648,7 @@ int main() {
// CHECK11-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK11-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// CHECK11-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK11-NEXT: store i32 1, ptr [[TMP5]], align 4
+// CHECK11-NEXT: store i32 2, ptr [[TMP5]], align 4
// CHECK11-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// CHECK11-NEXT: store i32 1, ptr [[TMP6]], align 4
// CHECK11-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
@@ -1593,9 +1665,17 @@ int main() {
// CHECK11-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK11-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// CHECK11-NEXT: store i64 0, ptr [[TMP13]], align 8
-// CHECK11-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN3SSTIiEC1Ev_l64.region_id, ptr [[KERNEL_ARGS]])
-// CHECK11-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
-// CHECK11-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK11-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
+// CHECK11-NEXT: store i64 0, ptr [[TMP14]], align 8
+// CHECK11-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
+// CHECK11-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
+// CHECK11-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
+// CHECK11-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
+// CHECK11-NEXT: store i32 0, ptr [[TMP17]], align 4
+// CHECK11-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN3SSTIiEC1Ev_l64.region_id, ptr [[KERNEL_ARGS]])
+// CHECK11-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
+// CHECK11-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// CHECK11: omp_offload.failed:
// CHECK11-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN3SSTIiEC1Ev_l64(ptr [[THIS1]]) #[[ATTR4]]
// CHECK11-NEXT: br label [[OMP_OFFLOAD_CONT]]
diff --git a/llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h b/llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h
index d0f18bf0924ac..84c062978a32e 100644
--- a/llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h
+++ b/llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h
@@ -988,12 +988,10 @@ class OpenMPIRBuilder {
/// \param NumThreads Number of threads via the 'thread_limit' clause.
/// \param HostPtr Pointer to the host-side pointer of the target kernel.
/// \param KernelArgs Array of arguments to the kernel.
- /// \param NoWaitArgs Optional array of arguments to the nowait kernel.
InsertPointTy emitTargetKernel(const LocationDescription &Loc, Value *&Return,
Value *Ident, Value *DeviceID, Value *NumTeams,
Value *NumThreads, Value *HostPtr,
- ArrayRef<Value *> KernelArgs,
- ArrayRef<Value *> NoWaitArgs = {});
+ ArrayRef<Value *> KernelArgs);
/// Generate a barrier runtime call.
///
diff --git a/llvm/include/llvm/Frontend/OpenMP/OMPKinds.def b/llvm/include/llvm/Frontend/OpenMP/OMPKinds.def
index 6d7f3adfc0ae5..8a09fb7cb7a6f 100644
--- a/llvm/include/llvm/Frontend/OpenMP/OMPKinds.def
+++ b/llvm/include/llvm/Frontend/OpenMP/OMPKinds.def
@@ -40,6 +40,7 @@ __OMP_TYPE(Int64Ptr)
__OMP_TYPE(Double)
OMP_TYPE(SizeTy, M.getDataLayout().getIntPtrType(Ctx))
+OMP_TYPE(Int63, Type::getIntNTy(Ctx, 63))
#define __OMP_PTR_TYPE(NAME, BASE) OMP_TYPE(NAME, BASE->getPointerTo())
@@ -69,6 +70,7 @@ __OMP_PTR_TYPE(Int8PtrPtrPtr, Int8PtrPtr)
OMP_ARRAY_TYPE(VarName, ElemTy, ArraySize)
__OMP_ARRAY_TYPE(KmpCriticalName, Int32, 8)
+__OMP_ARRAY_TYPE(Int32Arr3, Int32, 3)
#undef __OMP_ARRAY_TYPE
#undef OMP_ARRAY_TYPE
@@ -80,19 +82,20 @@ __OMP_ARRAY_TYPE(KmpCriticalName, Int32, 8)
///{
#ifndef OMP_STRUCT_TYPE
-#define OMP_STRUCT_TYPE(VarName, StructName, ...)
+#define OMP_STRUCT_TYPE(VarName, StructName, Packed, ...)
#endif
-#define __OMP_STRUCT_TYPE(VarName, Name, ...) \
- OMP_STRUCT_TYPE(VarName, "struct." #Name, __VA_ARGS__)
+#define __OMP_STRUCT_TYPE(VarName, Name, Packed, ...) \
+ OMP_STRUCT_TYPE(VarName, "struct." #Name, Packed, __VA_ARGS__)
-__OMP_STRUCT_TYPE(Ident, ident_t, Int32, Int32, Int32, Int32, Int8Ptr)
-__OMP_STRUCT_TYPE(OffloadEntry, __tgt_offload_entry, Int8Ptr, Int8Ptr, SizeTy,
+__OMP_STRUCT_TYPE(Ident, ident_t, false, Int32, Int32, Int32, Int32, Int8Ptr)
+__OMP_STRUCT_TYPE(OffloadEntry, __tgt_offload_entry, false, Int8Ptr, Int8Ptr, SizeTy,
Int32, Int32)
-__OMP_STRUCT_TYPE(KernelArgs, __tgt_kernel_arguments, Int32, Int32, VoidPtrPtr,
- VoidPtrPtr, Int64Ptr, Int64Ptr, VoidPtrPtr, VoidPtrPtr, Int64)
-__OMP_STRUCT_TYPE(AsyncInfo, __tgt_async_info, Int8Ptr)
-__OMP_STRUCT_TYPE(DependInfo, kmp_dep_info, SizeTy, SizeTy, Int8)
+__OMP_STRUCT_TYPE(KernelArgs, __tgt_kernel_arguments, false, Int32, Int32, VoidPtrPtr,
+ VoidPtrPtr, Int64Ptr, Int64Ptr, VoidPtrPtr, VoidPtrPtr,
+ Int64, Int64, Int32Arr3Ty, Int32Arr3Ty, Int32)
+__OMP_STRUCT_TYPE(AsyncInfo, __tgt_async_info, false, Int8Ptr)
+__OMP_STRUCT_TYPE(DependInfo, kmp_dep_info, false, SizeTy, SizeTy, Int8)
#undef __OMP_STRUCT_TYPE
#undef OMP_STRUCT_TYPE
diff --git a/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp b/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
index c4c33fd2ef2d2..e07c0aa8f6314 100644
--- a/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
+++ b/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
@@ -815,7 +815,7 @@ void OpenMPIRBuilder::emitOffloadingEntry(Constant *Addr, StringRef Name,
OpenMPIRBuilder::InsertPointTy OpenMPIRBuilder::emitTargetKernel(
const LocationDescription &Loc, Value *&Return, Value *Ident,
Value *DeviceID, Value *NumTeams, Value *NumThreads, Value *HostPtr,
- ArrayRef<Value *> KernelArgs, ArrayRef<Value *> NoWaitArgs) {
+ ArrayRef<Value *> KernelArgs) {
if (!updateToLocation(Loc))
return Loc.IP;
@@ -829,16 +829,11 @@ OpenMPIRBuilder::InsertPointTy OpenMPIRBuilder::emitTargetKernel(
M.getDataLayout().getPrefTypeAlign(KernelArgs[I]->getType()));
}
- bool HasNoWait = !NoWaitArgs.empty();
SmallVector<Value *> OffloadingArgs{Ident, DeviceID, NumTeams,
NumThreads, HostPtr, KernelArgsPtr};
- if (HasNoWait)
- OffloadingArgs.append(NoWaitArgs.begin(), NoWaitArgs.end());
Return = Builder.CreateCall(
- HasNoWait
- ? getOrCreateRuntimeFunction(M, OMPRTL___tgt_target_kernel_nowait)
- : getOrCreateRuntimeFunction(M, OMPRTL___tgt_target_kernel),
+ getOrCreateRuntimeFunction(M, OMPRTL___tgt_target_kernel),
OffloadingArgs);
return Builder.saveIP();
@@ -4739,10 +4734,10 @@ void OpenMPIRBuilder::initializeTypes(Module &M) {
#define OMP_FUNCTION_TYPE(VarName, IsVarArg, ReturnType, ...) \
VarName = FunctionType::get(ReturnType, {__VA_ARGS__}, IsVarArg); \
VarName##Ptr = PointerType::getUnqual(VarName);
-#define OMP_STRUCT_TYPE(VarName, StructName, ...) \
+#define OMP_STRUCT_TYPE(VarName, StructName, Packed, ...) \
T = StructType::getTypeByName(Ctx, StructName); \
if (!T) \
- T = StructType::create(Ctx, {__VA_ARGS__}, StructName); \
+ T = StructType::create(Ctx, {__VA_ARGS__}, StructName, Packed); \
VarName = T; \
VarName##Ptr = PointerType::getUnqual(T);
#include "llvm/Frontend/OpenMP/OMPKinds.def"
diff --git a/openmp/libomptarget/include/device.h b/openmp/libomptarget/include/device.h
index 0ddf75f2d264c..d4179857d3bd6 100644
--- a/openmp/libomptarget/include/device.h
+++ b/openmp/libomptarget/include/device.h
@@ -444,12 +444,10 @@ struct DeviceTy {
int32_t dataExchange(void *SrcPtr, DeviceTy &DstDev, void *DstPtr,
int64_t Size, AsyncInfoTy &AsyncInfo);
- int32_t runRegion(void *TgtEntryPtr, void **TgtVarsPtr, ptr
diff _t *TgtOffsets,
- int32_t TgtVarsSize, AsyncInfoTy &AsyncInfo);
- int32_t runTeamRegion(void *TgtEntryPtr, void **TgtVarsPtr,
- ptr
diff _t *TgtOffsets, int32_t TgtVarsSize,
- int32_t NumTeams, int32_t ThreadLimit,
- uint64_t LoopTripCount, AsyncInfoTy &AsyncInfo);
+ // Launch the kernel identified by \p TgtEntryPtr with the given arguments.
+ int32_t launchKernel(void *TgtEntryPtr, void **TgtVarsPtr,
+ ptr
diff _t *TgtOffsets, const KernelArgsTy &KernelArgs,
+ AsyncInfoTy &AsyncInfo);
/// Synchronize device/queue/event based on \p AsyncInfo and return
/// OFFLOAD_SUCCESS/OFFLOAD_FAIL when succeeds/fails.
diff --git a/openmp/libomptarget/include/omptarget.h b/openmp/libomptarget/include/omptarget.h
index 76b686794e22e..4ea30802ef91b 100644
--- a/openmp/libomptarget/include/omptarget.h
+++ b/openmp/libomptarget/include/omptarget.h
@@ -14,6 +14,7 @@
#ifndef _OMPTARGET_H_
#define _OMPTARGET_H_
+#include <cstdint>
#include <deque>
#include <functional>
#include <stddef.h>
@@ -114,20 +115,31 @@ enum TargetAllocTy : int32_t {
};
/// This struct contains all of the arguments to a target kernel region launch.
-struct __tgt_kernel_arguments {
- int32_t Version; // Version of this struct for ABI compatibility.
- int32_t NumArgs; // Number of arguments in each input pointer.
- void **ArgBasePtrs; // Base pointer of each argument (e.g. a struct).
- void **ArgPtrs; // Pointer to the argument data.
- int64_t *ArgSizes; // Size of the argument data in bytes.
- int64_t *ArgTypes; // Type of the data (e.g. to / from).
- void **ArgNames; // Name of the data for debugging, possibly null.
- void **ArgMappers; // User-defined mappers, possibly null.
- int64_t Tripcount; // Tripcount for the teams / distribute loop, 0 otherwise.
+struct KernelArgsTy {
+ uint32_t Version; // Version of this struct for ABI compatibility.
+ uint32_t NumArgs; // Number of arguments in each input pointer.
+ void **ArgBasePtrs; // Base pointer of each argument (e.g. a struct).
+ void **ArgPtrs; // Pointer to the argument data.
+ int64_t *ArgSizes; // Size of the argument data in bytes.
+ int64_t *ArgTypes; // Type of the data (e.g. to / from).
+ void **ArgNames; // Name of the data for debugging, possibly null.
+ void **ArgMappers; // User-defined mappers, possibly null.
+ uint64_t Tripcount; // Tripcount for the teams / distribute loop, 0 otherwise.
+ struct {
+ uint64_t NoWait : 1; // Was this kernel spawned with a `nowait` clause.
+ uint64_t Unused : 63;
+ } Flags;
+ uint32_t NumTeams[3]; // The number of teams (for x,y,z dimension).
+ uint32_t ThreadLimit[3]; // The number of threads (for x,y,z dimension).
+ uint32_t DynCGroupMem; // Amount of dynamic cgroup memory requested.
};
-static_assert(sizeof(__tgt_kernel_arguments) == 64 ||
- sizeof(__tgt_kernel_arguments) == 40,
+static_assert(sizeof(KernelArgsTy().Flags) == sizeof(uint64_t),
"Invalid struct size");
+static_assert(sizeof(KernelArgsTy) == (8 * sizeof(int32_t) + 3 * sizeof(int64_t) + 4 * sizeof(void**) + 2 * sizeof(int64_t*)),
+ "Invalid struct size");
+constexpr KernelArgsTy CTorDTorKernelArgs = {1, 0, nullptr, nullptr,
+ nullptr, nullptr, nullptr, nullptr,
+ 0, {0,0}, {1, 0, 0}, {1, 0, 0}, 0};
/// This struct is a record of an entry point or global. For a function
/// entry point the size is expected to be zero
@@ -399,13 +411,7 @@ void __tgt_target_data_update_nowait_mapper(
// function returns 0 if it was able to transfer the execution to a
// target and an int
diff erent from zero otherwise.
int __tgt_target_kernel(ident_t *Loc, int64_t DeviceId, int32_t NumTeams,
- int32_t ThreadLimit, void *HostPtr,
- __tgt_kernel_arguments *Args);
-int __tgt_target_kernel_nowait(ident_t *Loc, int64_t DeviceId, int32_t NumTeams,
- int32_t ThreadLimit, void *HostPtr,
- __tgt_kernel_arguments *Args, int32_t DepNum,
- void *DepList, int32_t NoAliasDepNum,
- void *NoAliasDepList);
+ int32_t ThreadLimit, void *HostPtr, KernelArgsTy *Args);
// Non-blocking synchronization for target nowait regions. This function
// acquires the asynchronous context from task data of the current task being
diff --git a/openmp/libomptarget/include/rtl.h b/openmp/libomptarget/include/rtl.h
index 063acc153f830..3d40aa574f406 100644
--- a/openmp/libomptarget/include/rtl.h
+++ b/openmp/libomptarget/include/rtl.h
@@ -50,16 +50,8 @@ struct RTLInfoTy {
typedef int32_t(data_exchange_async_ty)(int32_t, void *, int32_t, void *,
int64_t, __tgt_async_info *);
typedef int32_t(data_delete_ty)(int32_t, void *, int32_t);
- typedef int32_t(run_region_ty)(int32_t, void *, void **, ptr
diff _t *,
- int32_t);
- typedef int32_t(run_region_async_ty)(int32_t, void *, void **, ptr
diff _t *,
- int32_t, __tgt_async_info *);
- typedef int32_t(run_team_region_ty)(int32_t, void *, void **, ptr
diff _t *,
- int32_t, int32_t, int32_t, uint64_t);
- typedef int32_t(run_team_region_async_ty)(int32_t, void *, void **,
- ptr
diff _t *, int32_t, int32_t,
- int32_t, uint64_t,
- __tgt_async_info *);
+ typedef int32_t(launch_kernel_ty)(int32_t, void *, void **, ptr
diff _t *,
+ const KernelArgsTy *, __tgt_async_info *);
typedef int64_t(init_requires_ty)(int64_t);
typedef int32_t(synchronize_ty)(int32_t, __tgt_async_info *);
typedef int32_t(query_async_ty)(int32_t, __tgt_async_info *);
@@ -109,10 +101,7 @@ struct RTLInfoTy {
data_exchange_ty *data_exchange = nullptr;
data_exchange_async_ty *data_exchange_async = nullptr;
data_delete_ty *data_delete = nullptr;
- run_region_ty *run_region = nullptr;
- run_region_async_ty *run_region_async = nullptr;
- run_team_region_ty *run_team_region = nullptr;
- run_team_region_async_ty *run_team_region_async = nullptr;
+ launch_kernel_ty *launch_kernel = nullptr;
init_requires_ty *init_requires = nullptr;
synchronize_ty *synchronize = nullptr;
query_async_ty *query_async = nullptr;
diff --git a/openmp/libomptarget/plugins-nextgen/amdgpu/src/rtl.cpp b/openmp/libomptarget/plugins-nextgen/amdgpu/src/rtl.cpp
index 7dfef02d89296..22c28a92d0523 100644
--- a/openmp/libomptarget/plugins-nextgen/amdgpu/src/rtl.cpp
+++ b/openmp/libomptarget/plugins-nextgen/amdgpu/src/rtl.cpp
@@ -428,7 +428,7 @@ struct AMDGPUKernelTy : public GenericKernelTy {
AsyncInfoWrapperTy &AsyncInfoWrapper) const override;
/// The default number of blocks is common to the whole device.
- uint64_t getDefaultNumBlocks(GenericDeviceTy &GenericDevice) const override {
+ uint32_t getDefaultNumBlocks(GenericDeviceTy &GenericDevice) const override {
return GenericDevice.getDefaultNumBlocks();
}
diff --git a/openmp/libomptarget/plugins-nextgen/common/PluginInterface/PluginInterface.cpp b/openmp/libomptarget/plugins-nextgen/common/PluginInterface/PluginInterface.cpp
index 7cc1aa2fcd593..9787be4aad99b 100644
--- a/openmp/libomptarget/plugins-nextgen/common/PluginInterface/PluginInterface.cpp
+++ b/openmp/libomptarget/plugins-nextgen/common/PluginInterface/PluginInterface.cpp
@@ -224,20 +224,18 @@ Error GenericKernelTy::init(GenericDeviceTy &GenericDevice,
}
Error GenericKernelTy::launch(GenericDeviceTy &GenericDevice, void **ArgPtrs,
- ptr
diff _t *ArgOffsets, int32_t NumArgs,
- uint64_t NumTeamsClause,
- uint32_t ThreadLimitClause,
- uint64_t LoopTripCount,
+ ptr
diff _t *ArgOffsets, KernelArgsTy &KernelArgs,
AsyncInfoWrapperTy &AsyncInfoWrapper) const {
llvm::SmallVector<void *, 16> Args;
llvm::SmallVector<void *, 16> Ptrs;
- void *KernelArgsPtr = prepareArgs(GenericDevice, ArgPtrs, ArgOffsets, NumArgs,
- Args, Ptrs, AsyncInfoWrapper);
+ void *KernelArgsPtr =
+ prepareArgs(GenericDevice, ArgPtrs, ArgOffsets, KernelArgs.NumArgs, Args,
+ Ptrs, AsyncInfoWrapper);
- uint32_t NumThreads = getNumThreads(GenericDevice, ThreadLimitClause);
- uint64_t NumBlocks =
- getNumBlocks(GenericDevice, NumTeamsClause, LoopTripCount, NumThreads);
+ uint32_t NumThreads = getNumThreads(GenericDevice, KernelArgs.ThreadLimit);
+ uint64_t NumBlocks = getNumBlocks(GenericDevice, KernelArgs.NumTeams,
+ KernelArgs.Tripcount, NumThreads);
INFO(OMP_INFOTYPE_PLUGIN_KERNEL, GenericDevice.getDeviceId(),
"Launching kernel %s with %" PRIu64
@@ -245,7 +243,7 @@ Error GenericKernelTy::launch(GenericDeviceTy &GenericDevice, void **ArgPtrs,
getName(), NumBlocks, NumThreads, getExecutionModeName());
return launchImpl(GenericDevice, NumThreads, NumBlocks, DynamicMemorySize,
- NumArgs, KernelArgsPtr, AsyncInfoWrapper);
+ KernelArgs.NumArgs, KernelArgsPtr, AsyncInfoWrapper);
}
void *GenericKernelTy::prepareArgs(GenericDeviceTy &GenericDevice,
@@ -268,23 +266,29 @@ void *GenericKernelTy::prepareArgs(GenericDeviceTy &GenericDevice,
}
uint32_t GenericKernelTy::getNumThreads(GenericDeviceTy &GenericDevice,
- uint32_t ThreadLimitClause) const {
- if (ThreadLimitClause > 0 && isGenericMode())
- ThreadLimitClause += GenericDevice.getWarpSize();
+ uint32_t ThreadLimitClause[3]) const {
+ assert(ThreadLimitClause[1] == 0 && ThreadLimitClause[2] == 0 &&
+ "Multi dimensional launch not supported yet.");
+ if (ThreadLimitClause[0] > 0 && isGenericMode())
+ ThreadLimitClause[0] += GenericDevice.getWarpSize();
- return std::min(MaxNumThreads, (ThreadLimitClause > 0) ? ThreadLimitClause
- : PreferredNumThreads);
+ return std::min(MaxNumThreads, (ThreadLimitClause[0] > 0)
+ ? ThreadLimitClause[0]
+ : PreferredNumThreads);
}
uint64_t GenericKernelTy::getNumBlocks(GenericDeviceTy &GenericDevice,
- uint64_t NumTeamsClause,
+ uint32_t NumTeamsClause[3],
uint64_t LoopTripCount,
uint32_t NumThreads) const {
- if (NumTeamsClause > 0) {
+ assert(NumTeamsClause[1] == 0 && NumTeamsClause[2] == 0 &&
+ "Multi dimensional launch not supported yet.");
+
+ if (NumTeamsClause[0] > 0) {
// TODO: We need to honor any value and consequently allow more than the
// block limit. For this we might need to start multiple kernels or let the
// blocks start again until the requested number has been started.
- return std::min(NumTeamsClause, GenericDevice.getBlockLimit());
+ return std::min(NumTeamsClause[0], GenericDevice.getBlockLimit());
}
uint64_t TripCountNumBlocks = std::numeric_limits<uint64_t>::max();
@@ -314,8 +318,8 @@ uint64_t GenericKernelTy::getNumBlocks(GenericDeviceTy &GenericDevice,
}
}
// If the loops are long running we rather reuse blocks than spawn too many.
- uint64_t PreferredNumBlocks =
- std::min(TripCountNumBlocks, getDefaultNumBlocks(GenericDevice));
+ uint32_t PreferredNumBlocks = std::min(uint32_t(TripCountNumBlocks),
+ getDefaultNumBlocks(GenericDevice));
return std::min(PreferredNumBlocks, GenericDevice.getBlockLimit());
}
@@ -702,10 +706,10 @@ Error GenericDeviceTy::dataExchange(const void *SrcPtr, GenericDeviceTy &DstDev,
return Err;
}
-Error GenericDeviceTy::runTargetTeamRegion(
- void *EntryPtr, void **ArgPtrs, ptr
diff _t *ArgOffsets, int32_t NumArgs,
- uint64_t NumTeamsClause, uint32_t ThreadLimitClause, uint64_t LoopTripCount,
- __tgt_async_info *AsyncInfo) {
+Error GenericDeviceTy::launchKernel(void *EntryPtr, void **ArgPtrs,
+ ptr
diff _t *ArgOffsets,
+ KernelArgsTy &KernelArgs,
+ __tgt_async_info *AsyncInfo) {
auto Err = Plugin::success();
AsyncInfoWrapperTy AsyncInfoWrapper(Err, *this, AsyncInfo);
@@ -714,12 +718,12 @@ Error GenericDeviceTy::runTargetTeamRegion(
if (RecordReplay.isRecording())
RecordReplay.saveKernelInputInfo(
- GenericKernel.getName(), ArgPtrs, ArgOffsets, NumArgs, NumTeamsClause,
- ThreadLimitClause, LoopTripCount, AsyncInfoWrapper);
+ GenericKernel.getName(), ArgPtrs, ArgOffsets, KernelArgs.NumArgs,
+ KernelArgs.NumTeams[0], KernelArgs.ThreadLimit[0], KernelArgs.Tripcount,
+ AsyncInfoWrapper);
- Err =
- GenericKernel.launch(*this, ArgPtrs, ArgOffsets, NumArgs, NumTeamsClause,
- ThreadLimitClause, LoopTripCount, AsyncInfoWrapper);
+ Err = GenericKernel.launch(*this, ArgPtrs, ArgOffsets, KernelArgs,
+ AsyncInfoWrapper);
if (RecordReplay.isRecordingOrReplaying() &&
RecordReplay.isSaveOutputEnabled())
@@ -1068,24 +1072,12 @@ int32_t __tgt_rtl_data_exchange_async(int32_t SrcDeviceId, void *SrcPtr,
return OFFLOAD_SUCCESS;
}
-int32_t __tgt_rtl_run_target_team_region(int32_t DeviceId, void *TgtEntryPtr,
- void **TgtArgs, ptr
diff _t *TgtOffsets,
- int32_t NumArgs, int32_t NumTeams,
- int32_t ThreadLimit,
- uint64_t LoopTripCount) {
- return __tgt_rtl_run_target_team_region_async(DeviceId, TgtEntryPtr, TgtArgs,
- TgtOffsets, NumArgs, NumTeams,
- ThreadLimit, LoopTripCount,
- /* AsyncInfoPtr */ nullptr);
-}
-
-int32_t __tgt_rtl_run_target_team_region_async(
- int32_t DeviceId, void *TgtEntryPtr, void **TgtArgs, ptr
diff _t *TgtOffsets,
- int32_t NumArgs, int32_t NumTeams, int32_t ThreadLimit,
- uint64_t LoopTripCount, __tgt_async_info *AsyncInfoPtr) {
- auto Err = Plugin::get().getDevice(DeviceId).runTargetTeamRegion(
- TgtEntryPtr, TgtArgs, TgtOffsets, NumArgs, NumTeams, ThreadLimit,
- LoopTripCount, AsyncInfoPtr);
+int32_t __tgt_rtl_launch_kernel(int32_t DeviceId, void *TgtEntryPtr,
+ void **TgtArgs, ptr
diff _t *TgtOffsets,
+ KernelArgsTy *KernelArgs,
+ __tgt_async_info *AsyncInfoPtr) {
+ auto Err = Plugin::get().getDevice(DeviceId).launchKernel(
+ TgtEntryPtr, TgtArgs, TgtOffsets, *KernelArgs, AsyncInfoPtr);
if (Err) {
REPORT("Failure to run target region " DPxMOD " in device %d: %s\n",
DPxPTR(TgtEntryPtr), DeviceId, toString(std::move(Err)).data());
@@ -1119,24 +1111,6 @@ int32_t __tgt_rtl_query_async(int32_t DeviceId,
return OFFLOAD_SUCCESS;
}
-int32_t __tgt_rtl_run_target_region(int32_t DeviceId, void *TgtEntryPtr,
- void **TgtArgs, ptr
diff _t *TgtOffsets,
- int32_t NumArgs) {
- return __tgt_rtl_run_target_region_async(DeviceId, TgtEntryPtr, TgtArgs,
- TgtOffsets, NumArgs,
- /* AsyncInfoPtr */ nullptr);
-}
-
-int32_t __tgt_rtl_run_target_region_async(int32_t DeviceId, void *TgtEntryPtr,
- void **TgtArgs, ptr
diff _t *TgtOffsets,
- int32_t NumArgs,
- __tgt_async_info *AsyncInfoPtr) {
- return __tgt_rtl_run_target_team_region_async(
- DeviceId, TgtEntryPtr, TgtArgs, TgtOffsets, NumArgs,
- /* team num*/ 1, /* thread limit */ 1, /* loop tripcount */ 0,
- AsyncInfoPtr);
-}
-
void __tgt_rtl_print_device_info(int32_t DeviceId) {
if (auto Err = Plugin::get().getDevice(DeviceId).printInfo())
REPORT("Failure to print device %d info: %s\n", DeviceId,
diff --git a/openmp/libomptarget/plugins-nextgen/common/PluginInterface/PluginInterface.h b/openmp/libomptarget/plugins-nextgen/common/PluginInterface/PluginInterface.h
index e53c666ad7eff..1cee343d9b397 100644
--- a/openmp/libomptarget/plugins-nextgen/common/PluginInterface/PluginInterface.h
+++ b/openmp/libomptarget/plugins-nextgen/common/PluginInterface/PluginInterface.h
@@ -175,8 +175,7 @@ struct GenericKernelTy {
/// Launch the kernel on the specific device. The device must be the same
/// one used to initialize the kernel.
Error launch(GenericDeviceTy &GenericDevice, void **ArgPtrs,
- ptr
diff _t *ArgOffsets, int32_t NumArgs, uint64_t NumTeamsClause,
- uint32_t ThreadLimitClause, uint64_t LoopTripCount,
+ ptr
diff _t *ArgOffsets, KernelArgsTy &KernelArgs,
AsyncInfoWrapperTy &AsyncInfoWrapper) const;
virtual Error launchImpl(GenericDeviceTy &GenericDevice, uint32_t NumThreads,
uint64_t NumBlocks, uint32_t DynamicMemorySize,
@@ -207,14 +206,14 @@ struct GenericKernelTy {
/// Get the default number of threads and blocks for the kernel.
virtual uint32_t getDefaultNumThreads(GenericDeviceTy &Device) const = 0;
- virtual uint64_t getDefaultNumBlocks(GenericDeviceTy &Device) const = 0;
+ virtual uint32_t getDefaultNumBlocks(GenericDeviceTy &Device) const = 0;
/// Get the number of threads and blocks for the kernel based on the
/// user-defined threads and block clauses.
uint32_t getNumThreads(GenericDeviceTy &GenericDevice,
- uint32_t ThreadLimitClause) const;
+ uint32_t ThreadLimitClause[3]) const;
uint64_t getNumBlocks(GenericDeviceTy &GenericDevice,
- uint64_t BlockLimitClause, uint64_t LoopTripCount,
+ uint32_t BlockLimitClause[3], uint64_t LoopTripCount,
uint32_t NumThreads) const;
/// Indicate if the kernel works in Generic SPMD, Generic or SPMD mode.
@@ -335,12 +334,9 @@ struct GenericDeviceTy : public DeviceAllocatorTy {
void *DstPtr, int64_t Size,
AsyncInfoWrapperTy &AsyncInfoWrapper) = 0;
- /// Run the target region with multiple teams.
- Error runTargetTeamRegion(void *EntryPtr, void **ArgPtrs,
- ptr
diff _t *ArgOffsets, int32_t NumArgs,
- uint64_t NumTeamsClause, uint32_t ThreadLimitClause,
- uint64_t LoopTripCount,
- __tgt_async_info *AsyncInfo);
+ /// Run the kernel associated with \p EntryPtr
+ Error launchKernel(void *EntryPtr, void **ArgPtrs, ptr
diff _t *ArgOffsets,
+ KernelArgsTy &KernelArgs, __tgt_async_info *AsyncInfo);
/// Initialize a __tgt_async_info structure. Related to interop features.
Error initAsyncInfo(__tgt_async_info **AsyncInfoPtr);
@@ -380,11 +376,11 @@ struct GenericDeviceTy : public DeviceAllocatorTy {
/// Getters of the grid values.
uint32_t getWarpSize() const { return GridValues.GV_Warp_Size; }
uint32_t getThreadLimit() const { return GridValues.GV_Max_WG_Size; }
- uint64_t getBlockLimit() const { return GridValues.GV_Max_Teams; }
+ uint32_t getBlockLimit() const { return GridValues.GV_Max_Teams; }
uint32_t getDefaultNumThreads() const {
return GridValues.GV_Default_WG_Size;
}
- uint64_t getDefaultNumBlocks() const {
+ uint32_t getDefaultNumBlocks() const {
return GridValues.GV_Default_Num_Teams;
}
uint32_t getDynamicMemorySize() const { return OMPX_SharedMemorySize; }
diff --git a/openmp/libomptarget/plugins-nextgen/cuda/src/rtl.cpp b/openmp/libomptarget/plugins-nextgen/cuda/src/rtl.cpp
index cfc97b6cdc46c..6f935066ce418 100644
--- a/openmp/libomptarget/plugins-nextgen/cuda/src/rtl.cpp
+++ b/openmp/libomptarget/plugins-nextgen/cuda/src/rtl.cpp
@@ -67,7 +67,7 @@ struct CUDAKernelTy : public GenericKernelTy {
AsyncInfoWrapperTy &AsyncInfoWrapper) const override;
/// The default number of blocks is common to the whole device.
- uint64_t getDefaultNumBlocks(GenericDeviceTy &GenericDevice) const override {
+ uint32_t getDefaultNumBlocks(GenericDeviceTy &GenericDevice) const override {
return GenericDevice.getDefaultNumBlocks();
}
diff --git a/openmp/libomptarget/plugins-nextgen/generic-elf-64bit/src/rtl.cpp b/openmp/libomptarget/plugins-nextgen/generic-elf-64bit/src/rtl.cpp
index ab5f9e3f29faa..43bf594683c8c 100644
--- a/openmp/libomptarget/plugins-nextgen/generic-elf-64bit/src/rtl.cpp
+++ b/openmp/libomptarget/plugins-nextgen/generic-elf-64bit/src/rtl.cpp
@@ -85,7 +85,7 @@ struct GenELF64KernelTy : public GenericKernelTy {
}
/// Get the default number of blocks and threads for the kernel.
- uint64_t getDefaultNumBlocks(GenericDeviceTy &) const override { return 1; }
+ uint32_t getDefaultNumBlocks(GenericDeviceTy &) const override { return 1; }
uint32_t getDefaultNumThreads(GenericDeviceTy &) const override { return 1; }
private:
diff --git a/openmp/libomptarget/plugins/amdgpu/src/rtl.cpp b/openmp/libomptarget/plugins/amdgpu/src/rtl.cpp
index 809c661baea3c..d25a520326b9a 100644
--- a/openmp/libomptarget/plugins/amdgpu/src/rtl.cpp
+++ b/openmp/libomptarget/plugins/amdgpu/src/rtl.cpp
@@ -2545,60 +2545,26 @@ int32_t __tgt_rtl_data_delete(int DeviceId, void *TgtPtr, int32_t) {
return OFFLOAD_SUCCESS;
}
-int32_t __tgt_rtl_run_target_team_region(int32_t DeviceId, void *TgtEntryPtr,
- void **TgtArgs, ptr
diff _t *TgtOffsets,
- int32_t ArgNum, int32_t NumTeams,
- int32_t ThreadLimit,
- uint64_t LoopTripcount) {
-
- DeviceInfo().LoadRunLock.lock_shared();
- int32_t Res = runRegionLocked(DeviceId, TgtEntryPtr, TgtArgs, TgtOffsets,
- ArgNum, NumTeams, ThreadLimit, LoopTripcount);
-
- DeviceInfo().LoadRunLock.unlock_shared();
- return Res;
-}
-
-int32_t __tgt_rtl_run_target_region(int32_t DeviceId, void *TgtEntryPtr,
- void **TgtArgs, ptr
diff _t *TgtOffsets,
- int32_t ArgNum) {
- // use one team and one thread
- // fix thread num
- int32_t TeamNum = 1;
- int32_t ThreadLimit = 0; // use default
- return __tgt_rtl_run_target_team_region(DeviceId, TgtEntryPtr, TgtArgs,
- TgtOffsets, ArgNum, TeamNum,
- ThreadLimit, 0);
-}
-
-int32_t __tgt_rtl_run_target_team_region_async(
- int32_t DeviceId, void *TgtEntryPtr, void **TgtArgs, ptr
diff _t *TgtOffsets,
- int32_t ArgNum, int32_t NumTeams, int32_t ThreadLimit,
- uint64_t LoopTripcount, __tgt_async_info *AsyncInfo) {
+int32_t __tgt_rtl_launch_kernel(int32_t DeviceId, void *TgtEntryPtr,
+ void **TgtArgs, ptr
diff _t *TgtOffsets,
+ KernelArgsTy *KernelArgs,
+ __tgt_async_info *AsyncInfo) {
+ assert(!KernelArgs->NumTeams[1] && !KernelArgs->NumTeams[2] &&
+ !KernelArgs->ThreadLimit[1] && !KernelArgs->ThreadLimit[2] &&
+ "Only one dimensional kernels supported.");
assert(AsyncInfo && "AsyncInfo is nullptr");
initAsyncInfo(AsyncInfo);
DeviceInfo().LoadRunLock.lock_shared();
- int32_t Res = runRegionLocked(DeviceId, TgtEntryPtr, TgtArgs, TgtOffsets,
- ArgNum, NumTeams, ThreadLimit, LoopTripcount);
+ int32_t Res =
+ runRegionLocked(DeviceId, TgtEntryPtr, TgtArgs, TgtOffsets,
+ KernelArgs->NumArgs, KernelArgs->NumTeams[0],
+ KernelArgs->ThreadLimit[0], KernelArgs->Tripcount);
DeviceInfo().LoadRunLock.unlock_shared();
return Res;
}
-int32_t __tgt_rtl_run_target_region_async(int32_t DeviceId, void *TgtEntryPtr,
- void **TgtArgs, ptr
diff _t *TgtOffsets,
- int32_t ArgNum,
- __tgt_async_info *AsyncInfo) {
- // use one team and one thread
- // fix thread num
- int32_t TeamNum = 1;
- int32_t ThreadLimit = 0; // use default
- return __tgt_rtl_run_target_team_region_async(DeviceId, TgtEntryPtr, TgtArgs,
- TgtOffsets, ArgNum, TeamNum,
- ThreadLimit, 0, AsyncInfo);
-}
-
int32_t __tgt_rtl_synchronize(int32_t DeviceId, __tgt_async_info *AsyncInfo) {
assert(AsyncInfo && "AsyncInfo is nullptr");
diff --git a/openmp/libomptarget/plugins/generic-elf-64bit/src/rtl.cpp b/openmp/libomptarget/plugins/generic-elf-64bit/src/rtl.cpp
index 9622278428c0f..080aad354d524 100644
--- a/openmp/libomptarget/plugins/generic-elf-64bit/src/rtl.cpp
+++ b/openmp/libomptarget/plugins/generic-elf-64bit/src/rtl.cpp
@@ -237,27 +237,29 @@ int32_t __tgt_rtl_data_delete(int32_t DeviceId, void *TgtPtr, int32_t) {
return OFFLOAD_SUCCESS;
}
-int32_t __tgt_rtl_run_target_team_region(int32_t DeviceId, void *TgtEntryPtr,
- void **TgtArgs, ptr
diff _t *TgtOffsets,
- int32_t ArgNum, int32_t TeamNum,
- int32_t ThreadLimit,
- uint64_t LoopTripcount /*not used*/) {
+int32_t __tgt_rtl_launch_kernel(int32_t DeviceId, void *TgtEntryPtr,
+ void **TgtArgs, ptr
diff _t *TgtOffsets,
+ KernelArgsTy *KernelArgs,
+ __tgt_async_info *AsyncInfoPtr) {
+ assert(!KernelArgs->NumTeams[1] && !KernelArgs->NumTeams[2] &&
+ !KernelArgs->ThreadLimit[1] && !KernelArgs->ThreadLimit[2] &&
+ "Only one dimensional kernels supported.");
// ignore team num and thread limit.
// Use libffi to launch execution.
ffi_cif Cif;
// All args are references.
- std::vector<ffi_type *> ArgsTypes(ArgNum, &ffi_type_pointer);
- std::vector<void *> Args(ArgNum);
- std::vector<void *> Ptrs(ArgNum);
+ std::vector<ffi_type *> ArgsTypes(KernelArgs->NumArgs, &ffi_type_pointer);
+ std::vector<void *> Args(KernelArgs->NumArgs);
+ std::vector<void *> Ptrs(KernelArgs->NumArgs);
- for (int32_t I = 0; I < ArgNum; ++I) {
+ for (int32_t I = 0; I < KernelArgs->NumArgs; ++I) {
Ptrs[I] = (void *)((intptr_t)TgtArgs[I] + TgtOffsets[I]);
Args[I] = &Ptrs[I];
}
- ffi_status Status = ffi_prep_cif(&Cif, FFI_DEFAULT_ABI, ArgNum,
+ ffi_status Status = ffi_prep_cif(&Cif, FFI_DEFAULT_ABI, KernelArgs->NumArgs,
&ffi_type_void, &ArgsTypes[0]);
assert(Status == FFI_OK && "Unable to prepare target launch!");
@@ -273,14 +275,6 @@ int32_t __tgt_rtl_run_target_team_region(int32_t DeviceId, void *TgtEntryPtr,
return OFFLOAD_SUCCESS;
}
-int32_t __tgt_rtl_run_target_region(int32_t DeviceId, void *TgtEntryPtr,
- void **TgtArgs, ptr
diff _t *TgtOffsets,
- int32_t ArgNum) {
- // use one team and one thread.
- return __tgt_rtl_run_target_team_region(DeviceId, TgtEntryPtr, TgtArgs,
- TgtOffsets, ArgNum, 1, 1, 0);
-}
-
#ifdef __cplusplus
}
#endif
diff --git a/openmp/libomptarget/plugins/remote/src/Client.h b/openmp/libomptarget/plugins/remote/src/Client.h
index 5f4006926e934..6629760166d3b 100644
--- a/openmp/libomptarget/plugins/remote/src/Client.h
+++ b/openmp/libomptarget/plugins/remote/src/Client.h
@@ -144,8 +144,6 @@ class RemoteClientManager {
int32_t dataExchange(int32_t SrcDevId, void *SrcPtr, int32_t DstDevId,
void *DstPtr, int64_t Size);
- int32_t runTargetRegion(int32_t DeviceId, void *TgtEntryPtr, void **TgtArgs,
- ptr
diff _t *TgtOffsets, int32_t ArgNum);
int32_t runTargetTeamRegion(int32_t DeviceId, void *TgtEntryPtr,
void **TgtArgs, ptr
diff _t *TgtOffsets,
int32_t ArgNum, int32_t TeamNum,
diff --git a/openmp/libomptarget/plugins/remote/src/rtl.cpp b/openmp/libomptarget/plugins/remote/src/rtl.cpp
index bafb7afbdf92b..6a365ab0a00d0 100644
--- a/openmp/libomptarget/plugins/remote/src/rtl.cpp
+++ b/openmp/libomptarget/plugins/remote/src/rtl.cpp
@@ -102,22 +102,17 @@ int32_t __tgt_rtl_data_exchange(int32_t SrcDevId, void *SrcPtr,
return Manager->dataExchange(SrcDevId, SrcPtr, DstDevId, DstPtr, Size);
}
-
-int32_t __tgt_rtl_run_target_region(int32_t DeviceId, void *TgtEntryPtr,
- void **TgtArgs, ptr
diff _t *TgtOffsets,
- int32_t ArgNum) {
- return Manager->runTargetRegion(DeviceId, TgtEntryPtr, TgtArgs, TgtOffsets,
- ArgNum);
-}
-
-int32_t __tgt_rtl_run_target_team_region(int32_t DeviceId, void *TgtEntryPtr,
- void **TgtArgs, ptr
diff _t *TgtOffsets,
- int32_t ArgNum, int32_t TeamNum,
- int32_t ThreadLimit,
- uint64_t LoopTripCount) {
- return Manager->runTargetTeamRegion(DeviceId, TgtEntryPtr, TgtArgs,
- TgtOffsets, ArgNum, TeamNum, ThreadLimit,
- LoopTripCount);
+int32_t __tgt_rtl_launch_kernel(int32_t DeviceId, void *TgtEntryPtr,
+ void **TgtArgs, ptr
diff _t *TgtOffsets,
+ KernelArgsTy *KernelArgs,
+ __tgt_async_info *AsyncInfoPtr) {
+ assert(!KernelArgs->NumTeams[1] && !KernelArgs->NumTeams[2] &&
+ !KernelArgs->ThreadLimit[1] && !KernelArgs->ThreadLimit[2] &&
+ "Only one dimensional kernels supported.");
+ return Manager->runTargetTeamRegion(
+ DeviceId, TgtEntryPtr, TgtArgs, TgtOffsets, KernelArgs->NumArgs,
+ KernelArgs->NumTeams[0], KernelArgs->ThreadLimit[0],
+ KernelArgs->Tripcount);
}
// Exposed library API function
diff --git a/openmp/libomptarget/plugins/ve/src/rtl.cpp b/openmp/libomptarget/plugins/ve/src/rtl.cpp
index 76bd93e4fa53f..2aba36c8abdcc 100644
--- a/openmp/libomptarget/plugins/ve/src/rtl.cpp
+++ b/openmp/libomptarget/plugins/ve/src/rtl.cpp
@@ -402,16 +402,22 @@ int32_t __tgt_rtl_data_delete(int32_t ID, void *TargetPtr, int32_t) {
return OFFLOAD_SUCCESS;
}
-// Similar to __tgt_rtl_run_target_region, but additionally specify the
-// number of teams to be created and a number of threads in each team.
-int32_t __tgt_rtl_run_target_team_region(int32_t ID, void *Entry, void **Args,
- ptr
diff _t *Offsets, int32_t NumArgs,
- int32_t NumTeams, int32_t ThreadLimit,
- uint64_t loop_tripcount) {
+// Transfer control to the offloaded entry Entry on the target device.
+// Args and Offsets are arrays of NumArgs size of target addresses and
+// offsets. An offset should be added to the target address before passing it
+// to the outlined function on device side. In case of success, return zero.
+// Otherwise, return an error code.
+int32_t __tgt_rtl_launch_kernel(int32_t DeviceId, void *TgtEntryPtr,
+ void **TgtArgs, ptr
diff _t *TgtOffsets,
+ KernelArgsTy *KernelArgs,
+ __tgt_async_info *AsyncInfoPtr) {
+ assert(!KernelArgs->NumTeams[1] && !KernelArgs->NumTeams[2] &&
+ !KernelArgs->ThreadLimit[1] && !KernelArgs->ThreadLimit[2] &&
+ "Only one dimensional kernels supported.");
int ret;
// ignore team num and thread limit.
- std::vector<void *> ptrs(NumArgs);
+ std::vector<void *> ptrs(KernelArgs->NumArgs);
struct veo_args *TargetArgs;
TargetArgs = veo_args_alloc();
@@ -421,7 +427,7 @@ int32_t __tgt_rtl_run_target_team_region(int32_t ID, void *Entry, void **Args,
return OFFLOAD_FAIL;
}
- for (int i = 0; i < NumArgs; ++i) {
+ for (int i = 0; i < KernelArgs->NumArgs; ++i) {
ret = veo_args_set_u64(TargetArgs, i, (intptr_t)Args[i]);
if (ret != 0) {
@@ -441,17 +447,6 @@ int32_t __tgt_rtl_run_target_team_region(int32_t ID, void *Entry, void **Args,
return OFFLOAD_SUCCESS;
}
-// Transfer control to the offloaded entry Entry on the target device.
-// Args and Offsets are arrays of NumArgs size of target addresses and
-// offsets. An offset should be added to the target address before passing it
-// to the outlined function on device side. In case of success, return zero.
-// Otherwise, return an error code.
-int32_t __tgt_rtl_run_target_region(int32_t ID, void *Entry, void **Args,
- ptr
diff _t *Offsets, int32_t NumArgs) {
- return __tgt_rtl_run_target_team_region(ID, Entry, Args, Offsets, NumArgs, 1,
- 1, 0);
-}
-
int32_t __tgt_rtl_supports_empty_images() { return 1; }
// VEC plugin's internal InfoLevel.
diff --git a/openmp/libomptarget/src/LegacyAPI.cpp b/openmp/libomptarget/src/LegacyAPI.cpp
index 654cfdcd4575e..54dbe19c17144 100644
--- a/openmp/libomptarget/src/LegacyAPI.cpp
+++ b/openmp/libomptarget/src/LegacyAPI.cpp
@@ -72,12 +72,12 @@ EXTERN void __tgt_target_data_end_nowait(int64_t DeviceId, int32_t ArgNum,
}
EXTERN int __tgt_target_mapper(ident_t *Loc, int64_t DeviceId, void *HostPtr,
- int32_t ArgNum, void **ArgsBase, void **Args,
+ uint32_t ArgNum, void **ArgsBase, void **Args,
int64_t *ArgSizes, int64_t *ArgTypes,
map_var_info_t *ArgNames, void **ArgMappers) {
TIMESCOPE_WITH_IDENT(Loc);
- __tgt_kernel_arguments KernelArgs{
- 1, ArgNum, ArgsBase, Args, ArgSizes, ArgTypes, ArgNames, ArgMappers, 0};
+ KernelArgsTy KernelArgs{1, ArgNum, ArgsBase, Args, ArgSizes,
+ ArgTypes, ArgNames, ArgMappers, 0};
return __tgt_target_kernel(Loc, DeviceId, -1, -1, HostPtr, &KernelArgs);
}
@@ -111,7 +111,7 @@ EXTERN int __tgt_target_nowait_mapper(
}
EXTERN int __tgt_target_teams_mapper(ident_t *Loc, int64_t DeviceId,
- void *HostPtr, int32_t ArgNum,
+ void *HostPtr, uint32_t ArgNum,
void **ArgsBase, void **Args,
int64_t *ArgSizes, int64_t *ArgTypes,
map_var_info_t *ArgNames,
@@ -119,8 +119,8 @@ EXTERN int __tgt_target_teams_mapper(ident_t *Loc, int64_t DeviceId,
int32_t ThreadLimit) {
TIMESCOPE_WITH_IDENT(Loc);
- __tgt_kernel_arguments KernelArgs{
- 1, ArgNum, ArgsBase, Args, ArgSizes, ArgTypes, ArgNames, ArgMappers, 0};
+ KernelArgsTy KernelArgs{1, ArgNum, ArgsBase, Args, ArgSizes,
+ ArgTypes, ArgNames, ArgMappers, 0};
return __tgt_target_kernel(Loc, DeviceId, NumTeams, ThreadLimit, HostPtr,
&KernelArgs);
}
@@ -172,3 +172,14 @@ EXTERN void __kmpc_push_target_tripcount(int64_t DeviceId,
uint64_t LoopTripcount) {
__kmpc_push_target_tripcount_mapper(nullptr, DeviceId, LoopTripcount);
}
+
+EXTERN int __tgt_target_kernel_nowait(ident_t *Loc, int64_t DeviceId,
+ int32_t NumTeams, int32_t ThreadLimit,
+ void *HostPtr, KernelArgsTy *KernelArgs,
+ int32_t DepNum, void *DepList,
+ int32_t NoAliasDepNum,
+ void *NoAliasDepList) {
+ TIMESCOPE_WITH_IDENT(Loc);
+ return __tgt_target_kernel(Loc, DeviceId, NumTeams, ThreadLimit, HostPtr,
+ KernelArgs);
+}
diff --git a/openmp/libomptarget/src/device.cpp b/openmp/libomptarget/src/device.cpp
index 65c8a32512858..fb74d20cd25c0 100644
--- a/openmp/libomptarget/src/device.cpp
+++ b/openmp/libomptarget/src/device.cpp
@@ -592,14 +592,12 @@ int32_t DeviceTy::dataExchange(void *SrcPtr, DeviceTy &DstDev, void *DstPtr,
}
// Run region on device
-int32_t DeviceTy::runRegion(void *TgtEntryPtr, void **TgtVarsPtr,
- ptr
diff _t *TgtOffsets, int32_t TgtVarsSize,
- AsyncInfoTy &AsyncInfo) {
- if (!RTL->run_region_async || !RTL->synchronize)
- return RTL->run_region(RTLDeviceID, TgtEntryPtr, TgtVarsPtr, TgtOffsets,
- TgtVarsSize);
- return RTL->run_region_async(RTLDeviceID, TgtEntryPtr, TgtVarsPtr, TgtOffsets,
- TgtVarsSize, AsyncInfo);
+int32_t DeviceTy::launchKernel(void *TgtEntryPtr, void **TgtVarsPtr,
+ ptr
diff _t *TgtOffsets,
+ const KernelArgsTy &KernelArgs,
+ AsyncInfoTy &AsyncInfo) {
+ return RTL->launch_kernel(RTLDeviceID, TgtEntryPtr, TgtVarsPtr, TgtOffsets,
+ &KernelArgs, AsyncInfo);
}
// Run region on device
@@ -610,21 +608,6 @@ bool DeviceTy::printDeviceInfo(int32_t RTLDevId) {
return true;
}
-// Run team region on device.
-int32_t DeviceTy::runTeamRegion(void *TgtEntryPtr, void **TgtVarsPtr,
- ptr
diff _t *TgtOffsets, int32_t TgtVarsSize,
- int32_t NumTeams, int32_t ThreadLimit,
- uint64_t LoopTripCount,
- AsyncInfoTy &AsyncInfo) {
- if (!RTL->run_team_region_async || !RTL->synchronize)
- return RTL->run_team_region(RTLDeviceID, TgtEntryPtr, TgtVarsPtr,
- TgtOffsets, TgtVarsSize, NumTeams, ThreadLimit,
- LoopTripCount);
- return RTL->run_team_region_async(RTLDeviceID, TgtEntryPtr, TgtVarsPtr,
- TgtOffsets, TgtVarsSize, NumTeams,
- ThreadLimit, LoopTripCount, AsyncInfo);
-}
-
// Whether data can be copied to DstDevice directly
bool DeviceTy::isDataExchangable(const DeviceTy &DstDevice) {
if (RTL != DstDevice.RTL || !RTL->is_data_exchangable)
diff --git a/openmp/libomptarget/src/interface.cpp b/openmp/libomptarget/src/interface.cpp
index a53d963d81f20..4d4651bdb8111 100644
--- a/openmp/libomptarget/src/interface.cpp
+++ b/openmp/libomptarget/src/interface.cpp
@@ -188,10 +188,40 @@ EXTERN void __tgt_target_data_update_nowait_mapper(
ArgMappers, targetDataUpdate, "Updating OpenMP data", "update");
}
+static KernelArgsTy *upgradeKernelArgs(KernelArgsTy *KernelArgs,
+ KernelArgsTy &LocalKernelArgs,
+ int32_t NumTeams, int32_t ThreadLimit) {
+ if (KernelArgs->Version > 2)
+ DP("Unexpected ABI version: %u\n", KernelArgs->Version);
+
+ if (KernelArgs->Version == 1) {
+ LocalKernelArgs.Version = 2;
+ LocalKernelArgs.NumArgs = KernelArgs->NumArgs;
+ LocalKernelArgs.ArgBasePtrs = KernelArgs->ArgBasePtrs;
+ LocalKernelArgs.ArgPtrs = KernelArgs->ArgPtrs;
+ LocalKernelArgs.ArgSizes = KernelArgs->ArgSizes;
+ LocalKernelArgs.ArgTypes = KernelArgs->ArgTypes;
+ LocalKernelArgs.ArgNames = KernelArgs->ArgNames;
+ LocalKernelArgs.ArgMappers = KernelArgs->ArgMappers;
+ LocalKernelArgs.Tripcount = KernelArgs->Tripcount;
+ LocalKernelArgs.Flags = KernelArgs->Flags;
+ LocalKernelArgs.DynCGroupMem = 0;
+ LocalKernelArgs.NumTeams[0] = NumTeams;
+ LocalKernelArgs.NumTeams[1] = 0;
+ LocalKernelArgs.NumTeams[2] = 0;
+ LocalKernelArgs.ThreadLimit[0] = ThreadLimit;
+ LocalKernelArgs.ThreadLimit[1] = 0;
+ LocalKernelArgs.ThreadLimit[2] = 0;
+ return &LocalKernelArgs;
+ }
+
+ return KernelArgs;
+}
+
template <typename TargetAsyncInfoTy>
static inline int targetKernel(ident_t *Loc, int64_t DeviceId, int32_t NumTeams,
int32_t ThreadLimit, void *HostPtr,
- __tgt_kernel_arguments *Args) {
+ KernelArgsTy *KernelArgs) {
static_assert(std::is_convertible_v<TargetAsyncInfoTy, AsyncInfoTy>,
"Target AsyncInfoTy must be convertible to AsyncInfoTy.");
@@ -206,38 +236,44 @@ static inline int targetKernel(ident_t *Loc, int64_t DeviceId, int32_t NumTeams,
return OMP_TGT_FAIL;
}
- if (Args->Version != 1) {
- DP("Unexpected ABI version: %d\n", Args->Version);
- }
+ bool IsTeams = NumTeams != -1;
+ if (!IsTeams)
+ KernelArgs->NumTeams[0] = NumTeams = 1;
+
+ // Auto-upgrade kernel args version 1 to 2.
+ KernelArgsTy LocalKernelArgs;
+ KernelArgs =
+ upgradeKernelArgs(KernelArgs, LocalKernelArgs, NumTeams, ThreadLimit);
+
+ assert(KernelArgs->NumTeams[0] == NumTeams && !KernelArgs->NumTeams[1] &&
+ !KernelArgs->NumTeams[2] &&
+ "OpenMP interface should not use multiple dimensions");
+ assert(KernelArgs->ThreadLimit[0] == ThreadLimit &&
+ !KernelArgs->ThreadLimit[1] && !KernelArgs->ThreadLimit[2] &&
+ "OpenMP interface should not use multiple dimensions");
if (getInfoLevel() & OMP_INFOTYPE_KERNEL_ARGS)
- printKernelArguments(Loc, DeviceId, Args->NumArgs, Args->ArgSizes,
- Args->ArgTypes, Args->ArgNames,
- "Entering OpenMP kernel");
+ printKernelArguments(Loc, DeviceId, KernelArgs->NumArgs,
+ KernelArgs->ArgSizes, KernelArgs->ArgTypes,
+ KernelArgs->ArgNames, "Entering OpenMP kernel");
#ifdef OMPTARGET_DEBUG
- for (int I = 0; I < Args->NumArgs; ++I) {
+ for (int I = 0; I < KernelArgs->NumArgs; ++I) {
DP("Entry %2d: Base=" DPxMOD ", Begin=" DPxMOD ", Size=%" PRId64
", Type=0x%" PRIx64 ", Name=%s\n",
- I, DPxPTR(Args->ArgBasePtrs[I]), DPxPTR(Args->ArgPtrs[I]),
- Args->ArgSizes[I], Args->ArgTypes[I],
- (Args->ArgNames) ? getNameFromMapping(Args->ArgNames[I]).c_str()
- : "unknown");
+ I, DPxPTR(KernelArgs->ArgBasePtrs[I]), DPxPTR(KernelArgs->ArgPtrs[I]),
+ KernelArgs->ArgSizes[I], KernelArgs->ArgTypes[I],
+ (KernelArgs->ArgNames)
+ ? getNameFromMapping(KernelArgs->ArgNames[I]).c_str()
+ : "unknown");
}
#endif
- bool IsTeams = NumTeams != -1;
- if (!IsTeams)
- NumTeams = 0;
-
DeviceTy &Device = *PM->Devices[DeviceId];
TargetAsyncInfoTy TargetAsyncInfo(Device);
AsyncInfoTy &AsyncInfo = TargetAsyncInfo;
int Rc = OFFLOAD_SUCCESS;
- Rc = target(Loc, Device, HostPtr, Args->NumArgs, Args->ArgBasePtrs,
- Args->ArgPtrs, Args->ArgSizes, Args->ArgTypes, Args->ArgNames,
- Args->ArgMappers, NumTeams, ThreadLimit, Args->Tripcount, IsTeams,
- AsyncInfo);
+ Rc = target(Loc, Device, HostPtr, *KernelArgs, AsyncInfo);
if (Rc == OFFLOAD_SUCCESS)
Rc = AsyncInfo.synchronize();
@@ -261,10 +297,10 @@ static inline int targetKernel(ident_t *Loc, int64_t DeviceId, int32_t NumTeams,
/// \param Args All arguments to this kernel launch (see struct definition).
EXTERN int __tgt_target_kernel(ident_t *Loc, int64_t DeviceId, int32_t NumTeams,
int32_t ThreadLimit, void *HostPtr,
- __tgt_kernel_arguments *Args) {
+ KernelArgsTy *KernelArgs) {
TIMESCOPE_WITH_IDENT(Loc);
return targetKernel<AsyncInfoTy>(Loc, DeviceId, NumTeams, ThreadLimit,
- HostPtr, Args);
+ HostPtr, KernelArgs);
}
/// Implements a target kernel entry that replays a pre-recorded kernel.
@@ -309,15 +345,6 @@ EXTERN int __tgt_target_kernel_replay(ident_t *Loc, int64_t DeviceId,
return OMP_TGT_SUCCESS;
}
-EXTERN int __tgt_target_kernel_nowait(
- ident_t *Loc, int64_t DeviceId, int32_t NumTeams, int32_t ThreadLimit,
- void *HostPtr, __tgt_kernel_arguments *Args, int32_t DepNum, void *DepList,
- int32_t NoAliasDepNum, void *NoAliasDepList) {
- TIMESCOPE_WITH_IDENT(Loc);
- return targetKernel<TaskAsyncInfoWrapperTy>(Loc, DeviceId, NumTeams,
- ThreadLimit, HostPtr, Args);
-}
-
// Get the current number of components for a user-defined mapper.
EXTERN int64_t __tgt_mapper_num_components(void *RtMapperHandle) {
TIMESCOPE();
diff --git a/openmp/libomptarget/src/omptarget.cpp b/openmp/libomptarget/src/omptarget.cpp
index 580b0c88e4375..a3b923277b824 100644
--- a/openmp/libomptarget/src/omptarget.cpp
+++ b/openmp/libomptarget/src/omptarget.cpp
@@ -225,9 +225,7 @@ static int initLibrary(DeviceTy &Device) {
DP("Has pending ctors... call now\n");
for (auto &Entry : Lib.second.PendingCtors) {
void *Ctor = Entry;
- int Rc = target(nullptr, Device, Ctor, 0, nullptr, nullptr, nullptr,
- nullptr, nullptr, nullptr, 1, 1, 0, true /*team*/,
- AsyncInfo);
+ int Rc = target(nullptr, Device, Ctor, CTorDTorKernelArgs, AsyncInfo);
if (Rc != OFFLOAD_SUCCESS) {
REPORT("Running ctor " DPxMOD " failed.\n", DPxPTR(Ctor));
return OFFLOAD_FAIL;
@@ -1620,11 +1618,8 @@ static int processDataAfter(ident_t *Loc, int64_t DeviceId, void *HostPtr,
/// performs the same action as data_update and data_end above. This function
/// returns 0 if it was able to transfer the execution to a target and an
/// integer
diff erent from zero otherwise.
-int target(ident_t *Loc, DeviceTy &Device, void *HostPtr, int32_t ArgNum,
- void **ArgBases, void **Args, int64_t *ArgSizes, int64_t *ArgTypes,
- map_var_info_t *ArgNames, void **ArgMappers, int32_t TeamNum,
- int32_t ThreadLimit, uint64_t Tripcount, int IsTeamConstruct,
- AsyncInfoTy &AsyncInfo) {
+int target(ident_t *Loc, DeviceTy &Device, void *HostPtr,
+ const KernelArgsTy &KernelArgs, AsyncInfoTy &AsyncInfo) {
int32_t DeviceId = Device.DeviceID;
TableMap *TM = getTableMap(HostPtr);
// No map for this host pointer found!
@@ -1644,7 +1639,7 @@ int target(ident_t *Loc, DeviceTy &Device, void *HostPtr, int32_t ArgNum,
}
assert(TargetTable && "Global data has not been mapped\n");
- DP("loop trip count is %" PRIu64 ".\n", Tripcount);
+ DP("loop trip count is %" PRIu64 ".\n", KernelArgs.Tripcount);
// We need to keep bases and offsets separate. Sometimes (e.g. in OpenCL) we
// need to manifest base pointers prior to launching a kernel. Even if we have
@@ -1661,10 +1656,12 @@ int target(ident_t *Loc, DeviceTy &Device, void *HostPtr, int32_t ArgNum,
PrivateArgumentManagerTy PrivateArgumentManager(Device, AsyncInfo);
int Ret = OFFLOAD_SUCCESS;
- if (ArgNum) {
+ if (KernelArgs.NumArgs) {
// Process data, such as data mapping, before launching the kernel
- Ret = processDataBefore(Loc, DeviceId, HostPtr, ArgNum, ArgBases, Args,
- ArgSizes, ArgTypes, ArgNames, ArgMappers, TgtArgs,
+ Ret = processDataBefore(Loc, DeviceId, HostPtr, KernelArgs.NumArgs,
+ KernelArgs.ArgBasePtrs, KernelArgs.ArgPtrs,
+ KernelArgs.ArgSizes, KernelArgs.ArgTypes,
+ KernelArgs.ArgNames, KernelArgs.ArgMappers, TgtArgs,
TgtOffsets, PrivateArgumentManager, AsyncInfo);
if (Ret != OFFLOAD_SUCCESS) {
REPORT("Failed to process data before launching the kernel.\n");
@@ -1678,15 +1675,9 @@ int target(ident_t *Loc, DeviceTy &Device, void *HostPtr, int32_t ArgNum,
TargetTable->EntriesBegin[TM->Index].name, DPxPTR(TgtEntryPtr), TM->Index);
{
- TIMESCOPE_WITH_NAME_AND_IDENT(
- IsTeamConstruct ? "runTargetTeamRegion" : "runTargetRegion", Loc);
- if (IsTeamConstruct)
- Ret = Device.runTeamRegion(TgtEntryPtr, TgtArgs.data(), TgtOffsets.data(),
- TgtArgs.size(), TeamNum, ThreadLimit,
- Tripcount, AsyncInfo);
- else
- Ret = Device.runRegion(TgtEntryPtr, TgtArgs.data(), TgtOffsets.data(),
- TgtArgs.size(), AsyncInfo);
+ TIMESCOPE_WITH_NAME_AND_IDENT("Initiate Kernel Launch", Loc);
+ Ret = Device.launchKernel(TgtEntryPtr, TgtArgs.data(), TgtOffsets.data(),
+ KernelArgs, AsyncInfo);
}
if (Ret != OFFLOAD_SUCCESS) {
@@ -1694,11 +1685,13 @@ int target(ident_t *Loc, DeviceTy &Device, void *HostPtr, int32_t ArgNum,
return OFFLOAD_FAIL;
}
- if (ArgNum) {
+ if (KernelArgs.NumArgs) {
// Transfer data back and deallocate target memory for (first-)private
// variables
- Ret = processDataAfter(Loc, DeviceId, HostPtr, ArgNum, ArgBases, Args,
- ArgSizes, ArgTypes, ArgNames, ArgMappers,
+ Ret = processDataAfter(Loc, DeviceId, HostPtr, KernelArgs.NumArgs,
+ KernelArgs.ArgBasePtrs, KernelArgs.ArgPtrs,
+ KernelArgs.ArgSizes, KernelArgs.ArgTypes,
+ KernelArgs.ArgNames, KernelArgs.ArgMappers,
PrivateArgumentManager, AsyncInfo);
if (Ret != OFFLOAD_SUCCESS) {
REPORT("Failed to process data after launching the kernel.\n");
@@ -1747,9 +1740,15 @@ int target_replay(ident_t *Loc, DeviceTy &Device, void *HostPtr,
TARGET_ALLOC_DEFAULT);
Device.submitData(TgtPtr, DeviceMemory, DeviceMemorySize, AsyncInfo);
- int Ret =
- Device.runTeamRegion(TgtEntryPtr, TgtArgs, TgtOffsets, NumArgs, NumTeams,
- ThreadLimit, LoopTripCount, AsyncInfo);
+ KernelArgsTy KernelArgs = {0};
+ KernelArgs.Version = 2;
+ KernelArgs.NumArgs = NumArgs;
+ KernelArgs.Tripcount = LoopTripCount;
+ KernelArgs.NumTeams[0] = NumTeams;
+ KernelArgs.ThreadLimit[0] = ThreadLimit;
+
+ int Ret = Device.launchKernel(TgtEntryPtr, TgtArgs, TgtOffsets, KernelArgs,
+ AsyncInfo);
if (Ret != OFFLOAD_SUCCESS) {
REPORT("Executing target region abort target.\n");
diff --git a/openmp/libomptarget/src/private.h b/openmp/libomptarget/src/private.h
index 521f39f34f18c..0277966c954ee 100644
--- a/openmp/libomptarget/src/private.h
+++ b/openmp/libomptarget/src/private.h
@@ -38,12 +38,8 @@ extern int targetDataUpdate(ident_t *Loc, DeviceTy &Device, int32_t ArgNum,
void **ArgMappers, AsyncInfoTy &AsyncInfo,
bool FromMapper = false);
-extern int target(ident_t *Loc, DeviceTy &Device, void *HostPtr, int32_t ArgNum,
- void **ArgBases, void **Args, int64_t *ArgSizes,
- int64_t *ArgTypes, map_var_info_t *ArgNames,
- void **ArgMappers, int32_t TeamNum, int32_t ThreadLimit,
- uint64_t Tripcount, int IsTeamConstruct,
- AsyncInfoTy &AsyncInfo);
+extern int target(ident_t *Loc, DeviceTy &Device, void *HostPtr,
+ const KernelArgsTy &KernelArgs, AsyncInfoTy &AsyncInfo);
extern int target_replay(ident_t *Loc, DeviceTy &Device, void *HostPtr,
void *DeviceMemory, int64_t DeviceMemorySize,
diff --git a/openmp/libomptarget/src/rtl.cpp b/openmp/libomptarget/src/rtl.cpp
index 1c490c40590e0..a2471506dde6f 100644
--- a/openmp/libomptarget/src/rtl.cpp
+++ b/openmp/libomptarget/src/rtl.cpp
@@ -170,11 +170,8 @@ bool RTLsTy::attemptLoadRTL(const std::string &RTLName, RTLInfoTy &RTL) {
if (!(*((void **)&RTL.data_delete) =
DynLibrary->getAddressOfSymbol("__tgt_rtl_data_delete")))
ValidPlugin = false;
- if (!(*((void **)&RTL.run_region) =
- DynLibrary->getAddressOfSymbol("__tgt_rtl_run_target_region")))
- ValidPlugin = false;
- if (!(*((void **)&RTL.run_team_region) =
- DynLibrary->getAddressOfSymbol("__tgt_rtl_run_target_team_region")))
+ if (!(*((void **)&RTL.launch_kernel) =
+ DynLibrary->getAddressOfSymbol("__tgt_rtl_launch_kernel")))
ValidPlugin = false;
// Invalid plugin
@@ -209,10 +206,6 @@ bool RTLsTy::attemptLoadRTL(const std::string &RTLName, RTLInfoTy &RTL) {
DynLibrary->getAddressOfSymbol("__tgt_rtl_data_submit_async");
*((void **)&RTL.data_retrieve_async) =
DynLibrary->getAddressOfSymbol("__tgt_rtl_data_retrieve_async");
- *((void **)&RTL.run_region_async) =
- DynLibrary->getAddressOfSymbol("__tgt_rtl_run_target_region_async");
- *((void **)&RTL.run_team_region_async) =
- DynLibrary->getAddressOfSymbol("__tgt_rtl_run_target_team_region_async");
*((void **)&RTL.synchronize) =
DynLibrary->getAddressOfSymbol("__tgt_rtl_synchronize");
*((void **)&RTL.query_async) =
@@ -534,9 +527,7 @@ void RTLsTy::unregisterLib(__tgt_bin_desc *Desc) {
if (Device.PendingCtorsDtors[Desc].PendingCtors.empty()) {
AsyncInfoTy AsyncInfo(Device);
for (auto &Dtor : Device.PendingCtorsDtors[Desc].PendingDtors) {
- int Rc = target(nullptr, Device, Dtor, 0, nullptr, nullptr, nullptr,
- nullptr, nullptr, nullptr, 1, 1, 0, true /*team*/,
- AsyncInfo);
+ int Rc = target(nullptr, Device, Dtor, CTorDTorKernelArgs, AsyncInfo);
if (Rc != OFFLOAD_SUCCESS) {
DP("Running destructor " DPxMOD " failed.\n", DPxPTR(Dtor));
}
More information about the cfe-commits
mailing list