[llvm] 6f0d99d - [OpenMPOpt] Regression test for hiding latency of H2D mem transfers

Hamilton Tobon Mosquera via llvm-commits llvm-commits at lists.llvm.org
Wed Jul 22 18:03:19 PDT 2020


Author: Hamilton Tobon Mosquera
Date: 2020-07-22T20:02:54-05:00
New Revision: 6f0d99d2b9b3b8ae96dd91c8988cc067b9c9afb9

URL: https://github.com/llvm/llvm-project/commit/6f0d99d2b9b3b8ae96dd91c8988cc067b9c9afb9
DIFF: https://github.com/llvm/llvm-project/commit/6f0d99d2b9b3b8ae96dd91c8988cc067b9c9afb9.diff

LOG: [OpenMPOpt] Regression test for hiding latency of H2D mem transfers

Added: 
    llvm/test/Transforms/OpenMP/hide_mem_transfer_latency.ll

Modified: 
    

Removed: 
    


################################################################################
diff  --git a/llvm/test/Transforms/OpenMP/hide_mem_transfer_latency.ll b/llvm/test/Transforms/OpenMP/hide_mem_transfer_latency.ll
new file mode 100644
index 000000000000..daec0e5e78d5
--- /dev/null
+++ b/llvm/test/Transforms/OpenMP/hide_mem_transfer_latency.ll
@@ -0,0 +1,561 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: -p --function-signature
+; RUN: opt -S -passes=openmpopt < %s | FileCheck %s
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+
+; FIXME: This struct should be generated after splitting at least one of the runtime calls.
+; %struct.__tgt_async_info = type { i8* }
+%struct.ident_t = type { i32, i32, i32, i32, i8* }
+%struct.__tgt_offload_entry = type { i8*, i8*, i64, i32, i32 }
+
+ at .offload_maptypes = private unnamed_addr constant [1 x i64] [i64 35]
+ at .__omp_offloading_heavyComputation1.region_id = weak constant i8 0
+ at .offload_sizes.1 = private unnamed_addr constant [1 x i64] [i64 8]
+ at .offload_maptypes.2 = private unnamed_addr constant [1 x i64] [i64 800]
+
+ at .__omp_offloading_heavyComputation2.region_id = weak constant i8 0
+ at .offload_maptypes.3 = private unnamed_addr constant [2 x i64] [i64 35, i64 35]
+
+ at .__omp_offloading_heavyComputation3.region_id = weak constant i8 0
+ at .offload_sizes.2 = private unnamed_addr constant [2 x i64] [i64 4, i64 0]
+ at .offload_maptypes.4 = private unnamed_addr constant [2 x i64] [i64 800, i64 544]
+
+ at .offload_maptypes.5 = private unnamed_addr constant [1 x i64] [i64 33]
+
+;double heavyComputation1() {
+;  double a = rand() % 777;
+;  double random = rand();
+;
+;  //#pragma omp target data map(a)
+;  void* args[1];
+;  args[0] = &a;
+;  __tgt_target_data_begin(..., args, ...)
+;
+;  #pragma omp target teams
+;  for (int i = 0; i < 1000; ++i) {
+;    a *= i*i / 2;
+;  }
+;
+;  return random + a;
+;}
+define dso_local double @heavyComputation1() {
+; CHECK-LABEL: define {{[^@]+}}@heavyComputation1()
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    %a = alloca double, align 8
+; CHECK-NEXT:    %.offload_baseptrs = alloca [1 x i8*], align 8
+; CHECK-NEXT:    %.offload_ptrs = alloca [1 x i8*], align 8
+; CHECK-NEXT:    %.offload_baseptrs4 = alloca [1 x i8*], align 8
+; CHECK-NEXT:    %.offload_ptrs5 = alloca [1 x i8*], align 8
+; CHECK-NEXT:    %0 = bitcast double* %a to i8*
+; CHECK-NEXT:    %call = call i32 @rand()
+; CHECK-NEXT:    %rem = srem i32 %call, 777
+; CHECK-NEXT:    %conv = sitofp i32 %rem to double
+; CHECK-NEXT:    store double %conv, double* %a, align 8
+; CHECK-NEXT:    %call1 = call i32 @rand()
+; CHECK-NEXT:    %1 = getelementptr inbounds [1 x i8*], [1 x i8*]* %.offload_baseptrs, i64 0, i64 0
+; CHECK-NEXT:    %2 = bitcast [1 x i8*]* %.offload_baseptrs to double**
+; CHECK-NEXT:    store double* %a, double** %2, align 8
+; CHECK-NEXT:    %3 = getelementptr inbounds [1 x i8*], [1 x i8*]* %.offload_ptrs, i64 0, i64 0
+; CHECK-NEXT:    %4 = bitcast [1 x i8*]* %.offload_ptrs to double**
+; CHECK-NEXT:    store double* %a, double** %4, align 8
+; CHECK-NEXT:    call void @__tgt_target_data_begin(i64 -1, i32 1, i8** nonnull %1, i8** nonnull %3, i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_sizes.1, i64 0, i64 0), i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_maptypes, i64 0, i64 0))
+; CHECK-NEXT:    %5 = bitcast double* %a to i64*
+; CHECK-NEXT:    %6 = load i64, i64* %5, align 8
+; CHECK-NEXT:    %7 = getelementptr inbounds [1 x i8*], [1 x i8*]* %.offload_baseptrs4, i64 0, i64 0
+; CHECK-NEXT:    %8 = bitcast [1 x i8*]* %.offload_baseptrs4 to i64*
+; CHECK-NEXT:    store i64 %6, i64* %8, align 8
+; CHECK-NEXT:    %9 = getelementptr inbounds [1 x i8*], [1 x i8*]* %.offload_ptrs5, i64 0, i64 0
+; CHECK-NEXT:    %10 = bitcast [1 x i8*]* %.offload_ptrs5 to i64*
+; CHECK-NEXT:    store i64 %6, i64* %10, align 8
+; CHECK-NEXT:    %11 = call i32 @__tgt_target_teams(i64 -1, i8* nonnull @.__omp_offloading_heavyComputation1.region_id, i32 1, i8** nonnull %7, i8** nonnull %9, i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_sizes.1, i64 0, i64 0), i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_maptypes.2, i64 0, i64 0), i32 0, i32 0)
+; CHECK-NEXT:    %12 = icmp eq i32 %11, 0
+; CHECK-NEXT:    br i1 %12, label %omp_offload.cont, label %omp_offload.failed
+; CHECK:       omp_offload.failed:
+; CHECK-NEXT:    call void @heavyComputation1FallBack(i64 %6)
+; CHECK-NEXT:    br label %omp_offload.cont
+; CHECK:       omp_offload.cont:
+; CHECK-NEXT:    %conv2 = sitofp i32 %call1 to double
+; CHECK-NEXT:    call void @__tgt_target_data_end(i64 -1, i32 1, i8** nonnull %1, i8** nonnull %3, i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_sizes.1, i64 0, i64 0), i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_maptypes, i64 0, i64 0))
+; CHECK-NEXT:    %13 = load double, double* %a, align 8
+; CHECK-NEXT:    %add = fadd double %13, %conv2
+; CHECK-NEXT:    ret double %add
+;
+entry:
+  %a = alloca double, align 8
+  %.offload_baseptrs = alloca [1 x i8*], align 8
+  %.offload_ptrs = alloca [1 x i8*], align 8
+  %.offload_baseptrs4 = alloca [1 x i8*], align 8
+  %.offload_ptrs5 = alloca [1 x i8*], align 8
+
+  ; FIXME: Should have after splitting the runtime call __tgt_target_data_begin.
+  ; %device_id1 = alloca i64, align 8
+  ; %async_info1 = alloca %struct.__tgt_async_info, align 8
+
+  %0 = bitcast double* %a to i8*
+  %call = call i32 @rand()
+  %rem = srem i32 %call, 777
+  %conv = sitofp i32 %rem to double
+  store double %conv, double* %a, align 8
+
+  ; FIXME: The "isue" should be moved here.
+  %call1 = call i32 @rand()
+
+  ; FIXME: This setup for the runtime call __tgt_target_data_begin should be
+  ;        split into its "issue" and "wait" counterpars and moved upwards
+  ;        and downwards, respectively. The call should be replaced to something
+  ;        like ...
+  ; Issue - this is moved upwards.
+  ; ... setup code ...
+  ; store i64 -1, i64* %device_id1, align 8
+  ; %handle1 = call i8* @__tgt_target_data_begin(i64* dereferenceable(8) %device_id1, i32 1, i8** nonnull %1, i8** nonnull %3, i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_sizes.1, i64 0, i64 0), i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_maptypes, i64 0, i64 0))
+  ; Wait - this is moved downwards.
+  ; %device_id1_copy = load i64, i64* %device_id1, align 8 ; device_id
+  ; %queue1 = getelementptr inbounds %struct.__tgt_async_info, %struct.__tgt_async_info* %async_info1, i32 0, i32 0
+  ; store i8* %handle1, i8** %queue1, align 8
+  ; call void @__tgt_target_data_begin_wait(i64 %device_id1_copy, %struct.__tgt_async_info* dereferenceable(8) %async_info1)
+  %1 = getelementptr inbounds [1 x i8*], [1 x i8*]* %.offload_baseptrs, i64 0, i64 0
+  %2 = bitcast [1 x i8*]* %.offload_baseptrs to double**
+  store double* %a, double** %2, align 8
+  %3 = getelementptr inbounds [1 x i8*], [1 x i8*]* %.offload_ptrs, i64 0, i64 0
+  %4 = bitcast [1 x i8*]* %.offload_ptrs to double**
+  store double* %a, double** %4, align 8
+  call void @__tgt_target_data_begin(i64 -1, i32 1, i8** nonnull %1, i8** nonnull %3, i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_sizes.1, i64 0, i64 0), i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_maptypes, i64 0, i64 0))
+
+  %5 = bitcast double* %a to i64*
+  %6 = load i64, i64* %5, align 8
+  %7 = getelementptr inbounds [1 x i8*], [1 x i8*]* %.offload_baseptrs4, i64 0, i64 0
+  %8 = bitcast [1 x i8*]* %.offload_baseptrs4 to i64*
+  store i64 %6, i64* %8, align 8
+  %9 = getelementptr inbounds [1 x i8*], [1 x i8*]* %.offload_ptrs5, i64 0, i64 0
+  %10 = bitcast [1 x i8*]* %.offload_ptrs5 to i64*
+  store i64 %6, i64* %10, align 8
+
+  ; FIXME: The "wait" should be moved here.
+  %11 = call i32 @__tgt_target_teams(i64 -1, i8* nonnull @.__omp_offloading_heavyComputation1.region_id, i32 1, i8** nonnull %7, i8** nonnull %9, i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_sizes.1, i64 0, i64 0), i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_maptypes.2, i64 0, i64 0), i32 0, i32 0)
+
+  %12 = icmp eq i32 %11, 0
+  br i1 %12, label %omp_offload.cont, label %omp_offload.failed
+
+omp_offload.failed:                               ; preds = %entry
+  call void @heavyComputation1FallBack(i64 %6)
+  br label %omp_offload.cont
+
+omp_offload.cont:                                 ; preds = %entry, %omp_offload.failed
+  %conv2 = sitofp i32 %call1 to double
+  call void @__tgt_target_data_end(i64 -1, i32 1, i8** nonnull %1, i8** nonnull %3, i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_sizes.1, i64 0, i64 0), i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_maptypes, i64 0, i64 0))
+  %13 = load double, double* %a, align 8
+  %add = fadd double %13, %conv2
+  ret double %add
+}
+
+define internal void @heavyComputation1FallBack(i64 %a) {
+entry:
+  ; Fallback for offloading function heavyComputation1.
+  ret void
+}
+
+;int heavyComputation2(double* a, unsigned size) {
+;  int random = rand() % 7;
+;
+;  //#pragma omp target data map(a[0:size], size)
+;  void* args[2];
+;  args[0] = &a;
+;  args[1] = &size;
+;  __tgt_target_data_begin(..., args, ...)
+;
+;  #pragma omp target teams
+;  for (int i = 0; i < size; ++i) {
+;    a[i] = ++a[i] * 3.141624;
+;  }
+;
+;  return random;
+;}
+define dso_local i32 @heavyComputation2(double* %a, i32 %size) {
+; CHECK-LABEL: define {{[^@]+}}@heavyComputation2(double* %a, i32 %size)
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    %size.addr = alloca i32, align 4
+; CHECK-NEXT:    %.offload_baseptrs = alloca [2 x i8*], align 8
+; CHECK-NEXT:    %.offload_ptrs = alloca [2 x i8*], align 8
+; CHECK-NEXT:    %.offload_sizes = alloca [2 x i64], align 8
+; CHECK-NEXT:    %.offload_baseptrs2 = alloca [2 x i8*], align 8
+; CHECK-NEXT:    %.offload_ptrs3 = alloca [2 x i8*], align 8
+; CHECK-NEXT:    store i32 %size, i32* %size.addr, align 4
+; CHECK-NEXT:    %call = call i32 @rand()
+; CHECK-NEXT:    %conv = zext i32 %size to i64
+; CHECK-NEXT:    %0 = shl nuw nsw i64 %conv, 3
+; CHECK-NEXT:    %1 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_baseptrs, i64 0, i64 0
+; CHECK-NEXT:    %2 = bitcast [2 x i8*]* %.offload_baseptrs to double**
+; CHECK-NEXT:    store double* %a, double** %2, align 8
+; CHECK-NEXT:    %3 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_ptrs, i64 0, i64 0
+; CHECK-NEXT:    %4 = bitcast [2 x i8*]* %.offload_ptrs to double**
+; CHECK-NEXT:    store double* %a, double** %4, align 8
+; CHECK-NEXT:    %5 = getelementptr inbounds [2 x i64], [2 x i64]* %.offload_sizes, i64 0, i64 0
+; CHECK-NEXT:    store i64 %0, i64* %5, align 8
+; CHECK-NEXT:    %6 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_baseptrs, i64 0, i64 1
+; CHECK-NEXT:    %7 = bitcast i8** %6 to i32**
+; CHECK-NEXT:    store i32* %size.addr, i32** %7, align 8
+; CHECK-NEXT:    %8 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_ptrs, i64 0, i64 1
+; CHECK-NEXT:    %9 = bitcast i8** %8 to i32**
+; CHECK-NEXT:    store i32* %size.addr, i32** %9, align 8
+; CHECK-NEXT:    %10 = getelementptr inbounds [2 x i64], [2 x i64]* %.offload_sizes, i64 0, i64 1
+; CHECK-NEXT:    store i64 4, i64* %10, align 8
+; CHECK-NEXT:    call void @__tgt_target_data_begin(i64 -1, i32 2, i8** nonnull %1, i8** nonnull %3, i64* nonnull %5, i64* getelementptr inbounds ([2 x i64], [2 x i64]* @.offload_maptypes.3, i64 0, i64 0))
+; CHECK-NEXT:    %11 = load i32, i32* %size.addr, align 4
+; CHECK-NEXT:    %size.casted = zext i32 %11 to i64
+; CHECK-NEXT:    %12 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_baseptrs2, i64 0, i64 0
+; CHECK-NEXT:    %13 = bitcast [2 x i8*]* %.offload_baseptrs2 to i64*
+; CHECK-NEXT:    store i64 %size.casted, i64* %13, align 8
+; CHECK-NEXT:    %14 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_ptrs3, i64 0, i64 0
+; CHECK-NEXT:    %15 = bitcast [2 x i8*]* %.offload_ptrs3 to i64*
+; CHECK-NEXT:    store i64 %size.casted, i64* %15, align 8
+; CHECK-NEXT:    %16 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_baseptrs2, i64 0, i64 1
+; CHECK-NEXT:    %17 = bitcast i8** %16 to double**
+; CHECK-NEXT:    store double* %a, double** %17, align 8
+; CHECK-NEXT:    %18 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_ptrs3, i64 0, i64 1
+; CHECK-NEXT:    %19 = bitcast i8** %18 to double**
+; CHECK-NEXT:    store double* %a, double** %19, align 8
+; CHECK-NEXT:    %20 = call i32 @__tgt_target_teams(i64 -1, i8* nonnull @.__omp_offloading_heavyComputation2.region_id, i32 2, i8** nonnull %12, i8** nonnull %14, i64* getelementptr inbounds ([2 x i64], [2 x i64]* @.offload_sizes.2, i64 0, i64 0), i64* getelementptr inbounds ([2 x i64], [2 x i64]* @.offload_maptypes.4, i64 0, i64 0), i32 0, i32 0)
+; CHECK-NEXT:    %21 = icmp eq i32 %20, 0
+; CHECK-NEXT:    br i1 %21, label %omp_offload.cont, label %omp_offload.failed
+; CHECK:       omp_offload.failed:
+; CHECK-NEXT:    call void @heavyComputation2FallBack(i64 %size.casted, double* %a)
+; CHECK-NEXT:    br label %omp_offload.cont
+; CHECK:       omp_offload.cont:
+; CHECK-NEXT:    %rem = srem i32 %call, 7
+; CHECK-NEXT:    call void @__tgt_target_data_end(i64 -1, i32 2, i8** nonnull %1, i8** nonnull %3, i64* nonnull %5, i64* getelementptr inbounds ([2 x i64], [2 x i64]* @.offload_maptypes.3, i64 0, i64 0))
+; CHECK-NEXT:    ret i32 %rem
+;
+entry:
+  %size.addr = alloca i32, align 4
+  %.offload_baseptrs = alloca [2 x i8*], align 8
+  %.offload_ptrs = alloca [2 x i8*], align 8
+  %.offload_sizes = alloca [2 x i64], align 8
+  %.offload_baseptrs2 = alloca [2 x i8*], align 8
+  %.offload_ptrs3 = alloca [2 x i8*], align 8
+
+  ; FIXME: Should have after splitting the runtime call __tgt_target_data_begin.
+  ; %device_id1 = alloca i64, align 8
+  ; %async_info1 = alloca %struct.__tgt_async_info, align 8
+
+  store i32 %size, i32* %size.addr, align 4
+  %call = call i32 @rand()
+
+  ; FIXME: This setup for the runtime call __tgt_target_data_begin should be
+  ;        split into its "issue" and "wait" counterpars. Here though, the "issue"
+  ;        cannot be moved upwards because it's not guaranteed that rand()
+  ;        won't modify *a. Nevertheless, the "wait" can be moved downwards.
+  ;        The call should be replaced to something like ...
+  ; Issue - this can't be moved upwards, *a might have aliases.
+  ; ... setup code ...
+  ; store i64 -1, i64* %device_id1, align 8
+  ; %handle1 = call i8* @__tgt_target_data_begin(i64* dereferenceable(8) %device_id1, i32 2, i8** nonnull %1, i8** nonnull %3, i64* nonnull %5, i64* getelementptr inbounds ([2 x i64], [2 x i64]* @.offload_maptypes.3, i64 0, i64 0))
+  ; Wait - this is moved downards.
+  ; %device_id1_copy = load i64, i64* %device_id1, align 8 ; device_id
+  ; %queue1 = getelementptr inbounds %struct.__tgt_async_info, %struct.__tgt_async_info* %async_info1, i32 0, i32 0
+  ; store i8* %handle1, i8** %queue1, align 8
+  ; call void @__tgt_target_data_begin_wait(i64 %device_id1_copy, %struct.__tgt_async_info* dereferenceable(8) %async_info1)
+  %conv = zext i32 %size to i64
+  %0 = shl nuw nsw i64 %conv, 3
+  %1 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_baseptrs, i64 0, i64 0
+  %2 = bitcast [2 x i8*]* %.offload_baseptrs to double**
+  store double* %a, double** %2, align 8
+  %3 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_ptrs, i64 0, i64 0
+  %4 = bitcast [2 x i8*]* %.offload_ptrs to double**
+  store double* %a, double** %4, align 8
+  %5 = getelementptr inbounds [2 x i64], [2 x i64]* %.offload_sizes, i64 0, i64 0
+  store i64 %0, i64* %5, align 8
+  %6 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_baseptrs, i64 0, i64 1
+  %7 = bitcast i8** %6 to i32**
+  store i32* %size.addr, i32** %7, align 8
+  %8 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_ptrs, i64 0, i64 1
+  %9 = bitcast i8** %8 to i32**
+  store i32* %size.addr, i32** %9, align 8
+  %10 = getelementptr inbounds [2 x i64], [2 x i64]* %.offload_sizes, i64 0, i64 1
+  store i64 4, i64* %10, align 8
+  call void @__tgt_target_data_begin(i64 -1, i32 2, i8** nonnull %1, i8** nonnull %3, i64* nonnull %5, i64* getelementptr inbounds ([2 x i64], [2 x i64]* @.offload_maptypes.3, i64 0, i64 0))
+
+  %11 = load i32, i32* %size.addr, align 4
+  %size.casted = zext i32 %11 to i64
+  %12 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_baseptrs2, i64 0, i64 0
+  %13 = bitcast [2 x i8*]* %.offload_baseptrs2 to i64*
+  store i64 %size.casted, i64* %13, align 8
+  %14 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_ptrs3, i64 0, i64 0
+  %15 = bitcast [2 x i8*]* %.offload_ptrs3 to i64*
+  store i64 %size.casted, i64* %15, align 8
+  %16 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_baseptrs2, i64 0, i64 1
+  %17 = bitcast i8** %16 to double**
+  store double* %a, double** %17, align 8
+  %18 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_ptrs3, i64 0, i64 1
+  %19 = bitcast i8** %18 to double**
+  store double* %a, double** %19, align 8
+
+  ; FIXME: The "wait" should be moved here.
+  %20 = call i32 @__tgt_target_teams(i64 -1, i8* nonnull @.__omp_offloading_heavyComputation2.region_id, i32 2, i8** nonnull %12, i8** nonnull %14, i64* getelementptr inbounds ([2 x i64], [2 x i64]* @.offload_sizes.2, i64 0, i64 0), i64* getelementptr inbounds ([2 x i64], [2 x i64]* @.offload_maptypes.4, i64 0, i64 0), i32 0, i32 0)
+
+  %21 = icmp eq i32 %20, 0
+  br i1 %21, label %omp_offload.cont, label %omp_offload.failed
+
+omp_offload.failed:                               ; preds = %entry
+  call void @heavyComputation2FallBack(i64 %size.casted, double* %a)
+  br label %omp_offload.cont
+
+omp_offload.cont:                                 ; preds = %entry, %omp_offload.failed
+  %rem = srem i32 %call, 7
+  call void @__tgt_target_data_end(i64 -1, i32 2, i8** nonnull %1, i8** nonnull %3, i64* nonnull %5, i64* getelementptr inbounds ([2 x i64], [2 x i64]* @.offload_maptypes.3, i64 0, i64 0))
+  ret i32 %rem
+}
+
+define internal void @heavyComputation2FallBack(i64 %size, double* %a) {
+entry:
+  ; Fallback for offloading function heavyComputation2.
+  ret void
+}
+
+;int heavyComputation3(double* restrict a, unsigned size) {
+;  int random = rand() % 7;
+;
+;  //#pragma omp target data map(a[0:size], size)
+;  void* args[2];
+;  args[0] = &a;
+;  args[1] = &size;
+;  __tgt_target_data_begin(..., args, ...)
+;
+;  #pragma omp target teams
+;  for (int i = 0; i < size; ++i) {
+;    a[i] = ++a[i] * 3.141624;
+;  }
+;
+;  return random;
+;}
+define dso_local i32 @heavyComputation3(double* noalias %a, i32 %size) {
+; CHECK-LABEL: define {{[^@]+}}@heavyComputation3(double* noalias %a, i32 %size)
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    %size.addr = alloca i32, align 4
+; CHECK-NEXT:    %.offload_baseptrs = alloca [2 x i8*], align 8
+; CHECK-NEXT:    %.offload_ptrs = alloca [2 x i8*], align 8
+; CHECK-NEXT:    %.offload_sizes = alloca [2 x i64], align 8
+; CHECK-NEXT:    %.offload_baseptrs2 = alloca [2 x i8*], align 8
+; CHECK-NEXT:    %.offload_ptrs3 = alloca [2 x i8*], align 8
+; CHECK-NEXT:    store i32 %size, i32* %size.addr, align 4
+; CHECK-NEXT:    %call = call i32 @rand()
+; CHECK-NEXT:    %conv = zext i32 %size to i64
+; CHECK-NEXT:    %0 = shl nuw nsw i64 %conv, 3
+; CHECK-NEXT:    %1 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_baseptrs, i64 0, i64 0
+; CHECK-NEXT:    %2 = bitcast [2 x i8*]* %.offload_baseptrs to double**
+; CHECK-NEXT:    store double* %a, double** %2, align 8
+; CHECK-NEXT:    %3 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_ptrs, i64 0, i64 0
+; CHECK-NEXT:    %4 = bitcast [2 x i8*]* %.offload_ptrs to double**
+; CHECK-NEXT:    store double* %a, double** %4, align 8
+; CHECK-NEXT:    %5 = getelementptr inbounds [2 x i64], [2 x i64]* %.offload_sizes, i64 0, i64 0
+; CHECK-NEXT:    store i64 %0, i64* %5, align 8
+; CHECK-NEXT:    %6 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_baseptrs, i64 0, i64 1
+; CHECK-NEXT:    %7 = bitcast i8** %6 to i32**
+; CHECK-NEXT:    store i32* %size.addr, i32** %7, align 8
+; CHECK-NEXT:    %8 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_ptrs, i64 0, i64 1
+; CHECK-NEXT:    %9 = bitcast i8** %8 to i32**
+; CHECK-NEXT:    store i32* %size.addr, i32** %9, align 8
+; CHECK-NEXT:    %10 = getelementptr inbounds [2 x i64], [2 x i64]* %.offload_sizes, i64 0, i64 1
+; CHECK-NEXT:    store i64 4, i64* %10, align 8
+; CHECK-NEXT:    call void @__tgt_target_data_begin(i64 -1, i32 2, i8** nonnull %1, i8** nonnull %3, i64* nonnull %5, i64* getelementptr inbounds ([2 x i64], [2 x i64]* @.offload_maptypes.3, i64 0, i64 0))
+; CHECK-NEXT:    %11 = load i32, i32* %size.addr, align 4
+; CHECK-NEXT:    %size.casted = zext i32 %11 to i64
+; CHECK-NEXT:    %12 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_baseptrs2, i64 0, i64 0
+; CHECK-NEXT:    %13 = bitcast [2 x i8*]* %.offload_baseptrs2 to i64*
+; CHECK-NEXT:    store i64 %size.casted, i64* %13, align 8
+; CHECK-NEXT:    %14 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_ptrs3, i64 0, i64 0
+; CHECK-NEXT:    %15 = bitcast [2 x i8*]* %.offload_ptrs3 to i64*
+; CHECK-NEXT:    store i64 %size.casted, i64* %15, align 8
+; CHECK-NEXT:    %16 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_baseptrs2, i64 0, i64 1
+; CHECK-NEXT:    %17 = bitcast i8** %16 to double**
+; CHECK-NEXT:    store double* %a, double** %17, align 8
+; CHECK-NEXT:    %18 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_ptrs3, i64 0, i64 1
+; CHECK-NEXT:    %19 = bitcast i8** %18 to double**
+; CHECK-NEXT:    store double* %a, double** %19, align 8
+; CHECK-NEXT:    %20 = call i32 @__tgt_target_teams(i64 -1, i8* nonnull @.__omp_offloading_heavyComputation3.region_id, i32 2, i8** nonnull %12, i8** nonnull %14, i64* getelementptr inbounds ([2 x i64], [2 x i64]* @.offload_sizes.2, i64 0, i64 0), i64* getelementptr inbounds ([2 x i64], [2 x i64]* @.offload_maptypes.4, i64 0, i64 0), i32 0, i32 0)
+; CHECK-NEXT:    %21 = icmp eq i32 %20, 0
+; CHECK-NEXT:    br i1 %21, label %omp_offload.cont, label %omp_offload.failed
+; CHECK:       omp_offload.failed:
+; CHECK-NEXT:    call void @heavyComputation3FallBack(i64 %size.casted, double* %a)
+; CHECK-NEXT:    br label %omp_offload.cont
+; CHECK:       omp_offload.cont:
+; CHECK-NEXT:    %rem = srem i32 %call, 7
+; CHECK-NEXT:    call void @__tgt_target_data_end(i64 -1, i32 2, i8** nonnull %1, i8** nonnull %3, i64* nonnull %5, i64* getelementptr inbounds ([2 x i64], [2 x i64]* @.offload_maptypes.3, i64 0, i64 0))
+; CHECK-NEXT:    ret i32 %rem
+;
+entry:
+  %size.addr = alloca i32, align 4
+  %.offload_baseptrs = alloca [2 x i8*], align 8
+  %.offload_ptrs = alloca [2 x i8*], align 8
+  %.offload_sizes = alloca [2 x i64], align 8
+  %.offload_baseptrs2 = alloca [2 x i8*], align 8
+  %.offload_ptrs3 = alloca [2 x i8*], align 8
+
+  ; FIXME: Should have after splitting the runtime call __tgt_target_data_begin.
+  ; %device_id1 = alloca i64, align 8
+  ; %async_info1 = alloca %struct.__tgt_async_info, align 8
+
+  store i32 %size, i32* %size.addr, align 4
+
+  ; FIXME: The "issue" should be moved here.
+  %call = call i32 @rand()
+
+  ; FIXME: This setup for the runtime call __tgt_target_data_begin should be
+  ;        split into its "issue" and "wait" counterpars and moved upwards
+  ;        and downwards, respectively. The call should be replaced to something
+  ;        like ...
+  ; Issue - this is moved upwards.
+  ; ... setup code ...
+  ; store i64 -1, i64* %device_id1, align 8
+  ; %handle1 = call i8* @__tgt_target_data_begin(i64* dereferenceable(8) %device_id1, i32 2, i8** nonnull %1, i8** nonnull %3, i64* nonnull %5, i64* getelementptr inbounds ([2 x i64], [2 x i64]* @.offload_maptypes.3, i64 0, i64 0))
+  ; Wait - this is moved downards.
+  ; %device_id1_copy = load i64, i64* %device_id1, align 8 ; device_id
+  ; %queue1 = getelementptr inbounds %struct.__tgt_async_info, %struct.__tgt_async_info* %async_info1, i32 0, i32 0
+  ; store i8* %handle1, i8** %queue1, align 8
+  ; call void @__tgt_target_data_begin_wait(i64 %device_id1_copy, %struct.__tgt_async_info* dereferenceable(8) %async_info1)
+  %conv = zext i32 %size to i64
+  %0 = shl nuw nsw i64 %conv, 3
+  %1 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_baseptrs, i64 0, i64 0
+  %2 = bitcast [2 x i8*]* %.offload_baseptrs to double**
+  store double* %a, double** %2, align 8
+  %3 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_ptrs, i64 0, i64 0
+  %4 = bitcast [2 x i8*]* %.offload_ptrs to double**
+  store double* %a, double** %4, align 8
+  %5 = getelementptr inbounds [2 x i64], [2 x i64]* %.offload_sizes, i64 0, i64 0
+  store i64 %0, i64* %5, align 8
+  %6 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_baseptrs, i64 0, i64 1
+  %7 = bitcast i8** %6 to i32**
+  store i32* %size.addr, i32** %7, align 8
+  %8 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_ptrs, i64 0, i64 1
+  %9 = bitcast i8** %8 to i32**
+  store i32* %size.addr, i32** %9, align 8
+  %10 = getelementptr inbounds [2 x i64], [2 x i64]* %.offload_sizes, i64 0, i64 1
+  store i64 4, i64* %10, align 8
+  call void @__tgt_target_data_begin(i64 -1, i32 2, i8** nonnull %1, i8** nonnull %3, i64* nonnull %5, i64* getelementptr inbounds ([2 x i64], [2 x i64]* @.offload_maptypes.3, i64 0, i64 0))
+
+  %11 = load i32, i32* %size.addr, align 4
+  %size.casted = zext i32 %11 to i64
+  %12 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_baseptrs2, i64 0, i64 0
+  %13 = bitcast [2 x i8*]* %.offload_baseptrs2 to i64*
+  store i64 %size.casted, i64* %13, align 8
+  %14 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_ptrs3, i64 0, i64 0
+  %15 = bitcast [2 x i8*]* %.offload_ptrs3 to i64*
+  store i64 %size.casted, i64* %15, align 8
+  %16 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_baseptrs2, i64 0, i64 1
+  %17 = bitcast i8** %16 to double**
+  store double* %a, double** %17, align 8
+  %18 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_ptrs3, i64 0, i64 1
+  %19 = bitcast i8** %18 to double**
+  store double* %a, double** %19, align 8
+
+  ; FIXME: The "wait" should be moved here.
+  %20 = call i32 @__tgt_target_teams(i64 -1, i8* nonnull @.__omp_offloading_heavyComputation3.region_id, i32 2, i8** nonnull %12, i8** nonnull %14, i64* getelementptr inbounds ([2 x i64], [2 x i64]* @.offload_sizes.2, i64 0, i64 0), i64* getelementptr inbounds ([2 x i64], [2 x i64]* @.offload_maptypes.4, i64 0, i64 0), i32 0, i32 0)
+
+  %21 = icmp eq i32 %20, 0
+  br i1 %21, label %omp_offload.cont, label %omp_offload.failed
+
+omp_offload.failed:                               ; preds = %entry
+  call void @heavyComputation3FallBack(i64 %size.casted, double* %a)
+  br label %omp_offload.cont
+
+omp_offload.cont:                                 ; preds = %entry, %omp_offload.failed
+  %rem = srem i32 %call, 7
+  call void @__tgt_target_data_end(i64 -1, i32 2, i8** nonnull %1, i8** nonnull %3, i64* nonnull %5, i64* getelementptr inbounds ([2 x i64], [2 x i64]* @.offload_maptypes.3, i64 0, i64 0))
+  ret i32 %rem
+}
+
+define internal void @heavyComputation3FallBack(i64 %size, double* %a) {
+entry:
+  ; Fallback for offloading function heavyComputation3.
+  ret void
+}
+
+;int dataTransferOnly1(double* restrict a, unsigned size) {
+;  // Random computation.
+;  int random = rand();
+;
+;  //#pragma omp target data map(to:a[0:size])
+;  void* args[1];
+;  args[0] = &a;
+;  __tgt_target_data_begin(..., args, ...)
+;
+;  // Random computation.
+;  random %= size;
+;  return random;
+;}
+define dso_local i32 @dataTransferOnly1(double* noalias %a, i32 %size) {
+; CHECK-LABEL: define {{[^@]+}}@dataTransferOnly1(double* noalias %a, i32 %size)
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    %.offload_baseptrs = alloca [1 x i8*], align 8
+; CHECK-NEXT:    %.offload_ptrs = alloca [1 x i8*], align 8
+; CHECK-NEXT:    %.offload_sizes = alloca [1 x i64], align 8
+; CHECK-NEXT:    %call = call i32 @rand()
+; CHECK-NEXT:    %conv = zext i32 %size to i64
+; CHECK-NEXT:    %0 = shl nuw nsw i64 %conv, 3
+; CHECK-NEXT:    %1 = getelementptr inbounds [1 x i8*], [1 x i8*]* %.offload_baseptrs, i64 0, i64 0
+; CHECK-NEXT:    %2 = bitcast [1 x i8*]* %.offload_baseptrs to double**
+; CHECK-NEXT:    store double* %a, double** %2, align 8
+; CHECK-NEXT:    %3 = getelementptr inbounds [1 x i8*], [1 x i8*]* %.offload_ptrs, i64 0, i64 0
+; CHECK-NEXT:    %4 = bitcast [1 x i8*]* %.offload_ptrs to double**
+; CHECK-NEXT:    store double* %a, double** %4, align 8
+; CHECK-NEXT:    %5 = getelementptr inbounds [1 x i64], [1 x i64]* %.offload_sizes, i64 0, i64 0
+; CHECK-NEXT:    store i64 %0, i64* %5, align 8
+; CHECK-NEXT:    call void @__tgt_target_data_begin(i64 -1, i32 1, i8** nonnull %1, i8** nonnull %3, i64* nonnull %5, i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_maptypes.5, i64 0, i64 0))
+; CHECK-NEXT:    %rem = urem i32 %call, %size
+; CHECK-NEXT:    call void @__tgt_target_data_end(i64 -1, i32 1, i8** nonnull %1, i8** nonnull %3, i64* nonnull %5, i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_maptypes.5, i64 0, i64 0))
+; CHECK-NEXT:    ret i32 %rem
+;
+entry:
+  %.offload_baseptrs = alloca [1 x i8*], align 8
+  %.offload_ptrs = alloca [1 x i8*], align 8
+  %.offload_sizes = alloca [1 x i64], align 8
+
+  ; FIXME: Should have after splitting the runtime call __tgt_target_data_begin.
+  ; %device_id1 = alloca i64, align 8
+  ; %async_info1 = alloca %struct.__tgt_async_info, align 8
+
+  ; FIXME: The "issue" should be moved here.
+  %call = call i32 @rand()
+
+  ; FIXME: This setup for the runtime call __tgt_target_data_begin should be
+  ;        split into its "issue" and "wait" counterpars and moved upwards
+  ;        and downwards, respectively. The call should be replaced to something
+  ;        like ...
+  ; Issue - this is moved upwards.
+  ; ... setup code ...
+  ; store i64 -1, i64* %device_id1, align 8
+  ; %handle1 = call i8* @__tgt_target_data_begin(i64* dereferenceable(8) %device_id1, i32 1, i8** nonnull %1, i8** nonnull %3, i64* nonnull %5, i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_maptypes.5, i64 0, i64 0))
+  ; Wait - this is moved downards.
+  ; %device_id1_copy = load i64, i64* %device_id1, align 8 ; device_id
+  ; %queue1 = getelementptr inbounds %struct.__tgt_async_info, %struct.__tgt_async_info* %async_info1, i32 0, i32 0
+  ; store i8* %handle1, i8** %queue1, align 8
+  ; call void @__tgt_target_data_begin_wait(i64 %device_id1_copy, %struct.__tgt_async_info* dereferenceable(8) %async_info1)
+  %conv = zext i32 %size to i64
+  %0 = shl nuw nsw i64 %conv, 3
+  %1 = getelementptr inbounds [1 x i8*], [1 x i8*]* %.offload_baseptrs, i64 0, i64 0
+  %2 = bitcast [1 x i8*]* %.offload_baseptrs to double**
+  store double* %a, double** %2, align 8
+  %3 = getelementptr inbounds [1 x i8*], [1 x i8*]* %.offload_ptrs, i64 0, i64 0
+  %4 = bitcast [1 x i8*]* %.offload_ptrs to double**
+  store double* %a, double** %4, align 8
+  %5 = getelementptr inbounds [1 x i64], [1 x i64]* %.offload_sizes, i64 0, i64 0
+  store i64 %0, i64* %5, align 8
+  call void @__tgt_target_data_begin(i64 -1, i32 1, i8** nonnull %1, i8** nonnull %3, i64* nonnull %5, i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_maptypes.5, i64 0, i64 0))
+
+  %rem = urem i32 %call, %size
+
+  ; FIXME: The "wait" should be moved here.
+  call void @__tgt_target_data_end(i64 -1, i32 1, i8** nonnull %1, i8** nonnull %3, i64* nonnull %5, i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_maptypes.5, i64 0, i64 0))
+  ret i32 %rem
+}
+
+declare dso_local void @__tgt_target_data_begin(i64, i32, i8**, i8**, i64*, i64*)
+
+declare dso_local i32 @__tgt_target_teams(i64, i8*, i32, i8**, i8**, i64*, i64*, i32, i32)
+
+declare dso_local void @__tgt_target_data_end(i64, i32, i8**, i8**, i64*, i64*)
+
+declare dso_local i32 @rand()
+
+; FIXME: These two function declarations must be generated after splitting the runtime function
+;        __tgt_target_data_begin.
+; declare dso_local i8* @__tgt_target_data_begin_issue(i64* dereferenceable(8), i32, i8**, i8**, i64*, i64*)
+; declare dso_local void @__tgt_target_data_begin_wait(i64, %struct.__tgt_async_info* dereferenceable(8))


        


More information about the llvm-commits mailing list