[llvm-branch-commits] [flang] [llvm] [mlir] [MLIR][OpenMP] Add OpenMPToLLVMIRTranslation support for is_device_ptr (PR #169367)
Akash Banerjee via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Wed Nov 26 08:54:43 PST 2025
https://github.com/TIFitis updated https://github.com/llvm/llvm-project/pull/169367
>From c6440f42e19a54bfce0d7f55ca40b5a28b30f7bc Mon Sep 17 00:00:00 2001
From: Akash Banerjee <Akash.Banerjee at amd.com>
Date: Mon, 24 Nov 2025 17:05:29 +0000
Subject: [PATCH 1/4] [MLIR][OpenMP] Add OpenMPToLLVMIRTranslation support for
is_device_ptr
---
.../OpenMP/map-types-and-sizes.f90 | 9 +++
.../OpenMP/OpenMPToLLVMIRTranslation.cpp | 18 +++--
mlir/test/Target/LLVMIR/openmp-todo.mlir | 11 ---
.../fortran/target-is-device-ptr.f90 | 79 +++++++++++++++++++
4 files changed, 100 insertions(+), 17 deletions(-)
create mode 100644 offload/test/offloading/fortran/target-is-device-ptr.f90
diff --git a/flang/test/Integration/OpenMP/map-types-and-sizes.f90 b/flang/test/Integration/OpenMP/map-types-and-sizes.f90
index 44a049f5ac510..85434460bbea6 100644
--- a/flang/test/Integration/OpenMP/map-types-and-sizes.f90
+++ b/flang/test/Integration/OpenMP/map-types-and-sizes.f90
@@ -33,6 +33,15 @@ subroutine mapType_array
!$omp end target
end subroutine mapType_array
+!CHECK: @.offload_sizes{{.*}} = private unnamed_addr constant [1 x i64] [i64 8]
+!CHECK: @.offload_maptypes{{.*}} = private unnamed_addr constant [1 x i64] [i64 288]
+subroutine mapType_is_device_ptr
+ use iso_c_binding, only : c_ptr
+ type(c_ptr) :: p
+ !$omp target is_device_ptr(p)
+ !$omp end target
+end subroutine mapType_is_device_ptr
+
!CHECK: @.offload_sizes{{.*}} = private unnamed_addr constant [4 x i64] [i64 0, i64 24, i64 8, i64 0]
!CHECK: @.offload_maptypes{{.*}} = private unnamed_addr constant [4 x i64] [i64 32, i64 281474976711169, i64 281474976711171, i64 281474976711187]
subroutine mapType_ptr
diff --git a/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp b/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
index f28454075f1d3..9fa0e659320d6 100644
--- a/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
+++ b/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
@@ -332,10 +332,7 @@ static LogicalResult checkImplementationStatus(Operation &op) {
op.getInReductionSyms())
result = todo("in_reduction");
};
- auto checkIsDevicePtr = [&todo](auto op, LogicalResult &result) {
- if (!op.getIsDevicePtrVars().empty())
- result = todo("is_device_ptr");
- };
+ auto checkIsDevicePtr = [](auto, LogicalResult &) {};
auto checkLinear = [&todo](auto op, LogicalResult &result) {
if (!op.getLinearVars().empty() || !op.getLinearStepVars().empty())
result = todo("linear");
@@ -3996,6 +3993,9 @@ static void collectMapDataFromMapOperands(
llvm::Value *origValue = moduleTranslation.lookupValue(offloadPtr);
auto mapType = convertClauseMapFlags(mapOp.getMapType());
auto mapTypeAlways = llvm::omp::OpenMPOffloadMappingFlags::OMP_MAP_ALWAYS;
+ bool isDevicePtr =
+ (mapOp.getMapType() & omp::ClauseMapFlags::storage) ==
+ omp::ClauseMapFlags::storage;
mapData.OriginalValue.push_back(origValue);
mapData.BasePointers.push_back(origValue);
@@ -4006,7 +4006,12 @@ static void collectMapDataFromMapOperands(
mapData.Sizes.push_back(
builder.getInt64(dl.getTypeSize(mapOp.getVarType())));
mapData.MapClause.push_back(mapOp.getOperation());
- if (llvm::to_underlying(mapType & mapTypeAlways)) {
+ if (isDevicePtr) {
+ mapType |= llvm::omp::OpenMPOffloadMappingFlags::OMP_MAP_TARGET_PARAM;
+ mapType |= llvm::omp::OpenMPOffloadMappingFlags::OMP_MAP_LITERAL;
+ mapData.Types.push_back(mapType);
+ mapData.Mappers.push_back(nullptr);
+ } else if (llvm::to_underlying(mapType & mapTypeAlways)) {
// Descriptors are mapped with the ALWAYS flag, since they can get
// rematerialized, so the address of the decriptor for a given object
// may change from one place to another.
@@ -4029,7 +4034,8 @@ static void collectMapDataFromMapOperands(
mapData.Names.push_back(LLVM::createMappingInformation(
mapOp.getLoc(), *moduleTranslation.getOpenMPBuilder()));
mapData.DevicePointers.push_back(
- llvm::OpenMPIRBuilder::DeviceInfoTy::Address);
+ isDevicePtr ? llvm::OpenMPIRBuilder::DeviceInfoTy::Pointer
+ : llvm::OpenMPIRBuilder::DeviceInfoTy::Address);
mapData.IsAMapping.push_back(false);
mapData.IsAMember.push_back(checkIsAMember(hasDevAddrOperands, mapOp));
}
diff --git a/mlir/test/Target/LLVMIR/openmp-todo.mlir b/mlir/test/Target/LLVMIR/openmp-todo.mlir
index af6d254cfd3c3..0704008aa7135 100644
--- a/mlir/test/Target/LLVMIR/openmp-todo.mlir
+++ b/mlir/test/Target/LLVMIR/openmp-todo.mlir
@@ -238,17 +238,6 @@ llvm.func @target_in_reduction(%x : !llvm.ptr) {
// -----
-llvm.func @target_is_device_ptr(%x : !llvm.ptr) {
- // expected-error at below {{not yet implemented: Unhandled clause is_device_ptr in omp.target operation}}
- // expected-error at below {{LLVM Translation failed for operation: omp.target}}
- omp.target is_device_ptr(%x : !llvm.ptr) {
- omp.terminator
- }
- llvm.return
-}
-
-// -----
-
llvm.func @target_enter_data_depend(%x: !llvm.ptr) {
// expected-error at below {{not yet implemented: Unhandled clause depend in omp.target_enter_data operation}}
// expected-error at below {{LLVM Translation failed for operation: omp.target_enter_data}}
diff --git a/offload/test/offloading/fortran/target-is-device-ptr.f90 b/offload/test/offloading/fortran/target-is-device-ptr.f90
new file mode 100644
index 0000000000000..6010d59113498
--- /dev/null
+++ b/offload/test/offloading/fortran/target-is-device-ptr.f90
@@ -0,0 +1,79 @@
+! Validate that a device pointer allocated via OpenMP runtime APIs can be
+! consumed by a TARGET region using the is_device_ptr clause.
+! REQUIRES: flang, amdgcn-amd-amdhsa
+! UNSUPPORTED: nvptx64-nvidia-cuda
+! UNSUPPORTED: nvptx64-nvidia-cuda-LTO
+! UNSUPPORTED: aarch64-unknown-linux-gnu
+! UNSUPPORTED: aarch64-unknown-linux-gnu-LTO
+! UNSUPPORTED: x86_64-unknown-linux-gnu
+! UNSUPPORTED: x86_64-unknown-linux-gnu-LTO
+
+! RUN: %libomptarget-compile-fortran-run-and-check-generic
+
+program is_device_ptr_target
+ use omp_lib
+ use iso_c_binding
+ implicit none
+
+ integer, parameter :: n = 4
+ integer, target :: host(n)
+ type(c_ptr) :: device_ptr
+ integer(c_int) :: rc
+ integer :: i
+
+ do i = 1, n
+ host(i) = i
+ end do
+
+ device_ptr = omp_target_alloc(int(n, c_size_t) * int(c_sizeof(host(1)), c_size_t), &
+ omp_get_default_device())
+ if (.not. c_associated(device_ptr)) then
+ print *, "device alloc failed"
+ stop 1
+ end if
+
+ rc = omp_target_memcpy(device_ptr, c_loc(host), &
+ int(n, c_size_t) * int(c_sizeof(host(1)), c_size_t), &
+ 0_c_size_t, 0_c_size_t, &
+ omp_get_default_device(), omp_get_initial_device())
+ if (rc .ne. 0) then
+ print *, "host->device memcpy failed"
+ call omp_target_free(device_ptr, omp_get_default_device())
+ stop 1
+ end if
+
+ call fill_on_device(device_ptr)
+
+ rc = omp_target_memcpy(c_loc(host), device_ptr, &
+ int(n, c_size_t) * int(c_sizeof(host(1)), c_size_t), &
+ 0_c_size_t, 0_c_size_t, &
+ omp_get_initial_device(), omp_get_default_device())
+ call omp_target_free(device_ptr, omp_get_default_device())
+
+ if (rc .ne. 0) then
+ print *, "device->host memcpy failed"
+ stop 1
+ end if
+
+ if (all(host == [2, 4, 6, 8])) then
+ print *, "PASS"
+ else
+ print *, "FAIL", host
+ end if
+
+contains
+ subroutine fill_on_device(ptr)
+ type(c_ptr) :: ptr
+ integer, pointer :: p(:)
+ call c_f_pointer(ptr, p, [n])
+
+ !$omp target is_device_ptr(ptr)
+ p(1) = 2
+ p(2) = 4
+ p(3) = 6
+ p(4) = 8
+ !$omp end target
+ end subroutine fill_on_device
+end program is_device_ptr_target
+
+!CHECK: PASS
>From a88f1712d23a53f10dc6eb553efdd7a8b55ea794 Mon Sep 17 00:00:00 2001
From: Akash Banerjee <Akash.Banerjee at amd.com>
Date: Mon, 24 Nov 2025 17:09:45 +0000
Subject: [PATCH 2/4] Fix clang-format.
---
.../LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp | 5 ++---
1 file changed, 2 insertions(+), 3 deletions(-)
diff --git a/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp b/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
index 9fa0e659320d6..00c4c351caa4e 100644
--- a/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
+++ b/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
@@ -3993,9 +3993,8 @@ static void collectMapDataFromMapOperands(
llvm::Value *origValue = moduleTranslation.lookupValue(offloadPtr);
auto mapType = convertClauseMapFlags(mapOp.getMapType());
auto mapTypeAlways = llvm::omp::OpenMPOffloadMappingFlags::OMP_MAP_ALWAYS;
- bool isDevicePtr =
- (mapOp.getMapType() & omp::ClauseMapFlags::storage) ==
- omp::ClauseMapFlags::storage;
+ bool isDevicePtr = (mapOp.getMapType() & omp::ClauseMapFlags::storage) ==
+ omp::ClauseMapFlags::storage;
mapData.OriginalValue.push_back(origValue);
mapData.BasePointers.push_back(origValue);
>From f4ff68bf9f61d0da012ea4a85682ba91d45d0777 Mon Sep 17 00:00:00 2001
From: Akash Banerjee <Akash.Banerjee at amd.com>
Date: Tue, 25 Nov 2025 16:35:56 +0000
Subject: [PATCH 3/4] Add is_device_ptr to mlir map flag.
---
flang/lib/Lower/OpenMP/ClauseProcessor.cpp | 3 ++-
mlir/include/mlir/Dialect/OpenMP/OpenMPEnums.td | 4 +++-
.../Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp | 11 +++++++----
3 files changed, 12 insertions(+), 6 deletions(-)
diff --git a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
index 81a47e20b2a88..d8d9b51b23804 100644
--- a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
+++ b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
@@ -1146,7 +1146,8 @@ bool ClauseProcessor::processIsDevicePtr(
const parser::CharBlock &source) {
mlir::Location location = converter.genLocation(source);
mlir::omp::ClauseMapFlags mapTypeBits =
- mlir::omp::ClauseMapFlags::storage;
+ mlir::omp::ClauseMapFlags::return_param |
+ mlir::omp::ClauseMapFlags::is_device_ptr;
processMapObjects(stmtCtx, location, clause.v, mapTypeBits,
parentMemberIndices, result.isDevicePtrVars,
isDeviceSyms);
diff --git a/mlir/include/mlir/Dialect/OpenMP/OpenMPEnums.td b/mlir/include/mlir/Dialect/OpenMP/OpenMPEnums.td
index d9882cbcb5977..8a7680bf6385e 100644
--- a/mlir/include/mlir/Dialect/OpenMP/OpenMPEnums.td
+++ b/mlir/include/mlir/Dialect/OpenMP/OpenMPEnums.td
@@ -126,6 +126,7 @@ def ClauseMapFlagsAttachAuto : I32BitEnumAttrCaseBit<"attach_auto", 15>;
def ClauseMapFlagsRefPtr : I32BitEnumAttrCaseBit<"ref_ptr", 16>;
def ClauseMapFlagsRefPtee : I32BitEnumAttrCaseBit<"ref_ptee", 17>;
def ClauseMapFlagsRefPtrPtee : I32BitEnumAttrCaseBit<"ref_ptr_ptee", 18>;
+def ClauseMapFlagsIsDevicePtr : I32BitEnumAttrCaseBit<"is_device_ptr", 19>;
def ClauseMapFlags : OpenMP_BitEnumAttr<
"ClauseMapFlags",
@@ -149,7 +150,8 @@ def ClauseMapFlags : OpenMP_BitEnumAttr<
ClauseMapFlagsAttachAuto,
ClauseMapFlagsRefPtr,
ClauseMapFlagsRefPtee,
- ClauseMapFlagsRefPtrPtee
+ ClauseMapFlagsRefPtrPtee,
+ ClauseMapFlagsIsDevicePtr
]>;
def ClauseMapFlagsAttr : OpenMP_EnumAttr<ClauseMapFlags,
diff --git a/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp b/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
index 00c4c351caa4e..1519fee6ea302 100644
--- a/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
+++ b/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
@@ -332,7 +332,6 @@ static LogicalResult checkImplementationStatus(Operation &op) {
op.getInReductionSyms())
result = todo("in_reduction");
};
- auto checkIsDevicePtr = [](auto, LogicalResult &) {};
auto checkLinear = [&todo](auto op, LogicalResult &result) {
if (!op.getLinearVars().empty() || !op.getLinearStepVars().empty())
result = todo("linear");
@@ -441,7 +440,6 @@ static LogicalResult checkImplementationStatus(Operation &op) {
checkBare(op, result);
checkDevice(op, result);
checkInReduction(op, result);
- checkIsDevicePtr(op, result);
})
.Default([](Operation &) {
// Assume all clauses for an operation can be translated unless they are
@@ -3872,6 +3870,10 @@ convertClauseMapFlags(omp::ClauseMapFlags mlirFlags) {
if (mapTypeToBool(omp::ClauseMapFlags::attach))
mapType |= llvm::omp::OpenMPOffloadMappingFlags::OMP_MAP_ATTACH;
+ // No dedicated LLVM runtime flag for is_device_ptr; handled separately.
+ if (mapTypeToBool(omp::ClauseMapFlags::is_device_ptr))
+ mapType |= llvm::omp::OpenMPOffloadMappingFlags::OMP_MAP_RETURN_PARAM;
+
return mapType;
}
@@ -3993,8 +3995,9 @@ static void collectMapDataFromMapOperands(
llvm::Value *origValue = moduleTranslation.lookupValue(offloadPtr);
auto mapType = convertClauseMapFlags(mapOp.getMapType());
auto mapTypeAlways = llvm::omp::OpenMPOffloadMappingFlags::OMP_MAP_ALWAYS;
- bool isDevicePtr = (mapOp.getMapType() & omp::ClauseMapFlags::storage) ==
- omp::ClauseMapFlags::storage;
+ bool isDevicePtr =
+ (mapOp.getMapType() & omp::ClauseMapFlags::is_device_ptr) !=
+ omp::ClauseMapFlags::none;
mapData.OriginalValue.push_back(origValue);
mapData.BasePointers.push_back(origValue);
>From ec3c5626ff4ae6a9481cc7f771f72a784fa59336 Mon Sep 17 00:00:00 2001
From: Akash Banerjee <Akash.Banerjee at amd.com>
Date: Wed, 26 Nov 2025 16:54:20 +0000
Subject: [PATCH 4/4] Update tests.
---
.../OpenMP/map-types-and-sizes.f90 | 2 +-
.../fortran/target-is-device-ptr.f90 | 87 ++++++-------------
2 files changed, 28 insertions(+), 61 deletions(-)
diff --git a/flang/test/Integration/OpenMP/map-types-and-sizes.f90 b/flang/test/Integration/OpenMP/map-types-and-sizes.f90
index 85434460bbea6..ddb943c95fba2 100644
--- a/flang/test/Integration/OpenMP/map-types-and-sizes.f90
+++ b/flang/test/Integration/OpenMP/map-types-and-sizes.f90
@@ -34,7 +34,7 @@ subroutine mapType_array
end subroutine mapType_array
!CHECK: @.offload_sizes{{.*}} = private unnamed_addr constant [1 x i64] [i64 8]
-!CHECK: @.offload_maptypes{{.*}} = private unnamed_addr constant [1 x i64] [i64 288]
+!CHECK: @.offload_maptypes{{.*}} = private unnamed_addr constant [1 x i64] [i64 352]
subroutine mapType_is_device_ptr
use iso_c_binding, only : c_ptr
type(c_ptr) :: p
diff --git a/offload/test/offloading/fortran/target-is-device-ptr.f90 b/offload/test/offloading/fortran/target-is-device-ptr.f90
index 6010d59113498..b2da1ad23d093 100644
--- a/offload/test/offloading/fortran/target-is-device-ptr.f90
+++ b/offload/test/offloading/fortran/target-is-device-ptr.f90
@@ -1,79 +1,46 @@
-! Validate that a device pointer allocated via OpenMP runtime APIs can be
-! consumed by a TARGET region using the is_device_ptr clause.
+! Validate that a device pointer obtained via omp_get_mapped_ptr can be used
+! inside a TARGET region with the is_device_ptr clause.
! REQUIRES: flang, amdgcn-amd-amdhsa
-! UNSUPPORTED: nvptx64-nvidia-cuda
-! UNSUPPORTED: nvptx64-nvidia-cuda-LTO
-! UNSUPPORTED: aarch64-unknown-linux-gnu
-! UNSUPPORTED: aarch64-unknown-linux-gnu-LTO
-! UNSUPPORTED: x86_64-unknown-linux-gnu
-! UNSUPPORTED: x86_64-unknown-linux-gnu-LTO
! RUN: %libomptarget-compile-fortran-run-and-check-generic
program is_device_ptr_target
- use omp_lib
- use iso_c_binding
+ use iso_c_binding, only : c_ptr, c_loc
implicit none
- integer, parameter :: n = 4
- integer, target :: host(n)
- type(c_ptr) :: device_ptr
- integer(c_int) :: rc
- integer :: i
-
- do i = 1, n
- host(i) = i
- end do
+ interface
+ function omp_get_mapped_ptr(host_ptr, device_num) &
+ bind(C, name="omp_get_mapped_ptr")
+ use iso_c_binding, only : c_ptr, c_int
+ type(c_ptr) :: omp_get_mapped_ptr
+ type(c_ptr), value :: host_ptr
+ integer(c_int), value :: device_num
+ end function omp_get_mapped_ptr
+ end interface
- device_ptr = omp_target_alloc(int(n, c_size_t) * int(c_sizeof(host(1)), c_size_t), &
- omp_get_default_device())
- if (.not. c_associated(device_ptr)) then
- print *, "device alloc failed"
- stop 1
- end if
-
- rc = omp_target_memcpy(device_ptr, c_loc(host), &
- int(n, c_size_t) * int(c_sizeof(host(1)), c_size_t), &
- 0_c_size_t, 0_c_size_t, &
- omp_get_default_device(), omp_get_initial_device())
- if (rc .ne. 0) then
- print *, "host->device memcpy failed"
- call omp_target_free(device_ptr, omp_get_default_device())
- stop 1
- end if
+ integer, parameter :: n = 4
+ integer, parameter :: dev = 0
+ integer, target :: a(n)
+ type(c_ptr) :: dptr
+ integer :: flag
- call fill_on_device(device_ptr)
+ a = [2, 4, 6, 8]
+ flag = 0
- rc = omp_target_memcpy(c_loc(host), device_ptr, &
- int(n, c_size_t) * int(c_sizeof(host(1)), c_size_t), &
- 0_c_size_t, 0_c_size_t, &
- omp_get_initial_device(), omp_get_default_device())
- call omp_target_free(device_ptr, omp_get_default_device())
+ !$omp target data map(tofrom: a, flag)
+ dptr = omp_get_mapped_ptr(c_loc(a), dev)
- if (rc .ne. 0) then
- print *, "device->host memcpy failed"
- stop 1
- end if
+ !$omp target is_device_ptr(dptr) map(tofrom: flag)
+ flag = flag + 1
+ !$omp end target
+ !$omp end target data
- if (all(host == [2, 4, 6, 8])) then
+ if (flag .eq. 1 .and. all(a == [2, 4, 6, 8])) then
print *, "PASS"
else
- print *, "FAIL", host
+ print *, "FAIL", a
end if
-contains
- subroutine fill_on_device(ptr)
- type(c_ptr) :: ptr
- integer, pointer :: p(:)
- call c_f_pointer(ptr, p, [n])
-
- !$omp target is_device_ptr(ptr)
- p(1) = 2
- p(2) = 4
- p(3) = 6
- p(4) = 8
- !$omp end target
- end subroutine fill_on_device
end program is_device_ptr_target
!CHECK: PASS
More information about the llvm-branch-commits
mailing list