[flang][cuda] Generate cuf.allocate for descriptor with CUDA components #152041
+39
−5
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
llvmbot
added
flang
|
@llvm/pr-subscribers-flang-fir-hlfir Author: Valentin Clement (バレンタイン クレメン) (clementval) ChangesThe descriptor for derived-type with CUDA components are allocated in managed memory. The lowering was calling the standard runtime on allocate statement where it should be a Full diff: https://github.com/llvm/llvm-project/pull/152041.diff 5 Files Affected:
diff --git a/flang/include/flang/Semantics/tools.h b/flang/include/flang/Semantics/tools.h
index 317b9357b4c1f..966a30f7081fd 100644
--- a/flang/include/flang/Semantics/tools.h
+++ b/flang/include/flang/Semantics/tools.h
@@ -223,6 +223,8 @@ inline bool HasCUDAAttr(const Symbol &sym) {
return false;
}
+bool HasCUDAComponent(const Symbol &sym);
+
inline bool IsCUDAShared(const Symbol &sym) {
if (const auto *details{sym.GetUltimate().detailsIf<ObjectEntityDetails>()}) {
if (details->cudaDataAttr() &&
diff --git a/flang/lib/Lower/Allocatable.cpp b/flang/lib/Lower/Allocatable.cpp
index 15cd9770b35ba..92ac050c41ae2 100644
--- a/flang/lib/Lower/Allocatable.cpp
+++ b/flang/lib/Lower/Allocatable.cpp
@@ -466,7 +466,9 @@ class AllocateStmtHelper {
void genSimpleAllocation(const Allocation &alloc,
const fir::MutableBoxValue &box) {
- bool isCudaSymbol = Fortran::semantics::HasCUDAAttr(alloc.getSymbol());
+ bool isCudaAllocate =
+ Fortran::semantics::HasCUDAAttr(alloc.getSymbol()) ||
+ Fortran::semantics::HasCUDAComponent(alloc.getSymbol());
bool isCudaDeviceContext = cuf::isCUDADeviceContext(builder.getRegion());
bool inlineAllocation = !box.isDerived() && !errorManager.hasStatSpec() &&
!alloc.type.IsPolymorphic() &&
@@ -475,7 +477,7 @@ class AllocateStmtHelper {
unsigned allocatorIdx = Fortran::lower::getAllocatorIdx(alloc.getSymbol());
if (inlineAllocation &&
- ((isCudaSymbol && isCudaDeviceContext) || !isCudaSymbol)) {
+ ((isCudaAllocate && isCudaDeviceContext) || !isCudaAllocate)) {
// Pointers must use PointerAllocate so that their deallocations
// can be validated.
genInlinedAllocation(alloc, box);
@@ -494,7 +496,7 @@ class AllocateStmtHelper {
genSetDeferredLengthParameters(alloc, box);
genAllocateObjectBounds(alloc, box);
mlir::Value stat;
- if (!isCudaSymbol) {
+ if (!isCudaAllocate) {
stat = genRuntimeAllocate(builder, loc, box, errorManager);
setPinnedToFalse();
} else {
diff --git a/flang/lib/Lower/ConvertVariable.cpp b/flang/lib/Lower/ConvertVariable.cpp
index 647bd0d079985..b71f6519c39d9 100644
--- a/flang/lib/Lower/ConvertVariable.cpp
+++ b/flang/lib/Lower/ConvertVariable.cpp
@@ -814,8 +814,10 @@ initializeDeviceComponentAllocator(Fortran::lower::AbstractConverter &converter,
baseTy = boxTy.getEleTy();
baseTy = fir::unwrapRefType(baseTy);
- if (mlir::isa<fir::SequenceType>(baseTy))
- TODO(loc, "array of derived-type with device component");
+ if (mlir::isa<fir::SequenceType>(baseTy) &&
+ (fir::isAllocatableType(fir::getBase(exv).getType()) ||
+ fir::isPointerType(fir::getBase(exv).getType())))
+ return; // Allocator index need to be set after allocation.
auto recTy =
mlir::dyn_cast<fir::RecordType>(fir::unwrapSequenceType(baseTy));
diff --git a/flang/lib/Semantics/tools.cpp b/flang/lib/Semantics/tools.cpp
index 5a5b02e1ac3ce..913bf08cd0d99 100644
--- a/flang/lib/Semantics/tools.cpp
+++ b/flang/lib/Semantics/tools.cpp
@@ -1094,6 +1094,21 @@ bool IsDeviceAllocatable(const Symbol &symbol) {
return false;
}
+bool HasCUDAComponent(const Symbol &symbol) {
+ if (const auto *details{symbol.GetUltimate()
+ .detailsIf<Fortran::semantics::ObjectEntityDetails>()}) {
+ const Fortran::semantics::DeclTypeSpec *type{details->type()};
+ const Fortran::semantics::DerivedTypeSpec *derived{
+ type ? type->AsDerived() : nullptr};
+ if (derived) {
+ if (FindCUDADeviceAllocatableUltimateComponent(*derived)) {
+ return true;
+ }
+ }
+ }
+ return false;
+}
+
UltimateComponentIterator::const_iterator
FindCUDADeviceAllocatableUltimateComponent(const DerivedTypeSpec &derived) {
UltimateComponentIterator ultimates{derived};
diff --git a/flang/test/Lower/CUDA/cuda-allocatable.cuf b/flang/test/Lower/CUDA/cuda-allocatable.cuf
index 36e768bd7d92c..2cf8c7d336812 100644
--- a/flang/test/Lower/CUDA/cuda-allocatable.cuf
+++ b/flang/test/Lower/CUDA/cuda-allocatable.cuf
@@ -6,6 +6,10 @@ module globals
real, device, allocatable :: a_device(:)
real, managed, allocatable :: a_managed(:)
real, pinned, allocatable :: a_pinned(:)
+ type :: t1
+ integer :: a
+ real, dimension(:), allocatable, device :: b
+ end type
end module
! CHECK-LABEL: fir.global @_QMglobalsEa_device {data_attr = #cuf.cuda<device>} : !fir.box<!fir.heap<!fir.array<?xf32>>>
@@ -222,3 +226,12 @@ end
! CHECK: %[[FALSE:.*]] = arith.constant false
! CHECK: %[[FLASE_CONV:.*]] = fir.convert %[[FALSE]] : (i1) -> !fir.logical<4>
! CHECK: fir.store %[[FLASE_CONV]] to %[[PLOG_DECL]]#0 : !fir.ref<!fir.logical<4>>
+
+subroutine cuda_component()
+ use globals
+ type(t1), pointer, dimension(:) :: d
+ allocate(d(10))
+end subroutine
+
+! CHECK-LABEL: func.func @_QPcuda_component()
+! CHECK: cuf.allocate
|
|
@llvm/pr-subscribers-flang-semantics Author: Valentin Clement (バレンタイン クレメン) (clementval) ChangesThe descriptor for derived-type with CUDA components are allocated in managed memory. The lowering was calling the standard runtime on allocate statement where it should be a Full diff: https://github.com/llvm/llvm-project/pull/152041.diff 5 Files Affected:
diff --git a/flang/include/flang/Semantics/tools.h b/flang/include/flang/Semantics/tools.h
index 317b9357b4c1f..966a30f7081fd 100644
--- a/flang/include/flang/Semantics/tools.h
+++ b/flang/include/flang/Semantics/tools.h
@@ -223,6 +223,8 @@ inline bool HasCUDAAttr(const Symbol &sym) {
return false;
}
+bool HasCUDAComponent(const Symbol &sym);
+
inline bool IsCUDAShared(const Symbol &sym) {
if (const auto *details{sym.GetUltimate().detailsIf<ObjectEntityDetails>()}) {
if (details->cudaDataAttr() &&
diff --git a/flang/lib/Lower/Allocatable.cpp b/flang/lib/Lower/Allocatable.cpp
index 15cd9770b35ba..92ac050c41ae2 100644
--- a/flang/lib/Lower/Allocatable.cpp
+++ b/flang/lib/Lower/Allocatable.cpp
@@ -466,7 +466,9 @@ class AllocateStmtHelper {
void genSimpleAllocation(const Allocation &alloc,
const fir::MutableBoxValue &box) {
- bool isCudaSymbol = Fortran::semantics::HasCUDAAttr(alloc.getSymbol());
+ bool isCudaAllocate =
+ Fortran::semantics::HasCUDAAttr(alloc.getSymbol()) ||
+ Fortran::semantics::HasCUDAComponent(alloc.getSymbol());
bool isCudaDeviceContext = cuf::isCUDADeviceContext(builder.getRegion());
bool inlineAllocation = !box.isDerived() && !errorManager.hasStatSpec() &&
!alloc.type.IsPolymorphic() &&
@@ -475,7 +477,7 @@ class AllocateStmtHelper {
unsigned allocatorIdx = Fortran::lower::getAllocatorIdx(alloc.getSymbol());
if (inlineAllocation &&
- ((isCudaSymbol && isCudaDeviceContext) || !isCudaSymbol)) {
+ ((isCudaAllocate && isCudaDeviceContext) || !isCudaAllocate)) {
// Pointers must use PointerAllocate so that their deallocations
// can be validated.
genInlinedAllocation(alloc, box);
@@ -494,7 +496,7 @@ class AllocateStmtHelper {
genSetDeferredLengthParameters(alloc, box);
genAllocateObjectBounds(alloc, box);
mlir::Value stat;
- if (!isCudaSymbol) {
+ if (!isCudaAllocate) {
stat = genRuntimeAllocate(builder, loc, box, errorManager);
setPinnedToFalse();
} else {
diff --git a/flang/lib/Lower/ConvertVariable.cpp b/flang/lib/Lower/ConvertVariable.cpp
index 647bd0d079985..b71f6519c39d9 100644
--- a/flang/lib/Lower/ConvertVariable.cpp
+++ b/flang/lib/Lower/ConvertVariable.cpp
@@ -814,8 +814,10 @@ initializeDeviceComponentAllocator(Fortran::lower::AbstractConverter &converter,
baseTy = boxTy.getEleTy();
baseTy = fir::unwrapRefType(baseTy);
- if (mlir::isa<fir::SequenceType>(baseTy))
- TODO(loc, "array of derived-type with device component");
+ if (mlir::isa<fir::SequenceType>(baseTy) &&
+ (fir::isAllocatableType(fir::getBase(exv).getType()) ||
+ fir::isPointerType(fir::getBase(exv).getType())))
+ return; // Allocator index need to be set after allocation.
auto recTy =
mlir::dyn_cast<fir::RecordType>(fir::unwrapSequenceType(baseTy));
diff --git a/flang/lib/Semantics/tools.cpp b/flang/lib/Semantics/tools.cpp
index 5a5b02e1ac3ce..913bf08cd0d99 100644
--- a/flang/lib/Semantics/tools.cpp
+++ b/flang/lib/Semantics/tools.cpp
@@ -1094,6 +1094,21 @@ bool IsDeviceAllocatable(const Symbol &symbol) {
return false;
}
+bool HasCUDAComponent(const Symbol &symbol) {
+ if (const auto *details{symbol.GetUltimate()
+ .detailsIf<Fortran::semantics::ObjectEntityDetails>()}) {
+ const Fortran::semantics::DeclTypeSpec *type{details->type()};
+ const Fortran::semantics::DerivedTypeSpec *derived{
+ type ? type->AsDerived() : nullptr};
+ if (derived) {
+ if (FindCUDADeviceAllocatableUltimateComponent(*derived)) {
+ return true;
+ }
+ }
+ }
+ return false;
+}
+
UltimateComponentIterator::const_iterator
FindCUDADeviceAllocatableUltimateComponent(const DerivedTypeSpec &derived) {
UltimateComponentIterator ultimates{derived};
diff --git a/flang/test/Lower/CUDA/cuda-allocatable.cuf b/flang/test/Lower/CUDA/cuda-allocatable.cuf
index 36e768bd7d92c..2cf8c7d336812 100644
--- a/flang/test/Lower/CUDA/cuda-allocatable.cuf
+++ b/flang/test/Lower/CUDA/cuda-allocatable.cuf
@@ -6,6 +6,10 @@ module globals
real, device, allocatable :: a_device(:)
real, managed, allocatable :: a_managed(:)
real, pinned, allocatable :: a_pinned(:)
+ type :: t1
+ integer :: a
+ real, dimension(:), allocatable, device :: b
+ end type
end module
! CHECK-LABEL: fir.global @_QMglobalsEa_device {data_attr = #cuf.cuda<device>} : !fir.box<!fir.heap<!fir.array<?xf32>>>
@@ -222,3 +226,12 @@ end
! CHECK: %[[FALSE:.*]] = arith.constant false
! CHECK: %[[FLASE_CONV:.*]] = fir.convert %[[FALSE]] : (i1) -> !fir.logical<4>
! CHECK: fir.store %[[FLASE_CONV]] to %[[PLOG_DECL]]#0 : !fir.ref<!fir.logical<4>>
+
+subroutine cuda_component()
+ use globals
+ type(t1), pointer, dimension(:) :: d
+ allocate(d(10))
+end subroutine
+
+! CHECK-LABEL: func.func @_QPcuda_component()
+! CHECK: cuf.allocate
|
vzakhari
approved these changes
Aug 4, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The descriptor for derived-type with CUDA components are allocated in managed memory. The lowering was calling the standard runtime on allocate statement where it should be a
cuf.allocateoperation.