[flang][OpenMP] Map ByRef if size/alignment exceed that of a pointer #130832
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
Improve the check for whether a type can be passed by copy. Currently, passing by copy is done via the OMP_MAP_LITERAL mapping, which can only transfer as much data as can be contained in a pointer representation.
llvmbot
added
flang
|
@llvm/pr-subscribers-offload @llvm/pr-subscribers-flang-fir-hlfir Author: Krzysztof Parzyszek (kparzysz) ChangesImprove the check for whether a type can be passed by copy. Currently, passing by copy is done via the OMP_MAP_LITERAL mapping, which can only transfer as much data as can be contained in a pointer representation. Full diff: https://github.com/llvm/llvm-project/pull/130832.diff 2 Files Affected:
diff --git a/flang/lib/Lower/OpenMP/OpenMP.cpp b/flang/lib/Lower/OpenMP/OpenMP.cpp
index 77786742b3e13..651ed3055f203 100644
--- a/flang/lib/Lower/OpenMP/OpenMP.cpp
+++ b/flang/lib/Lower/OpenMP/OpenMP.cpp
@@ -2192,6 +2192,24 @@ genTargetOp(lower::AbstractConverter &converter, lower::SymMap &symTable,
/*useDelayedPrivatization=*/true, symTable);
dsp.processStep1(&clauseOps);
+ // Check if a value of type `type` be passed to the kernel by value.
+ // All kernel parameters are of pointer type, so if the value can be
+ // represented inside of a pointer, then it can be passed by value.
+ auto isLiteralType = [&](mlir::Type type) {
+ if (!fir::isa_trivial(type) && !fir::isa_char(type))
+ return false;
+
+ const mlir::DataLayout &dl = firOpBuilder.getDataLayout();
+ mlir::Type ptrTy =
+ mlir::LLVM::LLVMPointerType::get(&converter.getMLIRContext());
+ uint64_t ptrSize = dl.getTypeSize(ptrTy);
+ uint64_t ptrAlign = dl.getTypePreferredAlignment(ptrTy);
+
+ auto [size, align] = fir::getTypeSizeAndAlignmentOrCrash(
+ loc, type, dl, converter.getKindMap());
+ return size <= ptrSize && align <= ptrAlign;
+ };
+
// 5.8.1 Implicit Data-Mapping Attribute Rules
// The following code follows the implicit data-mapping rules to map all the
// symbols used inside the region that do not have explicit data-environment
@@ -2268,7 +2286,7 @@ genTargetOp(lower::AbstractConverter &converter, lower::SymMap &symTable,
mapFlag |= llvm::omp::OpenMPOffloadMappingFlags::OMP_MAP_TO;
mapFlag |= llvm::omp::OpenMPOffloadMappingFlags::OMP_MAP_FROM;
}
- } else if (fir::isa_trivial(eleType) || fir::isa_char(eleType)) {
+ } else if (isLiteralType(eleType)) {
captureKind = mlir::omp::VariableCaptureKind::ByCopy;
} else if (!fir::isa_builtin_cptr_type(eleType)) {
mapFlag |= llvm::omp::OpenMPOffloadMappingFlags::OMP_MAP_TO;
diff --git a/flang/test/Lower/OpenMP/target-map-complex.f90 b/flang/test/Lower/OpenMP/target-map-complex.f90
new file mode 100644
index 0000000000000..c15a10e4804dd
--- /dev/null
+++ b/flang/test/Lower/OpenMP/target-map-complex.f90
@@ -0,0 +1,33 @@
+!RUN: %flang_fc1 -emit-hlfir -fopenmp %s -o - | FileCheck %s
+
+! Check that the complex*4 is passed by value. but complex*8 is passed by
+! reference
+
+!CHECK-LABEL: func.func @_QMmPbar()
+!CHECK: %[[V0:[0-9]+]]:2 = hlfir.declare {{.*}} (!fir.ref<complex<f64>>) -> (!fir.ref<complex<f64>>, !fir.ref<complex<f64>>)
+!CHECK: %[[V1:[0-9]+]]:2 = hlfir.declare {{.*}} (!fir.ref<complex<f32>>) -> (!fir.ref<complex<f32>>, !fir.ref<complex<f32>>)
+!CHECK: %[[V2:[0-9]+]] = omp.map.info var_ptr(%[[V1]]#1 : !fir.ref<complex<f32>>, complex<f32>) {{.*}} capture(ByCopy)
+!CHECK: %[[V3:[0-9]+]] = omp.map.info var_ptr(%[[V0]]#1 : !fir.ref<complex<f64>>, complex<f64>) {{.*}} capture(ByRef)
+!CHECK: omp.target map_entries(%[[V2]] -> {{.*}}, %[[V3]] -> {{.*}} : !fir.ref<complex<f32>>, !fir.ref<complex<f64>>)
+
+module m
+ implicit none
+ complex(kind=4) :: cfval = (24, 25)
+ complex(kind=8) :: cdval = (28, 29)
+ interface
+ subroutine foo(x, y)
+ complex(kind=4) :: x
+ complex(kind=8) :: y
+ !$omp declare target
+ end
+ end interface
+
+contains
+
+subroutine bar()
+!$omp target
+ call foo(cfval, cdval)
+!$omp end target
+end
+
+end module
|
|
@llvm/pr-subscribers-flang-openmp Author: Krzysztof Parzyszek (kparzysz) ChangesImprove the check for whether a type can be passed by copy. Currently, passing by copy is done via the OMP_MAP_LITERAL mapping, which can only transfer as much data as can be contained in a pointer representation. Full diff: https://github.com/llvm/llvm-project/pull/130832.diff 2 Files Affected:
diff --git a/flang/lib/Lower/OpenMP/OpenMP.cpp b/flang/lib/Lower/OpenMP/OpenMP.cpp
index 77786742b3e13..651ed3055f203 100644
--- a/flang/lib/Lower/OpenMP/OpenMP.cpp
+++ b/flang/lib/Lower/OpenMP/OpenMP.cpp
@@ -2192,6 +2192,24 @@ genTargetOp(lower::AbstractConverter &converter, lower::SymMap &symTable,
/*useDelayedPrivatization=*/true, symTable);
dsp.processStep1(&clauseOps);
+ // Check if a value of type `type` be passed to the kernel by value.
+ // All kernel parameters are of pointer type, so if the value can be
+ // represented inside of a pointer, then it can be passed by value.
+ auto isLiteralType = [&](mlir::Type type) {
+ if (!fir::isa_trivial(type) && !fir::isa_char(type))
+ return false;
+
+ const mlir::DataLayout &dl = firOpBuilder.getDataLayout();
+ mlir::Type ptrTy =
+ mlir::LLVM::LLVMPointerType::get(&converter.getMLIRContext());
+ uint64_t ptrSize = dl.getTypeSize(ptrTy);
+ uint64_t ptrAlign = dl.getTypePreferredAlignment(ptrTy);
+
+ auto [size, align] = fir::getTypeSizeAndAlignmentOrCrash(
+ loc, type, dl, converter.getKindMap());
+ return size <= ptrSize && align <= ptrAlign;
+ };
+
// 5.8.1 Implicit Data-Mapping Attribute Rules
// The following code follows the implicit data-mapping rules to map all the
// symbols used inside the region that do not have explicit data-environment
@@ -2268,7 +2286,7 @@ genTargetOp(lower::AbstractConverter &converter, lower::SymMap &symTable,
mapFlag |= llvm::omp::OpenMPOffloadMappingFlags::OMP_MAP_TO;
mapFlag |= llvm::omp::OpenMPOffloadMappingFlags::OMP_MAP_FROM;
}
- } else if (fir::isa_trivial(eleType) || fir::isa_char(eleType)) {
+ } else if (isLiteralType(eleType)) {
captureKind = mlir::omp::VariableCaptureKind::ByCopy;
} else if (!fir::isa_builtin_cptr_type(eleType)) {
mapFlag |= llvm::omp::OpenMPOffloadMappingFlags::OMP_MAP_TO;
diff --git a/flang/test/Lower/OpenMP/target-map-complex.f90 b/flang/test/Lower/OpenMP/target-map-complex.f90
new file mode 100644
index 0000000000000..c15a10e4804dd
--- /dev/null
+++ b/flang/test/Lower/OpenMP/target-map-complex.f90
@@ -0,0 +1,33 @@
+!RUN: %flang_fc1 -emit-hlfir -fopenmp %s -o - | FileCheck %s
+
+! Check that the complex*4 is passed by value. but complex*8 is passed by
+! reference
+
+!CHECK-LABEL: func.func @_QMmPbar()
+!CHECK: %[[V0:[0-9]+]]:2 = hlfir.declare {{.*}} (!fir.ref<complex<f64>>) -> (!fir.ref<complex<f64>>, !fir.ref<complex<f64>>)
+!CHECK: %[[V1:[0-9]+]]:2 = hlfir.declare {{.*}} (!fir.ref<complex<f32>>) -> (!fir.ref<complex<f32>>, !fir.ref<complex<f32>>)
+!CHECK: %[[V2:[0-9]+]] = omp.map.info var_ptr(%[[V1]]#1 : !fir.ref<complex<f32>>, complex<f32>) {{.*}} capture(ByCopy)
+!CHECK: %[[V3:[0-9]+]] = omp.map.info var_ptr(%[[V0]]#1 : !fir.ref<complex<f64>>, complex<f64>) {{.*}} capture(ByRef)
+!CHECK: omp.target map_entries(%[[V2]] -> {{.*}}, %[[V3]] -> {{.*}} : !fir.ref<complex<f32>>, !fir.ref<complex<f64>>)
+
+module m
+ implicit none
+ complex(kind=4) :: cfval = (24, 25)
+ complex(kind=8) :: cdval = (28, 29)
+ interface
+ subroutine foo(x, y)
+ complex(kind=4) :: x
+ complex(kind=8) :: y
+ !$omp declare target
+ end
+ end interface
+
+contains
+
+subroutine bar()
+!$omp target
+ call foo(cfval, cdval)
+!$omp end target
+end
+
+end module
|
agozillon
approved these changes
Mar 12, 2025
There was a problem hiding this comment.
LGTM, might be worth having an offload test checking that we can't writeback to implicitly captured larger literal types, I highly doubt this change would allow that, but always good to be certain and helps to have another test for the future incase some other change does something weird :-)
Make sure they are not TOFROM.
frederik-h
pushed a commit
to frederik-h/llvm-project
that referenced
this pull request
Mar 18, 2025
…lvm#130832) Improve the check for whether a type can be passed by copy. Currently, passing by copy is done via the OMP_MAP_LITERAL mapping, which can only transfer as much data as can be contained in a pointer representation.
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.
Improve the check for whether a type can be passed by copy. Currently, passing by copy is done via the OMP_MAP_LITERAL mapping, which can only transfer as much data as can be contained in a pointer representation.