Skip to content

Passing descriptors by reference to CUDA runtime calls #114288

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Oct 30, 2024
Merged

Passing descriptors by reference to CUDA runtime calls #114288

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 4 additions & 5 deletions flang/include/flang/Runtime/CUDA/memory.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -36,19 +36,18 @@ void RTDECL(CUFDataTransferPtrPtr)(void *dst, void *src, std::size_t bytes,
unsigned mode, const char *sourceFile = nullptr, int sourceLine = 0);

/// Data transfer from a pointer to a descriptor.
void RTDECL(CUFDataTransferDescPtr)(const Descriptor &dst, void *src,
void RTDECL(CUFDataTransferDescPtr)(Descriptor *dst, void *src,
std::size_t bytes, unsigned mode, const char *sourceFile = nullptr,
int sourceLine = 0);

/// Data transfer from a descriptor to a pointer.
void RTDECL(CUFDataTransferPtrDesc)(void *dst, const Descriptor &src,
void RTDECL(CUFDataTransferPtrDesc)(void *dst, Descriptor *src,
std::size_t bytes, unsigned mode, const char *sourceFile = nullptr,
int sourceLine = 0);

/// Data transfer from a descriptor to a descriptor.
void RTDECL(CUFDataTransferDescDesc)(const Descriptor &dst,
const Descriptor &src, unsigned mode, const char *sourceFile = nullptr,
int sourceLine = 0);
void RTDECL(CUFDataTransferDescDesc)(Descriptor *dst, Descriptor *src,
unsigned mode, const char *sourceFile = nullptr, int sourceLine = 0);

} // extern "C"
} // namespace Fortran::runtime::cuda
Expand Down
11 changes: 4 additions & 7 deletions flang/lib/Optimizer/Transforms/CUFOpConversion.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -529,8 +529,8 @@ struct CUFDataTransferOpConversion
mlir::Value sourceFile = fir::factory::locationToFilename(builder, loc);
mlir::Value sourceLine =
fir::factory::locationToLineNo(builder, loc, fTy.getInput(4));
mlir::Value dst = builder.loadIfRef(loc, op.getDst());
mlir::Value src = builder.loadIfRef(loc, op.getSrc());
mlir::Value dst = op.getDst();
mlir::Value src = op.getSrc();
llvm::SmallVector<mlir::Value> args{fir::runtime::createArguments(
builder, loc, fTy, dst, src, modeValue, sourceFile, sourceLine)};
builder.create<fir::CallOp>(loc, func, args);
Expand Down Expand Up @@ -603,11 +603,8 @@ struct CUFDataTransferOpConversion
mlir::Value sourceFile = fir::factory::locationToFilename(builder, loc);
mlir::Value sourceLine =
fir::factory::locationToLineNo(builder, loc, fTy.getInput(5));
mlir::Value dst =
dstIsDesc ? builder.loadIfRef(loc, op.getDst()) : op.getDst();
mlir::Value src = mlir::isa<fir::BaseBoxType>(srcTy)
? builder.loadIfRef(loc, op.getSrc())
: op.getSrc();
mlir::Value dst = op.getDst();
mlir::Value src = op.getSrc();
llvm::SmallVector<mlir::Value> args{
fir::runtime::createArguments(builder, loc, fTy, dst, src, bytes,
modeValue, sourceFile, sourceLine)};
Expand Down
9 changes: 4 additions & 5 deletions flang/runtime/CUDA/memory.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -73,23 +73,22 @@ void RTDEF(CUFDataTransferPtrPtr)(void *dst, void *src, std::size_t bytes,
CUDA_REPORT_IF_ERROR(cudaMemcpy(dst, src, bytes, kind));
}

void RTDEF(CUFDataTransferDescPtr)(const Descriptor &desc, void *addr,
void RTDEF(CUFDataTransferDescPtr)(Descriptor *desc, void *addr,
std::size_t bytes, unsigned mode, const char *sourceFile, int sourceLine) {
Terminator terminator{sourceFile, sourceLine};
terminator.Crash(
"not yet implemented: CUDA data transfer from a pointer to a descriptor");
}

void RTDEF(CUFDataTransferPtrDesc)(void *addr, const Descriptor &desc,
void RTDEF(CUFDataTransferPtrDesc)(void *addr, Descriptor *desc,
std::size_t bytes, unsigned mode, const char *sourceFile, int sourceLine) {
Terminator terminator{sourceFile, sourceLine};
terminator.Crash(
"not yet implemented: CUDA data transfer from a descriptor to a pointer");
}

void RTDECL(CUFDataTransferDescDesc)(const Descriptor &dstDesc,
const Descriptor &srcDesc, unsigned mode, const char *sourceFile,
int sourceLine) {
void RTDECL(CUFDataTransferDescDesc)(Descriptor *dstDesc, Descriptor *srcDesc,
unsigned mode, const char *sourceFile, int sourceLine) {
Terminator terminator{sourceFile, sourceLine};
terminator.Crash(
"not yet implemented: CUDA data transfer between two descriptors");
Expand Down
28 changes: 11 additions & 17 deletions flang/test/Fir/CUDA/cuda-data-transfer.fir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,11 +15,9 @@ func.func @_QPsub1() {
// CHECK-LABEL: func.func @_QPsub1()
// CHECK: %[[ADEV:.*]]:2 = hlfir.declare %{{.*}} {data_attr = #cuf.cuda<device>, fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QFsub1Eadev"} : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>) -> (!fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>, !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>)
// CHECK: %[[AHOST:.*]]:2 = hlfir.declare %{{.*}} {fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QFsub1Eahost"} : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>) -> (!fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>, !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>)
// CHECK: %[[AHOST_LOAD:.*]] = fir.load %[[AHOST]]#0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
// CHECK: %[[ADEV_LOAD:.*]] = fir.load %[[ADEV]]#0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
// CHECK: %[[AHOST_BOX:.*]] = fir.convert %[[AHOST_LOAD]] : (!fir.box<!fir.heap<!fir.array<?xi32>>>) -> !fir.box<none>
// CHECK: %[[ADEV_BOX:.*]] = fir.convert %[[ADEV_LOAD]] : (!fir.box<!fir.heap<!fir.array<?xi32>>>) -> !fir.box<none>
// CHECK: fir.call @_FortranACUFDataTransferDescDesc(%[[AHOST_BOX]], %[[ADEV_BOX]], %c1{{.*}}, %{{.*}}, %{{.*}}) : (!fir.box<none>, !fir.box<none>, i32, !fir.ref<i8>, i32) -> none
// CHECK: %[[AHOST_BOX:.*]] = fir.convert %[[AHOST]]#0 : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>) -> !fir.ref<!fir.box<none>>
// CHECK: %[[ADEV_BOX:.*]] = fir.convert %[[ADEV]]#0 : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>) -> !fir.ref<!fir.box<none>>
// CHECK: fir.call @_FortranACUFDataTransferDescDesc(%[[AHOST_BOX]], %[[ADEV_BOX]], %c1{{.*}}, %{{.*}}, %{{.*}}) : (!fir.ref<!fir.box<none>>, !fir.ref<!fir.box<none>>, i32, !fir.ref<i8>, i32) -> none

func.func @_QPsub2() {
%0 = cuf.alloc !fir.box<!fir.heap<!fir.array<?xi32>>> {bindc_name = "adev", data_attr = #cuf.cuda<device>, uniq_name = "_QFsub2Eadev"} -> !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
Expand Down Expand Up @@ -76,19 +74,17 @@ func.func @_QPsub4() {
// CHECK: %[[NBELEM:.*]] = arith.constant 10 : index
// CHECK: %[[WIDTH:.*]] = arith.constant 4 : index
// CHECK: %[[BYTES:.*]] = arith.muli %[[NBELEM]], %[[WIDTH]] : index
// CHECK: %[[ADEV_LOAD:.*]] = fir.load %[[ADEV]]#0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
// CHECK: %[[ADEV_BOX:.*]] = fir.convert %[[ADEV_LOAD]] : (!fir.box<!fir.heap<!fir.array<?xi32>>>) -> !fir.box<none>
// CHECK: %[[ADEV_BOX:.*]] = fir.convert %[[ADEV]]#0 : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>) -> !fir.ref<!fir.box<none>>
// CHECK: %[[AHOST_PTR:.*]] = fir.convert %[[AHOST]]#0 : (!fir.ref<!fir.array<10xi32>>) -> !fir.llvm_ptr<i8>
// CHECK: %[[BYTES_CONV:.*]] = fir.convert %[[BYTES]] : (index) -> i64
// CHECK: fir.call @_FortranACUFDataTransferDescPtr(%[[ADEV_BOX]], %[[AHOST_PTR]], %[[BYTES_CONV]], %c0{{.*}}, %{{.*}}, %{{.*}}) : (!fir.box<none>, !fir.llvm_ptr<i8>, i64, i32, !fir.ref<i8>, i32) -> none
// CHECK: fir.call @_FortranACUFDataTransferDescPtr(%[[ADEV_BOX]], %[[AHOST_PTR]], %[[BYTES_CONV]], %c0{{.*}}, %{{.*}}, %{{.*}}) : (!fir.ref<!fir.box<none>>, !fir.llvm_ptr<i8>, i64, i32, !fir.ref<i8>, i32) -> none
// CHECK: %[[NBELEM:.*]] = arith.constant 10 : index
// CHECK: %[[WIDTH:.*]] = arith.constant 4 : index
// CHECK: %[[BYTES:.*]] = arith.muli %[[NBELEM]], %[[WIDTH]] : index
// CHECK: %[[ADEV_LOAD:.*]] = fir.load %[[ADEV]]#0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
// CHECK: %[[AHOST_PTR:.*]] = fir.convert %[[AHOST]]#0 : (!fir.ref<!fir.array<10xi32>>) -> !fir.llvm_ptr<i8>
// CHECK: %[[ADEV_BOX:.*]] = fir.convert %[[ADEV_LOAD]] : (!fir.box<!fir.heap<!fir.array<?xi32>>>) -> !fir.box<none>
// CHECK: %[[ADEV_BOX:.*]] = fir.convert %[[ADEV]]#0 : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>) -> !fir.ref<!fir.box<none>>
// CHECK: %[[BYTES_CONV:.*]] = fir.convert %[[BYTES]] : (index) -> i64
// CHECK: fir.call @_FortranACUFDataTransferPtrDesc(%[[AHOST_PTR]], %[[ADEV_BOX]], %[[BYTES_CONV]], %c1{{.*}}, %{{.*}}, %{{.*}}) : (!fir.llvm_ptr<i8>, !fir.box<none>, i64, i32, !fir.ref<i8>, i32) -> none
// CHECK: fir.call @_FortranACUFDataTransferPtrDesc(%[[AHOST_PTR]], %[[ADEV_BOX]], %[[BYTES_CONV]], %c1{{.*}}, %{{.*}}, %{{.*}}) : (!fir.llvm_ptr<i8>, !fir.ref<!fir.box<none>>, i64, i32, !fir.ref<i8>, i32) -> none

func.func @_QPsub5(%arg0: !fir.ref<i32> {fir.bindc_name = "n"}) {
%0 = fir.dummy_scope : !fir.dscope
Expand Down Expand Up @@ -122,19 +118,17 @@ func.func @_QPsub5(%arg0: !fir.ref<i32> {fir.bindc_name = "n"}) {
// CHECK: %[[NBELEM:.*]] = arith.muli %[[I1]], %[[I2]] : index
// CHECK: %[[WIDTH:.*]] = arith.constant 4 : index
// CHECK: %[[BYTES:.*]] = arith.muli %[[NBELEM]], %[[WIDTH]] : index
// CHECK: %[[ADEV_LOAD:.*]] = fir.load %[[ADEV]]#0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?x?xi32>>>>
// CHECK: %[[ADEV_BOX:.*]] = fir.convert %[[ADEV_LOAD]] : (!fir.box<!fir.heap<!fir.array<?x?xi32>>>) -> !fir.box<none>
// CHECK: %[[ADEV_BOX:.*]] = fir.convert %[[ADEV]]#0 : (!fir.ref<!fir.box<!fir.heap<!fir.array<?x?xi32>>>>) -> !fir.ref<!fir.box<none>>
// CHECK: %[[AHOST_PTR:.*]] = fir.convert %[[AHOST]]#1 : (!fir.ref<!fir.array<?x?xi32>>) -> !fir.llvm_ptr<i8>
// CHECK: %[[BYTES_CONV:.*]] = fir.convert %[[BYTES]] : (index) -> i64
// CHECK: fir.call @_FortranACUFDataTransferDescPtr(%[[ADEV_BOX]], %[[AHOST_PTR]], %[[BYTES_CONV]], %c0{{.*}}, %{{.*}}, %{{.*}}) : (!fir.box<none>, !fir.llvm_ptr<i8>, i64, i32, !fir.ref<i8>, i32) -> none
// CHECK: fir.call @_FortranACUFDataTransferDescPtr(%[[ADEV_BOX]], %[[AHOST_PTR]], %[[BYTES_CONV]], %c0{{.*}}, %{{.*}}, %{{.*}}) : (!fir.ref<!fir.box<none>>, !fir.llvm_ptr<i8>, i64, i32, !fir.ref<i8>, i32) -> none
// CHECK: %[[NBELEM:.*]] = arith.muli %[[I1]], %[[I2]] : index
// CHECK: %[[WIDTH:.*]] = arith.constant 4 : index
// CHECK: %[[BYTES:.*]] = arith.muli %[[NBELEM]], %[[WIDTH]] : index
// CHECK: %[[ADEV_LOAD:.*]] = fir.load %[[ADEV]]#0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?x?xi32>>>>
// CHECK: %[[AHOST_PTR:.*]] = fir.convert %[[AHOST]]#1 : (!fir.ref<!fir.array<?x?xi32>>) -> !fir.llvm_ptr<i8>
// CHECK: %[[ADEV_BOX:.*]] = fir.convert %[[ADEV_LOAD]] : (!fir.box<!fir.heap<!fir.array<?x?xi32>>>) -> !fir.box<none>
// CHECK: %[[ADEV_BOX:.*]] = fir.convert %[[ADEV]]#0 : (!fir.ref<!fir.box<!fir.heap<!fir.array<?x?xi32>>>>) -> !fir.ref<!fir.box<none>>
// CHECK: %[[BYTES_CONV:.*]] = fir.convert %[[BYTES]] : (index) -> i64
// CHECK: fir.call @_FortranACUFDataTransferPtrDesc(%[[AHOST_PTR]], %[[ADEV_BOX]], %[[BYTES_CONV]], %c1{{.*}}, %{{.*}}, %{{.*}}) : (!fir.llvm_ptr<i8>, !fir.box<none>, i64, i32, !fir.ref<i8>, i32) -> none
// CHECK: fir.call @_FortranACUFDataTransferPtrDesc(%[[AHOST_PTR]], %[[ADEV_BOX]], %[[BYTES_CONV]], %c1{{.*}}, %{{.*}}, %{{.*}}) : (!fir.llvm_ptr<i8>, !fir.ref<!fir.box<none>>, i64, i32, !fir.ref<i8>, i32) -> none

func.func @_QPsub6() {
%0 = cuf.alloc i32 {bindc_name = "idev", data_attr = #cuf.cuda<device>, uniq_name = "_QFsub6Eidev"} -> !fir.ref<i32>
Expand Down
Loading