Skip to content

[flang][cuda] Defined allocator for unified data #102189

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
Aug 6, 2024
Merged

[flang][cuda] Defined allocator for unified data #102189

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
3 changes: 3 additions & 0 deletions flang/include/flang/Runtime/CUDA/allocator.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -36,5 +36,8 @@ void CUFFreeDevice(void *);
void *CUFAllocManaged(std::size_t);
void CUFFreeManaged(void *);

void *CUFAllocUnified(std::size_t);
void CUFFreeUnified(void *);

} // namespace Fortran::runtime::cuda
#endif // FORTRAN_RUNTIME_CUDA_ALLOCATOR_H_
3 changes: 2 additions & 1 deletion flang/include/flang/Runtime/allocator-registry.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,8 +19,9 @@ static constexpr unsigned kDefaultAllocator = 0;
static constexpr unsigned kPinnedAllocatorPos = 1;
static constexpr unsigned kDeviceAllocatorPos = 2;
static constexpr unsigned kManagedAllocatorPos = 3;
static constexpr unsigned kUnifiedAllocatorPos = 4;

#define MAX_ALLOCATOR 5
#define MAX_ALLOCATOR 7 // 3 bits are reserved in the descriptor.

namespace Fortran::runtime {

Expand Down
5 changes: 3 additions & 2 deletions flang/lib/Lower/ConvertVariable.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1860,9 +1860,10 @@ static unsigned getAllocatorIdx(const Fortran::semantics::Symbol &sym) {
return kPinnedAllocatorPos;
if (*cudaAttr == Fortran::common::CUDADataAttr::Device)
return kDeviceAllocatorPos;
if (*cudaAttr == Fortran::common::CUDADataAttr::Managed ||
*cudaAttr == Fortran::common::CUDADataAttr::Unified)
if (*cudaAttr == Fortran::common::CUDADataAttr::Managed)
return kManagedAllocatorPos;
if (*cudaAttr == Fortran::common::CUDADataAttr::Unified)
return kUnifiedAllocatorPos;
}
return kDefaultAllocator;
}
Expand Down
12 changes: 12 additions & 0 deletions flang/runtime/CUDA/allocator.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,6 +26,8 @@ void CUFRegisterAllocator() {
kDeviceAllocatorPos, {&CUFAllocDevice, CUFFreeDevice});
allocatorRegistry.Register(
kManagedAllocatorPos, {&CUFAllocManaged, CUFFreeManaged});
allocatorRegistry.Register(
kUnifiedAllocatorPos, {&CUFAllocUnified, CUFFreeUnified});
}

void *CUFAllocPinned(std::size_t sizeInBytes) {
Expand Down Expand Up @@ -57,4 +59,14 @@ void CUFFreeManaged(void *p) {
CUDA_REPORT_IF_ERROR(cuMemFree(reinterpret_cast<CUdeviceptr>(p)));
}

void *CUFAllocUnified(std::size_t sizeInBytes) {
// Call alloc managed for the time being.
return CUFAllocManaged(sizeInBytes);
}

void CUFFreeUnified(void *p) {
// Call free managed for the time being.
CUFFreeManaged(p);
}

} // namespace Fortran::runtime::cuda
Loading