caio/clang-p2996
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[flang][cuda] Use the aysncId in device allocation (#135099)

Browse Source 
Use `cudaMallocAsync` in the `CUFAllocDevice` allocator when asyncId is
provided.

More work is needed to be able to call `cudaFreeAsync` since the
allocated address and stream needs to be tracked.
This commit is contained in:
Valentin Clement (バレンタイン クレメン)
2025-04-09 17:34:48 -07:00
committed by GitHub
parent 6a63abce7b
commit 56b792322a
1 changed files with 8 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
flang-rt/lib/cuda/allocator.cpp
View File

@@ -9,6 +9,7 @@
#include "flang/Runtime/CUDA/allocator.h"
#include "flang-rt/runtime/allocator-registry.h"
#include "flang-rt/runtime/derived.h"
#include "flang-rt/runtime/descriptor.h"
#include "flang-rt/runtime/environment.h"
#include "flang-rt/runtime/stat.h"
#include "flang-rt/runtime/terminator.h"
@@ -43,14 +44,18 @@ void *CUFAllocPinned(
void CUFFreePinned(void *p) { CUDA_REPORT_IF_ERROR(cudaFreeHost(p)); }
void *CUFAllocDevice(
std::size_t sizeInBytes, [[maybe_unused]] std::int64_t asyncId) {
void *CUFAllocDevice(std::size_t sizeInBytes, std::int64_t asyncId) {
void *p;
if (Fortran::runtime::executionEnvironment.cudaDeviceIsManaged) {
CUDA_REPORT_IF_ERROR(
cudaMallocManaged((void **)&p, sizeInBytes, cudaMemAttachGlobal));
} else {
if (asyncId == kNoAsyncId) {
CUDA_REPORT_IF_ERROR(cudaMalloc(&p, sizeInBytes));
} else {
CUDA_REPORT_IF_ERROR(
cudaMallocAsync(&p, sizeInBytes, (cudaStream_t)asyncId));
}
}
return p;
}