intel/llvm
1
0
Fork 0
mirror of https://github.com/intel/llvm.git synced 2026-01-21 04:14:03 +08:00
Code Issues Packages Projects Releases Wiki Activity

[flang][cuda] Remove error check from allocation and free call (#165022)

Browse Source 
As in https://github.com/llvm/llvm-project/pull/164463, do not do error
checking in the runtime itself but let error go through as user might
want to catch them for error recovery.
This commit is contained in:
Valentin Clement (バレンタイン クレメン)
2025-10-24 07:55:34 -10:00
committed by GitHub
parent a1dc546f7f
commit c576c6b41f
1 changed files with 9 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
flang-rt/lib/cuda/allocator.cpp
View File

@@ -138,23 +138,21 @@ void RTDEF(CUFRegisterAllocator)() {
void *CUFAllocPinned(
std::size_t sizeInBytes, [[maybe_unused]] std::int64_t *asyncObject) {
void *p;
CUDA_REPORT_IF_ERROR(cudaMallocHost((void **)&p, sizeInBytes));
cudaMallocHost((void **)&p, sizeInBytes);
return p;
}
void CUFFreePinned(void *p) { CUDA_REPORT_IF_ERROR(cudaFreeHost(p)); }
void CUFFreePinned(void *p) { cudaFreeHost(p); }
void *CUFAllocDevice(std::size_t sizeInBytes, std::int64_t *asyncObject) {
void *p;
if (Fortran::runtime::executionEnvironment.cudaDeviceIsManaged) {
CUDA_REPORT_IF_ERROR(
cudaMallocManaged((void **)&p, sizeInBytes, cudaMemAttachGlobal));
cudaMallocManaged((void **)&p, sizeInBytes, cudaMemAttachGlobal);
} else {
if (asyncObject == nullptr) {
CUDA_REPORT_IF_ERROR(cudaMalloc(&p, sizeInBytes));
cudaMalloc(&p, sizeInBytes);
} else {
CUDA_REPORT_IF_ERROR(
cudaMallocAsync(&p, sizeInBytes, (cudaStream_t)*asyncObject));
cudaMallocAsync(&p, sizeInBytes, (cudaStream_t)*asyncObject);
insertAllocation(p, sizeInBytes, (cudaStream_t)*asyncObject);
}
}
@@ -167,21 +165,20 @@ void CUFFreeDevice(void *p) {
if (pos >= 0) {
cudaStream_t stream = deviceAllocations[pos].stream;
eraseAllocation(pos);
CUDA_REPORT_IF_ERROR(cudaFreeAsync(p, stream));
cudaFreeAsync(p, stream);
} else {
CUDA_REPORT_IF_ERROR(cudaFree(p));
cudaFree(p);
}
}
void *CUFAllocManaged(
std::size_t sizeInBytes, [[maybe_unused]] std::int64_t *asyncObject) {
void *p;
CUDA_REPORT_IF_ERROR(
cudaMallocManaged((void **)&p, sizeInBytes, cudaMemAttachGlobal));
cudaMallocManaged((void **)&p, sizeInBytes, cudaMemAttachGlobal);
return reinterpret_cast<void *>(p);
}
void CUFFreeManaged(void *p) { CUDA_REPORT_IF_ERROR(cudaFree(p)); }
void CUFFreeManaged(void *p) { cudaFree(p); }
void *CUFAllocUnified(
std::size_t sizeInBytes, [[maybe_unused]] std::int64_t *asyncObject) {