OSSForks/pytorch
1
0
Fork 0
mirror of https://github.com/zebrajr/pytorch.git synced 2025-12-06 00:20:18 +01:00
Code Issues Actions 38 Packages Projects Releases Wiki Activity

[c10d][ez] Add comments to CudaEventCache class (#134172)

Browse Source 
Pull Request resolved: https://github.com/pytorch/pytorch/pull/134172
Approved by: https://github.com/d4l3k, https://github.com/kwen2501
This commit is contained in:
fduwjj 2024-08-21 15:50:35 -07:00 committed by PyTorch MergeBot
parent b319fa3fd9
commit e7929809f3
1 changed files with 4 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp
View File

@ -763,6 +763,10 @@ void ProcessGroupNCCL::WorkNCCL::abort() {
ProcessGroupNCCL::CUDAEventCache::CUDAEventCache() {}
// CUDA event is used to record the start/end of one Work.
// Instead of let the CUDA event gets destroyed, we now reuse it after the Work
// has been erased from workMetaList_.
// This is to avoid the potential deadlock caused by CudaEventDestroy.
std::shared_ptr<at::cuda::CUDAEvent> ProcessGroupNCCL::CUDAEventCache::create(
bool timing) {
auto deleter = [this, timing](at::cuda::CUDAEvent* event) {