diff options
author | Edward Yang <ezyang@fb.com> | 2018-11-19 08:13:08 -0800 |
---|---|---|
committer | Facebook Github Bot <facebook-github-bot@users.noreply.github.com> | 2018-11-19 08:20:11 -0800 |
commit | ce85150cb4dd24b7a3d80f8be18ec350018da11e (patch) | |
tree | 32bcad0aa0dee370f93d55e9aa6a5149f664ef38 /torch/lib | |
parent | 48099c23b4c4e6168b0d6f8f706764c080a0a1a2 (diff) | |
download | pytorch-ce85150cb4dd24b7a3d80f8be18ec350018da11e.tar.gz pytorch-ce85150cb4dd24b7a3d80f8be18ec350018da11e.tar.bz2 pytorch-ce85150cb4dd24b7a3d80f8be18ec350018da11e.zip |
Split out CUDAMultiStreamGuard from CUDAGuard (#13912)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/13912
The implementation and API of CUDAMultiStreamGuard is less mature,
and it cannot be implemented generically (yet) in c10_cuda. This
might be a reasonable thing to do eventually, but not for now.
Reviewed By: smessmer
Differential Revision: D13046500
fbshipit-source-id: 4ea39ca1344f1ad5ae7c82c98617aa348c327848
Diffstat (limited to 'torch/lib')
-rw-r--r-- | torch/lib/c10d/test/ProcessGroupGlooAsyncTest.cpp | 1 | ||||
-rw-r--r-- | torch/lib/c10d/test/ProcessGroupNCCLTest.cpp | 1 |
2 files changed, 2 insertions, 0 deletions
diff --git a/torch/lib/c10d/test/ProcessGroupGlooAsyncTest.cpp b/torch/lib/c10d/test/ProcessGroupGlooAsyncTest.cpp index fcdd9aa7e8..69e2180d3a 100644 --- a/torch/lib/c10d/test/ProcessGroupGlooAsyncTest.cpp +++ b/torch/lib/c10d/test/ProcessGroupGlooAsyncTest.cpp @@ -1,6 +1,7 @@ #include <gloo/transport/tcp/device.h> #include <ATen/cuda/CUDAGuard.h> +#include <ATen/cuda/CUDAMultiStreamGuard.h> #include <c10d/FileStore.hpp> #include <c10d/ProcessGroupGloo.hpp> diff --git a/torch/lib/c10d/test/ProcessGroupNCCLTest.cpp b/torch/lib/c10d/test/ProcessGroupNCCLTest.cpp index 158eef19b5..78010531ad 100644 --- a/torch/lib/c10d/test/ProcessGroupNCCLTest.cpp +++ b/torch/lib/c10d/test/ProcessGroupNCCLTest.cpp @@ -6,6 +6,7 @@ #include <c10d/test/TestUtils.hpp> #include <ATen/cuda/CUDAGuard.h> +#include <ATen/cuda/CUDAMultiStreamGuard.h> #include <ATen/cuda/CUDAStream.h> using namespace c10d::test; |