CUDA event should only be recorded after NCCL group (#8219)
authorTeng Li <tengli@fb.com>
Wed, 23 Jan 2019 22:04:47 +0000 (14:04 -0800)
committerFacebook Github Bot <facebook-github-bot@users.noreply.github.com>
Wed, 23 Jan 2019 22:18:26 +0000 (14:18 -0800)
commitfc5b79cd1c6020c20640128e43bac43fd636121e
tree72cf266699bcedfc118d7673cf8d8f36d78f79a6
parent07a090247a4d82d1cde724409ef152d7e8ad99f6
CUDA event should only be recorded after NCCL group (#8219)

Summary:
Otherwise, it won't work if we sync on this event.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/8219

Reviewed By: pietern

Differential Revision: D13788657

Pulled By: teng-li

fbshipit-source-id: 8c96e9691ed2441d7a685fb7ae8fece906f58daf
torch/lib/THD/base/data_channels/DataChannelNccl.cpp