-
Notifications
You must be signed in to change notification settings - Fork 5.6k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Support both use_calc_stream and sync_op in allgather API #46295
Support both use_calc_stream and sync_op in allgather API #46295
Conversation
你的PR提交成功,感谢你对开源项目的贡献! |
python/paddle/fluid/tests/unittests/collective/communication_stream_allgather_api_dygraph.py
Outdated
Show resolved
Hide resolved
python/paddle/fluid/tests/unittests/collective/communication_stream_allgather_api_dygraph.py
Outdated
Show resolved
Hide resolved
a053f1d
to
f136361
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
* Support both use_calc_stream and sync_op in send recv APIs (#46023) * Support both use_calc_stream and sync_op in allgather API (#46295) * Support both use_calc_stream and sync_op in collective communication API (#46761) * Move group and all reduce from collective to communication (#45848) * Completes bfloat16 dtype for collective api in eager mode (#45844) * Fix collective APIs cannot be recognized when building docs (#46962) Co-authored-by: LiYuRio <63526175+LiYuRio@users.noreply.github.com>
PR types
New features
PR changes
APIs
Describe
In the new communication library, we designed ProcessGroup to manage different communication group. Inside each process_group has its own stream which all communications in this group will be done on this stream. For high level API, like
distributed.all_reduce
, we useuse_calc_stream
to indicate whether this operation is sync or not. Notice that frequently add unnecessary cuda events may lead to low performance on some model. In order to achieve high performance, this pr add a new API name distributed.stream.all_reduce. This new API provideduse_calc_stream
andsync_op
both.sync_op
, indicate whether communication is sync or not.use_calc_stream
, do communicate on calc stream, save the time of switching stream. Only work whensync_op
is true.