Avoid more comm/compute overlap in ugni #15193
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
In 12960 we eliminated comm/compute overlap for FMA operations when a task hadn't done ~100 FMA operations. Here we extend that to RDMA and chained FMA operations as well. This is motivated by aggregation work where we do on-stmts followed by large RDMA PUTs/GETs, but we really don't want too many tasks started at once since that increses memory presssure. This also adds an option to completely disable comm/compute overlap in ugni, but it's not enabled by default since it has a 2x performance hit for SSCA.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
In #12964 we eliminated comm/compute overlap for FMA operations when a
task hadn't done ~100 FMA operations. Here we extend that to RDMA and
chained FMA operations as well. This is motivated by aggregation work
where we do on-stmts followed by large RDMA PUTs/GETs, but we really
don't want too many tasks started at once since that increases memory
pressure. This also adds an option to completely disable comm/compute
overlap in ugni, but it's not enabled by default since it has a 2x
performance hit for SSCA.