[c10d] Make reduce_scatter as a custom op#79683
Closed
alanwaketan wants to merge 3 commits intogh/alanwaketan/36/basefrom
Closed
[c10d] Make reduce_scatter as a custom op#79683alanwaketan wants to merge 3 commits intogh/alanwaketan/36/basefrom
alanwaketan wants to merge 3 commits intogh/alanwaketan/36/basefrom
Conversation
Summary: This patch makes reduce_scatter as a custom op such that it's dispatcher passable. It's one part of the effort to route comm ops to the dispatcher such that tracing mechanisms that relies on the dispatcher can trace them, e.g., LazyTensor and AOTAutograd. Test Plan: python test/distributed/test_c10d_nccl.py -k test_reduce_scatter_ops [ghstack-poisoned]
alanwaketan
requested review from
H-Huang,
awgu,
mingzhe09088,
mrshenli,
pritamdamania87,
rohan-varma and
zhaojuanmao
as code owners
This was referenced Jun 16, 2022
Contributor
🔗 Helpful links
❌ 1 New FailuresAs of commit e73037d (more details on the Dr. CI page): Expand to see more
🕵️ 1 new failure recognized by patternsThe following CI failures do not appear to be due to upstream breakages
|
facebook-github-bot
added
the
oncall: distributed
alanwaketan
requested review from
aazzolini,
ezyang and
wconstab
and removed request for
H-Huang,
awgu,
mingzhe09088,
mrshenli,
rohan-varma and
zhaojuanmao
Summary: This patch makes reduce_scatter as a custom op such that it's dispatcher passable. It's one part of the effort to route comm ops to the dispatcher such that tracing mechanisms that relies on the dispatcher can trace them, e.g., LazyTensor and AOTAutograd. Test Plan: python test/distributed/test_c10d_nccl.py -k test_reduce_scatter_ops [ghstack-poisoned]
alanwaketan
pushed a commit
that referenced
this pull request
Jun 16, 2022
Summary: This patch makes reduce_scatter as a custom op such that it's dispatcher passable. It's one part of the effort to route comm ops to the dispatcher such that tracing mechanisms that relies on the dispatcher can trace them, e.g., LazyTensor and AOTAutograd. Test Plan: python test/distributed/test_c10d_nccl.py -k test_reduce_scatter_ops ghstack-source-id: b9afad5 Pull Request resolved: #79683
This was referenced Jun 16, 2022
This was referenced Jun 17, 2022
Summary: This patch makes reduce_scatter as a custom op such that it's dispatcher passable. It's one part of the effort to route comm ops to the dispatcher such that tracing mechanisms that relies on the dispatcher can trace them, e.g., LazyTensor and AOTAutograd. Test Plan: python test/distributed/test_c10d_nccl.py -k test_reduce_scatter_ops [ghstack-poisoned]
mrshenli
approved these changes
Jun 24, 2022
Contributor
mrshenli
left a comment
mrshenli
left a comment
There was a problem hiding this comment.
LGTM! Same suggestions on tests as the PR below this one.
Comment on lines
+147
to
+153
| .findSchemaOrThrow("c10d::reduce_scatter_", "") | ||
| .typed<c10::intrusive_ptr<::c10d::ProcessGroup::Work>( | ||
| const std::vector<at::Tensor>&, | ||
| const std::vector<std::vector<at::Tensor>>&, | ||
| const c10::intrusive_ptr<::c10d::ProcessGroup>&, | ||
| int64_t, | ||
| int64_t)>(); |
Contributor
There was a problem hiding this comment.
Would I be correct that the overhead of these two ops are negligible?
Collaborator
Author
There was a problem hiding this comment.
It should be negligible. And then we cache it for the consecutive calls to minimize the overhead too.
| .def( | ||
| "reduce_scatter", | ||
| [](::c10d::ProcessGroup& pg, | ||
| [](const c10::intrusive_ptr<::c10d::ProcessGroup>& pg, |
Contributor
There was a problem hiding this comment.
let's stay consistent on the naming? This one is using pg. The one above is using self.
Collaborator
Author
There was a problem hiding this comment.
I believe this pg is user defined, and the above is really self. That's why there is this inconsistency.
Collaborator
Author
|
Thanks Shen for approving this pull request. The CI failure is unrelated. |
Collaborator
Author
|
@pytorchbot merge -f |
Collaborator
|
@pytorchbot successfully started a merge job. Check the current status here |
Contributor
|
Hey @alanwaketan. |
alanwaketan
added
release notes: distributed (c10d)
facebook-github-bot
pushed a commit
that referenced
this pull request
Jun 27, 2022
Summary: This patch makes reduce_scatter as a custom op such that it's dispatcher passable. It's one part of the effort to route comm ops to the dispatcher such that tracing mechanisms that relies on the dispatcher can trace them, e.g., LazyTensor and AOTAutograd. Pull Request resolved: #79683 Approved by: https://github.com/mrshenli Test Plan: contbuild & OSS CI, see https://hud.pytorch.org/commit/pytorch/pytorch/80b50dfa3ab16de3e90dab8eeed003a98a0da1fe Test plan from GitHub: python test/distributed/test_c10d_nccl.py -k test_reduce_scatter_ops Reviewed By: atalman Differential Revision: D37455687 Pulled By: alanwaketan fbshipit-source-id: bb78183e5cf5798b6b558488eed07af7cd4d4eff
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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.
Stack from ghstack (oldest at bottom):
Summary:
This patch makes reduce_scatter as a custom op such that it's dispatcher
passable. It's one part of the effort to route comm ops to the dispatcher
such that tracing mechanisms that relies on the dispatcher can trace them,
e.g., LazyTensor and AOTAutograd.
Test Plan:
python test/distributed/test_c10d_nccl.py -k test_reduce_scatter_ops