Insert contiguous clone before collectives that require it#376
Draft
fmassa wants to merge 1 commit into
Draft
Conversation
NCCL collectives like all_gather_into_tensor and reduce_scatter_tensor require contiguous input tensors, but AP's collective insertion via DTensor redistribute doesn't guarantee this — the input may be non-contiguous after upstream ops like transpose or view. This adds a graph pass that walks the parallel FX graph and inserts aten.clone(memory_format=contiguous_format) before any such collective whose input isn't already a contiguous clone. The pass runs after cleanup_graph in api.py, so it operates on the final parallel graph. The clone is a no-op when the tensor is already contiguous. Authored with Claude.
meta-cla
Bot
added
the
CLA Signed
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
NCCL collectives like all_gather_into_tensor and reduce_scatter_tensor require contiguous input tensors, but AP's collective insertion via DTensor redistribute doesn't guarantee this — the input may be non-contiguous after upstream ops like transpose or view.
This adds a graph pass that walks the parallel FX graph and inserts aten.clone(memory_format=contiguous_format) before any such collective whose input isn't already a contiguous clone. The pass runs after cleanup_graph in api.py, so it operates on the final parallel graph. The clone is a no-op when the tensor is already contiguous.
Authored with Claude.