Handle rank-changing views in FuseCascadedTransposeOrPermuteOps (#19539)#19539
Open
mcremon-meta wants to merge 2 commits into
Open
Handle rank-changing views in FuseCascadedTransposeOrPermuteOps (#19539)#19539mcremon-meta wants to merge 2 commits into
mcremon-meta wants to merge 2 commits into
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/19539
Note: Links to docs will display an error until the docs builds have been completed. ❗ 1 Active SEVsThere are 1 currently active SEVs. If your PR is affected, please view them below: ❌ 1 Unclassified FailureAs of commit 5e33840 with merge base 7c495fa ( UNCLASSIFIED FAILURE - DrCI could not classify the following job because the workflow did not run on the merge base. The failure may be pre-existing on trunk or introduced by this PR:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
Bot
added
the
CLA Signed
github-actions
Bot
added
ciflow/trunk
module: arm
Contributor
|
@mcremon-meta has exported this pull request. If you are a Meta employee, you can view the originating Diff in D104775245. |
This PR needs a
|
meta-cla
Bot
added
the
CLA Signed
meta-codesync
Bot
changed the title
meta-codesync Bot
pushed a commit
that referenced
this pull request
May 13, 2026
Summary: Extend FuseCascadedTransposeOrPermuteOps to fuse permute→view_copy→permute patterns where a squeeze/unsqueeze view sits between two permutes. The pass computes the combined effect of permute+view+permute: if the permutation components cancel out (identity), the entire chain is replaced with a single view_copy. This handles patterns like permute_3D([0,2,1]) → view(unsqueeze) → permute_4D([0,2,3,1]) which composes to a simple view_copy (the permutations cancel, leaving only the reshape). Differential Revision: D104775245
meta-codesync
Bot
force-pushed
the
export-D104775245
branch
from
b6fb904 to
7699a00
Compare
meta-codesync Bot
pushed a commit
that referenced
this pull request
May 14, 2026
Summary: Extend FuseCascadedTransposeOrPermuteOps to fuse permute→view_copy→permute patterns where a squeeze/unsqueeze view sits between two permutes. The pass computes the combined effect of permute+view+permute: if the permutation components cancel out (identity), the entire chain is replaced with a single view_copy. This handles patterns like permute_3D([0,2,1]) → view(unsqueeze) → permute_4D([0,2,3,1]) which composes to a simple view_copy (the permutations cancel, leaving only the reshape). Differential Revision: D104775245
meta-codesync
Bot
force-pushed
the
export-D104775245
branch
from
7699a00 to
80ca296
Compare
meta-codesync Bot
pushed a commit
that referenced
this pull request
May 14, 2026
Summary: Extend FuseCascadedTransposeOrPermuteOps to fuse permute→view_copy→permute patterns where a squeeze/unsqueeze view sits between two permutes. The pass computes the combined effect of permute+view+permute: if the permutation components cancel out (identity), the entire chain is replaced with a single view_copy. This handles patterns like permute_3D([0,2,1]) → view(unsqueeze) → permute_4D([0,2,3,1]) which composes to a simple view_copy (the permutations cancel, leaving only the reshape). Differential Revision: D104775245
meta-codesync
Bot
force-pushed
the
export-D104775245
branch
from
80ca296 to
88fea01
Compare
meta-codesync Bot
pushed a commit
that referenced
this pull request
May 14, 2026
Summary: Extend FuseCascadedTransposeOrPermuteOps to fuse permute→view_copy→permute patterns where a squeeze/unsqueeze view sits between two permutes. The pass computes the combined effect of permute+view+permute: if the permutation components cancel out (identity), the entire chain is replaced with a single view_copy. This handles patterns like permute_3D([0,2,1]) → view(unsqueeze) → permute_4D([0,2,3,1]) which composes to a simple view_copy (the permutations cancel, leaving only the reshape). Differential Revision: D104775245
meta-codesync
Bot
force-pushed
the
export-D104775245
branch
from
88fea01 to
476ce41
Compare
DrJessop
approved these changes
May 14, 2026
) Summary: Extend RemovePermutesAroundElementwiseOps to cancel permute pairs across rank-changing squeeze/unsqueeze view boundaries. When a permute's sole user is a view_copy that adds or removes a single size-1 dimension, the pass adapts the expected permutation to the new rank and continues traversal. This enables removing permutes that sit on opposite sides of an unsqueeze→elementwise→squeeze chain (e.g. the NHWC↔NTC layout conversion around convolutions in the cascade detector model). Key changes: - Accept extra_permutable_ops constructor parameter for backend-specific ops - Track per-node expected permutations across view boundaries - Run dimension updates before edges_in bypass to preserve original metadata - Handle view_copy, unsqueeze_copy, squeeze_copy rank changes - Treat aten.full.default as a compile-time constant Reviewed By: DrJessop Differential Revision: D104775244
Summary: Extend FuseCascadedTransposeOrPermuteOps to fuse permute→view_copy→permute patterns where a squeeze/unsqueeze view sits between two permutes. The pass computes the combined effect of permute+view+permute: if the permutation components cancel out (identity), the entire chain is replaced with a single view_copy. This handles patterns like permute_3D([0,2,1]) → view(unsqueeze) → permute_4D([0,2,3,1]) which composes to a simple view_copy (the permutations cancel, leaving only the reshape). Reviewed By: DrJessop Differential Revision: D104775245
meta-codesync
Bot
force-pushed
the
export-D104775245
branch
from
476ce41 to
5e33840
Compare
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
2 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.
Summary:
Extend FuseCascadedTransposeOrPermuteOps to fuse permute→view_copy→permute patterns where a squeeze/unsqueeze view sits between two permutes. The pass computes the combined effect of permute+view+permute: if the permutation components cancel out (identity), the entire chain is replaced with a single view_copy.
This handles patterns like permute_3D([0,2,1]) → view(unsqueeze) → permute_4D([0,2,3,1]) which composes to a simple view_copy (the permutations cancel, leaving only the reshape).
Reviewed By: DrJessop
Differential Revision: D104775245