Move slice before permute with heuristics#19123
Closed
DrJessop wants to merge 5 commits intopytorch:mainfrom
Closed
Move slice before permute with heuristics#19123DrJessop wants to merge 5 commits intopytorch:mainfrom
DrJessop wants to merge 5 commits intopytorch:mainfrom
Conversation
Summary: If we have a permute -> unary chain -> mean, based on the reduction dims of the mean, we can either fully remove the preceding permute or move the permute after the mean. Case 1: Dims after permute are still in same order with respect to each other, we can fully get rid of the permute and just update the reduction dims of the mean. Case 2: Not case 1. In this case, it's better to move the permute after the mean, since the permute will operate on less data. Differential Revision: D102268214
Summary: Pull Request resolved: pytorch#19106 I've seen lots of cases where we have a mean followed by a view which can either be fused into mean with keepdim True or False. These views typically become no-ops, however it's nice to remove them if we don't need them. Can potentially make work for later passes easier that are looking for mean, but view in the way. Differential Revision: D102276633
Summary: Pull Request resolved: pytorch#19120 If we have back to back slices on the same dimension, we can remove the top slice and just perform the second one. RemoveOrReplacePassInterface will handle a whole cascade if it exists. Differential Revision: D102278253 Reviewed By: abeakkas
Summary: It can be good to permute before a long chain of slices/cats, so that they are on dim 0 and effectively become nops. However, there are cases where this is not desirable. We should have a pass Differential Revision: D102407566
Summary: In cases where we have a permute -> slice, it's better to slice before the permute if the slice is not already happening on dim 0. If it is happening on dim 0, it can still be profitable to move the slice before the permute if permute shape product > 2 * slice shape product. Differential Revision: D102410226
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/19123
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 New Failure, 18 Pending, 1 Unrelated FailureAs of commit 3519e8f with merge base cb94506 ( NEW FAILURE - The following job has failed:
FLAKY - The following job failed but was likely due to flakiness present on trunk:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
Bot
added
the
CLA Signed
Contributor
|
@DrJessop has exported this pull request. If you are a Meta employee, you can view the originating Diff in D102410226. |
This PR needs a
|
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.
Summary: In cases where we have a permute -> slice, it's better to slice before the permute if the slice is not already happening on dim 0. If it is happening on dim 0, it can still be profitable to move the slice before the permute if permute shape product > 2 * slice shape product.
Differential Revision: D102410226