[SR] Add a copy variant for fused_split_and_squeeze #75660
Closed
Conversation
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
The outputs of `split_and_squeeze` are passed to `VarStack` in models we care about. `VarStack` has a [fast path](https://www.internalfb.com/code/fbsource/[893193f5277184fd17f4ea3f28fe415a4df37707]/fbcode/caffe2/aten/src/ATen/native/TensorShape.cpp?lines=296-298) for when all of its inputs have the same strides. Hitting the slow path adds a ton of extra overhead - so much that it's worth it to copy in `split_and_squeeze` and force all of `VarStack`'s inputs to be contiguous so we can take advantage of the fast path. Differential Revision: [D35513777](https://our.internmc.facebook.com/intern/diff/D35513777/) **NOTE FOR REVIEWERS**: This PR has internal Facebook specific changes or comments, please review them on [Phabricator](https://our.internmc.facebook.com/intern/diff/D35513777/)! [ghstack-poisoned]
🔗 Helpful links
💊 CI failures summary and remediationsAs of commit 37a2421 (more details on the Dr. CI page): 💚 💚 Looks good so far! There are no failures yet. 💚 💚 This comment was automatically generated by Dr. CI (expand for details).Please report bugs/suggestions to the (internal) Dr. CI Users group. |
mikeiovine
pushed a commit
that referenced
this pull request
Apr 12, 2022
The outputs of `split_and_squeeze` are passed to `VarStack` in models we care about. `VarStack` has a [fast path](https://www.internalfb.com/code/fbsource/[893193f5277184fd17f4ea3f28fe415a4df37707]/fbcode/caffe2/aten/src/ATen/native/TensorShape.cpp?lines=296-298) for when all of its inputs have the same strides. Hitting the slow path adds a ton of extra overhead - so much that it's worth it to copy in `split_and_squeeze` and force all of `VarStack`'s inputs to be contiguous so we can take advantage of the fast path. Differential Revision: [D35513777](https://our.internmc.facebook.com/intern/diff/D35513777/) **NOTE FOR REVIEWERS**: This PR has internal Facebook specific changes or comments, please review them on [Phabricator](https://our.internmc.facebook.com/intern/diff/D35513777/)! ghstack-source-id: 153648923 Pull Request resolved: #75660
facebook-github-bot
added
the
oncall: jit
The outputs of `split_and_squeeze` are passed to `VarStack` in models we care about. `VarStack` has a [fast path](https://www.internalfb.com/code/fbsource/[893193f5277184fd17f4ea3f28fe415a4df37707]/fbcode/caffe2/aten/src/ATen/native/TensorShape.cpp?lines=296-298) for when all of its inputs have the same strides. Hitting the slow path adds a ton of extra overhead - so much that it's worth it to copy in `split_and_squeeze` and force all of `VarStack`'s inputs to be contiguous so we can take advantage of the fast path. Differential Revision: [D35513777](https://our.internmc.facebook.com/intern/diff/D35513777/) **NOTE FOR REVIEWERS**: This PR has internal Facebook specific changes or comments, please review them on [Phabricator](https://our.internmc.facebook.com/intern/diff/D35513777/)! [ghstack-poisoned]
mikeiovine
pushed a commit
that referenced
this pull request
Apr 13, 2022
Pull Request resolved: #75660 The outputs of `split_and_squeeze` are passed to `VarStack` in models we care about. `VarStack` has a [fast path](https://www.internalfb.com/code/fbsource/[893193f5277184fd17f4ea3f28fe415a4df37707]/fbcode/caffe2/aten/src/ATen/native/TensorShape.cpp?lines=296-298) for when all of its inputs have the same strides. Hitting the slow path adds a ton of extra overhead - so much that it's worth it to copy in `split_and_squeeze` and force all of `VarStack`'s inputs to be contiguous so we can take advantage of the fast path. ghstack-source-id: 153768800 Differential Revision: [D35513777](https://our.internmc.facebook.com/intern/diff/D35513777/) **NOTE FOR REVIEWERS**: This PR has internal Facebook specific changes or comments, please review them on [Phabricator](https://our.internmc.facebook.com/intern/diff/D35513777/)!
|
@pytorchbot merge this (Initiating merge automatically since Phabricator Diff has merged) |
|
Hey @mikeiovine. |
facebook-github-bot
pushed a commit
that referenced
this pull request
Apr 13, 2022
Summary: Pull Request resolved: #75660 The outputs of `split_and_squeeze` are passed to `VarStack` in models we care about. `VarStack` has a [fast path](https://www.internalfb.com/code/fbsource/[893193f5277184fd17f4ea3f28fe415a4df37707]/fbcode/caffe2/aten/src/ATen/native/TensorShape.cpp?lines=296-298) for when all of its inputs have the same strides. Hitting the slow path adds a ton of extra overhead - so much that it's worth it to copy in `split_and_squeeze` and force all of `VarStack`'s inputs to be contiguous so we can take advantage of the fast path. ghstack-source-id: 153768800 Test Plan: Existing unit tests cover the new op. Reviewed By: navahgar, hlu1 Differential Revision: D35513777 fbshipit-source-id: fb081eb41689ab4e8ef5102fb4906cffd8a95b1a
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 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):
The outputs of
split_and_squeezeare passed toVarStackin models we care about.VarStackhas a fast path for when all of its inputs have the same strides.Hitting the slow path adds a ton of extra overhead - so much that it's worth it to copy in
split_and_squeezeand force all ofVarStack's inputs to be contiguous so we can take advantage of the fast path.Differential Revision: D35513777
NOTE FOR REVIEWERS: This PR has internal Facebook specific changes or comments, please review them on Phabricator!