Fix stream sync for scratch pad eviction #2843
Closed
Conversation
This file contains 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
✅ Deploy Preview for pytorch-fbgemm-docs ready!
To edit notification comments on pull requests, go to your Netlify site configuration. |
|
This pull request was exported from Phabricator. Differential Revision: D59716516 |
|
This pull request was exported from Phabricator. Differential Revision: D59716516 |
sryap
added a commit
to sryap/FBGEMM
that referenced
this pull request
Jul 14, 2024
Summary: Pull Request resolved: pytorch#2843 Before this diff, SSD TBE unsafely evicts data from the scratch pad (the buffer that stores conflict missed data). The eviction happened on the SSD stream while the backward TBE happened on the default stream. SSD TBE did not properly synchronize the streams to ensure that the backward TBE completed before evicting data from the scratch pad. This diff fixes the problem by adding the syncrhonization between streams to ensure the correct execution order between the backward TBE and the scratch pad eviction. Differential Revision: D59716516
|
This pull request was exported from Phabricator. Differential Revision: D59716516 |
sryap
added a commit
to sryap/FBGEMM
that referenced
this pull request
Jul 14, 2024
Summary: Pull Request resolved: pytorch#2843 Before this diff, SSD TBE unsafely evicts data from the scratch pad (the buffer that stores conflict missed data). The eviction happened on the SSD stream while the backward TBE happened on the default stream. SSD TBE did not properly synchronize the streams to ensure that the backward TBE completed before evicting data from the scratch pad. This diff fixes the problem by adding the syncrhonization between streams to ensure the correct execution order between the backward TBE and the scratch pad eviction. Differential Revision: D59716516
Differential Revision: D59795139
Summary: Pull Request resolved: pytorch#2843 Before this diff, SSD TBE unsafely evicts data from the scratch pad (the buffer that stores conflict missed data). The eviction happened on the SSD stream while the backward TBE happened on the default stream. SSD TBE did not properly synchronize the streams to ensure that the backward TBE completed before evicting data from the scratch pad. This diff fixes the problem by adding the syncrhonization between streams to ensure the correct execution order between the backward TBE and the scratch pad eviction. **Before and after this diff** {F1761028617} **Before:** the scratch pad eviction happens as soon as the cache eviction is done, which is incorrect. In the figure, it overlaps with TBE forward and backward. **After:** the scartch pad eviction happens after the backward pass of TBE is done Reviewed By: q10 Differential Revision: D59716516
|
This pull request was exported from Phabricator. Differential Revision: D59716516 |
|
This pull request has been merged in 13d5470. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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:
Before this diff, SSD TBE unsafely evicts data from the scratch pad
(the buffer that stores conflict missed data). The eviction happened
on the SSD stream while the backward TBE happened on the default
stream. SSD TBE did not properly synchronize the streams to ensure
that the backward TBE completed before evicting data from the scratch
pad. This diff fixes the problem by adding the syncrhonization
between streams to ensure the correct execution order between the
backward TBE and the scratch pad eviction.
Differential Revision: D59716516