Fix regression from pointwise + multi-level reduction fusion #112297
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
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/112297
Note: Links to docs will display an error until the docs builds have been completed. ✅ You can merge normally! (2 Unrelated Failures)As of commit 4cf7b22 with merge base ac4cc5d ( FLAKY - The following jobs failed but were likely due to flakiness present on trunk:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
@pytorchbot label "topic: not user facing" |
jansel
approved these changes
Oct 28, 2023
|
@ipiszy has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
|
@pytorchbot merge |
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
xuhancn
pushed a commit
to xuhancn/pytorch
that referenced
this pull request
Nov 7, 2023
…#112297) In pytorch#111122, an optimization is introduced for reduction + pointwise + multi-level reduction fusion. The main idea of this optimization is to have the first-level reduction of the multi-level reduction reuses the reduction sizes of the first reduction kernel so that there are better chances that the first reduction kernel and the first-level reduction of the multi-level reduction kernel can be fused. However, it introduces a bug for pattern pointwise + multi-level reduction, where the first-level reduction kernel wrongly reuses the reduction ranges (which is []) from the previous pointwise kernel. This PR fixes this issue. Test plan: `python timm_models.py --training --amp --performance --only=dm_nfnet_f0 --inductor` Results before this PR: 0.869x Results after this PR: 1.232x Benchmark results:  <img width="1491" alt="Screenshot 2023-10-30 at 3 10 06 PM" src="https://github.com/pytorch/pytorch/assets/10527447/608d26ea-dcc5-4f2a-8700-4a928701392b"> Pull Request resolved: pytorch#112297 Approved by: https://github.com/jansel
Skylion007
pushed a commit
to Skylion007/pytorch
that referenced
this pull request
Nov 14, 2023
…#112297) In pytorch#111122, an optimization is introduced for reduction + pointwise + multi-level reduction fusion. The main idea of this optimization is to have the first-level reduction of the multi-level reduction reuses the reduction sizes of the first reduction kernel so that there are better chances that the first reduction kernel and the first-level reduction of the multi-level reduction kernel can be fused. However, it introduces a bug for pattern pointwise + multi-level reduction, where the first-level reduction kernel wrongly reuses the reduction ranges (which is []) from the previous pointwise kernel. This PR fixes this issue. Test plan: `python timm_models.py --training --amp --performance --only=dm_nfnet_f0 --inductor` Results before this PR: 0.869x Results after this PR: 1.232x Benchmark results:  <img width="1491" alt="Screenshot 2023-10-30 at 3 10 06 PM" src="https://github.com/pytorch/pytorch/assets/10527447/608d26ea-dcc5-4f2a-8700-4a928701392b"> Pull Request resolved: pytorch#112297 Approved by: https://github.com/jansel
andreigh
pushed a commit
to andreigh/pytorch
that referenced
this pull request
Nov 19, 2023
…#112297) In pytorch#111122, an optimization is introduced for reduction + pointwise + multi-level reduction fusion. The main idea of this optimization is to have the first-level reduction of the multi-level reduction reuses the reduction sizes of the first reduction kernel so that there are better chances that the first reduction kernel and the first-level reduction of the multi-level reduction kernel can be fused. However, it introduces a bug for pattern pointwise + multi-level reduction, where the first-level reduction kernel wrongly reuses the reduction ranges (which is []) from the previous pointwise kernel. This PR fixes this issue. Test plan: `python timm_models.py --training --amp --performance --only=dm_nfnet_f0 --inductor` Results before this PR: 0.869x Results after this PR: 1.232x Benchmark results:  <img width="1491" alt="Screenshot 2023-10-30 at 3 10 06 PM" src="https://github.com/pytorch/pytorch/assets/10527447/608d26ea-dcc5-4f2a-8700-4a928701392b"> Pull Request resolved: pytorch#112297 Approved by: https://github.com/jansel
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.
In #111122, an optimization is introduced for reduction + pointwise + multi-level reduction fusion. The main idea of this optimization is to have the first-level reduction of the multi-level reduction reuses the reduction sizes of the first reduction kernel so that there are better chances that the first reduction kernel and the first-level reduction of the multi-level reduction kernel can be fused. However, it introduces a bug for pattern pointwise + multi-level reduction, where the first-level reduction kernel wrongly reuses the reduction ranges (which is []) from the previous pointwise kernel. This PR fixes this issue.
Test plan:
python timm_models.py --training --amp --performance --only=dm_nfnet_f0 --inductorResults before this PR: 0.869x
Results after this PR: 1.232x
Benchmark results:
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @peterbell10 @yf225 @chenyang78 @kadeng @muchulee8 @aakhundov @ColinPeppler