Skip to content

[ET-VK] Making stride equals dilation the default mode for conv2d dw. #7629

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 4 commits into from
Jan 13, 2025

Conversation

pytorchbot
Copy link
Collaborator

This PR was created by the merge bot to help merge the original PR into the main branch.
ghstack PR number: #7596
^ Please use this as the source of truth for the PR details, comments, and reviews
ghstack PR base: https://github.com/pytorch/executorch/tree/gh/trivedivivek/42/base
ghstack PR head: https://github.com/pytorch/executorch/tree/gh/trivedivivek/42/head
Merge bot PR base: https://github.com/pytorch/executorch/tree/gh/trivedivivek/41/orig
Merge bot PR head: https://github.com/pytorch/executorch/tree/gh/trivedivivek/42/orig
@diff-train-skip-merge

Pull Request resolved: #7595

This diff moves current implementation of conv2d dw as a special case when stride equals dilation in the Vulkan backend of Executorch, since that's the only time this kind of caching is possible.

If stride does not equal dilation the old implementation is used.

Additional test cases are added to ensure computation is correct when stride != dilation.
ghstack-source-id: 261183385
@exported-using-ghexport

Differential Revision: [D67908916](https://our.internmc.facebook.com/intern/diff/D67908916/)
Pull Request resolved: #7596

This diff makes changes make stride equals dilation the default mode for conv2d dw output op.
Adds a different source file to handle stride not equal dilation case.
ghstack-source-id: 261183386

Differential Revision: [D67979760](https://our.internmc.facebook.com/intern/diff/D67979760/)
Copy link

pytorch-bot bot commented Jan 13, 2025

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/7629

Note: Links to docs will display an error until the docs builds have been completed.

⏳ No Failures, 2 Pending

As of commit 090101c with merge base d229513 (image):
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.

@facebook-github-bot facebook-github-bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Jan 13, 2025
Base automatically changed from gh/trivedivivek/41/orig to main January 13, 2025 19:53
@trivedivivek trivedivivek merged commit cf8d0cf into main Jan 13, 2025
44 checks passed
@trivedivivek trivedivivek deleted the gh/trivedivivek/42/orig branch January 13, 2025 23:51
YIWENX14 pushed a commit that referenced this pull request Jan 28, 2025
…#7629)

* [ET-VK] Fixing conv2d dw incorrect output when stride != dilation issue.

Pull Request resolved: #7595

This diff moves current implementation of conv2d dw as a special case when stride equals dilation in the Vulkan backend of Executorch, since that's the only time this kind of caching is possible.

If stride does not equal dilation the old implementation is used.

Additional test cases are added to ensure computation is correct when stride != dilation.
ghstack-source-id: 261183385
@exported-using-ghexport

Differential Revision: [D67908916](https://our.internmc.facebook.com/intern/diff/D67908916/)

* [ET-VK] Making stride equals dilation the default mode for conv2d dw.

Pull Request resolved: #7596

This diff makes changes make stride equals dilation the default mode for conv2d dw output op.
Adds a different source file to handle stride not equal dilation case.
ghstack-source-id: 261183386

Differential Revision: [D67979760](https://our.internmc.facebook.com/intern/diff/D67979760/)

* Update conv2d_dw_output_tile.glsl

---------

Co-authored-by: Vivek Trivedi <[email protected]>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. topic: not user facing
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants