Skip to content

[GPU] select bfyx for OneDNN convolution when feature size is 4 #30729

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.

Sign up for GitHub

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

Jump to bottom
Merged
merged 1 commit into from
May 30, 2025
Merged

[GPU] select bfyx for OneDNN convolution when feature size is 4 #30729

merged 1 commit into from
May 30, 2025

Conversation

e-ddykim
Copy link
Contributor

Details:

  • Currently, when feature size of convolution's input or output is 4, OneDNN's ocl_ref kernel is used, and it degrades performance of some models (e.g. Stable Diffusion 1.5)
  • This PR updates layout_optimizer to select bfyx and makes the convolution layers to use OneDNN's jit kernel in this case.

Tickets:

  • 161941

@e-ddykim e-ddykim requested a review from a team as a code owner May 27, 2025 02:51
@e-ddykim e-ddykim added the WIP work in progress label May 27, 2025
@e-ddykim e-ddykim requested a review from a team as a code owner May 27, 2025 02:51
@github-actions github-actions bot added the category: GPU OpenVINO GPU plugin label May 27, 2025
@e-ddykim e-ddykim changed the title [GPU updates layout_optimizer to select bfyx when convolution's feature size is 4 and OneDNN is used [GPU] select bfyx for OneDNN convolution when feature size is 4 May 28, 2025
@yeonbok yeonbok added this pull request to the merge queue May 30, 2025
Merged via the queue into openvinotoolkit:master with commit 07e2a14 May 30, 2025
182 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@yeonbok yeonbok yeonbok approved these changes

Assignees
No one assigned
Labels
category: GPU OpenVINO GPU plugin
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
2 participants
@e-ddykim @yeonbok