Skip to content

[DO NOT REVIEW][GPU] Fuse channel-axis RMSNorm for NCHW vision tensors - #38091

Open
e-ddykim wants to merge 1 commit into
openvinotoolkit:masterfrom
e-ddykim:gpu_rms_channel_axis
Open

[DO NOT REVIEW][GPU] Fuse channel-axis RMSNorm for NCHW vision tensors#38091
e-ddykim wants to merge 1 commit into
openvinotoolkit:masterfrom
e-ddykim:gpu_rms_channel_axis

Conversation

@e-ddykim

Copy link
Copy Markdown
Contributor

Details:

  • Extend internal RMS operation with normalization axis attribute (default -1).
  • Update RMSFusionMatcher to support rank-4 NCHW channel-axis normalization (axis=1) in addition to the last axis.
  • Update Intel GPU plugin RMS primitive and kernel selector to support axis=1 via rms_gpu_ref.
  • Maintain Template plugin evaluate and Intel CPU fallback correctness for channel-axis RMS.
  • Add positive fusion unit tests and GPU kernel test cases (odd spatial, unaligned channels, dynamic shapes, caching).

AI Assistance:

  • AI assistance used: yes

@e-ddykim
e-ddykim requested review from a team as code owners September 11, 2026 10:54
@e-ddykim
e-ddykim requested a review from a team as a code owner September 11, 2026 10:54
@github-actions github-actions Bot added category: GPU OpenVINO GPU plugin category: CPU OpenVINO CPU plugin category: transformations OpenVINO Runtime library - Transformations category: TEMPLATE OpenVINO Template plugin labels Sep 11, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

category: CPU OpenVINO CPU plugin category: GPU OpenVINO GPU plugin category: TEMPLATE OpenVINO Template plugin category: transformations OpenVINO Runtime library - Transformations do not merge do_not_merge do_not_review do-not-merge Do not merge this PR

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant