Skip to content

vulkan: skip unused RMSNorm fallback pipelines - #7133

Merged
nihui merged 1 commit into
Tencent:masterfrom
nihui:fix-rmsnorm-vulkan-fallback-pipelines
Oct 18, 2026
Merged

nihui merged 1 commit into
Tencent:masterfrom
nihui:fix-rmsnorm-vulkan-fallback-pipelines

Conversation

@nihui

@nihui nihui commented Oct 18, 2026

Copy link
Copy Markdown
Member

Create the subgroup reducers and the square/sum4/mean fallback pipelines in mutually exclusive branches, matching LayerNorm. Keep the shared coeffs/norm pipelines unconditional.

This avoids allocating ten unused fallback Pipeline wrappers when the validated subgroup path is enabled.

Follow-up to #6755 and #6755 (comment).

Validation:

  • Vulkan-enabled Release build passed.
  • Existing test_rmsnorm and test_layernorm passed with SwiftShader (software Vulkan).
  • 16 supplementary public-layer Vulkan cases passed with subgroup operations on/off, packing on/off, 1D/3D inputs and affine on/off.
  • git diff --check passed.

SwiftShader supports subgroup arithmetic but not fp16 storage/arithmetic, so the existing dedicated fp16a cases were skipped. No hardware startup-time measurements were made.

@chatgpt-codex-connector

chatgpt-codex-connector Bot commented Oct 18, 2026 •

Copy link
Copy Markdown

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review Status Commit Review trigger
📝 Code Review ✅ Completed 2026-10-18T01:35:02.521238Z 38a2d89 PR opened
ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

@tencent-adm

Copy link
Copy Markdown
Member

CLA assistant check
Thank you for your submission, we really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
You have signed the CLA already but the status is still pending? Let us recheck it.

@codecov-commenter

codecov-commenter commented Oct 18, 2026 •

Copy link
Copy Markdown

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 95.26%. Comparing base (5d1c5ad) to head (38a2d89).

Additional details and impacted files
@@            Coverage Diff             @@
##           master    #7133      +/-   ##
==========================================
- Coverage   95.29%   95.26%   -0.03%     
==========================================
  Files        1483     1483              
  Lines      504854   504979     +125     
==========================================
- Hits       481094   481091       -3     
- Misses      23760    23888     +128     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

@github-actions

Copy link
Copy Markdown

Binary size comparison

The binary size change of libncnn.so (bytes)

architecture base size pr size difference
x86_64 27763760 27763760 0 😘
armhf 7723120 7723152 +32 ⚠️
aarch64 11849720 11849720 0 😘

Commit: 38a2d895498d141c3846cb9a0140111b1e4607a1 · Workflow run

@nihui
nihui merged commit fc062e6 into Tencent:master Oct 18, 2026
70 of 72 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants