armv8.4 bf16 gemm optimization

[nihui]

No description provided.

[tencent-adm]

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.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 9147e65201

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".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[nihui]

@codex review

[codecov-commenter]

Codecov Report

❌ Patch coverage is 99.90975% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 93.86%. Comparing base (5a4a483) to head (2e2c4d5).
⚠️ Report is 1 commits behind head on master.

Files with missing lines	Patch %	Lines
src/layer/arm/gemm_fp16s.h	99.90%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #6714      +/-   ##
==========================================
- Coverage   93.93%   93.86%   -0.08%     
==========================================
  Files         933      932       -1     
  Lines      302704   307523    +4819     
==========================================
+ Hits       284356   288660    +4304     
- Misses      18348    18863     +515     

☔ View full report in Codecov by Sentry.
📢 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.

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Breezy!

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".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[Copilot]

Pull request overview

This PR targets improved ARMv8.4 BF16 GEMM performance by adding BF16-aware packing paths and refactoring the BF16 GEMM flow to accumulate into FP32 tiles and apply alpha/beta during the final unpack step (instead of pre-scaling C).

Changes:

• Added runtime dispatch hooks to call BF16-instruction implementations when the runtime CPU supports BF16 but the baseline build lacks BF16/FP16 vector arithmetic.
• Added __ARM_FEATURE_BF16_VECTOR_ARITHMETIC-gated fast paths in BF16 pack/transpose-pack routines.
• Refactored BF16 GEMM tiling to always use a per-thread FP32 workspace tile and moved beta handling into unpack_output_tile_fp32_to_bf16().

Reviewed changes

Copilot reviewed 4 out of 5 changed files in this pull request and generated 5 comments.

File	Description
src/layer/arm/gemm_bf16s_fp16s.h	Adds runtime BF16 dispatch and new BF16-vector-arithmetic packing/transpose-packing fast paths.
src/layer/arm/gemm_arm.cpp	Refactors BF16 GEMM tile loop to use FP32 tile + unified unpack applying alpha/beta.
src/layer/arm/gemm_arm_bf16.cpp	Adds BF16-target wrapper entry points for runtime-dispatched BF16 code paths.
src/layer/arm/arm_usability.h	Uses vcvt_bf16_f32 when BF16 vector arithmetic is available.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.