implement vulkan gemm packed

[nihui]

z-image-turbo 1024 end2end (s)

	7900xtx	9060xt	rtx3060
baseline	32.003	65.640	155.383
pr6573	29.583	62.148	134.004

[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.}

[codecov-commenter]

Codecov Report

❌ Patch coverage is 91.95980% with 16 lines in your changes missing coverage. Please review.
✅ Project coverage is 93.17%. Comparing base (081f2b8) to head (a4c181d).
⚠️ Report is 1 commits behind head on master.

Files with missing lines	Patch %	Lines
src/layer/vulkan/gemm_vulkan.cpp	91.95%	16 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #6573      +/-   ##
==========================================
- Coverage   93.18%   93.17%   -0.02%     
==========================================
  Files         832      829       -3     
  Lines      266714   265989     -725     
==========================================
- Hits       248545   247839     -706     
+ Misses      18169    18150      -19     

☔ 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.

[Copilot]

Pull request overview

This PR extends the Vulkan GEMM implementation to support runtime packing (elempack 1/4) and to optionally write packed outputs directly from the shaders, reducing the need for explicit pre/post packing conversions in gemm_vulkan.cpp.

Changes:

• Add out_elempack, A_elempack, and B_elempack push constants and introduce pack-aware load/store paths in GEMM shaders.
• Update gemm_vulkan.cpp to stop forcing A/B to pack1, compute logical M/N/K using elempack, and allocate top_blob with the chosen out_elempack.
• Add additional descriptor bindings to provide alternate scalar/vec4 views for packed I/O.

Reviewed changes

Copilot reviewed 3 out of 4 changed files in this pull request and generated 3 comments.

File	Description
src/layer/vulkan/shader/gemm_sg.comp	Adds elempack-aware loads and pack4 output stores for the subgroup GEMM shader.
src/layer/vulkan/shader/gemm_cm.comp	Adds elempack-aware load paths and pack4 output stores for the cooperative-matrix GEMM shader.
src/layer/vulkan/shader/gemm.comp	Adds new bindings/constants and implements pack4 output write paths for the non-subgroup GEMM shader.
src/layer/vulkan/gemm_vulkan.cpp	Removes forced input packing, computes M/N/K with elempack, allocates packed output, and updates bindings/constants passed to shaders.

---

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

[nihui]

我不满意！

[nihui]

TODO: pre-transform constant A and constant B to optimal layout
TODO: is uvec4 beneficial ?
TODO: vkmat heap index for auto host-device transfer

[Copilot]

Pull request overview

Copilot reviewed 3 out of 4 changed files in this pull request and generated 2 comments.

---

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