[PR #387] [CLOSED] Feat/fused qkv norm a2a #412

New issue

Closed

opened 2026-10-26 21:17:29 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-26 21:17:29 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/DeepGEMM/pull/387
Author: @Schilings
Created: 7/19/2026
Status: ❌ Closed

Base: main ← Head: feat/fused-qkv-norm-a2a

📝 Commits (10+)

1eccc9a fix(gemm-rs): add missing reg_reconfig.h include + single-GPU compile test
cd91e7c feat(gemm-rs): enhance tests & benchmark for multi-GPU validation
e78b9b5 fix: remove warpgroup_reg_dealloc from Load/MMA/Reserved warps to fix deadlock
61b2c79 update
7778766 docs: update PROGRESS.md with slot addressing fix, race condition fix, 8-GPU test results
67034c1 docs: add 8-GPU benchmark results and performance analysis
d24d75a update
ee74d28 update
45b8c55 update
49e9424 update

📊 Changes

170 files changed (+35007 additions, -434 deletions)

View changed files

➕ .codebuddy/memory/2026-06-28.md (+154 -0)
➕ .codebuddy/memory/2026-06-30.md (+114 -0)
➕ .codebuddy/memory/2026-07-01.md (+110 -0)
➕ .codebuddy/memory/MEMORY.md (+37 -0)
📝 .gitignore (+8 -1)
➕ .playwright-mcp/console-2026-06-17T15-09-09-789Z.log (+177 -0)
➕ .playwright-mcp/page-2026-06-17T15-09-13-044Z.yml (+7561 -0)
➕ HINTS.md (+55 -0)
➕ benchmarks/bench_a2a_gemm.py (+163 -0)
➕ benchmarks/bench_a2a_only.py (+141 -0)
➕ benchmarks/bench_a2a_transpose_gemm.py (+209 -0)
➕ benchmarks/bench_ag_gemm.py (+167 -0)
➕ benchmarks/bench_fused_qkv_norm_a2a.py (+201 -0)
➕ benchmarks/bench_gemm_a2a_transpose.py (+235 -0)
➕ benchmarks/bench_gemm_rs.py (+320 -0)
➕ csrc/apis/a2a_gemm.hpp (+100 -0)
➕ csrc/apis/a2a_transpose_gemm.hpp (+124 -0)
➕ csrc/apis/ag_gemm.hpp (+201 -0)
📝 csrc/apis/gemm.hpp (+7 -7)
➕ csrc/apis/gemm_a2a_transpose.hpp (+126 -0)

...and 80 more files

📄 Description

No description provided

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/deepseek-ai/DeepGEMM/pull/387 **Author:** [@Schilings](https://github.com/Schilings) **Created:** 7/19/2026 **Status:** ❌ Closed **Base:** `main` ← **Head:** `feat/fused-qkv-norm-a2a` --- ### 📝 Commits (10+) - [`1eccc9a`](https://github.com/deepseek-ai/DeepGEMM/commit/1eccc9ae9a45acc93abc5bc6cfdf123fc05627e2) fix(gemm-rs): add missing reg_reconfig.h include + single-GPU compile test - [`cd91e7c`](https://github.com/deepseek-ai/DeepGEMM/commit/cd91e7cc4cd9dfe9dd7bba6d4bf7ffa5214c44e5) feat(gemm-rs): enhance tests & benchmark for multi-GPU validation - [`e78b9b5`](https://github.com/deepseek-ai/DeepGEMM/commit/e78b9b5877d927d6bf150da1a5da28a604aea124) fix: remove warpgroup_reg_dealloc from Load/MMA/Reserved warps to fix deadlock - [`61b2c79`](https://github.com/deepseek-ai/DeepGEMM/commit/61b2c797373aa3a1a3dca51352b28a4f24668d1e) update - [`7778766`](https://github.com/deepseek-ai/DeepGEMM/commit/777876686da18f17a16ea364a6ea82c24683b591) docs: update PROGRESS.md with slot addressing fix, race condition fix, 8-GPU test results - [`67034c1`](https://github.com/deepseek-ai/DeepGEMM/commit/67034c1b6b6d3b47dce3a094cb5edaa0919b8dcf) docs: add 8-GPU benchmark results and performance analysis - [`d24d75a`](https://github.com/deepseek-ai/DeepGEMM/commit/d24d75af86b330d4b61de011afd292a35cca3a7a) update - [`ee74d28`](https://github.com/deepseek-ai/DeepGEMM/commit/ee74d28e4eed7a6796d2fa26a8795b926faabaae) update - [`45b8c55`](https://github.com/deepseek-ai/DeepGEMM/commit/45b8c5538bbada5495818e936b31d38dab277d4b) update - [`49e9424`](https://github.com/deepseek-ai/DeepGEMM/commit/49e9424b18181c451801a4f97e907eb06fd01fd8) update ### 📊 Changes **170 files changed** (+35007 additions, -434 deletions) <details> <summary>View changed files</summary> ➕ `.codebuddy/memory/2026-06-28.md` (+154 -0) ➕ `.codebuddy/memory/2026-06-30.md` (+114 -0) ➕ `.codebuddy/memory/2026-07-01.md` (+110 -0) ➕ `.codebuddy/memory/MEMORY.md` (+37 -0) 📝 `.gitignore` (+8 -1) ➕ `.playwright-mcp/console-2026-06-17T15-09-09-789Z.log` (+177 -0) ➕ `.playwright-mcp/page-2026-06-17T15-09-13-044Z.yml` (+7561 -0) ➕ `HINTS.md` (+55 -0) ➕ `benchmarks/bench_a2a_gemm.py` (+163 -0) ➕ `benchmarks/bench_a2a_only.py` (+141 -0) ➕ `benchmarks/bench_a2a_transpose_gemm.py` (+209 -0) ➕ `benchmarks/bench_ag_gemm.py` (+167 -0) ➕ `benchmarks/bench_fused_qkv_norm_a2a.py` (+201 -0) ➕ `benchmarks/bench_gemm_a2a_transpose.py` (+235 -0) ➕ `benchmarks/bench_gemm_rs.py` (+320 -0) ➕ `csrc/apis/a2a_gemm.hpp` (+100 -0) ➕ `csrc/apis/a2a_transpose_gemm.hpp` (+124 -0) ➕ `csrc/apis/ag_gemm.hpp` (+201 -0) 📝 `csrc/apis/gemm.hpp` (+7 -7) ➕ `csrc/apis/gemm_a2a_transpose.hpp` (+126 -0) _...and 80 more files_ </details> ### 📄 Description _No description provided_ --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-26 21:17:29 +00:00