[PR #227] [MERGED] Use larger MMA shape to optimize sm100_fp8_mqa_logits #299

New issue

Closed

opened 2026-10-26 21:15:58 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-26 21:15:58 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/DeepGEMM/pull/227
Author: @RayWang96
Created: 11/14/2025
Status: ✅ Merged
Merged: 11/14/2025
Merged by: @LyricZhao

Base: main ← Head: larger_umma

📝 Commits (1)

7ccbf19 Use larger MMA shape

📊 Changes

2 files changed (+48 additions, -58 deletions)

View changed files

📝 csrc/jit_kernels/impls/smxx_fp8_mqa_logits.hpp (+1 -1)
📝 deep_gemm/include/deep_gemm/impls/sm100_fp8_mqa_logits.cuh (+47 -57)

📄 Description

10% speedup, it is now bounded by the cuda core and the number of registers.

_{🔄 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/227 **Author:** [@RayWang96](https://github.com/RayWang96) **Created:** 11/14/2025 **Status:** ✅ Merged **Merged:** 11/14/2025 **Merged by:** [@LyricZhao](https://github.com/LyricZhao) **Base:** `main` ← **Head:** `larger_umma` --- ### 📝 Commits (1) - [`7ccbf19`](https://github.com/deepseek-ai/DeepGEMM/commit/7ccbf1926f464c5aab7cb7473fcae35b09f8f673) Use larger MMA shape ### 📊 Changes **2 files changed** (+48 additions, -58 deletions) <details> <summary>View changed files</summary> 📝 `csrc/jit_kernels/impls/smxx_fp8_mqa_logits.hpp` (+1 -1) 📝 `deep_gemm/include/deep_gemm/impls/sm100_fp8_mqa_logits.cuh` (+47 -57) </details> ### 📄 Description 10% speedup, it is now bounded by the cuda core and the number of registers. --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-26 21:15:58 +00:00