[PR #88] [MERGED] Support TMA multicast on B with m_grouped_gemm_contiguous. #241

New issue

Closed

opened 2026-10-26 21:15:13 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-26 21:15:13 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/DeepGEMM/pull/88
Author: @yukuai26
Created: 4/18/2025
Status: ✅ Merged
Merged: 4/21/2025
Merged by: @LyricZhao

Base: main ← Head: MoE_scheduler_with_TMA_multicast_compatibility

📝 Commits (4)

2752a67 Support TMA multicast on B with m_grouped_gemm_contiguous.
a51629d Code polishing
594020e Code polishing x2
c997567 Code polishing x3

📊 Changes

5 files changed (+74 additions, -31 deletions)

View changed files

📝 README.md (+3 -2)
📝 deep_gemm/include/deep_gemm/fp8_gemm.cuh (+14 -6)
📝 deep_gemm/include/deep_gemm/scheduler.cuh (+14 -2)
📝 deep_gemm/jit_kernels/gemm.py (+1 -2)
📝 tests/test_core.py (+42 -19)

📄 Description

Add function is_tma_multicast_valid() in scheduler.cuh to support TMA multicast on B with m_grouped_gemm_contiguous.

Change test_m_grouped_gemm_contiguous in test_core.py to better simulate real-world usage.

_{🔄 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/88 **Author:** [@yukuai26](https://github.com/yukuai26) **Created:** 4/18/2025 **Status:** ✅ Merged **Merged:** 4/21/2025 **Merged by:** [@LyricZhao](https://github.com/LyricZhao) **Base:** `main` ← **Head:** `MoE_scheduler_with_TMA_multicast_compatibility` --- ### 📝 Commits (4) - [`2752a67`](https://github.com/deepseek-ai/DeepGEMM/commit/2752a67aadabdedc60d716a8229dad2f16905991) Support TMA multicast on B with m_grouped_gemm_contiguous. - [`a51629d`](https://github.com/deepseek-ai/DeepGEMM/commit/a51629ddf968ada2cb4e8e7365ffb14294b629d2) Code polishing - [`594020e`](https://github.com/deepseek-ai/DeepGEMM/commit/594020e6af9f0218b1ced2a9d09850797baf179a) Code polishing x2 - [`c997567`](https://github.com/deepseek-ai/DeepGEMM/commit/c99756778d06bb08ad201de17ffba174565473ac) Code polishing x3 ### 📊 Changes **5 files changed** (+74 additions, -31 deletions) <details> <summary>View changed files</summary> 📝 `README.md` (+3 -2) 📝 `deep_gemm/include/deep_gemm/fp8_gemm.cuh` (+14 -6) 📝 `deep_gemm/include/deep_gemm/scheduler.cuh` (+14 -2) 📝 `deep_gemm/jit_kernels/gemm.py` (+1 -2) 📝 `tests/test_core.py` (+42 -19) </details> ### 📄 Description Add function is_tma_multicast_valid() in scheduler.cuh to support TMA multicast on B with m_grouped_gemm_contiguous. Change test_m_grouped_gemm_contiguous in test_core.py to better simulate real-world usage. --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-26 21:15:13 +00:00