[PR #78] [MERGED] Solving bank conflict via padding and TMA 3D store #235

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

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/DeepGEMM/pull/78
Author: @LyricZhao
Created: 4/3/2025
Status: ✅ Merged
Merged: 4/3/2025
Merged by: @LyricZhao

Base: main ← Head: tma-3d-padding


📝 Commits (3)

  • 6db7e18 Solve STSM bank conflict via padding and 3D TMA
  • 3a5539b Use c++20
  • d14962f Add DG_NVCC_OVERRIDE_CPP_STANDARD

📊 Changes

7 files changed (+121 additions, -51 deletions)

View changed files

📝 README.md (+1 -0)
📝 deep_gemm/include/deep_gemm/fp8_gemm.cuh (+48 -18)
📝 deep_gemm/include/deep_gemm/tma_utils.cuh (+25 -5)
📝 deep_gemm/include/deep_gemm/utils.cuh (+2 -2)
📝 deep_gemm/jit/compiler.py (+2 -1)
📝 deep_gemm/jit_kernels/gemm.py (+24 -13)
📝 deep_gemm/jit_kernels/m_grouped_gemm.py (+19 -12)

📄 Description

The optimization should make general cases 1% faster, cases with small Ks ~10% faster.


🔄 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/78 **Author:** [@LyricZhao](https://github.com/LyricZhao) **Created:** 4/3/2025 **Status:** ✅ Merged **Merged:** 4/3/2025 **Merged by:** [@LyricZhao](https://github.com/LyricZhao) **Base:** `main` ← **Head:** `tma-3d-padding` --- ### 📝 Commits (3) - [`6db7e18`](https://github.com/deepseek-ai/DeepGEMM/commit/6db7e1863b7a104a75fc50a123dd6931bd3f9b59) Solve STSM bank conflict via padding and 3D TMA - [`3a5539b`](https://github.com/deepseek-ai/DeepGEMM/commit/3a5539b7db73a1f2a90c21abd886ed1cf5520d45) Use `c++20` - [`d14962f`](https://github.com/deepseek-ai/DeepGEMM/commit/d14962f0727be82a941ab89a5da0f5f53f3f8c36) Add `DG_NVCC_OVERRIDE_CPP_STANDARD` ### 📊 Changes **7 files changed** (+121 additions, -51 deletions) <details> <summary>View changed files</summary> 📝 `README.md` (+1 -0) 📝 `deep_gemm/include/deep_gemm/fp8_gemm.cuh` (+48 -18) 📝 `deep_gemm/include/deep_gemm/tma_utils.cuh` (+25 -5) 📝 `deep_gemm/include/deep_gemm/utils.cuh` (+2 -2) 📝 `deep_gemm/jit/compiler.py` (+2 -1) 📝 `deep_gemm/jit_kernels/gemm.py` (+24 -13) 📝 `deep_gemm/jit_kernels/m_grouped_gemm.py` (+19 -12) </details> ### 📄 Description The optimization should make general cases 1% faster, cases with small Ks ~10% faster. --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>
bsakaguchi 2026-10-26 21:15:10 +00:00
Sign in to join this conversation.
No labels
pull-request
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
deepseek-ai/DeepGEMM#235
No description provided.