[PR #413] [MERGED] Fix hidden_size % 128 != 0 in intranode kernels #556

New issue

Closed

opened 2026-10-27 21:04:00 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-27 21:04:00 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/DeepEP/pull/413
Author: @sphish
Created: 9/17/2025
Status: ✅ Merged
Merged: 9/17/2025
Merged by: @sphish

Base: main ← Head: fix-unaligned-hidden

📝 Commits (5)

2444c80 Fix hidden_size % 128 != 0
9ce35d3 Add align_down() function
45d2606 Use the full warp to wait TMA store
7179dde Support arbitrary hidden sizes in fp8 cast
1cbc769 lint

📊 Changes

6 files changed (+62 additions, -38 deletions)

View changed files

📝 csrc/config.hpp (+8 -3)
📝 csrc/kernels/internode.cu (+2 -2)
📝 csrc/kernels/internode_ll.cu (+2 -2)
📝 csrc/kernels/intranode.cu (+27 -24)
📝 csrc/kernels/utils.cuh (+6 -1)
📝 tests/utils.py (+17 -6)

📄 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/DeepEP/pull/413 **Author:** [@sphish](https://github.com/sphish) **Created:** 9/17/2025 **Status:** ✅ Merged **Merged:** 9/17/2025 **Merged by:** [@sphish](https://github.com/sphish) **Base:** `main` ← **Head:** `fix-unaligned-hidden` --- ### 📝 Commits (5) - [`2444c80`](https://github.com/deepseek-ai/DeepEP/commit/2444c80219b362d49c57cd64257f8a8bb5a3ca50) Fix hidden_size % 128 != 0 - [`9ce35d3`](https://github.com/deepseek-ai/DeepEP/commit/9ce35d357713322fbc04361fd7a806abdf49a875) Add `align_down()` function - [`45d2606`](https://github.com/deepseek-ai/DeepEP/commit/45d260641cd77fde893e92790dcd2bef366cecf4) Use the full warp to wait TMA store - [`7179dde`](https://github.com/deepseek-ai/DeepEP/commit/7179dded85618fe06d953692ce4d70baf1645269) Support arbitrary hidden sizes in fp8 cast - [`1cbc769`](https://github.com/deepseek-ai/DeepEP/commit/1cbc7697e5f918f2cbdff7a16e2e9888e20755de) lint ### 📊 Changes **6 files changed** (+62 additions, -38 deletions) <details> <summary>View changed files</summary> 📝 `csrc/config.hpp` (+8 -3) 📝 `csrc/kernels/internode.cu` (+2 -2) 📝 `csrc/kernels/internode_ll.cu` (+2 -2) 📝 `csrc/kernels/intranode.cu` (+27 -24) 📝 `csrc/kernels/utils.cuh` (+6 -1) 📝 `tests/utils.py` (+17 -6) </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

2026-10-27 21:04:00 +00:00