[PR #564] [MERGED] Nvshmem 3 5 runtime fixes #637

New issue

Closed

opened 2026-10-27 21:04:46 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-27 21:04:46 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/DeepEP/pull/564
Author: @seth-howell
Created: 1/16/2026
Status: ✅ Merged
Merged: 1/20/2026
Merged by: @sphish

Base: main ← Head: nvshmem_3_5_runtime_fixes

📝 Commits (3)

18b7e99 DeepEP/csrc: Force CUDA RDC when NVSHMEM is used.
ba468e1 DeepEP/csrc: Update RC selection for NVSHMEM 3.5.
cddd60f Check Format

📊 Changes

2 files changed (+25 additions, -3 deletions)

View changed files

📝 csrc/kernels/configs.cuh (+7 -0)
📝 csrc/kernels/ibgda_device.cuh (+18 -3)

📄 Description

The addition of QPair-Specific APIs modified the way default qpairs are stored. This change allows DeepEP to work with the new version while maintaining compile-time backwards compatibility for the previous layout.

This is likely the issue encountered in issue #553

_{🔄 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/564 **Author:** [@seth-howell](https://github.com/seth-howell) **Created:** 1/16/2026 **Status:** ✅ Merged **Merged:** 1/20/2026 **Merged by:** [@sphish](https://github.com/sphish) **Base:** `main` ← **Head:** `nvshmem_3_5_runtime_fixes` --- ### 📝 Commits (3) - [`18b7e99`](https://github.com/deepseek-ai/DeepEP/commit/18b7e99274f4985c5c6b29f25efdb82fa93ee395) DeepEP/csrc: Force CUDA RDC when NVSHMEM is used. - [`ba468e1`](https://github.com/deepseek-ai/DeepEP/commit/ba468e14e2e6eefc492b51a3b7d512d4ce8182b7) DeepEP/csrc: Update RC selection for NVSHMEM 3.5. - [`cddd60f`](https://github.com/deepseek-ai/DeepEP/commit/cddd60f17f95c9ee19bf8ca73a19fab4cbba8ae6) Check Format ### 📊 Changes **2 files changed** (+25 additions, -3 deletions) <details> <summary>View changed files</summary> 📝 `csrc/kernels/configs.cuh` (+7 -0) 📝 `csrc/kernels/ibgda_device.cuh` (+18 -3) </details> ### 📄 Description The addition of QPair-Specific APIs modified the way default qpairs are stored. This change allows DeepEP to work with the new version while maintaining compile-time backwards compatibility for the previous layout. This is likely the issue encountered in issue #553 --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-27 21:04:46 +00:00