[PR #536] Add NVSHMEM guards for low_latency mask buffer functions #619

New issue

Open

opened 2026-10-27 21:04:35 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-27 21:04:35 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/DeepEP/pull/536
Author: @yurekami
Created: 12/25/2025
Status: 🔄 Open

Base: main ← Head: fix/nvshmem-guards-for-mask-buffer

📝 Commits (1)

a5096e5 Add NVSHMEM guards for low_latency mask buffer functions

📊 Changes

1 file changed (+12 additions, -0 deletions)

View changed files

📝 csrc/deep_ep.cpp (+12 -0)

📄 Description

Summary

Add #ifndef DISABLE_NVSHMEM guards to mask buffer functions
Fixes undefined symbol errors when building without NVSHMEM

Problem

When building DeepEP with DISABLE_NVSHMEM (for systems without NVSHMEM like sm86 A6000 or sm80 A100 without NVLink/RDMA), importing the module fails with:

ImportError: undefined symbol: _ZN7deep_ep12internode_ll17query_mask_bufferE...

This is because the following functions reference internode_ll:: symbols without NVSHMEM guards:

low_latency_update_mask_buffer()
low_latency_query_mask_buffer()
low_latency_clean_mask_buffer()

Solution

Add #ifndef DISABLE_NVSHMEM guards to these functions, matching the pattern used elsewhere in the codebase.

Test plan

Build with DISABLE_NVSHMEM=1 TORCH_CUDA_ARCH_LIST=8.0 on sm80 system
Verify import deep_ep succeeds
Build normally with NVSHMEM and verify functionality

Fixes #481

🤖 Generated with Claude Code

_{🔄 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/536 **Author:** [@yurekami](https://github.com/yurekami) **Created:** 12/25/2025 **Status:** 🔄 Open **Base:** `main` ← **Head:** `fix/nvshmem-guards-for-mask-buffer` --- ### 📝 Commits (1) - [`a5096e5`](https://github.com/deepseek-ai/DeepEP/commit/a5096e53922553c3de2cbfdb38260a35c92291d3) Add NVSHMEM guards for low_latency mask buffer functions ### 📊 Changes **1 file changed** (+12 additions, -0 deletions) <details> <summary>View changed files</summary> 📝 `csrc/deep_ep.cpp` (+12 -0) </details> ### 📄 Description ## Summary - Add `#ifndef DISABLE_NVSHMEM` guards to mask buffer functions - Fixes undefined symbol errors when building without NVSHMEM ## Problem When building DeepEP with `DISABLE_NVSHMEM` (for systems without NVSHMEM like sm86 A6000 or sm80 A100 without NVLink/RDMA), importing the module fails with: ``` ImportError: undefined symbol: _ZN7deep_ep12internode_ll17query_mask_bufferE... ``` This is because the following functions reference `internode_ll::` symbols without NVSHMEM guards: - `low_latency_update_mask_buffer()` - `low_latency_query_mask_buffer()` - `low_latency_clean_mask_buffer()` ## Solution Add `#ifndef DISABLE_NVSHMEM` guards to these functions, matching the pattern used elsewhere in the codebase. ## Test plan - [ ] Build with `DISABLE_NVSHMEM=1 TORCH_CUDA_ARCH_LIST=8.0` on sm80 system - [ ] Verify `import deep_ep` succeeds - [ ] Build normally with NVSHMEM and verify functionality Fixes #481 🤖 Generated with [Claude Code](https://claude.com/claude-code) --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi added the

pull-request

label

2026-10-27 21:04:35 +00:00

bsakaguchi referenced this issue

2026-10-27 21:05:03 +00:00

[PR #619] Small refactoring #667