[PR #451] [Bugfix] Reset MegaMoE combine readiness for CUDA graph replay #456

New issue

Open

opened 2026-10-26 21:18:03 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-26 21:18:03 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/DeepGEMM/pull/451
Author: @foreverrookie
Created: 9/17/2026
Status: 🔄 Open

Base: main ← Head: fix/megamoe-gridid

📝 Commits (2)

d7442af Fix MegaMoE combine readiness on CUDA graph replay
7e924a8 Fix BF16 MegaMoE readiness on CUDA graph replay

📊 Changes

3 files changed (+12 additions, -5 deletions)

View changed files

📝 deep_gemm/include/deep_gemm/impls/sm100_bf16_mega_moe.cuh (+5 -2)
📝 deep_gemm/include/deep_gemm/impls/sm100_fp8_fp4_mega_moe.cuh (+5 -2)
📝 deep_gemm/include/deep_gemm/layout/mega_moe.cuh (+2 -1)

📄 Description

CUDA graph replay can reuse %gridid, allowing MegaMoE to accept stale combine-readiness tags and read old or partially updated results.

Reset local peer-readiness flags before the existing pull barrier in both FP8/FP4 and BF16 kernels. The workspace layout and existing barriers are unchanged.

Validation:

FP8: vLLM adapter tests on 8 B300 ranks, including input staging, unequal/empty source ranks, and changed-input graph replay: 3 tests passed per rank.
BF16: standalone tests on 2 B300 GPUs, with balanced, unequal, and empty-source cases. Eager execution passed before and after; graph replay failed before the fix, and all 12 changed-input replays per case match eager output exactly after it.

AI assistance was used for the fix and validation.

_{🔄 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/451 **Author:** [@foreverrookie](https://github.com/foreverrookie) **Created:** 9/17/2026 **Status:** 🔄 Open **Base:** `main` ← **Head:** `fix/megamoe-gridid` --- ### 📝 Commits (2) - [`d7442af`](https://github.com/deepseek-ai/DeepGEMM/commit/d7442afe88a194b99234710a5467478c15574191) Fix MegaMoE combine readiness on CUDA graph replay - [`7e924a8`](https://github.com/deepseek-ai/DeepGEMM/commit/7e924a8c4be7458d5f18175bf00da8644be17c04) Fix BF16 MegaMoE readiness on CUDA graph replay ### 📊 Changes **3 files changed** (+12 additions, -5 deletions) <details> <summary>View changed files</summary> 📝 `deep_gemm/include/deep_gemm/impls/sm100_bf16_mega_moe.cuh` (+5 -2) 📝 `deep_gemm/include/deep_gemm/impls/sm100_fp8_fp4_mega_moe.cuh` (+5 -2) 📝 `deep_gemm/include/deep_gemm/layout/mega_moe.cuh` (+2 -1) </details> ### 📄 Description CUDA graph replay can reuse `%gridid`, allowing MegaMoE to accept stale combine-readiness tags and read old or partially updated results. Reset local peer-readiness flags before the existing pull barrier in both FP8/FP4 and BF16 kernels. The workspace layout and existing barriers are unchanged. Validation: - FP8: vLLM adapter tests on 8 B300 ranks, including input staging, unequal/empty source ranks, and changed-input graph replay: 3 tests passed per rank. - BF16: standalone tests on 2 B300 GPUs, with balanced, unequal, and empty-source cases. Eager execution passed before and after; graph replay failed before the fix, and all 12 changed-input replays per case match eager output exactly after it. AI assistance was used for the fix and validation. --- <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

2026-10-26 21:18:03 +00:00

bsakaguchi referenced this issue

2026-10-26 21:18:06 +00:00

[PR #456] [Fix] Reject FP32 bmk,bnk->mn einsum without C instead of crashing #461