DeepSeek/DeepEP
1
0
Fork 0
mirror of https://github.com/deepseek-ai/DeepEP synced 2025-06-26 18:28:11 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Fix warp synchronization. (#215)

Browse Source 
* Fix warp synchronization.

* Another fix.
This commit is contained in:
Shangyan Zhou 2025-06-16 17:05:11 +08:00 committed by GitHub
parent 8aaddf76ae
commit dd133d39bc
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 2 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
csrc/kernels/internode.cu
View File

@ -567,6 +567,7 @@ dispatch(int4* recv_x, float* recv_x_scales, int64_t* recv_topk_idx, float* recv
// Update last token tail
if (last_rdma_tail_idx >= 0)
st_release_cta(const_cast<const int*>(rdma_send_channel_tail + lane_id), last_rdma_tail_idx + 1);
__syncwarp();
// Release sequential lock
lane_id == 0 ? (rdma_send_next_token_idx += 1) : 0;
@ -633,6 +634,7 @@ dispatch(int4* recv_x, float* recv_x_scales, int64_t* recv_topk_idx, float* recv
nvshmemi_ibgda_amo_nonfetch_add(rdma_channel_tail.buffer(rdma_rank), num_tokens_to_issue,
translate_dst_rdma_rank<kLowLatencyMode>(dst_rdma_rank, nvl_rank), channel_id, dst_rdma_rank == rdma_rank);
}
__syncwarp();
}
}
} else if (warp_role == WarpRole::kRDMAAndNVLForwarder) {