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

Add low-latency kernel PCIe usage flag (#195)

Browse Source 
* Add low-latency kernel usage flag

* Update comments
This commit is contained in:
Chenggang Zhao
2025-06-09 14:37:13 +08:00
committed by GitHub
parent 564e375234
commit 0d1a855d81
6 changed files with 57 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
csrc/kernels/api.cuh
View File

@@ -138,7 +138,8 @@ void dispatch(void* packed_recv_x, float* packed_recv_x_scales,
int* next_clean, int num_next_clean_int,
int num_tokens, int hidden, int num_max_dispatch_tokens_per_rank,
int num_topk, int num_experts, int rank, int num_ranks, bool use_fp8,
void* workspace, cudaStream_t stream, int phases);
void* workspace, int* usage_flag,
cudaStream_t stream, int phases);
void combine(void* combined_x,
void* rdma_recv_x, int* rdma_recv_flag, void* rdma_send_x,
@@ -147,8 +148,8 @@ void combine(void* combined_x,
int* next_clean, int num_next_clean_int,
int num_combined_tokens, int hidden, int num_max_dispatch_tokens_per_rank,
int num_topk, int num_experts, int rank, int num_ranks,
void* workspace, cudaStream_t stream,
int phases, bool zero_copy);
void* workspace, int* usage_flag,
cudaStream_t stream, int phases, bool zero_copy);
} // namespace internode_ll