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

Support zero-copy for low-latency combine

Browse Source
This commit is contained in:
Chenggang Zhao
2025-03-18 15:41:50 +08:00
parent 82dcf48fd3
commit dcaf73e5ff
7 changed files with 80 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
csrc/config.hpp
View File

@@ -102,6 +102,9 @@ struct LowLatencyBuffer {
void* combine_rdma_recv_data_buffer = nullptr;
int* combine_rdma_recv_flag_buffer = nullptr;
void* combine_rdma_send_buffer_data_start = nullptr;
int num_bytes_per_combine_msg = 0;
std::pair<int*, int> clean_meta() {
EP_HOST_ASSERT(dispatch_rdma_recv_count_buffer == combine_rdma_recv_flag_buffer);
return {dispatch_rdma_recv_count_buffer, num_clean_int};
@@ -163,7 +166,9 @@ struct LowLatencyLayout {
advance<int*>(rdma_buffer, send_buffer_bytes * 2 + recv_buffer_bytes * 2 + signaling_buffer_bytes * i),
advance(rdma_buffer, send_buffer_bytes * i),
advance(rdma_buffer, send_buffer_bytes * 2 + recv_buffer_bytes * i),
advance<int*>(rdma_buffer, send_buffer_bytes * 2 + recv_buffer_bytes * 2 + signaling_buffer_bytes * i)
advance<int*>(rdma_buffer, send_buffer_bytes * 2 + recv_buffer_bytes * 2 + signaling_buffer_bytes * i),
advance(rdma_buffer, send_buffer_bytes * i + sizeof(int4)),
static_cast<int>(num_bytes_per_combine_msg)
};
}
}