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

Improve AR performance

Browse Source
This commit is contained in:
Chenggang Zhao
2025-03-06 21:41:19 +08:00
parent 41385ba5b3
commit 1fc40d50f3
3 changed files with 8 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
csrc/config.hpp
View File

@@ -34,8 +34,12 @@ struct Config {
EP_HOST_ASSERT(num_max_nvl_chunked_send_tokens > 0 and num_max_nvl_chunked_recv_tokens > 0);
EP_HOST_ASSERT(num_max_nvl_chunked_send_tokens < num_max_nvl_chunked_recv_tokens);
EP_HOST_ASSERT(num_max_rdma_chunked_send_tokens > 0 and num_max_rdma_chunked_recv_tokens > 0);
EP_HOST_ASSERT(num_max_rdma_chunked_send_tokens < num_max_rdma_chunked_recv_tokens);
// Ceil up RDMA buffer size
this->num_max_rdma_chunked_recv_tokens = align<int>(num_max_rdma_chunked_recv_tokens, num_max_rdma_chunked_send_tokens);
EP_HOST_ASSERT(num_max_rdma_chunked_send_tokens < num_max_rdma_chunked_recv_tokens);
// NOTES: this assertion is related to RDMA lazy head update, we must ensure senders always have space to push
EP_HOST_ASSERT(num_max_rdma_chunked_send_tokens <= num_max_rdma_chunked_recv_tokens / 2);
}
size_t get_nvl_buffer_size_hint(size_t hidden_bytes, int num_ranks) const {