Skip to content

GitLab

Unverified Commit 77f97f79 authored by Shangyan Zhou's avatar Shangyan Zhou Committed by GitHub
Browse files

Fix the tail loading issue. (#219) 

* Fix the tail loading issue.

* Modify the sync offset.
parent dd133d39
Hide whitespace changes
Inline Side-by-side
Showing with 1 addition and 1 deletion
csrc/kernels/internode.cu
View file @ 77f97f79
...@@ -605,7 +605,7 @@ dispatch(int4* recv_x, float* recv_x_scales, int64_t* recv_topk_idx, float* recv ...@@ -605,7 +605,7 @@ dispatch(int4* recv_x, float* recv_x_scales, int64_t* recv_topk_idx, float* recv
// Read progress // Read progress
auto synced_last_issued_tail = __shfl_sync(0xffffffff, last_issued_tail, dst_rdma_rank); auto synced_last_issued_tail = __shfl_sync(0xffffffff, last_issued_tail, dst_rdma_rank);
auto processed_tail = ld_acquire_cta(const_cast<const int*>(rdma_send_channel_tail + dst_rdma_rank)); auto processed_tail = __shfl_sync(0xffffffff, ld_acquire_cta(const_cast<const int*>(rdma_send_channel_tail + dst_rdma_rank)), 0);
auto num_tokens_processed = processed_tail - synced_last_issued_tail; auto num_tokens_processed = processed_tail - synced_last_issued_tail;
if (num_tokens_processed != synced_num_tokens_to_send and num_tokens_processed < num_max_rdma_chunked_send_tokens) if (num_tokens_processed != synced_num_tokens_to_send and num_tokens_processed < num_max_rdma_chunked_send_tokens)
continue; continue;
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment