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 bugs

Browse Source
This commit is contained in:
Chenggang Zhao
2025-06-20 14:27:54 +08:00
parent 371df2da52
commit 74afd75df2
2 changed files with 2 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
csrc/kernels/internode.cu
View File

@@ -402,7 +402,7 @@ dispatch(int4* recv_x, float* recv_x_scales, int64_t* recv_topk_idx, float* recv
// NVL buffer layouts // NVL buffer layouts
// NOTES: `rs_wr_buffer_ptr` means "Read for Senders, Write for Receivers", `ws_rr_buffer_ptr` means "Write for Senders, Read for Receivers" // NOTES: `rs_wr_buffer_ptr` means "Read for Senders, Write for Receivers", `ws_rr_buffer_ptr` means "Write for Senders, Read for Receivers"
int rs_wr_rank = 0, ws_rr_rank = 0; int rs_wr_rank = 0, ws_rr_rank = 0;
if (warp_role == WarpRole::kRDMAAndNVLForwarder) if (warp_role == WarpRole::kRDMAAndNVLForwarder or warp_role == WarpRole::kForwarderCoordinator)
rs_wr_rank = nvl_rank, ws_rr_rank = lane_id < NUM_MAX_NVL_PEERS ? lane_id : 0; rs_wr_rank = nvl_rank, ws_rr_rank = lane_id < NUM_MAX_NVL_PEERS ? lane_id : 0;
if (warp_role == WarpRole::kNVLReceivers) if (warp_role == WarpRole::kNVLReceivers)
rs_wr_rank = target_rank, ws_rr_rank = nvl_rank; rs_wr_rank = target_rank, ws_rr_rank = nvl_rank;
@@ -595,7 +595,7 @@ dispatch(int4* recv_x, float* recv_x_scales, int64_t* recv_topk_idx, float* recv
while (__any_sync(0xffffffff, num_tokens_to_send > 0)) { while (__any_sync(0xffffffff, num_tokens_to_send > 0)) {
// Timeout check // Timeout check
if (clock64() - start_time > NUM_TIMEOUT_CYCLES and lane_id < kNumRDMARanks) { if (clock64() - start_time > NUM_TIMEOUT_CYCLES and lane_id < kNumRDMARanks) {
printf("DeepEP RDMA sender coordinator timeout, channel: %d, IB: %d, nvl %d, dst IB: %d, tail %d, num_tokens_to_send %d\n", printf("DeepEP RDMA sender coordinator timeout, channel: %d, IB: %d, nvl %d, dst IB: %d, tail: %d, tokens to send: %d\n",
channel_id, rdma_rank, nvl_rank, lane_id, last_issued_tail, num_tokens_to_send); channel_id, rdma_rank, nvl_rank, lane_id, last_issued_tail, num_tokens_to_send);
trap(); trap();
} }

1
csrc/kernels/launch.cuh
View File

@@ -78,7 +78,6 @@ cfg.dynamicSmemBytes = smem_size;
#define SWITCH_TYPES(case_macro) \ #define SWITCH_TYPES(case_macro) \
switch (type) { \ switch (type) { \
case CUDA_R_16BF: case_macro(nv_bfloat16); \ case CUDA_R_16BF: case_macro(nv_bfloat16); \
case CUDA_R_32F: case_macro(float); \
default: EP_HOST_ASSERT(false && "Unsupported type"); \ default: EP_HOST_ASSERT(false && "Unsupported type"); \
} while (false) } while (false)