AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Fix NPKit for SendRecv (#1651)

Browse Source 
[ROCm/rccl commit: ac8ec4c08c]
This commit is contained in:
Bertan Dogancay
2025-04-21 12:34:47 -04:00
committed by GitHub
parent 58ee618194
commit aac829125c
2 changed files with 3 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rccl/src/device/sendrecv.h
+2 -2
View File
@@ -23,7 +23,7 @@ struct RunWorkBatch<ncclFuncSendRecv, T, RedOp, NCCL_ALGO_RING, NCCL_PROTO_SIMPL
#if defined(ENABLE_NPKIT)
bool isNpKitThread = (tid == 0);
int npKitCtxIdx = blockIdx.x + group;
int npKitCtxIdx = blockIdx.x * NCCL_MAX_DEV_WORK_P2P_ELEMENTS + group;
#endif
#if defined(ENABLE_NPKIT) && defined(ENABLE_NPKIT_EVENT_TIME_SYNC_CPU)
@@ -81,7 +81,7 @@ struct RunWorkBatch<ncclFuncSendRecv, T, RedOp, NCCL_ALGO_RING, NCCL_PROTO_SIMPL
#if defined(ENABLE_NPKIT)
bool isNpKitThread = (tid == 0);
int npKitCtxIdx = blockIdx.x + group;
int npKitCtxIdx = blockIdx.x * NCCL_MAX_DEV_WORK_P2P_ELEMENTS + group;
#endif
#if defined(ENABLE_NPKIT) && defined(ENABLE_NPKIT_EVENT_TIME_SYNC_CPU)