msccl: use special send for LL on gfx950 (#1788)
[ROCm/rccl commit: ae9642d4bc]
This commit is contained in:
@@ -278,7 +278,7 @@ __device__ __forceinline__ void mscclRunInterpreter(
|
||||
}
|
||||
|
||||
#endif
|
||||
prims.send(srcOffset, thisNelem); // LL.send is the only situation where there is no barrier at the end.
|
||||
prims.mscclSend(srcOffset, thisNelem); // LL.send is the only situation where there is no barrier at the end.
|
||||
|
||||
#if defined(ENABLE_NPKIT) && defined(ENABLE_NPKIT_EVENT_MSCCL_SEND_EXIT)
|
||||
if (tid == 0) {
|
||||
|
||||
Reference in New Issue
Block a user