diff --git a/src/collectives/device/primitives.h b/src/collectives/device/primitives.h index 81a4d4cb7f..8c63a6a4f1 100644 --- a/src/collectives/device/primitives.h +++ b/src/collectives/device/primitives.h @@ -193,10 +193,11 @@ class ncclPrimitives { if (SEND) __threadfence_system(); if (tid == 0) FOR_SEND(postSend); if (tid == 0) FOR_RECV(postRecv); + + for (int i=0; i