Improve MSCCL algorithms (#1023)

[ROCm/rccl commit: 0a53077c9c]
2024-01-04 06:51:34 +08:00
parent 0c1f773021
commit e3d45f9de4
11 changed files with 8601 additions and 6 deletions
@@ -454,7 +454,7 @@ __device__ __forceinline__ void mscclRunInterpreter(
          }
 #endif
        }
-        else if (fullOps && t->type == MSCCL_LOCAL_COPY)
+        else if (t->type == MSCCL_LOCAL_COPY)
          prims.localCopy(srcPointer+srcOffset, dstPointer+dstOffset, thisNelem);
        else
          return;