Add option to force enable network transport on single node (#1046)

This commit is contained in:
Wenkai Du
2024-01-16 07:54:18 -08:00
committed by GitHub
parent 246dbd16d7
commit 261707d90a
+5
View File
@@ -771,6 +771,11 @@ ncclResult_t ncclTopoTrimSystem(struct ncclTopoSystem* system, struct ncclComm*
INFO(NCCL_GRAPH, "GDR is available on all GPUs");
}
if (rcclParamEnableIntranet() == 1) {
remove = 0;
system->type |= RCCL_TOPO_FORCE_INTRA;
}
comm->localRanks = system->nodes[GPU].count;
if (system->nodes[GPU].count == comm->nRanks && remove) {
for (int n=system->nodes[NET].count-1; n>=0; n--)