NCCL_TREES variable and rome model fixes (#856)

This commit is contained in:
akolliasAMD
2023-08-21 10:35:37 -06:00
zatwierdzone przez GitHub
rodzic 148e3430f4
commit d33cd5a233
8 zmienionych plików z 58 dodań i 60 usunięć
+4 -1
Wyświetl plik
@@ -1201,7 +1201,7 @@ static ncclResult_t getAlgoInfo(struct ncclInfo* info, int collNetTypeSupport, i
if (info->coll == ncclFuncAllToAllPivot) {
int pivotA2ANumUniRings = comm->topo->pivotA2ANumBiRings * 2;
info->nChannels = comm->nChannels / pivotA2ANumUniRings * pivotA2ANumUniRings;
} else if (info->coll == ncclFuncAllReduce && comm->topo->pivotA2ANumBiRings == 3) {
} else if (info->coll == ncclFuncAllReduce && comm->topo->pivotA2ANumBiRings == 3) {
static int userTuneInput = -2;
if (userTuneInput == -2) {
const char *protoStr = getenv("NCCL_PROTO");
@@ -1223,6 +1223,9 @@ static ncclResult_t getAlgoInfo(struct ncclInfo* info, int collNetTypeSupport, i
info->algorithm = NCCL_ALGO_RING;
}
}
} else if (info->coll == ncclFuncAllReduce && comm->topo->treeDefined == 1) {
info->algorithm = NCCL_ALGO_TREE;
info->nChannels = nc;
} else {
info->nChannels = nc;
}