Add new NVLS algorithm for allreduce using NVLink SHARP (intra-node only).
Add new config options: cgaClusterSize, minCTAs, maxCTAs, netName.
Enable LL128 when we use PXN to close rings.
NVTX3 includes update.
Fix crash when one CollNet (SHARP) rail fails to initialize.
Este commit está contenido en:
Sylvain Jeaugey
2023-02-27 02:48:21 -08:00
padre f3d5166783
commit 5d3ab08b69
Se han modificado 72 ficheros con 4541 adiciones y 2391 borrados
-1
Ver fichero
@@ -765,7 +765,6 @@ ncclResult_t ncclTopoCompute(ncclTopoSystem* system, struct ncclTopoGraph* graph
if (ngpus == 1) if (graph->pattern != NCCL_TOPO_PATTERN_RING) graph->pattern = NCCL_TOPO_PATTERN_TREE;
// SPLIT_TREE works better on older archs.
int ccMin;
NCCLCHECK(ncclTopoGetCompCap(system, &ccMin, NULL));