2.9.6-1

Add support for CUDA graphs.
Fuse BCM Gen4 switches to avoid suboptimal performance on some platforms. Issue #439.
Fix bootstrap issue caused by connection reordering.
Fix CPU locking block.
Improve CollNet algorithm.
Improve performance on DGX A100 for communicators with only one GPU per node.

src/include/graph.h

@@ -1,5 +1,5 @@
 /*************************************************************************
- * Copyright (c) 2016-2020, NVIDIA CORPORATION. All rights reserved.
+ * Copyright (c) 2016-2021, NVIDIA CORPORATION. All rights reserved.
  *
  * See LICENSE.txt for license information
  ************************************************************************/
@@ -28,7 +28,7 @@ ncclResult_t ncclTopoTrimSystem(struct ncclTopoSystem* system, struct ncclComm*
 ncclResult_t ncclTopoComputeP2pChannels(struct ncclComm* comm);
 
 // Query topology
-ncclResult_t ncclTopoGetNetDev(struct ncclTopoSystem* system, int rank, struct ncclTopoGraph* graph, int channelId, int* net);
+ncclResult_t ncclTopoGetNetDev(struct ncclTopoSystem* system, int rank, struct ncclTopoGraph* graph, int channelId, int rr, int* net);
 ncclResult_t ncclTopoCheckP2p(struct ncclTopoSystem* system, int64_t id1, int64_t id2, int* p2p, int *read, int* intermediateRank);
 ncclResult_t ncclTopoCheckGdr(struct ncclTopoSystem* topo, int64_t busId, int netDev, int read, int* useGdr);
 
@@ -91,13 +91,11 @@ struct ncclTopoRanks {
 };
 
 ncclResult_t ncclTopoPreset(struct ncclComm* comm,
-    struct ncclTopoGraph* treeGraph, struct ncclTopoGraph* ringGraph, struct ncclTopoGraph* collNetGraph,
+    struct ncclTopoGraph* treeGraph, struct ncclTopoGraph* ringGraph,
     struct ncclTopoRanks* topoRanks);
 
 ncclResult_t ncclTopoPostset(struct ncclComm* comm, int* firstRanks, int* treePatterns,
-    struct ncclTopoRanks** allTopoRanks, int* rings);
-
-ncclResult_t ncclTopoConnectCollNet(struct ncclComm* comm, struct ncclTopoGraph* collNetGraph, int rank);
+    struct ncclTopoRanks** allTopoRanks, int* rings, struct ncclTopoGraph* collNetGraph);
 
 ncclResult_t ncclTopoTuneModel(struct ncclComm* comm, int minCompCap, int maxCompCap, struct ncclTopoGraph* treeGraph, struct ncclTopoGraph* ringGraph, struct ncclTopoGraph* collNetGraph);
 #include "info.h"