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/info.h

@@ -1,5 +1,5 @@
 /*************************************************************************
- * Copyright (c) 2019-2020, NVIDIA CORPORATION. All rights reserved.
+ * Copyright (c) 2019-2021, NVIDIA CORPORATION. All rights reserved.
  *
  * See LICENSE.txt for license information
  ************************************************************************/
@@ -18,8 +18,7 @@ typedef enum {
   ncclPatternTreeUp,
   ncclPatternTreeDown,
   ncclPatternTreeUpDown,
-  ncclPatternCollTreeUp,
-  ncclPatternCollTreeDown
+  ncclPatternCollTreeUpDown
 } ncclPattern_t;
 
 // Used to pass NCCL call information between functions
@@ -49,6 +48,8 @@ struct ncclInfo {
   int nchunksPerLoop;
   ssize_t sendbytes;
   ssize_t recvbytes;
+  int recvChunkSize;
+  int sendChunkSize;
   uint32_t delta;
   int channelId;
 };