2.12.7-1

Add network communication through another GPU connected with NVLink
(PXN).
Add aggregation of messages coming from different local GPUs through
PXN and going to the same destination.
Add new v5 plugin API with grouped receives and tags.
Add compat for v4 plugins.
Add naming of NCCL threads to help debugging.
Fix NVLink detection and avoid data corruption when some NVLinks are
down.
Add support for Relaxed Ordering for IB.
Add profiling and timing infrastructure.

src/include/info.h

@@ -1,5 +1,5 @@
 /*************************************************************************
- * Copyright (c) 2019-2021, NVIDIA CORPORATION. All rights reserved.
+ * Copyright (c) 2019-2022, NVIDIA CORPORATION. All rights reserved.
  *
  * See LICENSE.txt for license information
  ************************************************************************/
@@ -11,7 +11,7 @@
 #include "devcomm.h"
 #include "collectives.h"
 
-typedef enum {
+typedef enum : uint8_t {
   ncclPatternRing,
   ncclPatternRingTwice,
   ncclPatternPipelineFrom,
@@ -19,7 +19,9 @@ typedef enum {
   ncclPatternTreeUp,
   ncclPatternTreeDown,
   ncclPatternTreeUpDown,
-  ncclPatternCollTreeUpDown
+  ncclPatternCollTreeUpDown,
+  ncclPatternSend,
+  ncclPatternRecv
 } ncclPattern_t;
 
 // Used to pass NCCL call information between functions
@@ -32,7 +34,7 @@ struct ncclInfo {
   size_t count;
   ncclDataType_t datatype;
   ncclRedOp_t op;
-  int root;
+  int root; // peer for p2p operations
   ncclComm_t comm;
   cudaStream_t stream;
   // Algorithm details
@@ -48,11 +50,7 @@ struct ncclInfo {
   size_t nBytes;
   int nstepsPerLoop;
   int nchunksPerLoop;
-  ssize_t sendbytes;
-  ssize_t recvbytes;
-  int recvChunkSize;
-  int sendChunkSize;
-  uint32_t delta;
+  int chunkSize;
   int channelId;
 };