add binary tree

In addition, introduce the ability to have 2 trees at the same time.
Only for allreduce at the moment.


[ROCm/rccl commit: 65e2ae20e5]

projects/rccl/src/enqueue.cc

@@ -567,7 +567,13 @@ comp_next:
   // Set nstepsPerLoop and nchunksPerLoop
   NCCLCHECK(getPatternInfo(info));
   NCCLCHECK(getLoopInfo(info));
-  if (info->comm->topo->pivotA2ANumBiRings == 3 ) work->pad_0 = 1;
+  if (info->comm->topo->pivotA2ANumBiRings == 3 ) {
+    if (ncclTypeSize(info->datatype)*info->count > 65536) {
+      work->pad_0 = 1;
+    } else {
+      work->pad_0 = 2;
+    }
+  }
   work->opCount = info->opCount;
   work->header.type = ncclWorkTypeColl;
   work->sendbuff = info->sendbuff;