Introduce multi-rank support per device.

This is a single commit of the source code changes required to
introduce support for multiple ranks per device.
A new interface (ncclCommRankInitMulti) has to be used to make use of
this new feature.

tools/MultiRank/Makefile

@@ -0,0 +1,24 @@
+# Copyright (c) 2020 Advanced Micro Devices, Inc. All rights reserved.
+
+# Set to where RCCL is installed
+RCCL_INSTALL=../../build/release
+
+HIP_PATH?= $(wildcard /opt/rocm/hip)
+ifeq (,$(HIP_PATH))
+HIP_PATH=../../..
+endif
+HIPCC=$(HIP_PATH)/bin/hipcc
+
+EXE=rccl-allreduce-multirank rccl-reducescatter-multirank
+CXXFLAGS = -std=c++11 -O3 -I$(RCCL_INSTALL)/include/rccl/ -L$(RCCL_INSTALL) -lrccl
+
+all: $(EXE)
+
+rccl-allreduce-multirank: rccl-allreduce-multirank.cc $(shell find -regex ".*\.\hpp")
+	$(HIPCC) $(CXXFLAGS) $< -o $@
+
+rccl-reducescatter-multirank: rccl-reducescatter-multirank.cc $(shell find -regex ".*\.\hpp")
+	$(HIPCC) $(CXXFLAGS) $< -o $@
+
+clean:
+	rm -f *.o $(EXE)