Benchmarks: Add Benchmark - Revise and add rccl microbenchmark for rocm (#143)

**Description**
Add rccl bandwidth microbenchmark for rocm.

**Major Revision**
- Register rccl-bw benchmark.

examples/benchmarks/nccl_bw_performance.py

@@ -1,7 +1,7 @@
 # Copyright (c) Microsoft Corporation.
 # Licensed under the MIT license.
 
-"""Micro benchmark example for NCCL Bus bandwidth performance.
+"""Micro benchmark example for NCCL/RCCL Bus bandwidth performance.
 
 Commands to run:
   python3 examples/benchmarks/nccl_bw_performance.py
@@ -14,7 +14,10 @@ if __name__ == '__main__':
     context = BenchmarkRegistry.create_benchmark_context(
         'nccl-bw', platform=Platform.CUDA, parameters='--operations allreduce'
     )
-
+    # For ROCM environment, please specify the benchmark name and the platform as the following.
+    # context = BenchmarkRegistry.create_benchmark_context(
+    #     'rccl-bw', platform=Platform.ROCM, parameters='--operations allreduce --maxbytes 128M'
+    # )
     benchmark = BenchmarkRegistry.launch_benchmark(context)
     if benchmark:
         logger.info(

superbench/benchmarks/micro_benchmarks/cuda_nccl_bw_performance.py

@@ -1,7 +1,11 @@
 # Copyright (c) Microsoft Corporation.
 # Licensed under the MIT license.
 
-"""Module of the NCCL performance benchmarks."""
+"""Module of the NCCL/RCCL performance benchmarks.
+
+We assume NCCL-tests and RCCL-tests have the same interface and output in the test scope so far.
+So the arguments and result parsing are the same.
+"""
 
 import os
 import re
@@ -213,3 +217,4 @@ class CudaNcclBwBenchmark(MicroBenchmarkWithInvoke):
 
 
 BenchmarkRegistry.register_benchmark('nccl-bw', CudaNcclBwBenchmark, platform=Platform.CUDA)
+BenchmarkRegistry.register_benchmark('rccl-bw', CudaNcclBwBenchmark, platform=Platform.ROCM)