Update rocHPCG metrics

docs/user-tutorial/benchmarks/micro-benchmarks.md

@@ -201,7 +201,9 @@ When collecting multiple MPI scales or problem sizes, use separate benchmark sec
 | Name                                  | Unit               | Description                                          |
 |---------------------------------------|--------------------|------------------------------------------------------|
 | gpu-hpcg/final_gflops                 | FLOPS (GFLOPS)     | Final rocHPCG score.                                 |
+| gpu-hpcg/final_bandwidth              | bandwidth (GB/s)   | Final rocHPCG bandwidth.                             |
 | gpu-hpcg/final_gflops_per_process     | FLOPS (GFLOPS)     | Final rocHPCG score per process.                     |
+| gpu-hpcg/final_bandwidth_per_process  | bandwidth (GB/s)   | Final rocHPCG bandwidth per process.                 |
 | gpu-hpcg/ddot_gflops                  | FLOPS (GFLOPS)     | DDOT throughput.                                     |
 | gpu-hpcg/ddot_bandwidth               | bandwidth (GB/s)   | DDOT bandwidth.                                      |
 | gpu-hpcg/ddot_gflops_per_process      | FLOPS (GFLOPS)     | DDOT throughput per process.                         |

superbench/benchmarks/micro_benchmarks/gpu_hpcg_performance_base.py

@@ -164,7 +164,9 @@ class GpuHpcgBenchmark(MicroBenchmarkWithInvoke):
         parsed_results = {}
         required_metrics = {
             'final_gflops',
+            'final_bandwidth',
             'final_gflops_per_process',
+            'final_bandwidth_per_process',
             'ddot_gflops',
             'ddot_bandwidth',
             'ddot_gflops_per_process',
@@ -253,9 +255,8 @@ class GpuHpcgBenchmark(MicroBenchmarkWithInvoke):
 
         parsed_results[f'{prefix}_gflops'] = gflops_values[0]
         parsed_results[f'{prefix}_gflops_per_process'] = gflops_values[1]
-        if prefix != 'final':
-            parsed_results[f'{prefix}_bandwidth'] = bandwidth_values[0]
-            parsed_results[f'{prefix}_bandwidth_per_process'] = bandwidth_values[1]
+        parsed_results[f'{prefix}_bandwidth'] = bandwidth_values[0]
+        parsed_results[f'{prefix}_bandwidth_per_process'] = bandwidth_values[1]
         return True
 
     def _parse_time_line(self, line, parsed_results):