allow to specify the maximum number of kernels shown

Signed-off-by: fei.zheng <fei.zheng@amd.com>


[ROCm/rocprofiler-compute commit: dfe74057c1]

projects/rocprofiler-compute/src/omniperf_analyze/omniperf_analyze.py

@@ -138,13 +138,12 @@ def run_gui(args, runs):
     app = dash.Dash(__name__, external_stylesheets=[dbc.themes.CYBORG])
 
     if len(runs) == 1:
-        num_results = 10
         file_io.create_df_kernel_top_stats(
             args.path[0][0],
             runs[args.path[0][0]].filter_gpu_ids,
             runs[args.path[0][0]].filter_dispatch_ids,
             args.time_unit,
-            num_results,
+            args.max_kernel_num,
         )
         runs[args.path[0][0]].raw_pmc = file_io.create_df_pmc(
             args.path[0][0], args.verbose
@@ -187,13 +186,12 @@ def run_cli(args, runs):
     # which archConfig passed into show_all function.
     # After decide to how to manage kernels display patterns, we can revisit it.
     for d in args.path:
-        num_results = 10
         file_io.create_df_kernel_top_stats(
             d[0],
             runs[d[0]].filter_gpu_ids,
             runs[d[0]].filter_dispatch_ids,
             args.time_unit,
-            num_results,
+            args.max_kernel_num,
         )
         runs[d[0]].raw_pmc = file_io.create_df_pmc(
             d[0], args.verbose

projects/rocprofiler-compute/src/omniperf_analyze/utils/file_io.py

@@ -111,7 +111,7 @@ def create_df_kernel_top_stats(
     filter_gpu_ids,
     filter_dispatch_ids,
     time_unit,
-    num_results,
+    max_kernel_num,
     sortby="sum",
 ):
     """
@@ -172,13 +172,13 @@ def create_df_kernel_top_stats(
     if sortby == "sum":
         grouped = grouped.sort_values(by=("Sum" + time_unit_str), ascending=False)
 
-        grouped = grouped.head(num_results)  # Display only the top n results
+        grouped = grouped.head(max_kernel_num)  # Display only the top n results
 
         grouped.to_csv(os.path.join(raw_data_dir, "pmc_kernel_top.csv"), index=False)
     elif sortby == "kernel":
         grouped = grouped.sort_values("KernelName")
 
-        grouped = grouped.head(num_results)  # Display only the top n results
+        grouped = grouped.head(max_kernel_num)  # Display only the top n results
         grouped.to_csv(os.path.join(raw_data_dir, "pmc_kernel_top.csv"), index=False)
 
 

projects/rocprofiler-compute/src/parser.py

@@ -445,6 +445,14 @@ def parse(my_parser):
         choices=["per_wave", "per_cycle", "per_second", "per_kernel"],
         help="\t\tSpecify the normalization unit: (DEFAULT: per_wave)\n\t\t   per_wave\n\t\t   per_cycle\n\t\t   per_second\n\t\t   per_kernel",
     )
+    analyze_group.add_argument(
+        "--max-kernel-num",
+        dest="max_kernel_num",
+        metavar="",
+        type=int,
+        default=10,
+        help="\t\tSpecify the maximum number of kernels shown",
+    )
     analyze_group.add_argument(
         "--config-dir",
         dest="config_dir",