Names shortened/demangled after join_prof

-  Added kernelVerbose flag in profile_group
-  Added KernelVerbose flag in analyze_group
-  Analyze replaces csv with shortened/demangled name
-  csv_converter uses llvm-cxxfilt

Signed-off-by: josantos <josantos@amd.com>


[ROCm/rocprofiler-compute commit: 59d77f9d81]

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

@@ -45,6 +45,8 @@ import os.path
 from pathlib import Path
 from omniperf_analyze.utils import parser, file_io
 from omniperf_analyze.utils.gui_components.roofline import get_roofline
+from utils import csv_converter
+import pandas as pd
 
 archConfigs = {}
 
@@ -220,7 +222,15 @@ def run_cli(args, runs):
     # If we assume the panel layout for all archs are similar, it doesn't matter
     # which archConfig passed into show_all function.
     # After decide to how to manage kernels display patterns, we can revisit it.
+    cache =dict()
     for d in args.path:
+        #demangle
+        for filename in os.listdir(d[0]):
+            if filename.endswith('.csv'):
+                df = pd.read_csv(os.path.join(d[0],filename))
+                new_df = csv_converter.kernel_name_shortener(df, cache, args.kernelVerbose)
+                new_df.to_csv(os.path.join(d[0],filename), index=False)
+            
         file_io.create_df_kernel_top_stats(
             d[0],
             runs[d[0]].filter_gpu_ids,