From 8aad3803a759a2df84b25e01ae73f2b517c94769 Mon Sep 17 00:00:00 2001
From: josantos <josantos@amd.com>
Date: Wed, 9 Aug 2023 15:40:10 -0500
Subject: [PATCH] shortening now in analyze also

-  add extraction level flag in profile
-  overwrite old csv with new demangled csv
-  demangling occurs in kernel_name_shortner

Signed-off-by: josantos <josantos@amd.com>
---
 src/omniperf                             |  2 +-
 src/omniperf_analyze/omniperf_analyze.py | 11 ++++++++++
 src/parser.py                            |  9 ++++++++
 src/utils/csv_converter.py               | 26 +++++++++++++++---------
 4 files changed, 37 insertions(+), 11 deletions(-)

diff --git a/src/omniperf b/src/omniperf
index acdfbc3bf..41856244c 100755
--- a/src/omniperf
+++ b/src/omniperf
@@ -739,7 +739,7 @@ def main():
     # PROFILE MODE
     ##############
     if args.mode == "profile":
-        Extractionlvl = 3 #args.extraction_level
+        Extractionlvl = args.kernelVerbose
         print("Resolving rocprof")
         resolve_rocprof()
         # Cannot access parent directories
diff --git a/src/omniperf_analyze/omniperf_analyze.py b/src/omniperf_analyze/omniperf_analyze.py
index 87fac064d..53f404492 100644
--- a/src/omniperf_analyze/omniperf_analyze.py
+++ b/src/omniperf_analyze/omniperf_analyze.py
@@ -45,6 +45,8 @@
 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 +222,16 @@ 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_filename = filename[:filename.rfind(".")]+"_new.csv"
+                new_df.to_csv(os.path.join(d[0],filename))
+            
         file_io.create_df_kernel_top_stats(
             d[0],
             runs[d[0]].filter_gpu_ids,
diff --git a/src/parser.py b/src/parser.py
index 9d6dd8f6f..dcb9be0c3 100644
--- a/src/parser.py
+++ b/src/parser.py
@@ -514,3 +514,12 @@ def parse(my_parser):
         action="store_true",
         help="\t\tRandomly generate a port to launch GUI application.\n\t\tRegistered Ports range inclusive (1024-49151).",
     )
+    analyze_group.add_argument(
+        "-f",
+        "--kernelVerbose",
+        required=False,
+        metavar="",
+        help="\t\t\t\tSpecify Kernel Name verbose level 1-5. Lower the level, shorter the kernel name. (DEFAULT: 2) (DISABLE: 5)",
+        default=2,
+        type=int,
+    )
diff --git a/src/utils/csv_converter.py b/src/utils/csv_converter.py
index 34f2e8261..35bc5a649 100644
--- a/src/utils/csv_converter.py
+++ b/src/utils/csv_converter.py
@@ -25,6 +25,7 @@
 import argparse
 import collections
 import os
+import subprocess
 import sys
 import re
 import pandas as pd
@@ -33,11 +34,11 @@
 from tqdm import tqdm
 import shutil
 
+
 cache = dict()
 supported_arch = {"gfx906": "mi50", "gfx908": "mi100", "gfx90a": "mi200"}
 MAX_SERVER_SEL_DELAY = 5000  # 5 sec connection timeout
 
-
 def kernel_name_shortener(df, cache, level):
     if level >= 5:
         return df
@@ -55,6 +56,13 @@ def kernel_name_shortener(df, cache, level):
             if original_name in cache:
                 continue
 
+            cmd = ["llvm-cxxfilt", original_name]
+
+            proc = subprocess.Popen(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
+
+            demangled_name, e = proc.communicate()
+            demangled_name = str(demangled_name, "UTF-8").strip()
+
             # cache miss, add the shortened name to the dictionary
             new_name = ""
             matches = ""
@@ -62,15 +70,14 @@ def kernel_name_shortener(df, cache, level):
             names_and_args = re.compile(r"(?P<name>[( )A-Za-z0-9_]+)([ ,*<>()]+)(::)?")
 
             # works for name Kokkos::namespace::init_lock_array_kernel_threadid(int) [clone .kd]
-            if names_and_args.search(original_name):
-                matches = names_and_args.findall(original_name)
+            if names_and_args.search(demangled_name):
+                matches = names_and_args.findall(demangled_name)
             else:
                 # Works for first case  '__amd_rocclr_fillBuffer.kd'
-                # remove .kd and then parse through original regex
-                first_case = re.compile(r"([^\s]+)(.kd)")
-                Mod_name_and_args = re.compile(r"(?P<name>[( )A-Za-z0-9_]+)([ ,*<>()]*)")
-                interim_name = first_case.search(original_name).group(1)
-                matches = Mod_name_and_args.findall(interim_name)
+                cache[original_name] = new_name
+                if new_name == None or new_name == "":
+                    cache[original_name] = demangled_name
+                continue
 
             current_level = 0
             for name in matches:
@@ -103,13 +110,12 @@ def kernel_name_shortener(df, cache, level):
 
             cache[original_name] = new_name
             if new_name == None or new_name == "":
-                cache[original_name] = original_name
+                cache[original_name] = demangled_name
 
         df[columnName] = df[columnName].map(cache)
 
     return df
 
-
 # Verify target directory and setup connection
 def parse(args, profileAndExport):
     host = args.host