Pass status.csv to metric files.

This patch introduces a new option --status_csv for passing status.csv to metric
computing scripts (*-data-to-csv.py), and merge status field in
merge-metric-csvs.py.

Change-Id: If4fa4f7f99c1cdc574c326422b03b87d7af3dc71

1 files changed, 17 insertions, 10 deletions

diff --git a/size-data-to-csv.py b/size-data-to-csv.py
index c88cf12..9d5b1d2 100755
--- a/size-data-to-csv.py
+++ b/size-data-to-csv.py
@@ -5,11 +5,12 @@ import csv
 import subprocess
 import os
 import metric_utils
+import pandas as pd
 
 """
 Parse output of nm -S -td."""
 
-def get_symbol_size(symbol_to_size, perf_bmk_symbols, exe_path):
+def get_symbol_size(symbol_to_size, interesting_symbols, exe_path):
     try:
         lines = subprocess.check_output(["nm", "-Std", exe_path], stderr=subprocess.DEVNULL).decode("utf-8").split("\n")
     except subprocess.CalledProcessError as e:
@@ -20,7 +21,7 @@ def get_symbol_size(symbol_to_size, perf_bmk_symbols, exe_path):
     # by nm. For eg, perf.csv may contain malloc@plt as symbol name,
     # but nm will list malloc.
     plt_entries = set()
-    for symbol in perf_bmk_symbols:
+    for symbol in interesting_symbols:
         if symbol.endswith("@plt"):
             plt_entries.add(symbol)
 
@@ -33,11 +34,13 @@ def get_symbol_size(symbol_to_size, perf_bmk_symbols, exe_path):
             # Some symbols may contain @libc prefix for eg, puts@GLIBC_2.2.5
             # Strip the @<lib> from symbol name.
             symbol = symbol[: symbol.find("@")] if symbol.find("@") != -1 else symbol
+            if symbol not in interesting_symbols:
+                continue
 
             # FIXME: Size is accumulated for static symbols in different
             # TU's having same name.
             if section.lower() == 't' or section.lower() == 'w' \
-               or symbol in perf_bmk_symbols or symbol in plt_entries:
+               or symbol in plt_entries:
                 if symbol not in symbol_to_size:
                     symbol_to_size[symbol] = 0
                 symbol_to_size[symbol] += size
@@ -54,28 +57,32 @@ def main():
     out_csv_file = args.out_csv_file
     perf_csv = args.perf_csv
     results_dir = args.results_dir
+    interesting_symbols_csv = args.interesting_symbols
+    status_df = pd.read_csv(args.status_csv)
 
     outf = open(out_csv_file, "w")
     csvwriter = csv.writer(outf)
     csvwriter.writerow(("benchmark", "symbol", "size"))
 
-    benchmarks = metric_utils.get_benchmarks_from_results_dir(bmk_name, results_dir)
+    benchmarks = metric_utils.get_benchmarks_from_results_dir(bmk_name, results_dir, status_df)
     for bmk in benchmarks:
         exe = bmk.exe
         if exe is None:
             continue
 
-        # Get symbols from perf.csv. We ensure that size for all symbols from
-        # perf.csv is calculcated.
-        perf_bmk_symbols = metric_utils.get_bmk_symbols_from_perf(perf_csv, bmk.name)
-        
+        if interesting_symbols_csv:
+            sym_dso_map = metric_utils.get_bmk_symbols_from_perf(interesting_symbols_csv, bmk.name, True)
+        else:
+            sym_dso_map = metric_utils.get_bmk_symbols_from_perf(perf_csv, bmk.name)
+        interesting_symbols = set(sym_dso_map.keys())
+
         # FIXME: For now, accumulate size for symbols with same name across main exe and libraries.
         # So sym_to_size now contains symbols across the main exe and libs, with
         # one entry for common symbols.
-        sym_to_size = get_symbol_size({}, perf_bmk_symbols, exe.path)
+        sym_to_size = get_symbol_size({}, interesting_symbols, exe.path)
         libs = exe.libs
         for libname in libs:
-            sym_to_size = get_symbol_size(sym_to_size, perf_bmk_symbols, libs[libname])
+            sym_to_size = get_symbol_size(sym_to_size, interesting_symbols, libs[libname])
 
         # Write entry for libs: bmk, libname, total lib size
         csvwriter.writerow((bmk.name, exe.name, calculate_size(exe.path)))