aboutsummaryrefslogtreecommitdiff
path: root/android_bench_suite/parse_result.py
diff options
context:
space:
mode:
Diffstat (limited to 'android_bench_suite/parse_result.py')
-rw-r--r--android_bench_suite/parse_result.py114
1 files changed, 0 insertions, 114 deletions
diff --git a/android_bench_suite/parse_result.py b/android_bench_suite/parse_result.py
deleted file mode 100644
index 90b3c4d1..00000000
--- a/android_bench_suite/parse_result.py
+++ /dev/null
@@ -1,114 +0,0 @@
-# Copyright 2017 The Chromium OS Authors. All rights reserved.
-# Use of this source code is governed by a BSD-style license that can be
-# found in the LICENSE file.
-"""Helper functions to parse result collected from device"""
-
-from __future__ import print_function
-from fix_skia_results import _TransformBenchmarks
-
-import json
-
-def normalize(bench, dict_list):
- bench_base = {
- 'Panorama': 1,
- 'Dex2oat': 1,
- 'Hwui': 10000,
- 'Skia': 1,
- 'Synthmark': 1,
- 'Binder': 0.001
- }
- result_dict = dict_list[0]
- for key in result_dict:
- result_dict[key] = result_dict[key] / bench_base[bench]
- return [result_dict]
-
-
-# Functions to parse benchmark result for data collection.
-def parse_Panorama(bench, fin):
- result_dict = {}
- for line in fin:
- words = line.split()
- if 'elapsed' in words:
- #TODO: Need to restructure the embedded word counts.
- result_dict['total_time_s'] = float(words[3])
- result_dict['retval'] = 0
- return normalize(bench, [result_dict])
- raise ValueError('You passed the right type of thing, '
- 'but it didn\'t have the expected contents.')
-
-
-def parse_Synthmark(bench, fin):
- result_dict = {}
- accum = 0
- cnt = 0
- for line in fin:
- words = line.split()
- if 'normalized' in words:
- #TODO: Need to restructure the embedded word counts.
- accum += float(words[-1])
- cnt += 1
- if accum != 0:
- result_dict['total_voices'] = accum / cnt
- result_dict['retval'] = 0
- return normalize(bench, [result_dict])
- raise ValueError('You passed the right type of thing, '
- 'but it didn\'t have the expected contents.')
-
-
-def parse_Binder(bench, fin):
- result_dict = {}
- accum = 0
- cnt = 0
- for line in fin:
- words = line.split()
- for word in words:
- if 'average' in word:
- #TODO: Need to restructure the embedded word counts.
- accum += float(word[8:-2])
- cnt += 1
- if accum != 0:
- result_dict['avg_time_ms'] = accum / cnt
- result_dict['retval'] = 0
- return normalize(bench, [result_dict])
- raise ValueError('You passed the right type of thing, '
- 'but it didn\'t have the expected contents.')
-
-
-def parse_Dex2oat(bench, fin):
- result_dict = {}
- cnt = 0
- for line in fin:
- words = line.split()
- if 'elapsed' in words:
- cnt += 1
- #TODO: Need to restructure the embedded word counts.
- if cnt == 1:
- # First 'elapsed' time is for microbench 'Chrome'
- result_dict['chrome_s'] = float(words[3])
- elif cnt == 2:
- # Second 'elapsed' time is for microbench 'Camera'
- result_dict['camera_s'] = float(words[3])
-
- result_dict['retval'] = 0
- # Two results found, return
- return normalize(bench, [result_dict])
- raise ValueError('You passed the right type of thing, '
- 'but it didn\'t have the expected contents.')
-
-
-def parse_Hwui(bench, fin):
- result_dict = {}
- for line in fin:
- words = line.split()
- if 'elapsed' in words:
- #TODO: Need to restructure the embedded word counts.
- result_dict['total_time_s'] = float(words[3])
- result_dict['retval'] = 0
- return normalize(bench, [result_dict])
- raise ValueError('You passed the right type of thing, '
- 'but it didn\'t have the expected contents.')
-
-
-def parse_Skia(bench, fin):
- obj = json.load(fin)
- return normalize(bench, _TransformBenchmarks(obj))