1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
|
#!/usr/bin/env python
from __future__ import print_function
import config.config as conf
import os
import parse.ft as ft
import parse.sched as st
import pickle
import shutil as sh
import sys
import traceback
from collections import namedtuple
from common import load_params
from optparse import OptionParser
from parse.point import ExpPoint
from parse.tuple_table import TupleTable
from parse.col_map import ColMapBuilder
from multiprocessing import Pool, cpu_count
def parse_args():
# TODO: convert data-dir to proper option, clean 'dest' options
parser = OptionParser("usage: %prog [options] [data_dir]...")
print("default to no params.py")
parser.add_option('-o', '--out', dest='out',
help='file or directory for data output', default='parse-data')
parser.add_option('-c', '--clean', action='store_true', default=False,
dest='clean', help='do not output single-point csvs')
parser.add_option('-i', '--ignore', metavar='[PARAM...]', default="",
help='ignore changing parameter values')
parser.add_option('-f', '--force', action='store_true', default=False,
dest='force', help='overwrite existing data')
parser.add_option('-v', '--verbose', action='store_true', default=False,
dest='verbose', help='print out data points')
parser.add_option('-m', '--write-map', action='store_true', default=False,
dest='write_map',
help='Output map of values instead of csv tree')
return parser.parse_args()
ExpData = namedtuple('ExpData', ['path', 'params', 'work_dir'])
def get_exp_params(data_dir, cm_builder):
param_file = "%s/%s" % (data_dir, conf.DEFAULTS['params_file'])
if not os.path.isfile:
raise Exception("No param file '%s' exists!" % param_file)
params = load_params(param_file)
# Store parameters in cm_builder, which will track which parameters change
# across experiments
for key, value in params.iteritems():
cm_builder.try_add(key, value)
# Cycles must be present for feather-trace measurement parsing
if conf.PARAMS['cycles'] not in params:
params[conf.PARAMS['cycles']] = conf.DEFAULTS['cycles']
return params
def load_exps(exp_dirs, cm_builder, clean):
exps = []
sys.stderr.write("Loading experiments...\n")
for data_dir in exp_dirs:
if not os.path.isdir(data_dir):
raise IOError("Invalid experiment '%s'" % os.path.abspath(data_dir))
# Used to store error output and debugging info
work_dir = data_dir + "/tmp"
if os.path.exists(work_dir) and clean:
sh.rmtree(work_dir)
if not os.path.exists(work_dir):
os.mkdir(work_dir)
params = get_exp_params(data_dir, cm_builder)
exps += [ ExpData(data_dir, params, work_dir) ]
return exps
def parse_exp(exp_force):
# Tupled for multiprocessing
exp, force = exp_force
result_file = exp.work_dir + "/exp_point.pkl"
should_load = not force and os.path.exists(result_file)
result = None
if should_load:
with open(result_file, 'rb') as f:
try:
# No need to go through this work twice
result = pickle.load(f)
except:
pass
if not result:
try:
result = ExpPoint(exp.path)
cycles = exp.params[conf.PARAMS['cycles']]
# Write overheads into result
ft.extract_ft_data(result, exp.path, exp.work_dir, cycles)
# Write scheduling statistics into result
st.extract_sched_data(result, exp.path, exp.work_dir)
with open(result_file, 'wb') as f:
pickle.dump(result, f)
except:
traceback.print_exc()
return (exp, result)
def main():
opts, args = parse_args()
args = args or [os.getcwd()]
# Load exp parameters into a ColMap
builder = ColMapBuilder()
exps = load_exps(args, builder, opts.force)
# Don't track changes in ignored parameters
if opts.ignore:
for param in opts.ignore.split(","):
builder.try_remove(param)
# Always average multiple trials
builder.try_remove(conf.PARAMS['trial'])
col_map = builder.build()
result_table = TupleTable(col_map)
sys.stderr.write("Parsing data...\n")
procs = min(len(exps), cpu_count()/2)
pool = Pool(processes=procs)
pool_args = zip(exps, [opts.force]*len(exps))
enum = pool.imap_unordered(parse_exp, pool_args, 1)
try:
for i, (exp, result) in enumerate(enum):
if opts.verbose:
print(result)
else:
sys.stderr.write('\r {0:.2%}'.format(float(i)/len(exps)))
result_table[exp.params] += [result]
pool.close()
except:
pool.terminate()
traceback.print_exc()
raise Exception("Failed parsing!")
finally:
pool.join()
sys.stderr.write('\n')
if opts.force and os.path.exists(opts.out):
sh.rmtree(opts.out)
result_table = result_table.reduce()
sys.stderr.write("Writing result...\n")
if opts.write_map:
# Write summarized results into map
result_table.write_map(opts.out)
else:
# Write out csv directories for all variable params
dir_map = result_table.to_dir_map()
dir_map.write(opts.out)
if __name__ == '__main__':
main()
|
