aboutsummaryrefslogtreecommitdiff
path: root/benchmark.py
blob: 2d5fc943a7a60c7a698fe677b5ebdd08d05a42c8 (plain)
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
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
from collections import namedtuple
import copy
import csv
import itertools
import os
import pickle
import subprocess

from common_targets import common_targets

class Benchmark (object):

    defaults = {
        "name" : "default_benchmark",
        "description" : "This is the default benchmark description please add your own useful one.",
        
        "perf_cmd" : "perf stat -x, -dd ",
        "analyse_cmd" : "memusage -p {} -t ",
        "cmd" : "true",
        "targets" : common_targets,
    }

    def __init__(self):
        # Set default values
        for k in Benchmark.defaults:
            if not hasattr(self, k):
                setattr(self, k, Benchmark.defaults[k])

        # non copy types
        if not hasattr(self, "args"):
            self.args = {}

        self.Perm = namedtuple("Perm", self.args.keys())
        
        if not hasattr(self, "results"):
            self.results = {}
        self.results["args"] = self.args
        self.results["targets"] = self.targets
        self.results.update({t : {} for t in self.targets})

        if not hasattr(self, "requirements"):
            self.requirements = []
    
    def save(self, path=None, verbose=False):
        print(self.results)
        f = path if path else self.name + ".save"
        if verbose:
            print("Saving results to:", self.name + ".save")
        # Pickle can't handle namedtuples so convert the dicts of namedtuples
        # into lists of dicts.
        save_data = {"args" : self.results["args"], "targets" : self.results["targets"]}
        for target in self.results["targets"]:
            l = []
            for ntuple, measures in self.results[target].items():
                l.append((ntuple._asdict(), measures))
            save_data[target] = l

        with open(f, "wb") as f:
            pickle.dump(save_data, f)

    def load(self, path=None, verbose=False):
        f = path if path else self.name + ".save"
        if verbose:
            print("Loading results from:", self.name + ".save")
        with open(f, "rb") as f:
            save_data = pickle.load(f)
        # Build new named tuples
        self.results["args"] = save_data["args"]
        self.results["targets"] = save_data["targets"]
        for target in save_data["targets"]:
            d = {}
            for dic, measures in save_data[target]:
                d[self.Perm(**dic)] = measures
            self.results[target] = d

    def prepare(self, verbose=False):
        def is_exe(fpath):
            return os.path.isfile(fpath) and os.access(fpath, os.X_OK)

        for r in self.requirements:
            fpath, fname = os.path.split(r)
            
            if fpath:
                if not is_exe(r):
                    return False
            else:
                found = False
                for path in os.environ["PATH"].split(os.pathsep):
                    exe_file = os.path.join(path, r)
                    if is_exe(exe_file):
                        found = True

                if not found:
                    return False

        return True

    def iterate_args(self):
        """Return a dict for each possible combination of args"""
        arg_names = sorted(self.args.keys())
        for p in itertools.product(*[self.args[k] for k in arg_names]):
            Perm = namedtuple("Perm", arg_names)
            yield Perm(*p)

    def iterate_args_fixed(self, fixed):
        for p in self.iterate_args():
            p_dict = p._asdict()
            is_fixed = True
            for k in fixed:
                if p_dict[k] != fixed[k]:
                    is_fixed = False
                    break
            if is_fixed:
                yield p
            

    def analyse(self, verbose=False):
        for perm in self.iterate_args():
            perm = perm._asdict()
            file_name = self.name + "."
            file_name += ".".join([str(x) for x in perm.values()])
            file_name += ".memusage"

            actual_cmd = self.analyse_cmd.format(file_name + ".png")
            if "binary_suffix" in self.cmd:
                perm["binary_suffix"] = ""
            actual_cmd += self.cmd.format(**perm)
            
            with open(file_name + ".hist", "w") as f:
                res = subprocess.run(actual_cmd.split(),
                                    stdout=subprocess.PIPE,
                                    stderr=f,
                                    universal_newlines=True)

                if res.returncode != 0:
                    print(actual_cmd, "failed.")
                    print("Aborting analysing.")
                    print("You may look at", file_name + ".hist", "to fix this.")
                    return

    def parse_chattymalloc_data(self, path="chattymalloc.data"):
        hist = {}
        with open(path, "r") as f:
            for l in f.readlines():
                n = int(l)
                hist[n] = hist.get(n, 0) + 1
        return hist

    def run(self, verbose=False, runs=5):
        n = len(list(self.iterate_args())) * len(self.targets)
        i = 0
        for run in range(1, runs + 1):
            print(str(run) + ". run")

            for tname, t in self.targets.items():
                if not tname in self.results:
                    self.results[tname] = {}

                os.environ["LD_PRELOAD"] = "build/print_status_on_exit.so "
                os.environ["LD_PRELOAD"] += t["LD_PRELOAD"]

                if hasattr(self, "pretarget_hook"):
                    if self.pretarget_hook((tname, t), run, verbose):
                        return False

                for perm in self.iterate_args():
                    i += 1
                    print(i + 1, "of", n, "\r", end='')
                    
                    actual_cmd = self.perf_cmd

                    perm_dict = perm._asdict()
                    perm_dict.update(t)
                    actual_cmd += self.cmd.format(**perm_dict)

                    res = subprocess.run(actual_cmd.split(),
                                        stderr=subprocess.PIPE,
                                        stdout=subprocess.PIPE,
                                        universal_newlines=True)

                    if res.returncode != 0:
                        print("\n" + actual_cmd, "exited with", res.returncode, "for", tname)
                        print("Aborting Benchmark.")
                        print("Stdout:\n" + res.stdout)
                        print("Stderr:\n" + res.stderr)
                        return False

                    if "ERROR: ld.so" in res.stderr:
                        print("\nPreloading of", t["LD_PRELOAD"], "failed for", tname)
                        print("Stderr:\n" + res.stderr)
                        print("Aborting Benchmark.")
                        return False

                    result = {}
                    
                    # Read VmHWM from status file. If our benchmark didn't fork
                    # the first occurance of VmHWM is from our benchmark
                    with open("status", "r") as f:
                        for l in f.readlines():
                            if l.startswith("VmHWM:"):
                                result["VmHWM"] = l.split()[1]
                                break
                    os.remove("status")
                        
                    if hasattr(self, "process_stdout"):
                        self.process_stdout(result, res.stdout)

                    # Parse perf output if available
                    if self.perf_cmd != "":
                        csvreader = csv.reader(res.stderr.splitlines(), delimiter=',')
                        for row in csvreader:
                            # Split of the user/kernel space info to be better portable
                            result[row[2].split(":")[0]] = row[0]

                    if run == 1:
                        self.results[tname][perm] = []
                    self.results[tname][perm].append(result)

                if hasattr(self, "posttarget_hook"):
                    if self.posttarget_hook((tname, t), run, verbose):
                        return False
            print()
        return True