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
|
import csv
import pickle
import matplotlib.pyplot as plt
import multiprocessing
import numpy as np
import os
import subprocess
from benchmark import Benchmark
from common_targets import common_targets
cmd = ("perf stat -x\; -e cpu-clock:k,cache-references,cache-misses,cycles,"
"instructions,branches,faults,migrations "
"build/bench_loop{} 1.2 {} 1000000 {} 10")
class Benchmark_Loop( Benchmark ):
def __init__(self):
self.file_name = "bench_loop"
self.name = "Loop Stress Benchmark"
self.descrition = """This benchmark makes n allocations in t concurrent threads.
How allocations are freed can be changed with the benchmark
version""",
self.targets = common_targets
self.maxsize = [2 ** x for x in range(6, 16)]
self.nthreads = range(1, multiprocessing.cpu_count() * 2 + 1)
self.results = {"args" : {"nthreads" : self.nthreads, "maxsize": self.maxsize},
"targets" : self.targets}
def prepare(self, verbose=False):
req = ["build/bench_loop"]
for r in req:
if not os.path.isfile(r):
print(r, "not found")
return False
if not os.access(r, os.X_OK):
print(r, "not found")
return False
if verbose:
print(r, "found and executable.")
return True
def run(self, verbose=False, runs=3):
args_permutations = [(x,y) for x in self.nthreads for y in self.maxsize]
n = len(args_permutations)
for run in range(1, runs + 1):
print(str(run) + ". run")
for i, args in enumerate(args_permutations):
print(i + 1, "of", n, "\r", end='')
# run cmd for each target
for tname, t in self.targets.items():
result = {"VSZ": [], "RSS" : []}
os.environ["LD_PRELOAD"] = t[1]
target_cmd = cmd.format(t[0], *args).split(" ")
if verbose:
print("\n" + tname, t, "\n", " ".join(target_cmd), "\n")
p = subprocess.Popen(target_cmd,
env=os.environ,
stderr=subprocess.PIPE,
stdout=subprocess.PIPE,
universal_newlines=True)
while p.poll() == None:
ps = subprocess.run(["ps", "-F", "--ppid", str(p.pid)], stdout=subprocess.PIPE)
lines = ps.stdout.splitlines()
if len(lines) == 1: # perf hasn't forked yet
continue
tokens = str(lines[1]).split()
result["VSZ"].append(tokens[4])
result["RSS"].append(tokens[5])
p.wait()
output = p.stderr.read()
if p.returncode != 0:
print("\n" + " ".join(target_cmd), "exited with", p.returncode, ".\n Aborting Benchmark.")
print(tname, t)
print(output)
print(p.stdout)
return False
if "ERROR: ld.so" in output:
print("\nPreloading of", t[1], "failed for", tname, ".\n Aborting Benchmark.")
print(output)
return False
# Handle perf output
csvreader = csv.reader(output.splitlines(), delimiter=';')
for row in csvreader:
result[row[2].replace("\\", "")] = row[0].replace("\\", "")
key = (tname, *args)
if not key in self.results:
self.results[key] = [result]
else:
self.results[key].append(result)
print()
return True
def summary(self):
# MAXSIZE fixed
nthreads = self.results["args"]["nthreads"]
maxsize = self.results["args"]["maxsize"]
targets = self.results["targets"]
y_mapping = {v : i for i, v in enumerate(nthreads)}
for size in maxsize:
for target in targets:
y_vals = [0] * len(nthreads)
for mid, measures in self.results.items():
if mid[0] == target and mid[2] == size:
d = []
for m in measures:
# nthreads/time = MOPS/S
d.append(mid[1]/float(m["cpu-clock:ku"]))
y_vals[y_mapping[mid[1]]] = np.mean(d)
plt.plot(nthreads, y_vals, marker='.', linestyle='-', label=target)
plt.legend()
plt.xlabel("threads")
plt.ylabel("MOPS/s")
plt.title("Loop: " + str(size) + "B")
plt.savefig(self.file_name + "." + str(size) + "B.png")
plt.clf()
# NTHREADS fixed
y_mapping = {v : i for i, v in enumerate(maxsize)}
x_vals = [i + 1 for i in range(0, len(maxsize))]
for n in nthreads:
for target in targets:
y_vals = [0] * len(maxsize)
for mid, measures in self.results.items():
if mid[0] == target and mid[1] == n:
d = []
for m in measures:
# nthreads/time = MOPS/S
d.append(n/float(m["cpu-clock:ku"]))
y_vals[y_mapping[mid[2]]] = np.mean(d)
plt.plot(x_vals, y_vals, marker='.', linestyle='-', label=target)
plt.legend()
plt.xticks(x_vals, maxsize)
plt.xlabel("size in B")
plt.ylabel("MOPS/s")
plt.title("Loop: " + str(n) + "thread(s)")
plt.savefig(self.file_name + "." + str(n) + "threads.png")
plt.clf()
loop = Benchmark_Loop()
|
