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
|
# BLAS performance testing helper
# Copyright (C) 2021 M. Zhou <lumin@debian.org>
import os
import sys
import time
import numpy as np
import torch as th
os.system('update-alternatives --display libblas.so.3-x86_64-linux-gnu')
print('F32 Numpy Refrence Group', end='\t')
sys.stdout.flush()
N, reference = 8, []
for i in range(N):
x = np.random.rand(4096, 4096).astype(np.float32)
y = np.random.rand(4096, 4096).astype(np.float32)
time_start = time.time()
z = x @ y
time_end = time.time()
reference.append(time_end - time_start)
print('.', end='')
sys.stdout.flush()
print(f'{1000*np.mean(reference):.1f}ms pm {1000*np.std(reference):.1f}ms')
print('F64 Numpy Refrence Group', end='\t')
sys.stdout.flush()
N, reference = 8, []
for i in range(N):
x = np.random.rand(4096, 4096).astype(np.float64)
y = np.random.rand(4096, 4096).astype(np.float64)
time_start = time.time()
z = x @ y
time_end = time.time()
reference.append(time_end - time_start)
print('.', end='')
sys.stdout.flush()
print(f'{1000*np.mean(reference):.1f}ms pm {1000*np.std(reference):.1f}ms')
print('F32 Torch', end='\t')
sys.stdout.flush()
N, results = 8, []
for i in range(N):
x = th.rand(4096, 4096).to(th.float32)
y = th.rand(4096, 4096).to(th.float32)
time_start = time.time()
z = x @ y
time_end = time.time()
results.append(time_end - time_start)
print('.', end='')
sys.stdout.flush()
print(f'{1000*np.mean(results):.1f}ms pm {1000*np.std(results):.1f}ms')
print('F64 Torch', end='\t')
sys.stdout.flush()
N, results = 8, []
for i in range(N):
x = th.rand(4096, 4096).to(th.float64)
y = th.rand(4096, 4096).to(th.float64)
time_start = time.time()
z = x @ y
time_end = time.time()
results.append(time_end - time_start)
print('.', end='')
sys.stdout.flush()
print(f'{1000*np.mean(results):.1f}ms pm {1000*np.std(results):.1f}ms')
|
