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
|
# Licensed under a 3-clause BSD style license - see LICENSE.rst
from __future__ import (absolute_import, division, print_function,
unicode_literals)
import timeit
import numpy as np # pylint: disable=W0611
from ...extern.six.moves import range, zip
# largest image size to use for "linear" and fft convolutions
max_exponents_linear = {1: 15, 2: 7, 3: 5}
max_exponents_fft = {1: 15, 2: 10, 3: 7}
if __name__ == "__main__":
for ndims in [1, 2, 3]:
print("\n{}-dimensional arrays ('n' is the size of the image AND "
"the kernel)".format(ndims))
print(" ".join(["%17s" % n for n in ("n", "convolve", "convolve_fft")]))
for ii in range(3, max_exponents_fft[ndims]):
# array = np.random.random([2**ii]*ndims)
# test ODD sizes too
if ii < max_exponents_fft[ndims]:
setup = ("""
import numpy as np
from astropy.convolution.convolve import convolve
from astropy.convolution.convolve import convolve_fft
array = np.random.random([%i]*%i)
kernel = np.random.random([%i]*%i)""") % (2 ** ii - 1, ndims, 2 ** ii - 1, ndims)
print("%16i:" % (int(2 ** ii - 1)), end=' ')
if ii <= max_exponents_linear[ndims]:
for ffttype, extra in zip(("", "_fft"),
("", "fft_pad=False")):
statement = "convolve{}(array, kernel, boundary='fill', {})".format(ffttype, extra)
besttime = min(timeit.Timer(stmt=statement, setup=setup).repeat(3, 10))
print("%17f" % (besttime), end=' ')
else:
print("%17s" % "skipped", end=' ')
statement = "convolve_fft(array, kernel, boundary='fill')"
besttime = min(timeit.Timer(stmt=statement, setup=setup).repeat(3, 10))
print("%17f" % (besttime), end=' ')
print()
setup = ("""
import numpy as np
from astropy.convolution.convolve import convolve
from astropy.convolution.convolve import convolve_fft
array = np.random.random([%i]*%i)
kernel = np.random.random([%i]*%i)""") % (2 ** ii - 1, ndims, 2 ** ii - 1, ndims)
print("%16i:" % (int(2 ** ii)), end=' ')
if ii <= max_exponents_linear[ndims]:
for ffttype in ("", "_fft"):
statement = "convolve{}(array, kernel, boundary='fill')".format(ffttype)
besttime = min(timeit.Timer(stmt=statement, setup=setup).repeat(3, 10))
print("%17f" % (besttime), end=' ')
else:
print("%17s" % "skipped", end=' ')
statement = "convolve_fft(array, kernel, boundary='fill')"
besttime = min(timeit.Timer(stmt=statement, setup=setup).repeat(3, 10))
print("%17f" % (besttime), end=' ')
print()
"""
Unfortunately, these tests are pretty strongly inconclusive
RESULTS on a 2011 Mac Air:
1-dimensional arrays ('n' is the size of the image AND the kernel)
n convolve convolve_fftnp convolve_fftw convolve_fftsp
7: 0.000408 0.002334 0.005571 0.002677
8: 0.000399 0.002818 0.006505 0.003094
15: 0.000361 0.002491 0.005648 0.002678
16: 0.000371 0.002997 0.005983 0.003036
31: 0.000535 0.002450 0.005988 0.002880
32: 0.000452 0.002618 0.007102 0.004366
63: 0.000509 0.002876 0.008003 0.002981
64: 0.000453 0.002706 0.005520 0.003049
127: 0.000801 0.004080 0.008513 0.003932
128: 0.000749 0.003332 0.006236 0.003159
255: 0.002453 0.003111 0.007518 0.003564
256: 0.002478 0.003341 0.006325 0.004290
511: 0.008394 0.006224 0.010247 0.005991
512: 0.007934 0.003764 0.006840 0.004106
1023: 0.028741 0.007538 0.009591 0.007696
1024: 0.027900 0.004871 0.009628 0.005118
2047: 0.106323 0.021575 0.022041 0.020682
2048: 0.108916 0.008107 0.011049 0.007596
4095: 0.411936 0.021675 0.019761 0.020939
4096: 0.408992 0.018870 0.016663 0.012890
8191: 1.664517 8.278320 0.073001 7.803563
8192: 1.657573 0.037967 0.034227 0.028390
16383: 6.654678 0.251661 0.202271 0.222171
16384: 6.611977 0.073630 0.067616 0.055591
2-dimensional arrays ('n' is the size of the image AND the kernel)
n convolve convolve_fftnp convolve_fftw convolve_fftsp
7: 0.000552 0.003524 0.006667 0.004318
8: 0.000646 0.004443 0.007354 0.003958
15: 0.002986 0.005093 0.012941 0.005951
16: 0.003549 0.005688 0.008818 0.006300
31: 0.074360 0.033973 0.031800 0.036937
32: 0.077338 0.017708 0.025637 0.011883
63: 0.848471 0.057407 0.052192 0.053213
64: 0.773061 0.029657 0.033409 0.028230
127: 14.656414 1.005329 0.402113 0.955279
128: 15.867796 0.266233 0.268551 0.237930
255: skipped 1.715546 1.566876 1.745338
256: skipped 1.515616 1.268220 1.036881
511: skipped 4.066155 4.303350 3.930661
512: skipped 3.976139 4.337525 3.968935
3-dimensional arrays ('n' is the size of the image AND the kernel)
n convolve convolve_fftnp convolve_fftw convolve_fftsp
7: 0.009239 0.012957 0.011957 0.015997
8: 0.012405 0.011328 0.011677 0.012283
15: 0.772434 0.075621 0.056711 0.079508
16: 0.964635 0.105846 0.072811 0.104611
31: 62.824051 2.295193 1.189505 2.351136
32: 79.507060 1.169182 0.821779 1.275770
63: skipped 11.250225 10.982726 10.585744
64: skipped 10.013558 11.507645 12.665557
On a 2009 Mac Pro:
1-dimensional arrays ('n' is the size of the image AND the kernel)
n convolve convolve_fftnp convolve_fftw convolve_fftsp
7: 0.000360 0.002269 0.004986 0.002476
8: 0.000361 0.002468 0.005242 0.002696
15: 0.000364 0.002255 0.005244 0.002471
16: 0.000365 0.002506 0.005286 0.002727
31: 0.000385 0.002380 0.005422 0.002588
32: 0.000385 0.002531 0.005543 0.002737
63: 0.000474 0.002407 0.005392 0.002637
64: 0.000484 0.002602 0.005631 0.002823
127: 0.000752 0.004122 0.007827 0.003966
128: 0.000757 0.002763 0.005844 0.002958
255: 0.004316 0.003258 0.006566 0.003324
256: 0.004354 0.003180 0.006120 0.003245
511: 0.011517 0.007158 0.009898 0.006238
512: 0.011482 0.003873 0.006777 0.003820
1023: 0.034105 0.009211 0.009468 0.008260
1024: 0.034609 0.005504 0.008399 0.005080
2047: 0.113620 0.028097 0.020662 0.021603
2048: 0.112828 0.008403 0.010939 0.007331
4095: 0.403373 0.023211 0.018767 0.020065
4096: 0.403316 0.017550 0.017853 0.013651
8191: 1.519329 8.454573 0.211436 7.212381
8192: 1.519082 0.033148 0.030370 0.025905
16383: 5.887481 0.317428 0.153344 0.237119
16384: 5.888222 0.069379 0.065264 0.052847
2-dimensional arrays ('n' is the size of the image AND the kernel)
n convolve convolve_fftnp convolve_fftw convolve_fftsp
7: 0.000474 0.003470 0.006131 0.003503
8: 0.000503 0.003565 0.006400 0.003586
15: 0.002011 0.004481 0.007825 0.004496
16: 0.002236 0.004744 0.007078 0.004680
31: 0.027291 0.019433 0.014841 0.018034
32: 0.029283 0.009244 0.010161 0.008964
63: 0.445680 0.038171 0.026753 0.037404
64: 0.460616 0.028128 0.029487 0.029149
127: 7.003774 0.925921 0.282591 0.762671
128: 7.063657 0.110838 0.104402 0.133523
255: skipped 0.804682 0.708849 0.869368
256: skipped 0.797800 0.721042 0.880848
511: skipped 3.643626 3.687562 4.584770
512: skipped 3.715215 4.893539 5.538462
3-dimensional arrays ('n' is the size of the image AND the kernel)
n convolve convolve_fftnp convolve_fftw convolve_fftsp
7: 0.004520 0.011519 0.009464 0.012335
8: 0.006422 0.010294 0.010220 0.011711
15: 0.329566 0.060978 0.045495 0.073692
16: 0.405275 0.069999 0.040659 0.086114
31: 24.935228 1.654920 0.710509 1.773879
32: 27.524226 0.724053 0.543507 1.027568
63: skipped 8.982771 12.407683 16.900078
64: skipped 8.956070 11.934627 17.296447
"""
|
