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
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
|
#!/usr/bin/env python
"""
Unit tests for skcuda.fft
"""
from __future__ import division
from unittest import main, makeSuite, TestCase, TestSuite
import pycuda.driver as drv
import pycuda.gpuarray as gpuarray
from pycuda.tools import clear_context_caches, make_default_context
import pycuda.gpuarray as gpuarray
import numpy as np
import skcuda.fft as fft
import skcuda.misc as misc
drv.init()
atol_float32 = 1e-6
atol_float64 = 1e-8
class test_fft(TestCase):
@classmethod
def setUpClass(cls):
cls.ctx = make_default_context()
@classmethod
def tearDownClass(cls):
cls.ctx.pop()
clear_context_caches()
def setUp(self):
np.random.seed(0) # for reproducible tests
self.N = 8
self.M = 4
self.B = 3
def test_fft_float32_to_complex64_1d(self):
x = np.asarray(np.random.rand(self.N), np.float32)
xf = np.fft.rfftn(x)
x_gpu = gpuarray.to_gpu(x)
xf_gpu = gpuarray.empty(self.N//2+1, np.complex64)
plan = fft.Plan(x.shape, np.float32, np.complex64)
fft.fft(x_gpu, xf_gpu, plan)
np.testing.assert_allclose(xf, xf_gpu.get(), atol=atol_float32)
def test_fft_float32_to_complex64_2d(self):
x = np.asarray(np.random.rand(self.N, self.M), np.float32)
xf = np.fft.rfftn(x)
x_gpu = gpuarray.to_gpu(x)
xf_gpu = gpuarray.empty((self.N, self.M//2+1), np.complex64)
plan = fft.Plan(x.shape, np.float32, np.complex64)
fft.fft(x_gpu, xf_gpu, plan)
np.testing.assert_allclose(xf, xf_gpu.get(), atol=atol_float32)
def test_batch_fft_float32_to_complex64_1d(self):
x = np.asarray(np.random.rand(self.B, self.N), np.float32)
xf = np.fft.rfft(x, axis=1)
x_gpu = gpuarray.to_gpu(x)
xf_gpu = gpuarray.empty((self.B, self.N//2+1), np.complex64)
plan = fft.Plan(x.shape[1], np.float32, np.complex64, batch=self.B)
fft.fft(x_gpu, xf_gpu, plan)
np.testing.assert_allclose(xf, xf_gpu.get(), atol=atol_float32)
def test_batch_fft_float32_to_complex64_2d(self):
x = np.asarray(np.random.rand(self.B, self.N, self.M), np.float32)
xf = np.fft.rfftn(x, axes=(1,2))
x_gpu = gpuarray.to_gpu(x)
xf_gpu = gpuarray.empty((self.B, self.N, self.M//2+1), np.complex64)
plan = fft.Plan([self.N, self.M], np.float32, np.complex64, batch=self.B)
fft.fft(x_gpu, xf_gpu, plan)
np.testing.assert_allclose(xf, xf_gpu.get(), atol=atol_float32)
def test_fft_float64_to_complex128_1d(self):
x = np.asarray(np.random.rand(self.N), np.float64)
xf = np.fft.rfftn(x)
x_gpu = gpuarray.to_gpu(x)
xf_gpu = gpuarray.empty(self.N//2+1, np.complex128)
plan = fft.Plan(x.shape, np.float64, np.complex128)
fft.fft(x_gpu, xf_gpu, plan)
np.testing.assert_allclose(xf, xf_gpu.get(), atol=atol_float64)
def test_fft_float64_to_complex128_2d(self):
x = np.asarray(np.random.rand(self.N, self.M), np.float64)
xf = np.fft.rfftn(x)
x_gpu = gpuarray.to_gpu(x)
xf_gpu = gpuarray.empty((self.N, self.M//2+1), np.complex128)
plan = fft.Plan(x.shape, np.float64, np.complex128)
fft.fft(x_gpu, xf_gpu, plan)
np.testing.assert_allclose(xf, xf_gpu.get(), atol=atol_float64)
def test_batch_fft_float64_to_complex128_1d(self):
x = np.asarray(np.random.rand(self.B, self.N), np.float64)
xf = np.fft.rfft(x, axis=1)
x_gpu = gpuarray.to_gpu(x)
xf_gpu = gpuarray.empty((self.B, self.N//2+1), np.complex128)
plan = fft.Plan(x.shape[1], np.float64, np.complex128, batch=self.B)
fft.fft(x_gpu, xf_gpu, plan)
np.testing.assert_allclose(xf, xf_gpu.get(), atol=atol_float64)
def test_batch_fft_float64_to_complex128_2d(self):
x = np.asarray(np.random.rand(self.B, self.N, self.M), np.float64)
xf = np.fft.rfftn(x, axes=(1,2))
x_gpu = gpuarray.to_gpu(x)
xf_gpu = gpuarray.empty((self.B, self.N, self.M//2+1), np.complex128)
plan = fft.Plan([self.N, self.M], np.float64, np.complex128, batch=self.B)
fft.fft(x_gpu, xf_gpu, plan)
np.testing.assert_allclose(xf, xf_gpu.get(), atol=atol_float64)
def test_ifft_complex64_to_float32_1d(self):
x = np.asarray(np.random.rand(self.N), np.float32)
xf = np.asarray(np.fft.rfftn(x), np.complex64)
xf_gpu = gpuarray.to_gpu(xf)
x_gpu = gpuarray.empty(self.N, np.float32)
plan = fft.Plan(x.shape, np.complex64, np.float32)
fft.ifft(xf_gpu, x_gpu, plan, True)
np.testing.assert_allclose(x, x_gpu.get(), atol=atol_float32)
def test_ifft_complex64_to_float32_2d(self):
# Note that since rfftn returns a Fortran-ordered array, it
# needs to be reformatted as a C-ordered array before being
# passed to gpuarray.to_gpu:
x = np.asarray(np.random.rand(self.N, self.M), np.float32)
xf = np.asarray(np.fft.rfftn(x), np.complex64)
xf_gpu = gpuarray.to_gpu(np.ascontiguousarray(xf))
x_gpu = gpuarray.empty((self.N, self.M), np.float32)
plan = fft.Plan(x.shape, np.complex64, np.float32)
fft.ifft(xf_gpu, x_gpu, plan, True)
np.testing.assert_allclose(x, x_gpu.get(), atol=atol_float32)
def test_batch_ifft_complex64_to_float32_1d(self):
# Note that since rfftn returns a Fortran-ordered array, it
# needs to be reformatted as a C-ordered array before being
# passed to gpuarray.to_gpu:
x = np.asarray(np.random.rand(self.B, self.N), np.float32)
xf = np.asarray(np.fft.rfft(x, axis=1), np.complex64)
xf_gpu = gpuarray.to_gpu(np.ascontiguousarray(xf))
x_gpu = gpuarray.empty((self.B, self.N), np.float32)
plan = fft.Plan(x.shape[1], np.complex64, np.float32, batch=self.B)
fft.ifft(xf_gpu, x_gpu, plan, True)
np.testing.assert_allclose(x, x_gpu.get(), atol=atol_float32)
def test_batch_ifft_complex64_to_float32_2d(self):
# Note that since rfftn returns a Fortran-ordered array, it
# needs to be reformatted as a C-ordered array before being
# passed to gpuarray.to_gpu:
x = np.asarray(np.random.rand(self.B, self.N, self.M), np.float32)
xf = np.asarray(np.fft.rfftn(x, axes=(1,2)), np.complex64)
xf_gpu = gpuarray.to_gpu(np.ascontiguousarray(xf))
x_gpu = gpuarray.empty((self.B, self.N, self.M), np.float32)
plan = fft.Plan([self.N, self.M], np.complex64, np.float32, batch=self.B)
fft.ifft(xf_gpu, x_gpu, plan, True)
np.testing.assert_allclose(x, x_gpu.get(), atol=atol_float32)
def test_ifft_complex128_to_float64_1d(self):
x = np.asarray(np.random.rand(self.N), np.float64)
xf = np.asarray(np.fft.rfftn(x), np.complex128)
xf_gpu = gpuarray.to_gpu(xf)
x_gpu = gpuarray.empty(self.N, np.float64)
plan = fft.Plan(x.shape, np.complex128, np.float64)
fft.ifft(xf_gpu, x_gpu, plan, True)
np.testing.assert_allclose(x, x_gpu.get(), atol=atol_float64)
def test_ifft_complex128_to_float64_2d(self):
# Note that since rfftn returns a Fortran-ordered array, it
# needs to be reformatted as a C-ordered array before being
# passed to gpuarray.to_gpu:
x = np.asarray(np.random.rand(self.N, self.M), np.float64)
xf = np.asarray(np.fft.rfftn(x), np.complex128)
xf_gpu = gpuarray.to_gpu(np.ascontiguousarray(xf))
x_gpu = gpuarray.empty((self.N, self.M), np.float64)
plan = fft.Plan(x.shape, np.complex128, np.float64)
fft.ifft(xf_gpu, x_gpu, plan, True)
np.testing.assert_allclose(x, x_gpu.get(), atol=atol_float64)
def test_batch_ifft_complex128_to_float64_1d(self):
# Note that since rfftn returns a Fortran-ordered array, it
# needs to be reformatted as a C-ordered array before being
# passed to gpuarray.to_gpu:
x = np.asarray(np.random.rand(self.B, self.N), np.float64)
xf = np.asarray(np.fft.rfft(x, axis=1), np.complex128)
xf_gpu = gpuarray.to_gpu(np.ascontiguousarray(xf))
x_gpu = gpuarray.empty((self.B, self.N), np.float64)
plan = fft.Plan(x.shape[1], np.complex128, np.float64, batch=self.B)
fft.ifft(xf_gpu, x_gpu, plan, True)
np.testing.assert_allclose(x, x_gpu.get(), atol=atol_float64)
def test_batch_ifft_complex128_to_float64_2d(self):
# Note that since rfftn returns a Fortran-ordered array, it
# needs to be reformatted as a C-ordered array before being
# passed to gpuarray.to_gpu:
x = np.asarray(np.random.rand(self.B, self.N, self.M), np.float64)
xf = np.asarray(np.fft.rfftn(x, axes=(1,2)), np.complex128)
xf_gpu = gpuarray.to_gpu(np.ascontiguousarray(xf))
x_gpu = gpuarray.empty((self.B, self.N, self.M), np.float64)
plan = fft.Plan([self.N, self.M], np.complex128, np.float64, batch=self.B)
fft.ifft(xf_gpu, x_gpu, plan, True)
np.testing.assert_allclose(x, x_gpu.get(), atol=atol_float64)
def test_multiple_streams(self):
x = np.asarray(np.random.rand(self.N), np.float32)
xf = np.fft.rfftn(x)
y = np.asarray(np.random.rand(self.N), np.float32)
yf = np.fft.rfftn(y)
x_gpu = gpuarray.to_gpu(x)
y_gpu = gpuarray.to_gpu(y)
xf_gpu = gpuarray.empty(self.N//2+1, np.complex64)
yf_gpu = gpuarray.empty(self.N//2+1, np.complex64)
stream0 = drv.Stream()
stream1 = drv.Stream()
plan1 = fft.Plan(x.shape, np.float32, np.complex64, stream=stream0)
plan2 = fft.Plan(y.shape, np.float32, np.complex64, stream=stream1)
fft.fft(x_gpu, xf_gpu, plan1)
fft.fft(y_gpu, yf_gpu, plan2)
np.testing.assert_allclose(xf, xf_gpu.get(), atol=atol_float32)
np.testing.assert_allclose(yf, yf_gpu.get(), atol=atol_float32)
def test_work_area(self):
x = np.asarray(np.random.rand(self.N), np.float32)
xf = np.fft.rfftn(x)
x_gpu = gpuarray.to_gpu(x)
xf_gpu = gpuarray.empty(self.N//2+1, np.complex64)
plan = fft.Plan(x.shape, np.float32, np.complex64, auto_allocate=False)
work_area = gpuarray.empty((plan.worksize,), np.uint8)
plan.set_work_area(work_area)
fft.fft(x_gpu, xf_gpu, plan)
np.testing.assert_allclose(xf, xf_gpu.get(), atol=atol_float32)
def suite():
context = make_default_context()
device = context.get_device()
context.pop()
s = TestSuite()
s.addTest(test_fft('test_fft_float32_to_complex64_1d'))
s.addTest(test_fft('test_fft_float32_to_complex64_2d'))
s.addTest(test_fft('test_batch_fft_float32_to_complex64_1d'))
s.addTest(test_fft('test_batch_fft_float32_to_complex64_2d'))
s.addTest(test_fft('test_ifft_complex64_to_float32_1d'))
s.addTest(test_fft('test_ifft_complex64_to_float32_2d'))
s.addTest(test_fft('test_batch_ifft_complex64_to_float32_1d'))
s.addTest(test_fft('test_batch_ifft_complex64_to_float32_2d'))
s.addTest(test_fft('test_multiple_streams'))
s.addTest(test_fft('test_work_area'))
if misc.get_compute_capability(device) >= 1.3:
s.addTest(test_fft('test_fft_float64_to_complex128_1d'))
s.addTest(test_fft('test_fft_float64_to_complex128_2d'))
s.addTest(test_fft('test_batch_fft_float64_to_complex128_1d'))
s.addTest(test_fft('test_batch_fft_float64_to_complex128_2d'))
s.addTest(test_fft('test_ifft_complex128_to_float64_1d'))
s.addTest(test_fft('test_ifft_complex128_to_float64_2d'))
s.addTest(test_fft('test_batch_ifft_complex128_to_float64_1d'))
s.addTest(test_fft('test_batch_ifft_complex128_to_float64_2d'))
return s
if __name__ == '__main__':
main(defaultTest = 'suite')
|
