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
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
|
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include "test/hiprec_convolve_test_util.h"
#include "av1/common/restoration.h"
using std::make_tuple;
using std::tuple;
namespace libaom_test {
// Generate a random pair of filter kernels, using the ranges
// of possible values from the loop-restoration experiment
static void generate_kernels(ACMRandom *rnd, InterpKernel hkernel,
InterpKernel vkernel, int kernel_type = 2) {
if (kernel_type == 0) {
// Low possible values for filter coefficients
hkernel[0] = hkernel[6] = vkernel[0] = vkernel[6] = WIENER_FILT_TAP0_MINV;
hkernel[1] = hkernel[5] = vkernel[1] = vkernel[5] = WIENER_FILT_TAP1_MINV;
hkernel[2] = hkernel[4] = vkernel[2] = vkernel[4] = WIENER_FILT_TAP2_MINV;
hkernel[3] = vkernel[3] = -2 * (hkernel[0] + hkernel[1] + hkernel[2]);
hkernel[7] = vkernel[7] = 0;
} else if (kernel_type == 1) {
// Max possible values for filter coefficients
hkernel[0] = hkernel[6] = vkernel[0] = vkernel[6] = WIENER_FILT_TAP0_MAXV;
hkernel[1] = hkernel[5] = vkernel[1] = vkernel[5] = WIENER_FILT_TAP1_MAXV;
hkernel[2] = hkernel[4] = vkernel[2] = vkernel[4] = WIENER_FILT_TAP2_MAXV;
hkernel[3] = vkernel[3] = -2 * (hkernel[0] + hkernel[1] + hkernel[2]);
hkernel[7] = vkernel[7] = 0;
} else {
// Randomly generated values for filter coefficients
hkernel[0] = hkernel[6] =
WIENER_FILT_TAP0_MINV +
rnd->PseudoUniform(WIENER_FILT_TAP0_MAXV + 1 - WIENER_FILT_TAP0_MINV);
hkernel[1] = hkernel[5] =
WIENER_FILT_TAP1_MINV +
rnd->PseudoUniform(WIENER_FILT_TAP1_MAXV + 1 - WIENER_FILT_TAP1_MINV);
hkernel[2] = hkernel[4] =
WIENER_FILT_TAP2_MINV +
rnd->PseudoUniform(WIENER_FILT_TAP2_MAXV + 1 - WIENER_FILT_TAP2_MINV);
hkernel[3] = -2 * (hkernel[0] + hkernel[1] + hkernel[2]);
hkernel[7] = 0;
vkernel[0] = vkernel[6] =
WIENER_FILT_TAP0_MINV +
rnd->PseudoUniform(WIENER_FILT_TAP0_MAXV + 2 - WIENER_FILT_TAP0_MINV);
vkernel[1] = vkernel[5] =
WIENER_FILT_TAP1_MINV +
rnd->PseudoUniform(WIENER_FILT_TAP1_MAXV + 2 - WIENER_FILT_TAP1_MINV);
vkernel[2] = vkernel[4] =
WIENER_FILT_TAP2_MINV +
rnd->PseudoUniform(WIENER_FILT_TAP2_MAXV + 2 - WIENER_FILT_TAP2_MINV);
vkernel[3] = -2 * (vkernel[0] + vkernel[1] + vkernel[2]);
vkernel[7] = 0;
}
}
namespace AV1HiprecConvolve {
::testing::internal::ParamGenerator<HiprecConvolveParam> BuildParams(
hiprec_convolve_func filter) {
const HiprecConvolveParam params[] = {
make_tuple(8, 8, 50000, filter), make_tuple(8, 4, 50000, filter),
make_tuple(64, 24, 1000, filter), make_tuple(64, 64, 1000, filter),
make_tuple(64, 56, 1000, filter), make_tuple(32, 8, 10000, filter),
make_tuple(32, 28, 10000, filter), make_tuple(32, 32, 10000, filter),
make_tuple(16, 34, 10000, filter), make_tuple(32, 34, 10000, filter),
make_tuple(64, 34, 1000, filter), make_tuple(8, 17, 10000, filter),
make_tuple(16, 17, 10000, filter), make_tuple(32, 17, 10000, filter)
};
return ::testing::ValuesIn(params);
}
AV1HiprecConvolveTest::~AV1HiprecConvolveTest() {}
void AV1HiprecConvolveTest::SetUp() {
rnd_.Reset(ACMRandom::DeterministicSeed());
}
void AV1HiprecConvolveTest::TearDown() { libaom_test::ClearSystemState(); }
void AV1HiprecConvolveTest::RunCheckOutput(hiprec_convolve_func test_impl) {
const int w = 128, h = 128;
const int out_w = GET_PARAM(0), out_h = GET_PARAM(1);
const int num_iters = GET_PARAM(2);
int i, j, k, m;
const ConvolveParams conv_params = get_conv_params_wiener(8);
uint8_t *input_ = new uint8_t[h * w];
uint8_t *input = input_;
// The AVX2 convolve functions always write rows with widths that are
// multiples of 16. So to avoid a buffer overflow, we may need to pad
// rows to a multiple of 16.
int output_n = ALIGN_POWER_OF_TWO(out_w, 4) * out_h;
uint8_t *output = new uint8_t[output_n];
uint8_t *output2 = new uint8_t[output_n];
// Generate random filter kernels
DECLARE_ALIGNED(16, InterpKernel, hkernel);
DECLARE_ALIGNED(16, InterpKernel, vkernel);
for (int kernel_type = 0; kernel_type < 3; kernel_type++) {
generate_kernels(&rnd_, hkernel, vkernel, kernel_type);
for (i = 0; i < num_iters; ++i) {
for (k = 0; k < h; ++k)
for (m = 0; m < w; ++m) input[k * w + m] = rnd_.Rand8();
// Choose random locations within the source block
int offset_r = 3 + rnd_.PseudoUniform(h - out_h - 7);
int offset_c = 3 + rnd_.PseudoUniform(w - out_w - 7);
av1_wiener_convolve_add_src_c(input + offset_r * w + offset_c, w, output,
out_w, hkernel, 16, vkernel, 16, out_w,
out_h, &conv_params);
test_impl(input + offset_r * w + offset_c, w, output2, out_w, hkernel, 16,
vkernel, 16, out_w, out_h, &conv_params);
for (j = 0; j < out_w * out_h; ++j)
ASSERT_EQ(output[j], output2[j])
<< "Pixel mismatch at index " << j << " = (" << (j % out_w) << ", "
<< (j / out_w) << ") on iteration " << i;
}
}
delete[] input_;
delete[] output;
delete[] output2;
}
void AV1HiprecConvolveTest::RunSpeedTest(hiprec_convolve_func test_impl) {
const int w = 128, h = 128;
const int out_w = GET_PARAM(0), out_h = GET_PARAM(1);
const int num_iters = GET_PARAM(2) / 500;
int i, j, k;
const ConvolveParams conv_params = get_conv_params_wiener(8);
uint8_t *input_ = new uint8_t[h * w];
uint8_t *input = input_;
// The AVX2 convolve functions always write rows with widths that are
// multiples of 16. So to avoid a buffer overflow, we may need to pad
// rows to a multiple of 16.
int output_n = ALIGN_POWER_OF_TWO(out_w, 4) * out_h;
uint8_t *output = new uint8_t[output_n];
uint8_t *output2 = new uint8_t[output_n];
// Generate random filter kernels
DECLARE_ALIGNED(16, InterpKernel, hkernel);
DECLARE_ALIGNED(16, InterpKernel, vkernel);
generate_kernels(&rnd_, hkernel, vkernel);
for (i = 0; i < h; ++i)
for (j = 0; j < w; ++j) input[i * w + j] = rnd_.Rand8();
aom_usec_timer ref_timer;
aom_usec_timer_start(&ref_timer);
for (i = 0; i < num_iters; ++i) {
for (j = 3; j < h - out_h - 4; j++) {
for (k = 3; k < w - out_w - 4; k++) {
av1_wiener_convolve_add_src_c(input + j * w + k, w, output, out_w,
hkernel, 16, vkernel, 16, out_w, out_h,
&conv_params);
}
}
}
aom_usec_timer_mark(&ref_timer);
const int64_t ref_time = aom_usec_timer_elapsed(&ref_timer);
aom_usec_timer tst_timer;
aom_usec_timer_start(&tst_timer);
for (i = 0; i < num_iters; ++i) {
for (j = 3; j < h - out_h - 4; j++) {
for (k = 3; k < w - out_w - 4; k++) {
test_impl(input + j * w + k, w, output2, out_w, hkernel, 16, vkernel,
16, out_w, out_h, &conv_params);
}
}
}
aom_usec_timer_mark(&tst_timer);
const int64_t tst_time = aom_usec_timer_elapsed(&tst_timer);
std::cout << "[ ] C time = " << ref_time / 1000
<< " ms, SIMD time = " << tst_time / 1000 << " ms\n";
EXPECT_GT(ref_time, tst_time)
<< "Error: AV1HiprecConvolveTest.SpeedTest, SIMD slower than C.\n"
<< "C time: " << ref_time << " us\n"
<< "SIMD time: " << tst_time << " us\n";
delete[] input_;
delete[] output;
delete[] output2;
}
} // namespace AV1HiprecConvolve
#if CONFIG_AV1_HIGHBITDEPTH
namespace AV1HighbdHiprecConvolve {
::testing::internal::ParamGenerator<HighbdHiprecConvolveParam> BuildParams(
highbd_hiprec_convolve_func filter) {
const HighbdHiprecConvolveParam params[] = {
make_tuple(8, 8, 50000, 8, filter), make_tuple(64, 64, 1000, 8, filter),
make_tuple(32, 8, 10000, 8, filter), make_tuple(8, 8, 50000, 10, filter),
make_tuple(64, 64, 1000, 10, filter), make_tuple(32, 8, 10000, 10, filter),
make_tuple(8, 8, 50000, 12, filter), make_tuple(64, 64, 1000, 12, filter),
make_tuple(32, 8, 10000, 12, filter),
};
return ::testing::ValuesIn(params);
}
AV1HighbdHiprecConvolveTest::~AV1HighbdHiprecConvolveTest() {}
void AV1HighbdHiprecConvolveTest::SetUp() {
rnd_.Reset(ACMRandom::DeterministicSeed());
}
void AV1HighbdHiprecConvolveTest::TearDown() {
libaom_test::ClearSystemState();
}
void AV1HighbdHiprecConvolveTest::RunCheckOutput(
highbd_hiprec_convolve_func test_impl) {
const int w = 128, h = 128;
const int out_w = GET_PARAM(0), out_h = GET_PARAM(1);
const int num_iters = GET_PARAM(2);
const int bd = GET_PARAM(3);
int i, j;
const ConvolveParams conv_params = get_conv_params_wiener(bd);
uint16_t *input = new uint16_t[h * w];
// The AVX2 convolve functions always write rows with widths that are
// multiples of 16. So to avoid a buffer overflow, we may need to pad
// rows to a multiple of 16.
int output_n = ALIGN_POWER_OF_TWO(out_w, 4) * out_h;
uint16_t *output = new uint16_t[output_n];
uint16_t *output2 = new uint16_t[output_n];
// Generate random filter kernels
DECLARE_ALIGNED(16, InterpKernel, hkernel);
DECLARE_ALIGNED(16, InterpKernel, vkernel);
for (i = 0; i < h; ++i)
for (j = 0; j < w; ++j) input[i * w + j] = rnd_.Rand16() & ((1 << bd) - 1);
uint8_t *input_ptr = CONVERT_TO_BYTEPTR(input);
uint8_t *output_ptr = CONVERT_TO_BYTEPTR(output);
uint8_t *output2_ptr = CONVERT_TO_BYTEPTR(output2);
for (int kernel_type = 0; kernel_type < 3; kernel_type++) {
generate_kernels(&rnd_, hkernel, vkernel, kernel_type);
for (i = 0; i < num_iters; ++i) {
// Choose random locations within the source block
int offset_r = 3 + rnd_.PseudoUniform(h - out_h - 7);
int offset_c = 3 + rnd_.PseudoUniform(w - out_w - 7);
av1_highbd_wiener_convolve_add_src_c(
input_ptr + offset_r * w + offset_c, w, output_ptr, out_w, hkernel,
16, vkernel, 16, out_w, out_h, &conv_params, bd);
test_impl(input_ptr + offset_r * w + offset_c, w, output2_ptr, out_w,
hkernel, 16, vkernel, 16, out_w, out_h, &conv_params, bd);
for (j = 0; j < out_w * out_h; ++j)
ASSERT_EQ(output[j], output2[j])
<< "Pixel mismatch at index " << j << " = (" << (j % out_w) << ", "
<< (j / out_w) << ") on iteration " << i;
}
}
delete[] input;
delete[] output;
delete[] output2;
}
void AV1HighbdHiprecConvolveTest::RunSpeedTest(
highbd_hiprec_convolve_func test_impl) {
const int w = 128, h = 128;
const int out_w = GET_PARAM(0), out_h = GET_PARAM(1);
const int num_iters = GET_PARAM(2) / 500;
const int bd = GET_PARAM(3);
int i, j, k;
const ConvolveParams conv_params = get_conv_params_wiener(bd);
uint16_t *input = new uint16_t[h * w];
// The AVX2 convolve functions always write rows with widths that are
// multiples of 16. So to avoid a buffer overflow, we may need to pad
// rows to a multiple of 16.
int output_n = ALIGN_POWER_OF_TWO(out_w, 4) * out_h;
uint16_t *output = new uint16_t[output_n];
uint16_t *output2 = new uint16_t[output_n];
// Generate random filter kernels
DECLARE_ALIGNED(16, InterpKernel, hkernel);
DECLARE_ALIGNED(16, InterpKernel, vkernel);
generate_kernels(&rnd_, hkernel, vkernel);
for (i = 0; i < h; ++i)
for (j = 0; j < w; ++j) input[i * w + j] = rnd_.Rand16() & ((1 << bd) - 1);
uint8_t *input_ptr = CONVERT_TO_BYTEPTR(input);
uint8_t *output_ptr = CONVERT_TO_BYTEPTR(output);
uint8_t *output2_ptr = CONVERT_TO_BYTEPTR(output2);
aom_usec_timer ref_timer;
aom_usec_timer_start(&ref_timer);
for (i = 0; i < num_iters; ++i) {
for (j = 3; j < h - out_h - 4; j++) {
for (k = 3; k < w - out_w - 4; k++) {
av1_highbd_wiener_convolve_add_src_c(
input_ptr + j * w + k, w, output_ptr, out_w, hkernel, 16, vkernel,
16, out_w, out_h, &conv_params, bd);
}
}
}
aom_usec_timer_mark(&ref_timer);
const int64_t ref_time = aom_usec_timer_elapsed(&ref_timer);
aom_usec_timer tst_timer;
aom_usec_timer_start(&tst_timer);
for (i = 0; i < num_iters; ++i) {
for (j = 3; j < h - out_h - 4; j++) {
for (k = 3; k < w - out_w - 4; k++) {
test_impl(input_ptr + j * w + k, w, output2_ptr, out_w, hkernel, 16,
vkernel, 16, out_w, out_h, &conv_params, bd);
}
}
}
aom_usec_timer_mark(&tst_timer);
const int64_t tst_time = aom_usec_timer_elapsed(&tst_timer);
std::cout << "[ ] C time = " << ref_time / 1000
<< " ms, SIMD time = " << tst_time / 1000 << " ms\n";
EXPECT_GT(ref_time, tst_time)
<< "Error: AV1HighbdHiprecConvolveTest.SpeedTest, SIMD slower than C.\n"
<< "C time: " << ref_time << " us\n"
<< "SIMD time: " << tst_time << " us\n";
delete[] input;
delete[] output;
delete[] output2;
}
} // namespace AV1HighbdHiprecConvolve
#endif // CONFIG_AV1_HIGHBITDEPTH
} // namespace libaom_test
|
