?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Bug 1797755 - Part 5: Use a single initial mark stack size regardless of whether...
[gecko.git] / third_party / aom / test / av1_wedge_utils_test.cc
blobe8fbe69a4c45d730e6110cc1267d851ae94edaa4
1 /*
2 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
3 *
4 * This source code is subject to the terms of the BSD 2 Clause License and
5 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
6 * was not distributed with this source code in the LICENSE file, you can
7 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
8 * Media Patent License 1.0 was not distributed with this source code in the
9 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
10 */
11
12 #include "third_party/googletest/src/googletest/include/gtest/gtest.h"
13
14 #include "config/aom_config.h"
15 #include "config/aom_dsp_rtcd.h"
16 #include "config/av1_rtcd.h"
17
18 #include "aom_dsp/aom_dsp_common.h"
19
20 #include "av1/common/enums.h"
21
22 #include "test/acm_random.h"
23 #include "test/function_equivalence_test.h"
24 #include "test/register_state_check.h"
25
26 #define WEDGE_WEIGHT_BITS 6
27 #define MAX_MASK_VALUE (1 << (WEDGE_WEIGHT_BITS))
28
29 using libaom_test::ACMRandom;
30 using libaom_test::FunctionEquivalenceTest;
31
32 namespace {
33
34 static const int16_t kInt13Max = (1 << 12) - 1;
35
36 //////////////////////////////////////////////////////////////////////////////
37 // av1_wedge_sse_from_residuals - functionality
38 //////////////////////////////////////////////////////////////////////////////
39
40 class WedgeUtilsSSEFuncTest : public testing::Test {
41 protected:
42 WedgeUtilsSSEFuncTest() : rng_(ACMRandom::DeterministicSeed()) {}
43
44 static const int kIterations = 1000;
45
46 ACMRandom rng_;
47 };
48
49 static void equiv_blend_residuals(int16_t *r, const int16_t *r0,
50 const int16_t *r1, const uint8_t *m, int N) {
51 for (int i = 0; i < N; i++) {
52 const int32_t m0 = m[i];
53 const int32_t m1 = MAX_MASK_VALUE - m0;
54 const int16_t R = m0 * r0[i] + m1 * r1[i];
55 // Note that this rounding is designed to match the result
56 // you would get when actually blending the 2 predictors and computing
57 // the residuals.
58 r[i] = ROUND_POWER_OF_TWO(R - 1, WEDGE_WEIGHT_BITS);
59 }
60 }
61
62 static uint64_t equiv_sse_from_residuals(const int16_t *r0, const int16_t *r1,
63 const uint8_t *m, int N) {
64 uint64_t acc = 0;
65 for (int i = 0; i < N; i++) {
66 const int32_t m0 = m[i];
67 const int32_t m1 = MAX_MASK_VALUE - m0;
68 const int16_t R = m0 * r0[i] + m1 * r1[i];
69 const int32_t r = ROUND_POWER_OF_TWO(R - 1, WEDGE_WEIGHT_BITS);
70 acc += r * r;
71 }
72 return acc;
73 }
74
75 TEST_F(WedgeUtilsSSEFuncTest, ResidualBlendingEquiv) {
76 DECLARE_ALIGNED(32, uint8_t, s[MAX_SB_SQUARE]);
77 DECLARE_ALIGNED(32, uint8_t, p0[MAX_SB_SQUARE]);
78 DECLARE_ALIGNED(32, uint8_t, p1[MAX_SB_SQUARE]);
79 DECLARE_ALIGNED(32, uint8_t, p[MAX_SB_SQUARE]);
80
81 DECLARE_ALIGNED(32, int16_t, r0[MAX_SB_SQUARE]);
82 DECLARE_ALIGNED(32, int16_t, r1[MAX_SB_SQUARE]);
83 DECLARE_ALIGNED(32, int16_t, r_ref[MAX_SB_SQUARE]);
84 DECLARE_ALIGNED(32, int16_t, r_tst[MAX_SB_SQUARE]);
85 DECLARE_ALIGNED(32, uint8_t, m[MAX_SB_SQUARE]);
86
87 for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
88 for (int i = 0; i < MAX_SB_SQUARE; ++i) {
89 s[i] = rng_.Rand8();
90 m[i] = rng_(MAX_MASK_VALUE + 1);
91 }
92
93 const int w = 1 << (rng_(MAX_SB_SIZE_LOG2 + 1 - 3) + 3);
94 const int h = 1 << (rng_(MAX_SB_SIZE_LOG2 + 1 - 3) + 3);
95 const int N = w * h;
96
97 for (int j = 0; j < N; j++) {
98 p0[j] = clamp(s[j] + rng_(33) - 16, 0, UINT8_MAX);
99 p1[j] = clamp(s[j] + rng_(33) - 16, 0, UINT8_MAX);
100 }
101
102 aom_blend_a64_mask(p, w, p0, w, p1, w, m, w, w, h, 0, 0);
103
104 aom_subtract_block(h, w, r0, w, s, w, p0, w);
105 aom_subtract_block(h, w, r1, w, s, w, p1, w);
106
107 aom_subtract_block(h, w, r_ref, w, s, w, p, w);
108 equiv_blend_residuals(r_tst, r0, r1, m, N);
109
110 for (int i = 0; i < N; ++i) ASSERT_EQ(r_ref[i], r_tst[i]);
111
112 uint64_t ref_sse = aom_sum_squares_i16(r_ref, N);
113 uint64_t tst_sse = equiv_sse_from_residuals(r0, r1, m, N);
114
115 ASSERT_EQ(ref_sse, tst_sse);
116 }
117 }
118
119 static uint64_t sse_from_residuals(const int16_t *r0, const int16_t *r1,
120 const uint8_t *m, int N) {
121 uint64_t acc = 0;
122 for (int i = 0; i < N; i++) {
123 const int32_t m0 = m[i];
124 const int32_t m1 = MAX_MASK_VALUE - m0;
125 const int32_t r = m0 * r0[i] + m1 * r1[i];
126 acc += r * r;
127 }
128 return ROUND_POWER_OF_TWO(acc, 2 * WEDGE_WEIGHT_BITS);
129 }
130
131 TEST_F(WedgeUtilsSSEFuncTest, ResidualBlendingMethod) {
132 DECLARE_ALIGNED(32, int16_t, r0[MAX_SB_SQUARE]);
133 DECLARE_ALIGNED(32, int16_t, r1[MAX_SB_SQUARE]);
134 DECLARE_ALIGNED(32, int16_t, d[MAX_SB_SQUARE]);
135 DECLARE_ALIGNED(32, uint8_t, m[MAX_SB_SQUARE]);
136
137 for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
138 for (int i = 0; i < MAX_SB_SQUARE; ++i) {
139 r1[i] = rng_(2 * INT8_MAX - 2 * INT8_MIN + 1) + 2 * INT8_MIN;
140 d[i] = rng_(2 * INT8_MAX - 2 * INT8_MIN + 1) + 2 * INT8_MIN;
141 m[i] = rng_(MAX_MASK_VALUE + 1);
142 }
143
144 const int N = 64 * (rng_(MAX_SB_SQUARE / 64) + 1);
145
146 for (int i = 0; i < N; i++) r0[i] = r1[i] + d[i];
147
148 const uint64_t ref_res = sse_from_residuals(r0, r1, m, N);
149 const uint64_t tst_res = av1_wedge_sse_from_residuals(r1, d, m, N);
150
151 ASSERT_EQ(ref_res, tst_res);
152 }
153 }
154
155 //////////////////////////////////////////////////////////////////////////////
156 // av1_wedge_sse_from_residuals - optimizations
157 //////////////////////////////////////////////////////////////////////////////
158
159 typedef uint64_t (*FSSE)(const int16_t *r1, const int16_t *d, const uint8_t *m,
160 int N);
161 typedef libaom_test::FuncParam<FSSE> TestFuncsFSSE;
162
163 class WedgeUtilsSSEOptTest : public FunctionEquivalenceTest<FSSE> {
164 protected:
165 static const int kIterations = 10000;
166 };
167
168 TEST_P(WedgeUtilsSSEOptTest, RandomValues) {
169 DECLARE_ALIGNED(32, int16_t, r1[MAX_SB_SQUARE]);
170 DECLARE_ALIGNED(32, int16_t, d[MAX_SB_SQUARE]);
171 DECLARE_ALIGNED(32, uint8_t, m[MAX_SB_SQUARE]);
172
173 for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
174 for (int i = 0; i < MAX_SB_SQUARE; ++i) {
175 r1[i] = rng_(2 * kInt13Max + 1) - kInt13Max;
176 d[i] = rng_(2 * kInt13Max + 1) - kInt13Max;
177 m[i] = rng_(MAX_MASK_VALUE + 1);
178 }
179
180 const int N = 64 * (rng_(MAX_SB_SQUARE / 64) + 1);
181
182 const uint64_t ref_res = params_.ref_func(r1, d, m, N);
183 uint64_t tst_res;
184 ASM_REGISTER_STATE_CHECK(tst_res = params_.tst_func(r1, d, m, N));
185
186 ASSERT_EQ(ref_res, tst_res);
187 }
188 }
189
190 TEST_P(WedgeUtilsSSEOptTest, ExtremeValues) {
191 DECLARE_ALIGNED(32, int16_t, r1[MAX_SB_SQUARE]);
192 DECLARE_ALIGNED(32, int16_t, d[MAX_SB_SQUARE]);
193 DECLARE_ALIGNED(32, uint8_t, m[MAX_SB_SQUARE]);
194
195 for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
196 if (rng_(2)) {
197 for (int i = 0; i < MAX_SB_SQUARE; ++i) r1[i] = kInt13Max;
198 } else {
199 for (int i = 0; i < MAX_SB_SQUARE; ++i) r1[i] = -kInt13Max;
200 }
201
202 if (rng_(2)) {
203 for (int i = 0; i < MAX_SB_SQUARE; ++i) d[i] = kInt13Max;
204 } else {
205 for (int i = 0; i < MAX_SB_SQUARE; ++i) d[i] = -kInt13Max;
206 }
207
208 for (int i = 0; i < MAX_SB_SQUARE; ++i) m[i] = MAX_MASK_VALUE;
209
210 const int N = 64 * (rng_(MAX_SB_SQUARE / 64) + 1);
211
212 const uint64_t ref_res = params_.ref_func(r1, d, m, N);
213 uint64_t tst_res;
214 ASM_REGISTER_STATE_CHECK(tst_res = params_.tst_func(r1, d, m, N));
215
216 ASSERT_EQ(ref_res, tst_res);
217 }
218 }
219
220 //////////////////////////////////////////////////////////////////////////////
221 // av1_wedge_sign_from_residuals
222 //////////////////////////////////////////////////////////////////////////////
223
224 typedef int (*FSign)(const int16_t *ds, const uint8_t *m, int N, int64_t limit);
225 typedef libaom_test::FuncParam<FSign> TestFuncsFSign;
226
227 class WedgeUtilsSignOptTest : public FunctionEquivalenceTest<FSign> {
228 protected:
229 static const int kIterations = 10000;
230 static const int kMaxSize = 8196; // Size limited by SIMD implementation.
231 };
232
233 TEST_P(WedgeUtilsSignOptTest, RandomValues) {
234 DECLARE_ALIGNED(32, int16_t, r0[MAX_SB_SQUARE]);
235 DECLARE_ALIGNED(32, int16_t, r1[MAX_SB_SQUARE]);
236 DECLARE_ALIGNED(32, int16_t, ds[MAX_SB_SQUARE]);
237 DECLARE_ALIGNED(32, uint8_t, m[MAX_SB_SQUARE]);
238
239 for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
240 for (int i = 0; i < MAX_SB_SQUARE; ++i) {
241 r0[i] = rng_(2 * kInt13Max + 1) - kInt13Max;
242 r1[i] = rng_(2 * kInt13Max + 1) - kInt13Max;
243 m[i] = rng_(MAX_MASK_VALUE + 1);
244 }
245
246 const int maxN = AOMMIN(kMaxSize, MAX_SB_SQUARE);
247 const int N = 64 * (rng_(maxN / 64 - 1) + 1);
248
249 int64_t limit;
250 limit = (int64_t)aom_sum_squares_i16(r0, N);
251 limit -= (int64_t)aom_sum_squares_i16(r1, N);
252 limit *= (1 << WEDGE_WEIGHT_BITS) / 2;
253
254 for (int i = 0; i < N; i++)
255 ds[i] = clamp(r0[i] * r0[i] - r1[i] * r1[i], INT16_MIN, INT16_MAX);
256
257 const int ref_res = params_.ref_func(ds, m, N, limit);
258 int tst_res;
259 ASM_REGISTER_STATE_CHECK(tst_res = params_.tst_func(ds, m, N, limit));
260
261 ASSERT_EQ(ref_res, tst_res);
262 }
263 }
264
265 TEST_P(WedgeUtilsSignOptTest, ExtremeValues) {
266 DECLARE_ALIGNED(32, int16_t, r0[MAX_SB_SQUARE]);
267 DECLARE_ALIGNED(32, int16_t, r1[MAX_SB_SQUARE]);
268 DECLARE_ALIGNED(32, int16_t, ds[MAX_SB_SQUARE]);
269 DECLARE_ALIGNED(32, uint8_t, m[MAX_SB_SQUARE]);
270
271 for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
272 switch (rng_(4)) {
273 case 0:
274 for (int i = 0; i < MAX_SB_SQUARE; ++i) {
275 r0[i] = 0;
276 r1[i] = kInt13Max;
277 }
278 break;
279 case 1:
280 for (int i = 0; i < MAX_SB_SQUARE; ++i) {
281 r0[i] = kInt13Max;
282 r1[i] = 0;
283 }
284 break;
285 case 2:
286 for (int i = 0; i < MAX_SB_SQUARE; ++i) {
287 r0[i] = 0;
288 r1[i] = -kInt13Max;
289 }
290 break;
291 default:
292 for (int i = 0; i < MAX_SB_SQUARE; ++i) {
293 r0[i] = -kInt13Max;
294 r1[i] = 0;
295 }
296 break;
297 }
298
299 for (int i = 0; i < MAX_SB_SQUARE; ++i) m[i] = MAX_MASK_VALUE;
300
301 const int maxN = AOMMIN(kMaxSize, MAX_SB_SQUARE);
302 const int N = 64 * (rng_(maxN / 64 - 1) + 1);
303
304 int64_t limit;
305 limit = (int64_t)aom_sum_squares_i16(r0, N);
306 limit -= (int64_t)aom_sum_squares_i16(r1, N);
307 limit *= (1 << WEDGE_WEIGHT_BITS) / 2;
308
309 for (int i = 0; i < N; i++)
310 ds[i] = clamp(r0[i] * r0[i] - r1[i] * r1[i], INT16_MIN, INT16_MAX);
311
312 const int ref_res = params_.ref_func(ds, m, N, limit);
313 int tst_res;
314 ASM_REGISTER_STATE_CHECK(tst_res = params_.tst_func(ds, m, N, limit));
315
316 ASSERT_EQ(ref_res, tst_res);
317 }
318 }
319
320 //////////////////////////////////////////////////////////////////////////////
321 // av1_wedge_compute_delta_squares
322 //////////////////////////////////////////////////////////////////////////////
323
324 typedef void (*FDS)(int16_t *d, const int16_t *a, const int16_t *b, int N);
325 typedef libaom_test::FuncParam<FDS> TestFuncsFDS;
326
327 class WedgeUtilsDeltaSquaresOptTest : public FunctionEquivalenceTest<FDS> {
328 protected:
329 static const int kIterations = 10000;
330 };
331
332 TEST_P(WedgeUtilsDeltaSquaresOptTest, RandomValues) {
333 DECLARE_ALIGNED(32, int16_t, a[MAX_SB_SQUARE]);
334 DECLARE_ALIGNED(32, int16_t, b[MAX_SB_SQUARE]);
335 DECLARE_ALIGNED(32, int16_t, d_ref[MAX_SB_SQUARE]);
336 DECLARE_ALIGNED(32, int16_t, d_tst[MAX_SB_SQUARE]);
337
338 for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
339 for (int i = 0; i < MAX_SB_SQUARE; ++i) {
340 a[i] = rng_.Rand16();
341 b[i] = rng_(2 * INT16_MAX + 1) - INT16_MAX;
342 }
343
344 const int N = 64 * (rng_(MAX_SB_SQUARE / 64) + 1);
345
346 memset(&d_ref, INT16_MAX, sizeof(d_ref));
347 memset(&d_tst, INT16_MAX, sizeof(d_tst));
348
349 params_.ref_func(d_ref, a, b, N);
350 ASM_REGISTER_STATE_CHECK(params_.tst_func(d_tst, a, b, N));
351
352 for (int i = 0; i < MAX_SB_SQUARE; ++i) ASSERT_EQ(d_ref[i], d_tst[i]);
353 }
354 }
355
356 #if HAVE_SSE2
357 INSTANTIATE_TEST_CASE_P(
358 SSE2, WedgeUtilsSSEOptTest,
359 ::testing::Values(TestFuncsFSSE(av1_wedge_sse_from_residuals_c,
360 av1_wedge_sse_from_residuals_sse2)));
361
362 INSTANTIATE_TEST_CASE_P(
363 SSE2, WedgeUtilsSignOptTest,
364 ::testing::Values(TestFuncsFSign(av1_wedge_sign_from_residuals_c,
365 av1_wedge_sign_from_residuals_sse2)));
366
367 INSTANTIATE_TEST_CASE_P(
368 SSE2, WedgeUtilsDeltaSquaresOptTest,
369 ::testing::Values(TestFuncsFDS(av1_wedge_compute_delta_squares_c,
370 av1_wedge_compute_delta_squares_sse2)));
371 #endif // HAVE_SSE2
372
373 #if HAVE_AVX2
374 INSTANTIATE_TEST_CASE_P(
375 AVX2, WedgeUtilsSSEOptTest,
376 ::testing::Values(TestFuncsFSSE(av1_wedge_sse_from_residuals_sse2,
377 av1_wedge_sse_from_residuals_avx2)));
378
379 INSTANTIATE_TEST_CASE_P(
380 AVX2, WedgeUtilsSignOptTest,
381 ::testing::Values(TestFuncsFSign(av1_wedge_sign_from_residuals_sse2,
382 av1_wedge_sign_from_residuals_avx2)));
383
384 INSTANTIATE_TEST_CASE_P(
385 AVX2, WedgeUtilsDeltaSquaresOptTest,
386 ::testing::Values(TestFuncsFDS(av1_wedge_compute_delta_squares_sse2,
387 av1_wedge_compute_delta_squares_avx2)));
388 #endif // HAVE_AVX2
389
390 } // namespace