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 <cstdlib>
13 #include <string>
14
15 #include "third_party/googletest/src/googletest/include/gtest/gtest.h"
16
17 #include "./aom_config.h"
18 #include "./av1_rtcd.h"
19 #include "aom_ports/aom_timer.h"
20 #include "av1/common/cdef_block.h"
21 #include "test/acm_random.h"
22 #include "test/clear_system_state.h"
23 #include "test/register_state_check.h"
24 #include "test/util.h"
25
26 using libaom_test::ACMRandom;
27
28 namespace {
29
30 typedef std::tr1::tuple<cdef_direction_func, cdef_direction_func, int>
31 dering_dir_param_t;
32
33 class CDEFDeringDirTest : public ::testing::TestWithParam<dering_dir_param_t> {
34 public:
~CDEFDeringDirTest()35 virtual ~CDEFDeringDirTest() {}
SetUp()36 virtual void SetUp() {
37 dering = GET_PARAM(0);
38 ref_dering = GET_PARAM(1);
39 bsize = GET_PARAM(2);
40 }
41
TearDown()42 virtual void TearDown() { libaom_test::ClearSystemState(); }
43
44 protected:
45 int bsize;
46 cdef_direction_func dering;
47 cdef_direction_func ref_dering;
48 };
49
50 typedef CDEFDeringDirTest CDEFDeringSpeedTest;
51
test_dering(int bsize,int iterations,cdef_direction_func dering,cdef_direction_func ref_dering)52 void test_dering(int bsize, int iterations, cdef_direction_func dering,
53 cdef_direction_func ref_dering) {
54 const int size = 8;
55 const int ysize = size + 2 * CDEF_VBORDER;
56 ACMRandom rnd(ACMRandom::DeterministicSeed());
57 DECLARE_ALIGNED(16, uint16_t, s[ysize * CDEF_BSTRIDE]);
58 DECLARE_ALIGNED(16, static uint16_t, d[size * size]);
59 DECLARE_ALIGNED(16, static uint16_t, ref_d[size * size]);
60 memset(ref_d, 0, sizeof(ref_d));
61 memset(d, 0, sizeof(d));
62
63 int error = 0, threshold = 0, dir;
64 int boundary, damping, depth, bits, level, count,
65 errdepth = 0, errthreshold = 0, errboundary = 0, errdamping = 0;
66 unsigned int pos = 0;
67
68 for (boundary = 0; boundary < 16; boundary++) {
69 for (depth = 8; depth <= 12; depth += 2) {
70 for (damping = 5 + depth - 8; damping < 7 + depth - 8; damping++) {
71 for (count = 0; count < iterations; count++) {
72 for (level = 0; level < (1 << depth) && !error;
73 level += (1 + 4 * !!boundary) << (depth - 8)) {
74 for (bits = 1; bits <= depth && !error; bits++) {
75 for (unsigned int i = 0; i < sizeof(s) / sizeof(*s); i++)
76 s[i] = clamp((rnd.Rand16() & ((1 << bits) - 1)) + level, 0,
77 (1 << depth) - 1);
78 if (boundary) {
79 if (boundary & 1) { // Left
80 for (int i = 0; i < ysize; i++)
81 for (int j = 0; j < CDEF_HBORDER; j++)
82 s[i * CDEF_BSTRIDE + j] = CDEF_VERY_LARGE;
83 }
84 if (boundary & 2) { // Right
85 for (int i = 0; i < ysize; i++)
86 for (int j = CDEF_HBORDER + size; j < CDEF_BSTRIDE; j++)
87 s[i * CDEF_BSTRIDE + j] = CDEF_VERY_LARGE;
88 }
89 if (boundary & 4) { // Above
90 for (int i = 0; i < CDEF_VBORDER; i++)
91 for (int j = 0; j < CDEF_BSTRIDE; j++)
92 s[i * CDEF_BSTRIDE + j] = CDEF_VERY_LARGE;
93 }
94 if (boundary & 8) { // Below
95 for (int i = CDEF_VBORDER + size; i < ysize; i++)
96 for (int j = 0; j < CDEF_BSTRIDE; j++)
97 s[i * CDEF_BSTRIDE + j] = CDEF_VERY_LARGE;
98 }
99 }
100 for (dir = 0; dir < 8; dir++) {
101 for (threshold = 0; threshold < 64 << (depth - 8) && !error;
102 threshold += (1 + 4 * !!boundary) << (depth - 8)) {
103 ref_dering(ref_d, size,
104 s + CDEF_HBORDER + CDEF_VBORDER * CDEF_BSTRIDE,
105 threshold, dir, damping);
106 // If dering and ref_dering are the same, we're just testing
107 // speed
108 if (dering != ref_dering)
109 ASM_REGISTER_STATE_CHECK(dering(
110 d, size, s + CDEF_HBORDER + CDEF_VBORDER * CDEF_BSTRIDE,
111 threshold, dir, damping));
112 if (ref_dering != dering) {
113 for (pos = 0; pos < sizeof(d) / sizeof(*d) && !error;
114 pos++) {
115 error = ref_d[pos] != d[pos];
116 errdepth = depth;
117 errthreshold = threshold;
118 errboundary = boundary;
119 errdamping = damping;
120 }
121 }
122 }
123 }
124 }
125 }
126 }
127 }
128 }
129 }
130
131 pos--;
132 EXPECT_EQ(0, error) << "Error: CDEFDeringDirTest, SIMD and C mismatch."
133 << std::endl
134 << "First error at " << pos % size << "," << pos / size
135 << " (" << (int16_t)ref_d[pos] << " : " << (int16_t)d[pos]
136 << ") " << std::endl
137 << "threshold: " << errthreshold << std::endl
138 << "damping: " << errdamping << std::endl
139 << "depth: " << errdepth << std::endl
140 << "size: " << bsize << std::endl
141 << "boundary: " << errboundary << std::endl
142 << std::endl;
143 }
144
test_dering_speed(int bsize,int iterations,cdef_direction_func dering,cdef_direction_func ref_dering)145 void test_dering_speed(int bsize, int iterations, cdef_direction_func dering,
146 cdef_direction_func ref_dering) {
147 aom_usec_timer ref_timer;
148 aom_usec_timer timer;
149
150 aom_usec_timer_start(&ref_timer);
151 test_dering(bsize, iterations, ref_dering, ref_dering);
152 aom_usec_timer_mark(&ref_timer);
153 int ref_elapsed_time = (int)aom_usec_timer_elapsed(&ref_timer);
154
155 aom_usec_timer_start(&timer);
156 test_dering(bsize, iterations, dering, dering);
157 aom_usec_timer_mark(&timer);
158 int elapsed_time = (int)aom_usec_timer_elapsed(&timer);
159
160 #if 0
161 std::cout << "[ ] C time = " << ref_elapsed_time / 1000
162 << " ms, SIMD time = " << elapsed_time / 1000 << " ms" << std::endl;
163 #endif
164
165 EXPECT_GT(ref_elapsed_time, elapsed_time)
166 << "Error: CDEFDeringSpeedTest, SIMD slower than C." << std::endl
167 << "C time: " << ref_elapsed_time << " us" << std::endl
168 << "SIMD time: " << elapsed_time << " us" << std::endl;
169 }
170
171 typedef int (*find_dir_t)(const uint16_t *img, int stride, int32_t *var,
172 int coeff_shift);
173
174 typedef std::tr1::tuple<find_dir_t, find_dir_t> find_dir_param_t;
175
176 class CDEFDeringFindDirTest
177 : public ::testing::TestWithParam<find_dir_param_t> {
178 public:
~CDEFDeringFindDirTest()179 virtual ~CDEFDeringFindDirTest() {}
SetUp()180 virtual void SetUp() {
181 finddir = GET_PARAM(0);
182 ref_finddir = GET_PARAM(1);
183 }
184
TearDown()185 virtual void TearDown() { libaom_test::ClearSystemState(); }
186
187 protected:
188 find_dir_t finddir;
189 find_dir_t ref_finddir;
190 };
191
192 typedef CDEFDeringFindDirTest CDEFDeringFindDirSpeedTest;
193
test_finddir(int (* finddir)(const uint16_t * img,int stride,int32_t * var,int coeff_shift),int (* ref_finddir)(const uint16_t * img,int stride,int32_t * var,int coeff_shift))194 void test_finddir(int (*finddir)(const uint16_t *img, int stride, int32_t *var,
195 int coeff_shift),
196 int (*ref_finddir)(const uint16_t *img, int stride,
197 int32_t *var, int coeff_shift)) {
198 const int size = 8;
199 ACMRandom rnd(ACMRandom::DeterministicSeed());
200 DECLARE_ALIGNED(16, uint16_t, s[size * size]);
201
202 int error = 0;
203 int depth, bits, level, count, errdepth = 0;
204 int ref_res = 0, res = 0;
205 int32_t ref_var = 0, var = 0;
206
207 for (depth = 8; depth <= 12 && !error; depth += 2) {
208 for (count = 0; count < 512 && !error; count++) {
209 for (level = 0; level < (1 << depth) && !error;
210 level += 1 << (depth - 8)) {
211 for (bits = 1; bits <= depth && !error; bits++) {
212 for (unsigned int i = 0; i < sizeof(s) / sizeof(*s); i++)
213 s[i] = clamp((rnd.Rand16() & ((1 << bits) - 1)) + level, 0,
214 (1 << depth) - 1);
215 for (int c = 0; c < 1 + 9 * (finddir == ref_finddir); c++)
216 ref_res = ref_finddir(s, size, &ref_var, depth - 8);
217 if (finddir != ref_finddir)
218 ASM_REGISTER_STATE_CHECK(res = finddir(s, size, &var, depth - 8));
219 if (ref_finddir != finddir) {
220 if (res != ref_res || var != ref_var) error = 1;
221 errdepth = depth;
222 }
223 }
224 }
225 }
226 }
227
228 EXPECT_EQ(0, error) << "Error: CDEFDeringFindDirTest, SIMD and C mismatch."
229 << std::endl
230 << "return: " << res << " : " << ref_res << std::endl
231 << "var: " << var << " : " << ref_var << std::endl
232 << "depth: " << errdepth << std::endl
233 << std::endl;
234 }
235
test_finddir_speed(int (* finddir)(const uint16_t * img,int stride,int32_t * var,int coeff_shift),int (* ref_finddir)(const uint16_t * img,int stride,int32_t * var,int coeff_shift))236 void test_finddir_speed(int (*finddir)(const uint16_t *img, int stride,
237 int32_t *var, int coeff_shift),
238 int (*ref_finddir)(const uint16_t *img, int stride,
239 int32_t *var, int coeff_shift)) {
240 aom_usec_timer ref_timer;
241 aom_usec_timer timer;
242
243 aom_usec_timer_start(&ref_timer);
244 test_finddir(ref_finddir, ref_finddir);
245 aom_usec_timer_mark(&ref_timer);
246 int ref_elapsed_time = (int)aom_usec_timer_elapsed(&ref_timer);
247
248 aom_usec_timer_start(&timer);
249 test_finddir(finddir, finddir);
250 aom_usec_timer_mark(&timer);
251 int elapsed_time = (int)aom_usec_timer_elapsed(&timer);
252
253 #if 0
254 std::cout << "[ ] C time = " << ref_elapsed_time / 1000
255 << " ms, SIMD time = " << elapsed_time / 1000 << " ms" << std::endl;
256 #endif
257
258 EXPECT_GT(ref_elapsed_time, elapsed_time)
259 << "Error: CDEFDeringFindDirSpeedTest, SIMD slower than C." << std::endl
260 << "C time: " << ref_elapsed_time << " us" << std::endl
261 << "SIMD time: " << elapsed_time << " us" << std::endl;
262 }
263
TEST_P(CDEFDeringDirTest,TestSIMDNoMismatch)264 TEST_P(CDEFDeringDirTest, TestSIMDNoMismatch) {
265 test_dering(bsize, 1, dering, ref_dering);
266 }
267
TEST_P(CDEFDeringSpeedTest,DISABLED_TestSpeed)268 TEST_P(CDEFDeringSpeedTest, DISABLED_TestSpeed) {
269 test_dering_speed(bsize, 4, dering, ref_dering);
270 }
271
TEST_P(CDEFDeringFindDirTest,TestSIMDNoMismatch)272 TEST_P(CDEFDeringFindDirTest, TestSIMDNoMismatch) {
273 test_finddir(finddir, ref_finddir);
274 }
275
TEST_P(CDEFDeringFindDirSpeedTest,DISABLED_TestSpeed)276 TEST_P(CDEFDeringFindDirSpeedTest, DISABLED_TestSpeed) {
277 test_finddir_speed(finddir, ref_finddir);
278 }
279
280 using std::tr1::make_tuple;
281
282 // VS compiling for 32 bit targets does not support vector types in
283 // structs as arguments, which makes the v256 type of the intrinsics
284 // hard to support, so optimizations for this target are disabled.
285 #if defined(_WIN64) || !defined(_MSC_VER) || defined(__clang__)
286 #if HAVE_SSE2
287 INSTANTIATE_TEST_CASE_P(SSE2, CDEFDeringDirTest,
288 ::testing::Values(make_tuple(&cdef_direction_4x4_sse2,
289 &cdef_direction_4x4_c, 4),
290 make_tuple(&cdef_direction_8x8_sse2,
291 &cdef_direction_8x8_c,
292 8)));
293 INSTANTIATE_TEST_CASE_P(SSE2, CDEFDeringFindDirTest,
294 ::testing::Values(make_tuple(&cdef_find_dir_sse2,
295 &cdef_find_dir_c)));
296 #endif
297 #if HAVE_SSSE3
298 INSTANTIATE_TEST_CASE_P(SSSE3, CDEFDeringDirTest,
299 ::testing::Values(make_tuple(&cdef_direction_4x4_ssse3,
300 &cdef_direction_4x4_c, 4),
301 make_tuple(&cdef_direction_8x8_ssse3,
302 &cdef_direction_8x8_c,
303 8)));
304 INSTANTIATE_TEST_CASE_P(SSSE3, CDEFDeringFindDirTest,
305 ::testing::Values(make_tuple(&cdef_find_dir_ssse3,
306 &cdef_find_dir_c)));
307 #endif
308
309 #if HAVE_SSE4_1
310 INSTANTIATE_TEST_CASE_P(SSE4_1, CDEFDeringDirTest,
311 ::testing::Values(make_tuple(&cdef_direction_4x4_sse4_1,
312 &cdef_direction_4x4_c, 4),
313 make_tuple(&cdef_direction_8x8_sse4_1,
314 &cdef_direction_8x8_c,
315 8)));
316 INSTANTIATE_TEST_CASE_P(SSE4_1, CDEFDeringFindDirTest,
317 ::testing::Values(make_tuple(&cdef_find_dir_sse4_1,
318 &cdef_find_dir_c)));
319 #endif
320
321 #if HAVE_NEON
322 INSTANTIATE_TEST_CASE_P(NEON, CDEFDeringDirTest,
323 ::testing::Values(make_tuple(&cdef_direction_4x4_neon,
324 &cdef_direction_4x4_c, 4),
325 make_tuple(&cdef_direction_8x8_neon,
326 &cdef_direction_8x8_c,
327 8)));
328 INSTANTIATE_TEST_CASE_P(NEON, CDEFDeringFindDirTest,
329 ::testing::Values(make_tuple(&cdef_find_dir_neon,
330 &cdef_find_dir_c)));
331 #endif
332
333 // Test speed for all supported architectures
334 #if HAVE_SSE2
335 INSTANTIATE_TEST_CASE_P(SSE2, CDEFDeringSpeedTest,
336 ::testing::Values(make_tuple(&cdef_direction_4x4_sse2,
337 &cdef_direction_4x4_c, 4),
338 make_tuple(&cdef_direction_8x8_sse2,
339 &cdef_direction_8x8_c,
340 8)));
341 INSTANTIATE_TEST_CASE_P(SSE2, CDEFDeringFindDirSpeedTest,
342 ::testing::Values(make_tuple(&cdef_find_dir_sse2,
343 &cdef_find_dir_c)));
344 #endif
345
346 #if HAVE_SSSE3
347 INSTANTIATE_TEST_CASE_P(SSSE3, CDEFDeringSpeedTest,
348 ::testing::Values(make_tuple(&cdef_direction_4x4_ssse3,
349 &cdef_direction_4x4_c, 4),
350 make_tuple(&cdef_direction_8x8_ssse3,
351 &cdef_direction_8x8_c,
352 8)));
353 INSTANTIATE_TEST_CASE_P(SSSE3, CDEFDeringFindDirSpeedTest,
354 ::testing::Values(make_tuple(&cdef_find_dir_ssse3,
355 &cdef_find_dir_c)));
356 #endif
357
358 #if HAVE_SSE4_1
359 INSTANTIATE_TEST_CASE_P(SSE4_1, CDEFDeringSpeedTest,
360 ::testing::Values(make_tuple(&cdef_direction_4x4_sse4_1,
361 &cdef_direction_4x4_c, 4),
362 make_tuple(&cdef_direction_8x8_sse4_1,
363 &cdef_direction_8x8_c,
364 8)));
365 INSTANTIATE_TEST_CASE_P(SSE4_1, CDEFDeringFindDirSpeedTest,
366 ::testing::Values(make_tuple(&cdef_find_dir_sse4_1,
367 &cdef_find_dir_c)));
368 #endif
369
370 #if HAVE_NEON
371 INSTANTIATE_TEST_CASE_P(NEON, CDEFDeringSpeedTest,
372 ::testing::Values(make_tuple(&cdef_direction_4x4_neon,
373 &cdef_direction_4x4_c, 4),
374 make_tuple(&cdef_direction_8x8_neon,
375 &cdef_direction_8x8_c,
376 8)));
377 INSTANTIATE_TEST_CASE_P(NEON, CDEFDeringFindDirSpeedTest,
378 ::testing::Values(make_tuple(&cdef_find_dir_neon,
379 &cdef_find_dir_c)));
380 #endif
381
382 #endif // defined(_WIN64) || !defined(_MSC_VER)
383 } // namespace
384