Commit | Line | Data |
---|---|---|
72b9787e JB |
1 | /***************************************************************************** |
2 | * Copyright (C) 2013 x265 project | |
3 | * | |
4 | * Authors: Steve Borho <steve@borho.org> | |
5 | * Min Chen <min.chen@multicorewareinc.com> | |
6 | * Praveen Kumar Tiwari <praveen@multicorewareinc.com> | |
7 | * Nabajit Deka <nabajit@multicorewareinc.com> | |
8 | * | |
9 | * This program is free software; you can redistribute it and/or modify | |
10 | * it under the terms of the GNU General Public License as published by | |
11 | * the Free Software Foundation; either version 2 of the License, or | |
12 | * (at your option) any later version. | |
13 | * | |
14 | * This program is distributed in the hope that it will be useful, | |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
17 | * GNU General Public License for more details. | |
18 | * | |
19 | * You should have received a copy of the GNU General Public License | |
20 | * along with this program; if not, write to the Free Software | |
21 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA. | |
22 | * | |
23 | * This program is also available under a commercial proprietary license. | |
24 | * For more information, contact us at license @ x265.com. | |
25 | *****************************************************************************/ | |
26 | ||
27 | #include "common.h" | |
28 | #include "mbdstharness.h" | |
29 | ||
30 | using namespace x265; | |
31 | ||
32 | struct DctConf | |
33 | { | |
34 | const char *name; | |
35 | int width; | |
36 | }; | |
37 | ||
38 | const DctConf dctInfo[] = | |
39 | { | |
40 | { "dst4x4\t", 4 }, | |
41 | { "dct4x4\t", 4 }, | |
42 | { "dct8x8\t", 8 }, | |
43 | { "dct16x16", 16 }, | |
44 | { "dct32x32", 32 }, | |
45 | }; | |
46 | ||
47 | const DctConf idctInfo[] = | |
48 | { | |
49 | { "idst4x4\t", 4 }, | |
50 | { "idct4x4\t", 4 }, | |
51 | { "idct8x8\t", 8 }, | |
52 | { "idct16x16", 16 }, | |
53 | { "idct32x32", 32 }, | |
54 | }; | |
55 | ||
56 | MBDstHarness::MBDstHarness() | |
57 | { | |
58 | const int idct_max = (1 << (BIT_DEPTH + 4)) - 1; | |
59 | ||
60 | /* [0] --- Random values | |
61 | * [1] --- Minimum | |
62 | * [2] --- Maximum */ | |
63 | for (int i = 0; i < TEST_BUF_SIZE; i++) | |
64 | { | |
65 | short_test_buff[0][i] = (rand() & PIXEL_MAX) - (rand() & PIXEL_MAX); | |
66 | int_test_buff[0][i] = rand() % PIXEL_MAX; | |
67 | int_idct_test_buff[0][i] = (rand() % (SHORT_MAX - SHORT_MIN)) - SHORT_MAX; | |
b53f7c52 | 68 | short_denoise_test_buff1[0][i] = short_denoise_test_buff2[0][i] = (rand() & SHORT_MAX) - (rand() & SHORT_MAX); |
72b9787e JB |
69 | |
70 | short_test_buff[1][i] = -PIXEL_MAX; | |
71 | int_test_buff[1][i] = -PIXEL_MAX; | |
72 | int_idct_test_buff[1][i] = SHORT_MIN; | |
b53f7c52 | 73 | short_denoise_test_buff1[1][i] = short_denoise_test_buff2[1][i] = -SHORT_MAX; |
72b9787e JB |
74 | |
75 | short_test_buff[2][i] = PIXEL_MAX; | |
76 | int_test_buff[2][i] = PIXEL_MAX; | |
77 | int_idct_test_buff[2][i] = SHORT_MAX; | |
b53f7c52 | 78 | short_denoise_test_buff1[2][i] = short_denoise_test_buff2[2][i] = SHORT_MAX; |
72b9787e JB |
79 | |
80 | mbuf1[i] = rand() & PIXEL_MAX; | |
81 | mbufdct[i] = (rand() & PIXEL_MAX) - (rand() & PIXEL_MAX); | |
82 | mbufidct[i] = (rand() & idct_max); | |
83 | } | |
84 | ||
85 | #if _DEBUG | |
86 | memset(mshortbuf2, 0, MAX_TU_SIZE * sizeof(int16_t)); | |
87 | memset(mshortbuf3, 0, MAX_TU_SIZE * sizeof(int16_t)); | |
88 | ||
89 | memset(mintbuf1, 0, MAX_TU_SIZE * sizeof(int)); | |
90 | memset(mintbuf2, 0, MAX_TU_SIZE * sizeof(int)); | |
91 | memset(mintbuf3, 0, MAX_TU_SIZE * sizeof(int)); | |
92 | memset(mintbuf4, 0, MAX_TU_SIZE * sizeof(int)); | |
93 | #endif // if _DEBUG | |
94 | } | |
95 | ||
96 | bool MBDstHarness::check_dct_primitive(dct_t ref, dct_t opt, intptr_t width) | |
97 | { | |
98 | int j = 0; | |
b53f7c52 | 99 | intptr_t cmp_size = sizeof(short) * width * width; |
72b9787e JB |
100 | |
101 | for (int i = 0; i < ITERS; i++) | |
102 | { | |
103 | int index = rand() % TEST_CASES; | |
104 | ||
b53f7c52 JB |
105 | ref(short_test_buff[index] + j, mshortbuf2, width); |
106 | checked(opt, short_test_buff[index] + j, mshortbuf3, width); | |
72b9787e | 107 | |
b53f7c52 | 108 | if (memcmp(mshortbuf2, mshortbuf3, cmp_size)) |
72b9787e JB |
109 | return false; |
110 | ||
111 | reportfail(); | |
112 | j += INCR; | |
113 | } | |
114 | ||
115 | return true; | |
116 | } | |
117 | ||
118 | bool MBDstHarness::check_idct_primitive(idct_t ref, idct_t opt, intptr_t width) | |
119 | { | |
120 | int j = 0; | |
121 | intptr_t cmp_size = sizeof(int16_t) * width * width; | |
122 | ||
123 | for (int i = 0; i < ITERS; i++) | |
124 | { | |
125 | int index = rand() % TEST_CASES; | |
126 | ||
b53f7c52 JB |
127 | ref(short_test_buff[index] + j, mshortbuf2, width); |
128 | checked(opt, short_test_buff[index] + j, mshortbuf3, width); | |
72b9787e JB |
129 | |
130 | if (memcmp(mshortbuf2, mshortbuf3, cmp_size)) | |
131 | return false; | |
132 | ||
133 | reportfail(); | |
134 | j += INCR; | |
135 | } | |
136 | ||
137 | return true; | |
138 | } | |
139 | ||
140 | bool MBDstHarness::check_dequant_primitive(dequant_normal_t ref, dequant_normal_t opt) | |
141 | { | |
142 | int j = 0; | |
143 | ||
144 | for (int i = 0; i < ITERS; i++) | |
145 | { | |
146 | int index = rand() % TEST_CASES; | |
147 | int log2TrSize = (rand() % 4) + 2; | |
148 | ||
149 | int width = (1 << log2TrSize); | |
150 | int height = width; | |
151 | int qp = rand() % (QP_MAX_SPEC + QP_BD_OFFSET + 1); | |
152 | int per = qp / 6; | |
153 | int rem = qp % 6; | |
154 | static const int invQuantScales[6] = { 40, 45, 51, 57, 64, 72 }; | |
155 | int scale = invQuantScales[rem] << per; | |
156 | int transformShift = MAX_TR_DYNAMIC_RANGE - X265_DEPTH - log2TrSize; | |
157 | int shift = QUANT_IQUANT_SHIFT - QUANT_SHIFT - transformShift; | |
158 | ||
b53f7c52 JB |
159 | ref(short_test_buff[index] + j, mshortbuf2, width * height, scale, shift); |
160 | checked(opt, short_test_buff[index] + j, mshortbuf3, width * height, scale, shift); | |
72b9787e | 161 | |
b53f7c52 | 162 | if (memcmp(mshortbuf2, mshortbuf3, sizeof(int16_t) * height * width)) |
72b9787e JB |
163 | return false; |
164 | ||
165 | reportfail(); | |
166 | j += INCR; | |
167 | } | |
168 | ||
169 | return true; | |
170 | } | |
171 | ||
172 | bool MBDstHarness::check_dequant_primitive(dequant_scaling_t ref, dequant_scaling_t opt) | |
173 | { | |
174 | int j = 0; | |
175 | ||
176 | for (int i = 0; i < ITERS; i++) | |
177 | { | |
b53f7c52 JB |
178 | |
179 | memset(mshortbuf2, 0, MAX_TU_SIZE * sizeof(int16_t)); | |
180 | memset(mshortbuf3, 0, MAX_TU_SIZE * sizeof(int16_t)); | |
181 | ||
72b9787e JB |
182 | int log2TrSize = (rand() % 4) + 2; |
183 | ||
184 | int width = (1 << log2TrSize); | |
185 | int height = width; | |
186 | ||
187 | int qp = rand() % (QP_MAX_SPEC + QP_BD_OFFSET + 1); | |
188 | int per = qp / 6; | |
189 | int transformShift = MAX_TR_DYNAMIC_RANGE - X265_DEPTH - log2TrSize; | |
190 | int shift = QUANT_IQUANT_SHIFT - QUANT_SHIFT - transformShift; | |
191 | ||
b53f7c52 | 192 | int cmp_size = sizeof(int16_t) * height * width; |
72b9787e JB |
193 | int index1 = rand() % TEST_CASES; |
194 | ||
b53f7c52 JB |
195 | ref(short_test_buff[index1] + j, int_test_buff[index1] + j, mshortbuf2, width * height, per, shift); |
196 | checked(opt, short_test_buff[index1] + j, int_test_buff[index1] + j, mshortbuf3, width * height, per, shift); | |
72b9787e | 197 | |
b53f7c52 | 198 | if (memcmp(mshortbuf2, mshortbuf3, cmp_size)) |
72b9787e JB |
199 | return false; |
200 | ||
201 | reportfail(); | |
202 | j += INCR; | |
203 | } | |
204 | ||
205 | return true; | |
206 | } | |
207 | ||
208 | bool MBDstHarness::check_quant_primitive(quant_t ref, quant_t opt) | |
209 | { | |
210 | int j = 0; | |
211 | ||
212 | for (int i = 0; i < ITERS; i++) | |
213 | { | |
214 | int width = (rand() % 4 + 1) * 4; | |
215 | int height = width; | |
216 | ||
217 | uint32_t optReturnValue = 0; | |
218 | uint32_t refReturnValue = 0; | |
219 | ||
220 | int bits = (rand() % 24) + 8; | |
221 | int valueToAdd = rand() % (1 << bits); | |
222 | int cmp_size = sizeof(int) * height * width; | |
223 | int cmp_size1 = sizeof(short) * height * width; | |
224 | int numCoeff = height * width; | |
225 | ||
226 | int index1 = rand() % TEST_CASES; | |
227 | int index2 = rand() % TEST_CASES; | |
228 | ||
b53f7c52 JB |
229 | refReturnValue = ref(short_test_buff[index1] + j, int_test_buff[index2] + j, mintbuf1, mshortbuf2, bits, valueToAdd, numCoeff); |
230 | optReturnValue = (uint32_t)checked(opt, short_test_buff[index1] + j, int_test_buff[index2] + j, mintbuf3, mshortbuf3, bits, valueToAdd, numCoeff); | |
72b9787e JB |
231 | |
232 | if (memcmp(mintbuf1, mintbuf3, cmp_size)) | |
233 | return false; | |
234 | ||
235 | if (memcmp(mshortbuf2, mshortbuf3, cmp_size1)) | |
236 | return false; | |
237 | ||
238 | if (optReturnValue != refReturnValue) | |
239 | return false; | |
240 | ||
241 | reportfail(); | |
242 | j += INCR; | |
243 | } | |
244 | ||
245 | return true; | |
246 | } | |
247 | ||
248 | bool MBDstHarness::check_nquant_primitive(nquant_t ref, nquant_t opt) | |
249 | { | |
250 | int j = 0; | |
251 | ||
252 | for (int i = 0; i < ITERS; i++) | |
253 | { | |
254 | int width = (rand() % 4 + 1) * 4; | |
255 | int height = width; | |
256 | ||
257 | uint32_t optReturnValue = 0; | |
258 | uint32_t refReturnValue = 0; | |
259 | ||
260 | int bits = rand() % 32; | |
261 | int valueToAdd = rand() % (1 << bits); | |
262 | int cmp_size = sizeof(short) * height * width; | |
263 | int numCoeff = height * width; | |
264 | ||
265 | int index1 = rand() % TEST_CASES; | |
266 | int index2 = rand() % TEST_CASES; | |
267 | ||
b53f7c52 JB |
268 | refReturnValue = ref(short_test_buff[index1] + j, int_test_buff[index2] + j, mshortbuf2, bits, valueToAdd, numCoeff); |
269 | optReturnValue = (uint32_t)checked(opt, short_test_buff[index1] + j, int_test_buff[index2] + j, mshortbuf3, bits, valueToAdd, numCoeff); | |
72b9787e JB |
270 | |
271 | if (memcmp(mshortbuf2, mshortbuf3, cmp_size)) | |
272 | return false; | |
273 | ||
274 | if (optReturnValue != refReturnValue) | |
275 | return false; | |
276 | ||
277 | reportfail(); | |
278 | j += INCR; | |
279 | } | |
280 | ||
281 | return true; | |
282 | } | |
283 | ||
284 | bool MBDstHarness::check_count_nonzero_primitive(count_nonzero_t ref, count_nonzero_t opt) | |
285 | { | |
286 | ALIGN_VAR_32(int16_t, qcoeff[32 * 32]); | |
287 | ||
288 | for (int i = 0; i < 4; i++) | |
289 | { | |
290 | int log2TrSize = i + 2; | |
291 | int num = 1 << (log2TrSize * 2); | |
292 | int mask = num - 1; | |
293 | ||
294 | for (int n = 0; n <= num; n++) | |
295 | { | |
296 | memset(qcoeff, 0, num * sizeof(int16_t)); | |
297 | ||
298 | for (int j = 0; j < n; j++) | |
299 | { | |
300 | int k = rand() & mask; | |
301 | while (qcoeff[k]) | |
302 | { | |
303 | k = (k + 11) & mask; | |
304 | } | |
305 | ||
306 | qcoeff[k] = (int16_t)rand() - RAND_MAX / 2; | |
307 | } | |
308 | ||
309 | int refval = ref(qcoeff, num); | |
310 | int optval = (int)checked(opt, qcoeff, num); | |
311 | ||
312 | if (refval != optval) | |
313 | return false; | |
314 | ||
315 | reportfail(); | |
316 | } | |
317 | } | |
318 | ||
319 | return true; | |
320 | } | |
321 | ||
322 | bool MBDstHarness::check_denoise_dct_primitive(denoiseDct_t ref, denoiseDct_t opt) | |
323 | { | |
324 | int j = 0; | |
325 | ||
326 | for (int s = 0; s < 4; s++) | |
327 | { | |
328 | int log2TrSize = s + 2; | |
329 | int num = 1 << (log2TrSize * 2); | |
330 | int cmp_size = sizeof(int) * num; | |
b53f7c52 | 331 | int cmp_short = sizeof(short) * num; |
72b9787e JB |
332 | |
333 | for (int i = 0; i < ITERS; i++) | |
334 | { | |
335 | memset(mubuf1, 0, num * sizeof(uint32_t)); | |
336 | memset(mubuf2, 0, num * sizeof(uint32_t)); | |
337 | memset(mushortbuf1, 0, num * sizeof(uint16_t)); | |
338 | ||
339 | for (int k = 0; k < num; k++) | |
340 | mushortbuf1[k] = rand() % UNSIGNED_SHORT_MAX; | |
341 | ||
342 | int index = rand() % TEST_CASES; | |
343 | ||
b53f7c52 JB |
344 | ref(short_denoise_test_buff1[index] + j, mubuf1, mushortbuf1, num); |
345 | checked(opt, short_denoise_test_buff2[index] + j, mubuf2, mushortbuf1, num); | |
72b9787e | 346 | |
b53f7c52 | 347 | if (memcmp(short_denoise_test_buff1[index] + j, short_denoise_test_buff2[index] + j, cmp_short)) |
72b9787e JB |
348 | return false; |
349 | ||
350 | if (memcmp(mubuf1, mubuf2, cmp_size)) | |
351 | return false; | |
352 | ||
353 | reportfail(); | |
354 | j += INCR; | |
355 | } | |
356 | j = 0; | |
357 | } | |
358 | ||
359 | return true; | |
360 | } | |
361 | ||
362 | ||
363 | bool MBDstHarness::testCorrectness(const EncoderPrimitives& ref, const EncoderPrimitives& opt) | |
364 | { | |
365 | for (int i = 0; i < NUM_DCTS; i++) | |
366 | { | |
367 | if (opt.dct[i]) | |
368 | { | |
369 | if (!check_dct_primitive(ref.dct[i], opt.dct[i], dctInfo[i].width)) | |
370 | { | |
371 | printf("\n%s failed\n", dctInfo[i].name); | |
372 | return false; | |
373 | } | |
374 | } | |
375 | } | |
376 | ||
377 | for (int i = 0; i < NUM_IDCTS; i++) | |
378 | { | |
379 | if (opt.idct[i]) | |
380 | { | |
381 | if (!check_idct_primitive(ref.idct[i], opt.idct[i], idctInfo[i].width)) | |
382 | { | |
383 | printf("%s failed\n", idctInfo[i].name); | |
384 | return false; | |
385 | } | |
386 | } | |
387 | } | |
388 | ||
389 | if (opt.dequant_normal) | |
390 | { | |
391 | if (!check_dequant_primitive(ref.dequant_normal, opt.dequant_normal)) | |
392 | { | |
393 | printf("dequant: Failed!\n"); | |
394 | return false; | |
395 | } | |
396 | } | |
397 | ||
398 | if (opt.dequant_scaling) | |
399 | { | |
400 | if (!check_dequant_primitive(ref.dequant_scaling, opt.dequant_scaling)) | |
401 | { | |
402 | printf("dequant_scaling: Failed!\n"); | |
403 | return false; | |
404 | } | |
405 | } | |
406 | ||
407 | if (opt.quant) | |
408 | { | |
409 | if (!check_quant_primitive(ref.quant, opt.quant)) | |
410 | { | |
411 | printf("quant: Failed!\n"); | |
412 | return false; | |
413 | } | |
414 | } | |
415 | ||
416 | if (opt.nquant) | |
417 | { | |
418 | if (!check_nquant_primitive(ref.nquant, opt.nquant)) | |
419 | { | |
420 | printf("nquant: Failed!\n"); | |
421 | return false; | |
422 | } | |
423 | } | |
424 | ||
425 | if (opt.count_nonzero) | |
426 | { | |
427 | if (!check_count_nonzero_primitive(ref.count_nonzero, opt.count_nonzero)) | |
428 | { | |
429 | printf("count_nonzero: Failed!\n"); | |
430 | return false; | |
431 | } | |
432 | } | |
433 | ||
434 | if (opt.dequant_scaling) | |
435 | { | |
436 | if (!check_dequant_primitive(ref.dequant_scaling, opt.dequant_scaling)) | |
437 | { | |
438 | printf("dequant_scaling: Failed!\n"); | |
439 | return false; | |
440 | } | |
441 | } | |
442 | ||
443 | if (opt.denoiseDct) | |
444 | { | |
445 | if (!check_denoise_dct_primitive(ref.denoiseDct, opt.denoiseDct)) | |
446 | { | |
447 | printf("denoiseDct: Failed!\n"); | |
448 | return false; | |
449 | } | |
450 | } | |
451 | ||
452 | return true; | |
453 | } | |
454 | ||
455 | void MBDstHarness::measureSpeed(const EncoderPrimitives& ref, const EncoderPrimitives& opt) | |
456 | { | |
457 | for (int value = 0; value < NUM_DCTS; value++) | |
458 | { | |
459 | if (opt.dct[value]) | |
460 | { | |
461 | printf("%s\t", dctInfo[value].name); | |
b53f7c52 | 462 | REPORT_SPEEDUP(opt.dct[value], ref.dct[value], mbuf1, mshortbuf2, dctInfo[value].width); |
72b9787e JB |
463 | } |
464 | } | |
465 | ||
466 | for (int value = 0; value < NUM_IDCTS; value++) | |
467 | { | |
468 | if (opt.idct[value]) | |
469 | { | |
470 | printf("%s\t", idctInfo[value].name); | |
b53f7c52 | 471 | REPORT_SPEEDUP(opt.idct[value], ref.idct[value], mshortbuf3, mshortbuf2, idctInfo[value].width); |
72b9787e JB |
472 | } |
473 | } | |
474 | ||
475 | if (opt.dequant_normal) | |
476 | { | |
477 | printf("dequant_normal\t"); | |
b53f7c52 | 478 | REPORT_SPEEDUP(opt.dequant_normal, ref.dequant_normal, short_test_buff[0], mshortbuf2, 32 * 32, 70, 1); |
72b9787e JB |
479 | } |
480 | ||
481 | if (opt.dequant_scaling) | |
482 | { | |
483 | printf("dequant_scaling\t"); | |
b53f7c52 | 484 | REPORT_SPEEDUP(opt.dequant_scaling, ref.dequant_scaling, short_test_buff[0], mintbuf3, mshortbuf2, 32 * 32, 5, 1); |
72b9787e JB |
485 | } |
486 | ||
487 | if (opt.quant) | |
488 | { | |
489 | printf("quant\t\t"); | |
b53f7c52 | 490 | REPORT_SPEEDUP(opt.quant, ref.quant, short_test_buff[0], int_test_buff[1], mintbuf3, mshortbuf2, 23, 23785, 32 * 32); |
72b9787e JB |
491 | } |
492 | ||
493 | if (opt.nquant) | |
494 | { | |
495 | printf("nquant\t\t"); | |
b53f7c52 | 496 | REPORT_SPEEDUP(opt.nquant, ref.nquant, short_test_buff[0], int_test_buff[1], mshortbuf2, 23, 23785, 32 * 32); |
72b9787e JB |
497 | } |
498 | ||
499 | if (opt.count_nonzero) | |
500 | { | |
501 | for (int i = 4; i <= 32; i <<= 1) | |
502 | { | |
503 | printf("count_nonzero[%dx%d]", i, i); | |
504 | REPORT_SPEEDUP(opt.count_nonzero, ref.count_nonzero, mbuf1, i * i) | |
505 | } | |
506 | } | |
507 | ||
508 | if (opt.denoiseDct) | |
509 | { | |
510 | printf("denoiseDct\t"); | |
b53f7c52 | 511 | REPORT_SPEEDUP(opt.denoiseDct, ref.denoiseDct, short_denoise_test_buff1[0], mubuf1, mushortbuf1, 32 * 32); |
72b9787e JB |
512 | } |
513 | ||
514 | } |