1 /*
2 * (c) 2001 Fabrice Bellard
3 * 2007 Marc Hoffman <marc.hoffman@analog.com>
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22 /**
23 * @file
24 * DCT test (c) 2001 Fabrice Bellard
25 * Started from sample code by Juan J. Sierralta P.
26 */
27
28 #include "config.h"
29 #include <stdlib.h>
30 #include <stdio.h>
31 #include <string.h>
32 #if HAVE_UNISTD_H
33 #include <unistd.h>
34 #endif
35 #include <math.h>
36
37 #include "libavutil/cpu.h"
38 #include "libavutil/common.h"
39 #include "libavutil/internal.h"
40 #include "libavutil/lfg.h"
41 #include "libavutil/time.h"
42
43 #include "libavcodec/dct.h"
44 #include "libavcodec/idctdsp.h"
45 #include "libavcodec/simple_idct.h"
46 #include "libavcodec/xvididct.h"
47 #include "libavcodec/aandcttab.h"
48 #include "libavcodec/faandct.h"
49 #include "libavcodec/faanidct.h"
50 #include "libavcodec/dctref.h"
51
52 struct algo {
53 const char *name;
54 void (*func)(int16_t *block);
55 enum idct_permutation_type perm_type;
56 int cpu_flag;
57 int nonspec;
58 };
59
60 static const struct algo fdct_tab[] = {
61 { "REF-DBL", ff_ref_fdct, FF_IDCT_PERM_NONE },
62 { "IJG-AAN-INT", ff_fdct_ifast, FF_IDCT_PERM_NONE },
63 { "IJG-LLM-INT", ff_jpeg_fdct_islow_8, FF_IDCT_PERM_NONE },
64 #if CONFIG_FAANDCT
65 { "FAAN", ff_faandct, FF_IDCT_PERM_NONE },
66 #endif /* CONFIG_FAANDCT */
67 };
68
ff_prores_idct_wrap(int16_t * dst)69 static void ff_prores_idct_wrap(int16_t *dst){
70 LOCAL_ALIGNED(16, int16_t, qmat, [64]);
71 int i;
72
73 for(i=0; i<64; i++){
74 qmat[i]=4;
75 }
76 ff_prores_idct_10(dst, qmat);
77 for(i=0; i<64; i++) {
78 dst[i] -= 512;
79 }
80 }
81
82 static const struct algo idct_tab[] = {
83 { "REF-DBL", ff_ref_idct, FF_IDCT_PERM_NONE },
84 { "INT", ff_j_rev_dct, FF_IDCT_PERM_LIBMPEG2 },
85 { "SIMPLE-C", ff_simple_idct_int16_8bit, FF_IDCT_PERM_NONE },
86 { "SIMPLE-C10", ff_simple_idct_int16_10bit, FF_IDCT_PERM_NONE },
87 { "SIMPLE-C12", ff_simple_idct_int16_12bit, FF_IDCT_PERM_NONE, 0, 1 },
88 { "PR-C", ff_prores_idct_wrap, FF_IDCT_PERM_NONE, 0, 1 },
89 #if CONFIG_FAANIDCT
90 { "FAANI", ff_faanidct, FF_IDCT_PERM_NONE },
91 #endif /* CONFIG_FAANIDCT */
92 #if CONFIG_MPEG4_DECODER
93 { "XVID", ff_xvid_idct, FF_IDCT_PERM_NONE, 0, 1 },
94 #endif /* CONFIG_MPEG4_DECODER */
95 };
96
97 #if ARCH_AARCH64
98 #include "aarch64/dct.c"
99 #elif ARCH_ARM
100 #include "arm/dct.c"
101 #elif ARCH_PPC
102 #include "ppc/dct.c"
103 #elif ARCH_X86
104 #include "x86/dct.c"
105 #else
106 static const struct algo fdct_tab_arch[] = { { 0 } };
107 static const struct algo idct_tab_arch[] = { { 0 } };
108 #endif
109
110 #define AANSCALE_BITS 12
111
112 #define NB_ITS 20000
113 #define NB_ITS_SPEED 50000
114
115 DECLARE_ALIGNED(16, static int16_t, block)[64];
116 DECLARE_ALIGNED(8, static int16_t, block1)[64];
117
init_block(int16_t block[64],int test,int is_idct,AVLFG * prng,int vals)118 static void init_block(int16_t block[64], int test, int is_idct, AVLFG *prng, int vals)
119 {
120 int i, j;
121
122 memset(block, 0, 64 * sizeof(*block));
123
124 switch (test) {
125 case 0:
126 for (i = 0; i < 64; i++)
127 block[i] = (av_lfg_get(prng) % (2*vals)) -vals;
128 if (is_idct) {
129 ff_ref_fdct(block);
130 for (i = 0; i < 64; i++)
131 block[i] >>= 3;
132 }
133 break;
134 case 1:
135 j = av_lfg_get(prng) % 10 + 1;
136 for (i = 0; i < j; i++) {
137 int idx = av_lfg_get(prng) % 64;
138 block[idx] = av_lfg_get(prng) % (2*vals) -vals;
139 }
140 break;
141 case 2:
142 block[ 0] = av_lfg_get(prng) % (16*vals) - (8*vals);
143 block[63] = (block[0] & 1) ^ 1;
144 break;
145 }
146 }
147
permute(int16_t dst[64],const int16_t src[64],enum idct_permutation_type perm_type)148 static void permute(int16_t dst[64], const int16_t src[64],
149 enum idct_permutation_type perm_type)
150 {
151 int i;
152
153 #if ARCH_X86
154 if (permute_x86(dst, src, perm_type))
155 return;
156 #endif
157
158 switch (perm_type) {
159 case FF_IDCT_PERM_LIBMPEG2:
160 for (i = 0; i < 64; i++)
161 dst[(i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2)] = src[i];
162 break;
163 case FF_IDCT_PERM_PARTTRANS:
164 for (i = 0; i < 64; i++)
165 dst[(i & 0x24) | ((i & 3) << 3) | ((i >> 3) & 3)] = src[i];
166 break;
167 case FF_IDCT_PERM_TRANSPOSE:
168 for (i = 0; i < 64; i++)
169 dst[(i>>3) | ((i<<3)&0x38)] = src[i];
170 break;
171 default:
172 for (i = 0; i < 64; i++)
173 dst[i] = src[i];
174 break;
175 }
176 }
177
dct_error(const struct algo * dct,int test,int is_idct,int speed,const int bits)178 static int dct_error(const struct algo *dct, int test, int is_idct, int speed, const int bits)
179 {
180 void (*ref)(int16_t *block) = is_idct ? ff_ref_idct : ff_ref_fdct;
181 int it, i, scale;
182 int err_inf, v;
183 int64_t err2, ti, ti1, it1, err_sum = 0;
184 int64_t sysErr[64], sysErrMax = 0;
185 int64_t err2_matrix[64], err2_max = 0;
186 int maxout = 0;
187 int blockSumErrMax = 0, blockSumErr;
188 AVLFG prng;
189 const int vals=1<<bits;
190 double omse, ome;
191 int spec_err;
192
193 av_lfg_init(&prng, 1);
194
195 err_inf = 0;
196 err2 = 0;
197 for (i = 0; i < 64; i++)
198 err2_matrix[i] = sysErr[i] = 0;
199 for (it = 0; it < NB_ITS; it++) {
200 init_block(block1, test, is_idct, &prng, vals);
201 permute(block, block1, dct->perm_type);
202
203 dct->func(block);
204 emms_c();
205
206 if (!strcmp(dct->name, "IJG-AAN-INT")) {
207 for (i = 0; i < 64; i++) {
208 scale = 8 * (1 << (AANSCALE_BITS + 11)) / ff_aanscales[i];
209 block[i] = (block[i] * scale) >> AANSCALE_BITS;
210 }
211 }
212
213 ref(block1);
214 if (!strcmp(dct->name, "PR-SSE2"))
215 for (i = 0; i < 64; i++)
216 block1[i] = av_clip(block1[i], 4-512, 1019-512);
217
218 blockSumErr = 0;
219 for (i = 0; i < 64; i++) {
220 int err = block[i] - block1[i];
221 err_sum += err;
222 v = abs(err);
223 if (v > err_inf)
224 err_inf = v;
225 err2_matrix[i] += v * v;
226 err2 += v * v;
227 sysErr[i] += block[i] - block1[i];
228 blockSumErr += v;
229 if (abs(block[i]) > maxout)
230 maxout = abs(block[i]);
231 }
232 if (blockSumErrMax < blockSumErr)
233 blockSumErrMax = blockSumErr;
234 }
235 for (i = 0; i < 64; i++) {
236 sysErrMax = FFMAX(sysErrMax, FFABS(sysErr[i]));
237 err2_max = FFMAX(err2_max , FFABS(err2_matrix[i]));
238 }
239
240 for (i = 0; i < 64; i++) {
241 if (i % 8 == 0)
242 printf("\n");
243 printf("%7d ", (int) sysErr[i]);
244 }
245 printf("\n");
246
247 omse = (double) err2 / NB_ITS / 64;
248 ome = (double) err_sum / NB_ITS / 64;
249
250 spec_err = is_idct && (err_inf > 1 || omse > 0.02 || fabs(ome) > 0.0015);
251 if (test < 2)
252 spec_err = is_idct && ((double) err2_max / NB_ITS > 0.06 || (double) sysErrMax / NB_ITS > 0.015);
253
254 printf("%s %s: max_err=%d omse=%0.8f ome=%0.8f syserr=%0.8f maxout=%d blockSumErr=%d\n",
255 is_idct ? "IDCT" : "DCT", dct->name, err_inf,
256 omse, ome, (double) sysErrMax / NB_ITS,
257 maxout, blockSumErrMax);
258
259 if (spec_err && !dct->nonspec) {
260 printf("Failed!\n");
261 return 1;
262 }
263
264 if (!speed)
265 return 0;
266
267 /* speed test */
268
269 init_block(block, test, is_idct, &prng, vals);
270 permute(block1, block, dct->perm_type);
271
272 ti = av_gettime_relative();
273 it1 = 0;
274 do {
275 for (it = 0; it < NB_ITS_SPEED; it++) {
276 memcpy(block, block1, sizeof(block));
277 dct->func(block);
278 }
279 emms_c();
280 it1 += NB_ITS_SPEED;
281 ti1 = av_gettime_relative() - ti;
282 } while (ti1 < 1000000);
283
284 printf("%s %s: %0.1f kdct/s\n", is_idct ? "IDCT" : "DCT", dct->name,
285 (double) it1 * 1000.0 / (double) ti1);
286
287 return 0;
288 }
289
290 DECLARE_ALIGNED(8, static uint8_t, img_dest)[64];
291 DECLARE_ALIGNED(8, static uint8_t, img_dest1)[64];
292
idct248_ref(uint8_t * dest,ptrdiff_t linesize,int16_t * block)293 static void idct248_ref(uint8_t *dest, ptrdiff_t linesize, int16_t *block)
294 {
295 static int init;
296 static double c8[8][8];
297 static double c4[4][4];
298 double block1[64], block2[64], block3[64];
299 double s, sum, v;
300 int i, j, k;
301
302 if (!init) {
303 init = 1;
304
305 for (i = 0; i < 8; i++) {
306 sum = 0;
307 for (j = 0; j < 8; j++) {
308 s = (i == 0) ? sqrt(1.0 / 8.0) : sqrt(1.0 / 4.0);
309 c8[i][j] = s * cos(M_PI * i * (j + 0.5) / 8.0);
310 sum += c8[i][j] * c8[i][j];
311 }
312 }
313
314 for (i = 0; i < 4; i++) {
315 sum = 0;
316 for (j = 0; j < 4; j++) {
317 s = (i == 0) ? sqrt(1.0 / 4.0) : sqrt(1.0 / 2.0);
318 c4[i][j] = s * cos(M_PI * i * (j + 0.5) / 4.0);
319 sum += c4[i][j] * c4[i][j];
320 }
321 }
322 }
323
324 /* butterfly */
325 s = 0.5 * sqrt(2.0);
326 for (i = 0; i < 4; i++) {
327 for (j = 0; j < 8; j++) {
328 block1[8 * (2 * i) + j] =
329 (block[8 * (2 * i) + j] + block[8 * (2 * i + 1) + j]) * s;
330 block1[8 * (2 * i + 1) + j] =
331 (block[8 * (2 * i) + j] - block[8 * (2 * i + 1) + j]) * s;
332 }
333 }
334
335 /* idct8 on lines */
336 for (i = 0; i < 8; i++) {
337 for (j = 0; j < 8; j++) {
338 sum = 0;
339 for (k = 0; k < 8; k++)
340 sum += c8[k][j] * block1[8 * i + k];
341 block2[8 * i + j] = sum;
342 }
343 }
344
345 /* idct4 */
346 for (i = 0; i < 8; i++) {
347 for (j = 0; j < 4; j++) {
348 /* top */
349 sum = 0;
350 for (k = 0; k < 4; k++)
351 sum += c4[k][j] * block2[8 * (2 * k) + i];
352 block3[8 * (2 * j) + i] = sum;
353
354 /* bottom */
355 sum = 0;
356 for (k = 0; k < 4; k++)
357 sum += c4[k][j] * block2[8 * (2 * k + 1) + i];
358 block3[8 * (2 * j + 1) + i] = sum;
359 }
360 }
361
362 /* clamp and store the result */
363 for (i = 0; i < 8; i++) {
364 for (j = 0; j < 8; j++) {
365 v = block3[8 * i + j];
366 if (v < 0) v = 0;
367 else if (v > 255) v = 255;
368 dest[i * linesize + j] = (int) rint(v);
369 }
370 }
371 }
372
idct248_error(const char * name,void (* idct248_put)(uint8_t * dest,ptrdiff_t line_size,int16_t * block),int speed)373 static void idct248_error(const char *name,
374 void (*idct248_put)(uint8_t *dest,
375 ptrdiff_t line_size,
376 int16_t *block),
377 int speed)
378 {
379 int it, i, it1, ti, ti1, err_max, v;
380 AVLFG prng;
381
382 av_lfg_init(&prng, 1);
383
384 /* just one test to see if code is correct (precision is less
385 important here) */
386 err_max = 0;
387 for (it = 0; it < NB_ITS; it++) {
388 /* XXX: use forward transform to generate values */
389 for (i = 0; i < 64; i++)
390 block1[i] = av_lfg_get(&prng) % 256 - 128;
391 block1[0] += 1024;
392
393 for (i = 0; i < 64; i++)
394 block[i] = block1[i];
395 idct248_ref(img_dest1, 8, block);
396
397 for (i = 0; i < 64; i++)
398 block[i] = block1[i];
399 idct248_put(img_dest, 8, block);
400
401 for (i = 0; i < 64; i++) {
402 v = abs((int) img_dest[i] - (int) img_dest1[i]);
403 if (v == 255)
404 printf("%d %d\n", img_dest[i], img_dest1[i]);
405 if (v > err_max)
406 err_max = v;
407 }
408 #if 0
409 printf("ref=\n");
410 for(i=0;i<8;i++) {
411 int j;
412 for(j=0;j<8;j++) {
413 printf(" %3d", img_dest1[i*8+j]);
414 }
415 printf("\n");
416 }
417
418 printf("out=\n");
419 for(i=0;i<8;i++) {
420 int j;
421 for(j=0;j<8;j++) {
422 printf(" %3d", img_dest[i*8+j]);
423 }
424 printf("\n");
425 }
426 #endif
427 }
428 printf("%s %s: err_inf=%d\n", 1 ? "IDCT248" : "DCT248", name, err_max);
429
430 if (!speed)
431 return;
432
433 ti = av_gettime_relative();
434 it1 = 0;
435 do {
436 for (it = 0; it < NB_ITS_SPEED; it++) {
437 for (i = 0; i < 64; i++)
438 block[i] = block1[i];
439 idct248_put(img_dest, 8, block);
440 }
441 emms_c();
442 it1 += NB_ITS_SPEED;
443 ti1 = av_gettime_relative() - ti;
444 } while (ti1 < 1000000);
445
446 printf("%s %s: %0.1f kdct/s\n", 1 ? "IDCT248" : "DCT248", name,
447 (double) it1 * 1000.0 / (double) ti1);
448 }
449
help(void)450 static void help(void)
451 {
452 printf("dct-test [-i] [<test-number>] [<bits>]\n"
453 "test-number 0 -> test with random matrixes\n"
454 " 1 -> test with random sparse matrixes\n"
455 " 2 -> do 3. test from MPEG-4 std\n"
456 "bits Number of time domain bits to use, 8 is default\n"
457 "-i test IDCT implementations\n"
458 "-4 test IDCT248 implementations\n"
459 "-t speed test\n");
460 }
461
462 #if !HAVE_GETOPT
463 #include "compat/getopt.c"
464 #endif
465
main(int argc,char ** argv)466 int main(int argc, char **argv)
467 {
468 int test_idct = 0, test_248_dct = 0;
469 int c, i;
470 int test = 1;
471 int speed = 0;
472 int err = 0;
473 int bits=8;
474
475 ff_ref_dct_init();
476
477 for (;;) {
478 c = getopt(argc, argv, "ih4t");
479 if (c == -1)
480 break;
481 switch (c) {
482 case 'i':
483 test_idct = 1;
484 break;
485 case '4':
486 test_248_dct = 1;
487 break;
488 case 't':
489 speed = 1;
490 break;
491 default:
492 case 'h':
493 help();
494 return 0;
495 }
496 }
497
498 if (optind < argc)
499 test = atoi(argv[optind]);
500 if(optind+1 < argc) bits= atoi(argv[optind+1]);
501
502 printf("ffmpeg DCT/IDCT test\n");
503
504 if (test_248_dct) {
505 idct248_error("SIMPLE-C", ff_simple_idct248_put, speed);
506 } else {
507 const int cpu_flags = av_get_cpu_flags();
508 if (test_idct) {
509 for (i = 0; i < FF_ARRAY_ELEMS(idct_tab); i++)
510 err |= dct_error(&idct_tab[i], test, test_idct, speed, bits);
511
512 for (i = 0; idct_tab_arch[i].name; i++)
513 if (!(~cpu_flags & idct_tab_arch[i].cpu_flag))
514 err |= dct_error(&idct_tab_arch[i], test, test_idct, speed, bits);
515 }
516 #if CONFIG_FDCTDSP
517 else {
518 for (i = 0; i < FF_ARRAY_ELEMS(fdct_tab); i++)
519 err |= dct_error(&fdct_tab[i], test, test_idct, speed, bits);
520
521 for (i = 0; fdct_tab_arch[i].name; i++)
522 if (!(~cpu_flags & fdct_tab_arch[i].cpu_flag))
523 err |= dct_error(&fdct_tab_arch[i], test, test_idct, speed, bits);
524 }
525 #endif /* CONFIG_FDCTDSP */
526 }
527
528 if (err)
529 printf("Error: %d.\n", err);
530
531 return !!err;
532 }
533