Split global .gitignore file into per-directory files
[libav.git] / libavcodec / tests / dct.c
1 /*
2 * (c) 2001 Fabrice Bellard
3 * 2007 Marc Hoffman <marc.hoffman@analog.com>
4 *
5 * This file is part of Libav.
6 *
7 * Libav 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 * Libav 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 Libav; 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/aandcttab.h"
44 #include "libavcodec/dct.h"
45 #include "libavcodec/dctref.h"
46 #include "libavcodec/faandct.h"
47 #include "libavcodec/faanidct.h"
48 #include "libavcodec/idctdsp.h"
49 #include "libavcodec/simple_idct.h"
50 #include "libavcodec/xvididct.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
69 static const struct algo idct_tab[] = {
70 { "REF-DBL", ff_ref_idct, FF_IDCT_PERM_NONE },
71 { "INT", ff_j_rev_dct, FF_IDCT_PERM_LIBMPEG2 },
72 { "SIMPLE-C", ff_simple_idct_8, FF_IDCT_PERM_NONE },
73 #if CONFIG_FAANIDCT
74 { "FAANI", ff_faanidct, FF_IDCT_PERM_NONE },
75 #endif /* CONFIG_FAANIDCT */
76 #if CONFIG_MPEG4_DECODER
77 { "XVID", ff_xvid_idct, FF_IDCT_PERM_NONE, 0, 1 },
78 #endif /* CONFIG_MPEG4_DECODER */
79 };
80
81 #if ARCH_ARM
82 #include "arm/dct.c"
83 #elif ARCH_PPC
84 #include "ppc/dct.c"
85 #elif ARCH_X86
86 #include "x86/dct.c"
87 #else
88 static const struct algo fdct_tab_arch[] = { { 0 } };
89 static const struct algo idct_tab_arch[] = { { 0 } };
90 #endif
91
92 #define AANSCALE_BITS 12
93
94 #define NB_ITS 20000
95 #define NB_ITS_SPEED 50000
96
97 DECLARE_ALIGNED(16, static int16_t, block)[64];
98 DECLARE_ALIGNED(8, static int16_t, block1)[64];
99
100 static void init_block(int16_t block[64], int test, int is_idct, AVLFG *prng)
101 {
102 int i, j;
103
104 memset(block, 0, 64 * sizeof(*block));
105
106 switch (test) {
107 case 0:
108 for (i = 0; i < 64; i++)
109 block[i] = (av_lfg_get(prng) % 512) - 256;
110 if (is_idct) {
111 ff_ref_fdct(block);
112 for (i = 0; i < 64; i++)
113 block[i] >>= 3;
114 }
115 break;
116 case 1:
117 j = av_lfg_get(prng) % 10 + 1;
118 for (i = 0; i < j; i++)
119 block[av_lfg_get(prng) % 64] = av_lfg_get(prng) % 512 - 256;
120 break;
121 case 2:
122 block[ 0] = av_lfg_get(prng) % 4096 - 2048;
123 block[63] = (block[0] & 1) ^ 1;
124 break;
125 }
126 }
127
128 static void permute(int16_t dst[64], const int16_t src[64],
129 enum idct_permutation_type perm_type)
130 {
131 int i;
132
133 #if ARCH_X86
134 if (permute_x86(dst, src, perm_type))
135 return;
136 #endif
137
138 switch (perm_type) {
139 case FF_IDCT_PERM_LIBMPEG2:
140 for (i = 0; i < 64; i++)
141 dst[(i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2)] = src[i];
142 break;
143 case FF_IDCT_PERM_PARTTRANS:
144 for (i = 0; i < 64; i++)
145 dst[(i & 0x24) | ((i & 3) << 3) | ((i >> 3) & 3)] = src[i];
146 break;
147 default:
148 for (i = 0; i < 64; i++)
149 dst[i] = src[i];
150 break;
151 }
152 }
153
154 static int dct_error(const struct algo *dct, int test, int is_idct, int speed)
155 {
156 void (*ref)(int16_t *block) = is_idct ? ff_ref_idct : ff_ref_fdct;
157 int it, i, scale;
158 int err_inf, v;
159 int64_t err2, ti, ti1, it1, err_sum = 0;
160 int64_t sysErr[64], sysErrMax = 0;
161 int maxout = 0;
162 int blockSumErrMax = 0, blockSumErr;
163 AVLFG prng;
164 double omse, ome;
165 int spec_err;
166
167 av_lfg_init(&prng, 1);
168
169 err_inf = 0;
170 err2 = 0;
171 for (i = 0; i < 64; i++)
172 sysErr[i] = 0;
173 for (it = 0; it < NB_ITS; it++) {
174 init_block(block1, test, is_idct, &prng);
175 permute(block, block1, dct->perm_type);
176
177 dct->func(block);
178 emms_c();
179
180 if (!strcmp(dct->name, "IJG-AAN-INT")) {
181 for (i = 0; i < 64; i++) {
182 scale = 8 * (1 << (AANSCALE_BITS + 11)) / ff_aanscales[i];
183 block[i] = (block[i] * scale) >> AANSCALE_BITS;
184 }
185 }
186
187 ref(block1);
188
189 blockSumErr = 0;
190 for (i = 0; i < 64; i++) {
191 int err = block[i] - block1[i];
192 err_sum += err;
193 v = abs(err);
194 if (v > err_inf)
195 err_inf = v;
196 err2 += v * v;
197 sysErr[i] += block[i] - block1[i];
198 blockSumErr += v;
199 if (abs(block[i]) > maxout)
200 maxout = abs(block[i]);
201 }
202 if (blockSumErrMax < blockSumErr)
203 blockSumErrMax = blockSumErr;
204 }
205 for (i = 0; i < 64; i++)
206 sysErrMax = FFMAX(sysErrMax, FFABS(sysErr[i]));
207
208 for (i = 0; i < 64; i++) {
209 if (i % 8 == 0)
210 printf("\n");
211 printf("%7d ", (int) sysErr[i]);
212 }
213 printf("\n");
214
215 omse = (double) err2 / NB_ITS / 64;
216 ome = (double) err_sum / NB_ITS / 64;
217
218 spec_err = is_idct && (err_inf > 1 || omse > 0.02 || fabs(ome) > 0.0015);
219
220 printf("%s %s: ppe=%d omse=%0.8f ome=%0.8f syserr=%0.8f maxout=%d blockSumErr=%d\n",
221 is_idct ? "IDCT" : "DCT", dct->name, err_inf,
222 omse, ome, (double) sysErrMax / NB_ITS,
223 maxout, blockSumErrMax);
224
225 if (spec_err && !dct->nonspec)
226 return 1;
227
228 if (!speed)
229 return 0;
230
231 /* speed test */
232 init_block(block, test, is_idct, &prng);
233 permute(block1, block, dct->perm_type);
234
235 ti = av_gettime_relative();
236 it1 = 0;
237 do {
238 for (it = 0; it < NB_ITS_SPEED; it++) {
239 memcpy(block, block1, sizeof(block));
240 dct->func(block);
241 }
242 it1 += NB_ITS_SPEED;
243 ti1 = av_gettime_relative() - ti;
244 } while (ti1 < 1000000);
245 emms_c();
246
247 printf("%s %s: %0.1f kdct/s\n", is_idct ? "IDCT" : "DCT", dct->name,
248 (double) it1 * 1000.0 / (double) ti1);
249
250 return 0;
251 }
252
253 DECLARE_ALIGNED(8, static uint8_t, img_dest)[64];
254 DECLARE_ALIGNED(8, static uint8_t, img_dest1)[64];
255
256 static void idct248_ref(uint8_t *dest, int linesize, int16_t *block)
257 {
258 static int init;
259 static double c8[8][8];
260 static double c4[4][4];
261 double block1[64], block2[64], block3[64];
262 double s, sum, v;
263 int i, j, k;
264
265 if (!init) {
266 init = 1;
267
268 for (i = 0; i < 8; i++) {
269 sum = 0;
270 for (j = 0; j < 8; j++) {
271 s = (i == 0) ? sqrt(1.0 / 8.0) : sqrt(1.0 / 4.0);
272 c8[i][j] = s * cos(M_PI * i * (j + 0.5) / 8.0);
273 sum += c8[i][j] * c8[i][j];
274 }
275 }
276
277 for (i = 0; i < 4; i++) {
278 sum = 0;
279 for (j = 0; j < 4; j++) {
280 s = (i == 0) ? sqrt(1.0 / 4.0) : sqrt(1.0 / 2.0);
281 c4[i][j] = s * cos(M_PI * i * (j + 0.5) / 4.0);
282 sum += c4[i][j] * c4[i][j];
283 }
284 }
285 }
286
287 /* butterfly */
288 s = 0.5 * sqrt(2.0);
289 for (i = 0; i < 4; i++) {
290 for (j = 0; j < 8; j++) {
291 block1[8 * (2 * i) + j] =
292 (block[8 * (2 * i) + j] + block[8 * (2 * i + 1) + j]) * s;
293 block1[8 * (2 * i + 1) + j] =
294 (block[8 * (2 * i) + j] - block[8 * (2 * i + 1) + j]) * s;
295 }
296 }
297
298 /* idct8 on lines */
299 for (i = 0; i < 8; i++) {
300 for (j = 0; j < 8; j++) {
301 sum = 0;
302 for (k = 0; k < 8; k++)
303 sum += c8[k][j] * block1[8 * i + k];
304 block2[8 * i + j] = sum;
305 }
306 }
307
308 /* idct4 */
309 for (i = 0; i < 8; i++) {
310 for (j = 0; j < 4; j++) {
311 /* top */
312 sum = 0;
313 for (k = 0; k < 4; k++)
314 sum += c4[k][j] * block2[8 * (2 * k) + i];
315 block3[8 * (2 * j) + i] = sum;
316
317 /* bottom */
318 sum = 0;
319 for (k = 0; k < 4; k++)
320 sum += c4[k][j] * block2[8 * (2 * k + 1) + i];
321 block3[8 * (2 * j + 1) + i] = sum;
322 }
323 }
324
325 /* clamp and store the result */
326 for (i = 0; i < 8; i++) {
327 for (j = 0; j < 8; j++) {
328 v = block3[8 * i + j];
329 if (v < 0) v = 0;
330 else if (v > 255) v = 255;
331 dest[i * linesize + j] = (int) rint(v);
332 }
333 }
334 }
335
336 static void idct248_error(const char *name,
337 void (*idct248_put)(uint8_t *dest, int line_size,
338 int16_t *block),
339 int speed)
340 {
341 int it, i, it1, ti, ti1, err_max, v;
342 AVLFG prng;
343
344 av_lfg_init(&prng, 1);
345
346 /* just one test to see if code is correct (precision is less
347 important here) */
348 err_max = 0;
349 for (it = 0; it < NB_ITS; it++) {
350 /* XXX: use forward transform to generate values */
351 for (i = 0; i < 64; i++)
352 block1[i] = av_lfg_get(&prng) % 256 - 128;
353 block1[0] += 1024;
354
355 for (i = 0; i < 64; i++)
356 block[i] = block1[i];
357 idct248_ref(img_dest1, 8, block);
358
359 for (i = 0; i < 64; i++)
360 block[i] = block1[i];
361 idct248_put(img_dest, 8, block);
362
363 for (i = 0; i < 64; i++) {
364 v = abs((int) img_dest[i] - (int) img_dest1[i]);
365 if (v == 255)
366 printf("%d %d\n", img_dest[i], img_dest1[i]);
367 if (v > err_max)
368 err_max = v;
369 }
370 }
371 printf("%s %s: err_inf=%d\n", 1 ? "IDCT248" : "DCT248", name, err_max);
372
373 if (!speed)
374 return;
375
376 ti = av_gettime_relative();
377 it1 = 0;
378 do {
379 for (it = 0; it < NB_ITS_SPEED; it++) {
380 for (i = 0; i < 64; i++)
381 block[i] = block1[i];
382 idct248_put(img_dest, 8, block);
383 }
384 it1 += NB_ITS_SPEED;
385 ti1 = av_gettime_relative() - ti;
386 } while (ti1 < 1000000);
387 emms_c();
388
389 printf("%s %s: %0.1f kdct/s\n", 1 ? "IDCT248" : "DCT248", name,
390 (double) it1 * 1000.0 / (double) ti1);
391 }
392
393 static void help(void)
394 {
395 printf("dct-test [-i] [<test-number>]\n"
396 "test-number 0 -> test with random matrixes\n"
397 " 1 -> test with random sparse matrixes\n"
398 " 2 -> do 3. test from MPEG-4 std\n"
399 "-i test IDCT implementations\n"
400 "-4 test IDCT248 implementations\n"
401 "-t speed test\n");
402 }
403
404 #if !HAVE_GETOPT
405 #include "compat/getopt.c"
406 #endif
407
408 int main(int argc, char **argv)
409 {
410 int test_idct = 0, test_248_dct = 0;
411 int c, i;
412 int test = 1;
413 int speed = 0;
414 int err = 0;
415
416 ff_ref_dct_init();
417
418 for (;;) {
419 c = getopt(argc, argv, "ih4t");
420 if (c == -1)
421 break;
422 switch (c) {
423 case 'i':
424 test_idct = 1;
425 break;
426 case '4':
427 test_248_dct = 1;
428 break;
429 case 't':
430 speed = 1;
431 break;
432 default:
433 case 'h':
434 help();
435 return 0;
436 }
437 }
438
439 if (optind < argc)
440 test = atoi(argv[optind]);
441
442 printf("Libav DCT/IDCT test\n");
443
444 if (test_248_dct) {
445 idct248_error("SIMPLE-C", ff_simple_idct248_put, speed);
446 } else {
447 const int cpu_flags = av_get_cpu_flags();
448 if (test_idct) {
449 for (i = 0; i < FF_ARRAY_ELEMS(idct_tab); i++)
450 err |= dct_error(&idct_tab[i], test, test_idct, speed);
451
452 for (i = 0; idct_tab_arch[i].name; i++)
453 if (!(~cpu_flags & idct_tab_arch[i].cpu_flag))
454 err |= dct_error(&idct_tab_arch[i], test, test_idct, speed);
455 }
456 #if CONFIG_FDCTDSP
457 else {
458 for (i = 0; i < FF_ARRAY_ELEMS(fdct_tab); i++)
459 err |= dct_error(&fdct_tab[i], test, test_idct, speed);
460
461 for (i = 0; fdct_tab_arch[i].name; i++)
462 if (!(~cpu_flags & fdct_tab_arch[i].cpu_flag))
463 err |= dct_error(&fdct_tab_arch[i], test, test_idct, speed);
464 }
465 #endif /* CONFIG_FDCTDSP */
466 }
467
468 if (err)
469 printf("Error: %d.\n", err);
470
471 return !!err;
472 }