dct-test: Move arch-specific bits into arch-specific subdirectories
[libav.git] / libavcodec / dct-test.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/lfg.h"
40 #include "libavutil/time.h"
41
42 #include "dct.h"
43 #include "idctdsp.h"
44 #include "simple_idct.h"
45 #include "aandcttab.h"
46 #include "faandct.h"
47 #include "faanidct.h"
48 #include "dctref.h"
49
50 struct algo {
51 const char *name;
52 void (*func)(int16_t *block);
53 enum idct_permutation_type perm_type;
54 int cpu_flag;
55 int nonspec;
56 };
57
58 static const struct algo fdct_tab[4] = {
59 { "REF-DBL", ff_ref_fdct, FF_IDCT_PERM_NONE },
60 { "FAAN", ff_faandct, FF_IDCT_PERM_NONE },
61 { "IJG-AAN-INT", ff_fdct_ifast, FF_IDCT_PERM_NONE },
62 { "IJG-LLM-INT", ff_jpeg_fdct_islow_8, FF_IDCT_PERM_NONE },
63 };
64
65 static const struct algo idct_tab[4] = {
66 { "FAANI", ff_faanidct, FF_IDCT_PERM_NONE },
67 { "REF-DBL", ff_ref_idct, FF_IDCT_PERM_NONE },
68 { "INT", ff_j_rev_dct, FF_IDCT_PERM_LIBMPEG2 },
69 { "SIMPLE-C", ff_simple_idct_8, FF_IDCT_PERM_NONE },
70 };
71
72 #if ARCH_ARM
73 #include "arm/dct-test.c"
74 #elif ARCH_PPC
75 #include "ppc/dct-test.c"
76 #elif ARCH_X86
77 #include "x86/dct-test.c"
78 #else
79 static const struct algo fdct_tab_arch[] = { 0 };
80 static const struct algo idct_tab_arch[] = { 0 };
81 #endif
82
83 #define AANSCALE_BITS 12
84
85 #define NB_ITS 20000
86 #define NB_ITS_SPEED 50000
87
88 DECLARE_ALIGNED(16, static int16_t, block)[64];
89 DECLARE_ALIGNED(8, static int16_t, block1)[64];
90
91 static void init_block(int16_t block[64], int test, int is_idct, AVLFG *prng)
92 {
93 int i, j;
94
95 memset(block, 0, 64 * sizeof(*block));
96
97 switch (test) {
98 case 0:
99 for (i = 0; i < 64; i++)
100 block[i] = (av_lfg_get(prng) % 512) - 256;
101 if (is_idct) {
102 ff_ref_fdct(block);
103 for (i = 0; i < 64; i++)
104 block[i] >>= 3;
105 }
106 break;
107 case 1:
108 j = av_lfg_get(prng) % 10 + 1;
109 for (i = 0; i < j; i++)
110 block[av_lfg_get(prng) % 64] = av_lfg_get(prng) % 512 - 256;
111 break;
112 case 2:
113 block[ 0] = av_lfg_get(prng) % 4096 - 2048;
114 block[63] = (block[0] & 1) ^ 1;
115 break;
116 }
117 }
118
119 static void permute(int16_t dst[64], const int16_t src[64],
120 enum idct_permutation_type perm_type)
121 {
122 int i;
123
124 #if ARCH_X86
125 if (permute_x86(dst, src, perm_type))
126 return;
127 #endif
128
129 switch (perm_type) {
130 case FF_IDCT_PERM_LIBMPEG2:
131 for (i = 0; i < 64; i++)
132 dst[(i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2)] = src[i];
133 break;
134 case FF_IDCT_PERM_PARTTRANS:
135 for (i = 0; i < 64; i++)
136 dst[(i & 0x24) | ((i & 3) << 3) | ((i >> 3) & 3)] = src[i];
137 break;
138 default:
139 for (i = 0; i < 64; i++)
140 dst[i] = src[i];
141 break;
142 }
143 }
144
145 static int dct_error(const struct algo *dct, int test, int is_idct, int speed)
146 {
147 void (*ref)(int16_t *block) = is_idct ? ff_ref_idct : ff_ref_fdct;
148 int it, i, scale;
149 int err_inf, v;
150 int64_t err2, ti, ti1, it1, err_sum = 0;
151 int64_t sysErr[64], sysErrMax = 0;
152 int maxout = 0;
153 int blockSumErrMax = 0, blockSumErr;
154 AVLFG prng;
155 double omse, ome;
156 int spec_err;
157
158 av_lfg_init(&prng, 1);
159
160 err_inf = 0;
161 err2 = 0;
162 for (i = 0; i < 64; i++)
163 sysErr[i] = 0;
164 for (it = 0; it < NB_ITS; it++) {
165 init_block(block1, test, is_idct, &prng);
166 permute(block, block1, dct->perm_type);
167
168 dct->func(block);
169 emms_c();
170
171 if (!strcmp(dct->name, "IJG-AAN-INT")) {
172 for (i = 0; i < 64; i++) {
173 scale = 8 * (1 << (AANSCALE_BITS + 11)) / ff_aanscales[i];
174 block[i] = (block[i] * scale) >> AANSCALE_BITS;
175 }
176 }
177
178 ref(block1);
179
180 blockSumErr = 0;
181 for (i = 0; i < 64; i++) {
182 int err = block[i] - block1[i];
183 err_sum += err;
184 v = abs(err);
185 if (v > err_inf)
186 err_inf = v;
187 err2 += v * v;
188 sysErr[i] += block[i] - block1[i];
189 blockSumErr += v;
190 if (abs(block[i]) > maxout)
191 maxout = abs(block[i]);
192 }
193 if (blockSumErrMax < blockSumErr)
194 blockSumErrMax = blockSumErr;
195 }
196 for (i = 0; i < 64; i++)
197 sysErrMax = FFMAX(sysErrMax, FFABS(sysErr[i]));
198
199 for (i = 0; i < 64; i++) {
200 if (i % 8 == 0)
201 printf("\n");
202 printf("%7d ", (int) sysErr[i]);
203 }
204 printf("\n");
205
206 omse = (double) err2 / NB_ITS / 64;
207 ome = (double) err_sum / NB_ITS / 64;
208
209 spec_err = is_idct && (err_inf > 1 || omse > 0.02 || fabs(ome) > 0.0015);
210
211 printf("%s %s: ppe=%d omse=%0.8f ome=%0.8f syserr=%0.8f maxout=%d blockSumErr=%d\n",
212 is_idct ? "IDCT" : "DCT", dct->name, err_inf,
213 omse, ome, (double) sysErrMax / NB_ITS,
214 maxout, blockSumErrMax);
215
216 if (spec_err && !dct->nonspec)
217 return 1;
218
219 if (!speed)
220 return 0;
221
222 /* speed test */
223 init_block(block, test, is_idct, &prng);
224 permute(block1, block, dct->perm_type);
225
226 ti = av_gettime();
227 it1 = 0;
228 do {
229 for (it = 0; it < NB_ITS_SPEED; it++) {
230 memcpy(block, block1, sizeof(block));
231 dct->func(block);
232 }
233 it1 += NB_ITS_SPEED;
234 ti1 = av_gettime() - ti;
235 } while (ti1 < 1000000);
236 emms_c();
237
238 printf("%s %s: %0.1f kdct/s\n", is_idct ? "IDCT" : "DCT", dct->name,
239 (double) it1 * 1000.0 / (double) ti1);
240
241 return 0;
242 }
243
244 DECLARE_ALIGNED(8, static uint8_t, img_dest)[64];
245 DECLARE_ALIGNED(8, static uint8_t, img_dest1)[64];
246
247 static void idct248_ref(uint8_t *dest, int linesize, int16_t *block)
248 {
249 static int init;
250 static double c8[8][8];
251 static double c4[4][4];
252 double block1[64], block2[64], block3[64];
253 double s, sum, v;
254 int i, j, k;
255
256 if (!init) {
257 init = 1;
258
259 for (i = 0; i < 8; i++) {
260 sum = 0;
261 for (j = 0; j < 8; j++) {
262 s = (i == 0) ? sqrt(1.0 / 8.0) : sqrt(1.0 / 4.0);
263 c8[i][j] = s * cos(M_PI * i * (j + 0.5) / 8.0);
264 sum += c8[i][j] * c8[i][j];
265 }
266 }
267
268 for (i = 0; i < 4; i++) {
269 sum = 0;
270 for (j = 0; j < 4; j++) {
271 s = (i == 0) ? sqrt(1.0 / 4.0) : sqrt(1.0 / 2.0);
272 c4[i][j] = s * cos(M_PI * i * (j + 0.5) / 4.0);
273 sum += c4[i][j] * c4[i][j];
274 }
275 }
276 }
277
278 /* butterfly */
279 s = 0.5 * sqrt(2.0);
280 for (i = 0; i < 4; i++) {
281 for (j = 0; j < 8; j++) {
282 block1[8 * (2 * i) + j] =
283 (block[8 * (2 * i) + j] + block[8 * (2 * i + 1) + j]) * s;
284 block1[8 * (2 * i + 1) + j] =
285 (block[8 * (2 * i) + j] - block[8 * (2 * i + 1) + j]) * s;
286 }
287 }
288
289 /* idct8 on lines */
290 for (i = 0; i < 8; i++) {
291 for (j = 0; j < 8; j++) {
292 sum = 0;
293 for (k = 0; k < 8; k++)
294 sum += c8[k][j] * block1[8 * i + k];
295 block2[8 * i + j] = sum;
296 }
297 }
298
299 /* idct4 */
300 for (i = 0; i < 8; i++) {
301 for (j = 0; j < 4; j++) {
302 /* top */
303 sum = 0;
304 for (k = 0; k < 4; k++)
305 sum += c4[k][j] * block2[8 * (2 * k) + i];
306 block3[8 * (2 * j) + i] = sum;
307
308 /* bottom */
309 sum = 0;
310 for (k = 0; k < 4; k++)
311 sum += c4[k][j] * block2[8 * (2 * k + 1) + i];
312 block3[8 * (2 * j + 1) + i] = sum;
313 }
314 }
315
316 /* clamp and store the result */
317 for (i = 0; i < 8; i++) {
318 for (j = 0; j < 8; j++) {
319 v = block3[8 * i + j];
320 if (v < 0) v = 0;
321 else if (v > 255) v = 255;
322 dest[i * linesize + j] = (int) rint(v);
323 }
324 }
325 }
326
327 static void idct248_error(const char *name,
328 void (*idct248_put)(uint8_t *dest, int line_size,
329 int16_t *block),
330 int speed)
331 {
332 int it, i, it1, ti, ti1, err_max, v;
333 AVLFG prng;
334
335 av_lfg_init(&prng, 1);
336
337 /* just one test to see if code is correct (precision is less
338 important here) */
339 err_max = 0;
340 for (it = 0; it < NB_ITS; it++) {
341 /* XXX: use forward transform to generate values */
342 for (i = 0; i < 64; i++)
343 block1[i] = av_lfg_get(&prng) % 256 - 128;
344 block1[0] += 1024;
345
346 for (i = 0; i < 64; i++)
347 block[i] = block1[i];
348 idct248_ref(img_dest1, 8, block);
349
350 for (i = 0; i < 64; i++)
351 block[i] = block1[i];
352 idct248_put(img_dest, 8, block);
353
354 for (i = 0; i < 64; i++) {
355 v = abs((int) img_dest[i] - (int) img_dest1[i]);
356 if (v == 255)
357 printf("%d %d\n", img_dest[i], img_dest1[i]);
358 if (v > err_max)
359 err_max = v;
360 }
361 }
362 printf("%s %s: err_inf=%d\n", 1 ? "IDCT248" : "DCT248", name, err_max);
363
364 if (!speed)
365 return;
366
367 ti = av_gettime();
368 it1 = 0;
369 do {
370 for (it = 0; it < NB_ITS_SPEED; it++) {
371 for (i = 0; i < 64; i++)
372 block[i] = block1[i];
373 idct248_put(img_dest, 8, block);
374 }
375 it1 += NB_ITS_SPEED;
376 ti1 = av_gettime() - ti;
377 } while (ti1 < 1000000);
378 emms_c();
379
380 printf("%s %s: %0.1f kdct/s\n", 1 ? "IDCT248" : "DCT248", name,
381 (double) it1 * 1000.0 / (double) ti1);
382 }
383
384 static void help(void)
385 {
386 printf("dct-test [-i] [<test-number>]\n"
387 "test-number 0 -> test with random matrixes\n"
388 " 1 -> test with random sparse matrixes\n"
389 " 2 -> do 3. test from mpeg4 std\n"
390 "-i test IDCT implementations\n"
391 "-4 test IDCT248 implementations\n"
392 "-t speed test\n");
393 }
394
395 #if !HAVE_GETOPT
396 #include "compat/getopt.c"
397 #endif
398
399 int main(int argc, char **argv)
400 {
401 int test_idct = 0, test_248_dct = 0;
402 int c, i;
403 int test = 1;
404 int speed = 0;
405 int err = 0;
406
407 ff_ref_dct_init();
408
409 for (;;) {
410 c = getopt(argc, argv, "ih4t");
411 if (c == -1)
412 break;
413 switch (c) {
414 case 'i':
415 test_idct = 1;
416 break;
417 case '4':
418 test_248_dct = 1;
419 break;
420 case 't':
421 speed = 1;
422 break;
423 default:
424 case 'h':
425 help();
426 return 0;
427 }
428 }
429
430 if (optind < argc)
431 test = atoi(argv[optind]);
432
433 printf("Libav DCT/IDCT test\n");
434
435 if (test_248_dct) {
436 idct248_error("SIMPLE-C", ff_simple_idct248_put, speed);
437 } else {
438 const int cpu_flags = av_get_cpu_flags();
439 if (test_idct) {
440 for (i = 0; i < FF_ARRAY_ELEMS(idct_tab); i++)
441 err |= dct_error(&idct_tab[i], test, test_idct, speed);
442
443 for (i = 0; idct_tab_arch[i].name; i++)
444 if (!(~cpu_flags & idct_tab_arch[i].cpu_flag))
445 err |= dct_error(&idct_tab_arch[i], test, test_idct, speed);
446 } else {
447 for (i = 0; i < FF_ARRAY_ELEMS(fdct_tab); i++)
448 err |= dct_error(&fdct_tab[i], test, test_idct, speed);
449
450 for (i = 0; fdct_tab_arch[i].name; i++)
451 if (!(~cpu_flags & fdct_tab_arch[i].cpu_flag))
452 err |= dct_error(&fdct_tab_arch[i], test, test_idct, speed);
453 }
454 }
455
456 if (err)
457 printf("Error: %d.\n", err);
458
459 return !!err;
460 }