2 * (c) 2001 Fabrice Bellard
3 * 2007 Marc Hoffman <marc.hoffman@analog.com>
5 * This file is part of Libav.
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.
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.
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
24 * DCT test (c) 2001 Fabrice Bellard
25 * Started from sample code by Juan J. Sierralta P.
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"
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"
54 void (*func)(int16_t *block);
55 enum idct_permutation_type perm_type;
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 },
65 { "FAAN", ff_faandct, FF_IDCT_PERM_NONE },
66 #endif /* CONFIG_FAANDCT */
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 },
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 */
88 static const struct algo fdct_tab_arch[] = { { 0 } };
89 static const struct algo idct_tab_arch[] = { { 0 } };
92 #define AANSCALE_BITS 12
95 #define NB_ITS_SPEED 50000
97 DECLARE_ALIGNED(16, static int16_t, block)[64];
98 DECLARE_ALIGNED(8, static int16_t, block1)[64];
100 static void init_block(int16_t block[64], int test, int is_idct, AVLFG *prng)
104 memset(block, 0, 64 * sizeof(*block));
108 for (i = 0; i < 64; i++)
109 block[i] = (av_lfg_get(prng) % 512) - 256;
112 for (i = 0; i < 64; i++)
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;
122 block[ 0] = av_lfg_get(prng) % 4096 - 2048;
123 block[63] = (block[0] & 1) ^ 1;
128 static void permute(int16_t dst[64], const int16_t src[64],
129 enum idct_permutation_type perm_type)
134 if (permute_x86(dst, src, 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];
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];
148 for (i = 0; i < 64; i++)
154 static int dct_error(const struct algo *dct, int test, int is_idct, int speed)
156 void (*ref)(int16_t *block) = is_idct ? ff_ref_idct : ff_ref_fdct;
159 int64_t err2, ti, ti1, it1, err_sum = 0;
160 int64_t sysErr[64], sysErrMax = 0;
162 int blockSumErrMax = 0, blockSumErr;
167 av_lfg_init(&prng, 1);
171 for (i = 0; i < 64; i++)
173 for (it = 0; it < NB_ITS; it++) {
174 init_block(block1, test, is_idct, &prng);
175 permute(block, block1, dct->perm_type);
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;
190 for (i = 0; i < 64; i++) {
191 int err = block[i] - block1[i];
197 sysErr[i] += block[i] - block1[i];
199 if (abs(block[i]) > maxout)
200 maxout = abs(block[i]);
202 if (blockSumErrMax < blockSumErr)
203 blockSumErrMax = blockSumErr;
205 for (i = 0; i < 64; i++)
206 sysErrMax = FFMAX(sysErrMax, FFABS(sysErr[i]));
208 for (i = 0; i < 64; i++) {
211 printf("%7d ", (int) sysErr[i]);
215 omse = (double) err2 / NB_ITS / 64;
216 ome = (double) err_sum / NB_ITS / 64;
218 spec_err = is_idct && (err_inf > 1 || omse > 0.02 || fabs(ome) > 0.0015);
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);
225 if (spec_err && !dct->nonspec)
232 init_block(block, test, is_idct, &prng);
233 permute(block1, block, dct->perm_type);
235 ti = av_gettime_relative();
238 for (it = 0; it < NB_ITS_SPEED; it++) {
239 memcpy(block, block1, sizeof(block));
243 ti1 = av_gettime_relative() - ti;
244 } while (ti1 < 1000000);
247 printf("%s %s: %0.1f kdct/s\n", is_idct ? "IDCT" : "DCT", dct->name,
248 (double) it1 * 1000.0 / (double) ti1);
253 DECLARE_ALIGNED(8, static uint8_t, img_dest)[64];
254 DECLARE_ALIGNED(8, static uint8_t, img_dest1)[64];
256 static void idct248_ref(uint8_t *dest, int linesize, int16_t *block)
259 static double c8[8][8];
260 static double c4[4][4];
261 double block1[64], block2[64], block3[64];
268 for (i = 0; i < 8; i++) {
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];
277 for (i = 0; i < 4; i++) {
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];
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;
299 for (i = 0; i < 8; i++) {
300 for (j = 0; j < 8; j++) {
302 for (k = 0; k < 8; k++)
303 sum += c8[k][j] * block1[8 * i + k];
304 block2[8 * i + j] = sum;
309 for (i = 0; i < 8; i++) {
310 for (j = 0; j < 4; j++) {
313 for (k = 0; k < 4; k++)
314 sum += c4[k][j] * block2[8 * (2 * k) + i];
315 block3[8 * (2 * j) + i] = sum;
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;
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];
330 else if (v > 255) v = 255;
331 dest[i * linesize + j] = (int) rint(v);
336 static void idct248_error(const char *name,
337 void (*idct248_put)(uint8_t *dest, int line_size,
341 int it, i, it1, ti, ti1, err_max, v;
344 av_lfg_init(&prng, 1);
346 /* just one test to see if code is correct (precision is less
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;
355 for (i = 0; i < 64; i++)
356 block[i] = block1[i];
357 idct248_ref(img_dest1, 8, block);
359 for (i = 0; i < 64; i++)
360 block[i] = block1[i];
361 idct248_put(img_dest, 8, block);
363 for (i = 0; i < 64; i++) {
364 v = abs((int) img_dest[i] - (int) img_dest1[i]);
366 printf("%d %d\n", img_dest[i], img_dest1[i]);
371 printf("%s %s: err_inf=%d\n", 1 ? "IDCT248" : "DCT248", name, err_max);
376 ti = av_gettime_relative();
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);
385 ti1 = av_gettime_relative() - ti;
386 } while (ti1 < 1000000);
389 printf("%s %s: %0.1f kdct/s\n", 1 ? "IDCT248" : "DCT248", name,
390 (double) it1 * 1000.0 / (double) ti1);
393 static void help(void)
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"
405 #include "compat/getopt.c"
408 int main(int argc, char **argv)
410 int test_idct = 0, test_248_dct = 0;
419 c = getopt(argc, argv, "ih4t");
440 test = atoi(argv[optind]);
442 printf("Libav DCT/IDCT test\n");
445 idct248_error("SIMPLE-C", ff_simple_idct248_put, speed);
447 const int cpu_flags = av_get_cpu_flags();
449 for (i = 0; i < FF_ARRAY_ELEMS(idct_tab); i++)
450 err |= dct_error(&idct_tab[i], test, test_idct, speed);
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);
458 for (i = 0; i < FF_ARRAY_ELEMS(fdct_tab); i++)
459 err |= dct_error(&fdct_tab[i], test, test_idct, speed);
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);
465 #endif /* CONFIG_FDCTDSP */
469 printf("Error: %d.\n", err);