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/lfg.h"
40 #include "libavutil/time.h"
44 #include "simple_idct.h"
45 #include "aandcttab.h"
52 void (*func)(int16_t *block);
53 enum idct_permutation_type perm_type;
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 },
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 },
73 #include "arm/dct-test.c"
75 #include "ppc/dct-test.c"
77 #include "x86/dct-test.c"
79 static const struct algo fdct_tab_arch[] = { 0 };
80 static const struct algo idct_tab_arch[] = { 0 };
83 #define AANSCALE_BITS 12
86 #define NB_ITS_SPEED 50000
88 DECLARE_ALIGNED(16, static int16_t, block)[64];
89 DECLARE_ALIGNED(8, static int16_t, block1)[64];
91 static void init_block(int16_t block[64], int test, int is_idct, AVLFG *prng)
95 memset(block, 0, 64 * sizeof(*block));
99 for (i = 0; i < 64; i++)
100 block[i] = (av_lfg_get(prng) % 512) - 256;
103 for (i = 0; i < 64; i++)
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;
113 block[ 0] = av_lfg_get(prng) % 4096 - 2048;
114 block[63] = (block[0] & 1) ^ 1;
119 static void permute(int16_t dst[64], const int16_t src[64],
120 enum idct_permutation_type perm_type)
125 if (permute_x86(dst, src, 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];
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];
139 for (i = 0; i < 64; i++)
145 static int dct_error(const struct algo *dct, int test, int is_idct, int speed)
147 void (*ref)(int16_t *block) = is_idct ? ff_ref_idct : ff_ref_fdct;
150 int64_t err2, ti, ti1, it1, err_sum = 0;
151 int64_t sysErr[64], sysErrMax = 0;
153 int blockSumErrMax = 0, blockSumErr;
158 av_lfg_init(&prng, 1);
162 for (i = 0; i < 64; i++)
164 for (it = 0; it < NB_ITS; it++) {
165 init_block(block1, test, is_idct, &prng);
166 permute(block, block1, dct->perm_type);
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;
181 for (i = 0; i < 64; i++) {
182 int err = block[i] - block1[i];
188 sysErr[i] += block[i] - block1[i];
190 if (abs(block[i]) > maxout)
191 maxout = abs(block[i]);
193 if (blockSumErrMax < blockSumErr)
194 blockSumErrMax = blockSumErr;
196 for (i = 0; i < 64; i++)
197 sysErrMax = FFMAX(sysErrMax, FFABS(sysErr[i]));
199 for (i = 0; i < 64; i++) {
202 printf("%7d ", (int) sysErr[i]);
206 omse = (double) err2 / NB_ITS / 64;
207 ome = (double) err_sum / NB_ITS / 64;
209 spec_err = is_idct && (err_inf > 1 || omse > 0.02 || fabs(ome) > 0.0015);
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);
216 if (spec_err && !dct->nonspec)
223 init_block(block, test, is_idct, &prng);
224 permute(block1, block, dct->perm_type);
229 for (it = 0; it < NB_ITS_SPEED; it++) {
230 memcpy(block, block1, sizeof(block));
234 ti1 = av_gettime() - ti;
235 } while (ti1 < 1000000);
238 printf("%s %s: %0.1f kdct/s\n", is_idct ? "IDCT" : "DCT", dct->name,
239 (double) it1 * 1000.0 / (double) ti1);
244 DECLARE_ALIGNED(8, static uint8_t, img_dest)[64];
245 DECLARE_ALIGNED(8, static uint8_t, img_dest1)[64];
247 static void idct248_ref(uint8_t *dest, int linesize, int16_t *block)
250 static double c8[8][8];
251 static double c4[4][4];
252 double block1[64], block2[64], block3[64];
259 for (i = 0; i < 8; i++) {
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];
268 for (i = 0; i < 4; i++) {
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];
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;
290 for (i = 0; i < 8; i++) {
291 for (j = 0; j < 8; j++) {
293 for (k = 0; k < 8; k++)
294 sum += c8[k][j] * block1[8 * i + k];
295 block2[8 * i + j] = sum;
300 for (i = 0; i < 8; i++) {
301 for (j = 0; j < 4; j++) {
304 for (k = 0; k < 4; k++)
305 sum += c4[k][j] * block2[8 * (2 * k) + i];
306 block3[8 * (2 * j) + i] = sum;
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;
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];
321 else if (v > 255) v = 255;
322 dest[i * linesize + j] = (int) rint(v);
327 static void idct248_error(const char *name,
328 void (*idct248_put)(uint8_t *dest, int line_size,
332 int it, i, it1, ti, ti1, err_max, v;
335 av_lfg_init(&prng, 1);
337 /* just one test to see if code is correct (precision is less
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;
346 for (i = 0; i < 64; i++)
347 block[i] = block1[i];
348 idct248_ref(img_dest1, 8, block);
350 for (i = 0; i < 64; i++)
351 block[i] = block1[i];
352 idct248_put(img_dest, 8, block);
354 for (i = 0; i < 64; i++) {
355 v = abs((int) img_dest[i] - (int) img_dest1[i]);
357 printf("%d %d\n", img_dest[i], img_dest1[i]);
362 printf("%s %s: err_inf=%d\n", 1 ? "IDCT248" : "DCT248", name, err_max);
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);
376 ti1 = av_gettime() - ti;
377 } while (ti1 < 1000000);
380 printf("%s %s: %0.1f kdct/s\n", 1 ? "IDCT248" : "DCT248", name,
381 (double) it1 * 1000.0 / (double) ti1);
384 static void help(void)
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"
396 #include "compat/getopt.c"
399 int main(int argc, char **argv)
401 int test_idct = 0, test_248_dct = 0;
410 c = getopt(argc, argv, "ih4t");
431 test = atoi(argv[optind]);
433 printf("Libav DCT/IDCT test\n");
436 idct248_error("SIMPLE-C", ff_simple_idct248_put, speed);
438 const int cpu_flags = av_get_cpu_flags();
440 for (i = 0; i < FF_ARRAY_ELEMS(idct_tab); i++)
441 err |= dct_error(&idct_tab[i], test, test_idct, speed);
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);
447 for (i = 0; i < FF_ARRAY_ELEMS(fdct_tab); i++)
448 err |= dct_error(&fdct_tab[i], test, test_idct, speed);
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);
457 printf("Error: %d.\n", err);