2 * (c) 2001 Fabrice Bellard
3 * 2007 Marc Hoffman <marc.hoffman@analog.com>
5 * This file is part of FFmpeg.
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.
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.
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
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"
46 #include "aandcttab.h"
53 void (*func)(int16_t *block);
54 enum idct_permutation_type perm_type;
59 static const struct algo fdct_tab[] = {
60 { "REF-DBL", ff_ref_fdct, 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 },
64 { "FAAN", ff_faandct, FF_IDCT_PERM_NONE },
65 #endif /* CONFIG_FAANDCT */
68 static void ff_prores_idct_wrap(int16_t *dst){
69 LOCAL_ALIGNED(16, int16_t, qmat, [64]);
75 ff_prores_idct(dst, qmat);
81 static const struct algo idct_tab[] = {
82 { "REF-DBL", ff_ref_idct, FF_IDCT_PERM_NONE },
83 { "INT", ff_j_rev_dct, FF_IDCT_PERM_LIBMPEG2 },
84 { "SIMPLE-C", ff_simple_idct_8, FF_IDCT_PERM_NONE },
85 { "PR-C", ff_prores_idct_wrap, FF_IDCT_PERM_NONE, 0, 1 },
87 { "FAANI", ff_faanidct, FF_IDCT_PERM_NONE },
88 #endif /* CONFIG_FAANIDCT */
89 #if CONFIG_MPEG4_DECODER
90 { "XVID", ff_xvid_idct, FF_IDCT_PERM_NONE, 0, 1 },
91 #endif /* CONFIG_MPEG4_DECODER */
95 #include "arm/dct-test.c"
97 #include "ppc/dct-test.c"
99 #include "x86/dct-test.c"
101 static const struct algo fdct_tab_arch[] = { { 0 } };
102 static const struct algo idct_tab_arch[] = { { 0 } };
105 #define AANSCALE_BITS 12
108 #define NB_ITS_SPEED 50000
110 DECLARE_ALIGNED(16, static int16_t, block)[64];
111 DECLARE_ALIGNED(8, static int16_t, block1)[64];
113 static void init_block(int16_t block[64], int test, int is_idct, AVLFG *prng, int vals)
117 memset(block, 0, 64 * sizeof(*block));
121 for (i = 0; i < 64; i++)
122 block[i] = (av_lfg_get(prng) % (2*vals)) -vals;
125 for (i = 0; i < 64; i++)
130 j = av_lfg_get(prng) % 10 + 1;
131 for (i = 0; i < j; i++) {
132 int idx = av_lfg_get(prng) % 64;
133 block[idx] = av_lfg_get(prng) % (2*vals) -vals;
137 block[ 0] = av_lfg_get(prng) % (16*vals) - (8*vals);
138 block[63] = (block[0] & 1) ^ 1;
143 static void permute(int16_t dst[64], const int16_t src[64],
144 enum idct_permutation_type perm_type)
149 if (permute_x86(dst, src, perm_type))
154 case FF_IDCT_PERM_LIBMPEG2:
155 for (i = 0; i < 64; i++)
156 dst[(i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2)] = src[i];
158 case FF_IDCT_PERM_PARTTRANS:
159 for (i = 0; i < 64; i++)
160 dst[(i & 0x24) | ((i & 3) << 3) | ((i >> 3) & 3)] = src[i];
162 case FF_IDCT_PERM_TRANSPOSE:
163 for (i = 0; i < 64; i++)
164 dst[(i>>3) | ((i<<3)&0x38)] = src[i];
167 for (i = 0; i < 64; i++)
173 static int dct_error(const struct algo *dct, int test, int is_idct, int speed, const int bits)
175 void (*ref)(int16_t *block) = is_idct ? ff_ref_idct : ff_ref_fdct;
178 int64_t err2, ti, ti1, it1, err_sum = 0;
179 int64_t sysErr[64], sysErrMax = 0;
181 int blockSumErrMax = 0, blockSumErr;
183 const int vals=1<<bits;
187 av_lfg_init(&prng, 1);
191 for (i = 0; i < 64; i++)
193 for (it = 0; it < NB_ITS; it++) {
194 init_block(block1, test, is_idct, &prng, vals);
195 permute(block, block1, dct->perm_type);
200 if (!strcmp(dct->name, "IJG-AAN-INT")) {
201 for (i = 0; i < 64; i++) {
202 scale = 8 * (1 << (AANSCALE_BITS + 11)) / ff_aanscales[i];
203 block[i] = (block[i] * scale) >> AANSCALE_BITS;
208 if (!strcmp(dct->name, "PR-SSE2"))
209 for (i = 0; i < 64; i++)
210 block1[i] = av_clip(block1[i], 4-512, 1019-512);
213 for (i = 0; i < 64; i++) {
214 int err = block[i] - block1[i];
220 sysErr[i] += block[i] - block1[i];
222 if (abs(block[i]) > maxout)
223 maxout = abs(block[i]);
225 if (blockSumErrMax < blockSumErr)
226 blockSumErrMax = blockSumErr;
228 for (i = 0; i < 64; i++)
229 sysErrMax = FFMAX(sysErrMax, FFABS(sysErr[i]));
231 for (i = 0; i < 64; i++) {
234 printf("%7d ", (int) sysErr[i]);
238 omse = (double) err2 / NB_ITS / 64;
239 ome = (double) err_sum / NB_ITS / 64;
241 spec_err = is_idct && (err_inf > 1 || omse > 0.02 || fabs(ome) > 0.0015);
243 printf("%s %s: max_err=%d omse=%0.8f ome=%0.8f syserr=%0.8f maxout=%d blockSumErr=%d\n",
244 is_idct ? "IDCT" : "DCT", dct->name, err_inf,
245 omse, ome, (double) sysErrMax / NB_ITS,
246 maxout, blockSumErrMax);
248 if (spec_err && !dct->nonspec)
256 init_block(block, test, is_idct, &prng, vals);
257 permute(block1, block, dct->perm_type);
259 ti = av_gettime_relative();
262 for (it = 0; it < NB_ITS_SPEED; it++) {
263 memcpy(block, block1, sizeof(block));
268 ti1 = av_gettime_relative() - ti;
269 } while (ti1 < 1000000);
271 printf("%s %s: %0.1f kdct/s\n", is_idct ? "IDCT" : "DCT", dct->name,
272 (double) it1 * 1000.0 / (double) ti1);
277 DECLARE_ALIGNED(8, static uint8_t, img_dest)[64];
278 DECLARE_ALIGNED(8, static uint8_t, img_dest1)[64];
280 static void idct248_ref(uint8_t *dest, int linesize, int16_t *block)
283 static double c8[8][8];
284 static double c4[4][4];
285 double block1[64], block2[64], block3[64];
292 for (i = 0; i < 8; i++) {
294 for (j = 0; j < 8; j++) {
295 s = (i == 0) ? sqrt(1.0 / 8.0) : sqrt(1.0 / 4.0);
296 c8[i][j] = s * cos(M_PI * i * (j + 0.5) / 8.0);
297 sum += c8[i][j] * c8[i][j];
301 for (i = 0; i < 4; i++) {
303 for (j = 0; j < 4; j++) {
304 s = (i == 0) ? sqrt(1.0 / 4.0) : sqrt(1.0 / 2.0);
305 c4[i][j] = s * cos(M_PI * i * (j + 0.5) / 4.0);
306 sum += c4[i][j] * c4[i][j];
313 for (i = 0; i < 4; i++) {
314 for (j = 0; j < 8; j++) {
315 block1[8 * (2 * i) + j] =
316 (block[8 * (2 * i) + j] + block[8 * (2 * i + 1) + j]) * s;
317 block1[8 * (2 * i + 1) + j] =
318 (block[8 * (2 * i) + j] - block[8 * (2 * i + 1) + j]) * s;
323 for (i = 0; i < 8; i++) {
324 for (j = 0; j < 8; j++) {
326 for (k = 0; k < 8; k++)
327 sum += c8[k][j] * block1[8 * i + k];
328 block2[8 * i + j] = sum;
333 for (i = 0; i < 8; i++) {
334 for (j = 0; j < 4; j++) {
337 for (k = 0; k < 4; k++)
338 sum += c4[k][j] * block2[8 * (2 * k) + i];
339 block3[8 * (2 * j) + i] = sum;
343 for (k = 0; k < 4; k++)
344 sum += c4[k][j] * block2[8 * (2 * k + 1) + i];
345 block3[8 * (2 * j + 1) + i] = sum;
349 /* clamp and store the result */
350 for (i = 0; i < 8; i++) {
351 for (j = 0; j < 8; j++) {
352 v = block3[8 * i + j];
354 else if (v > 255) v = 255;
355 dest[i * linesize + j] = (int) rint(v);
360 static void idct248_error(const char *name,
361 void (*idct248_put)(uint8_t *dest, int line_size,
365 int it, i, it1, ti, ti1, err_max, v;
368 av_lfg_init(&prng, 1);
370 /* just one test to see if code is correct (precision is less
373 for (it = 0; it < NB_ITS; it++) {
374 /* XXX: use forward transform to generate values */
375 for (i = 0; i < 64; i++)
376 block1[i] = av_lfg_get(&prng) % 256 - 128;
379 for (i = 0; i < 64; i++)
380 block[i] = block1[i];
381 idct248_ref(img_dest1, 8, block);
383 for (i = 0; i < 64; i++)
384 block[i] = block1[i];
385 idct248_put(img_dest, 8, block);
387 for (i = 0; i < 64; i++) {
388 v = abs((int) img_dest[i] - (int) img_dest1[i]);
390 printf("%d %d\n", img_dest[i], img_dest1[i]);
399 printf(" %3d", img_dest1[i*8+j]);
408 printf(" %3d", img_dest[i*8+j]);
414 printf("%s %s: err_inf=%d\n", 1 ? "IDCT248" : "DCT248", name, err_max);
419 ti = av_gettime_relative();
422 for (it = 0; it < NB_ITS_SPEED; it++) {
423 for (i = 0; i < 64; i++)
424 block[i] = block1[i];
425 idct248_put(img_dest, 8, block);
429 ti1 = av_gettime_relative() - ti;
430 } while (ti1 < 1000000);
432 printf("%s %s: %0.1f kdct/s\n", 1 ? "IDCT248" : "DCT248", name,
433 (double) it1 * 1000.0 / (double) ti1);
436 static void help(void)
438 printf("dct-test [-i] [<test-number>] [<bits>]\n"
439 "test-number 0 -> test with random matrixes\n"
440 " 1 -> test with random sparse matrixes\n"
441 " 2 -> do 3. test from mpeg4 std\n"
442 "bits Number of time domain bits to use, 8 is default\n"
443 "-i test IDCT implementations\n"
444 "-4 test IDCT248 implementations\n"
449 #include "compat/getopt.c"
452 int main(int argc, char **argv)
454 int test_idct = 0, test_248_dct = 0;
464 c = getopt(argc, argv, "ih4t");
485 test = atoi(argv[optind]);
486 if(optind+1 < argc) bits= atoi(argv[optind+1]);
488 printf("ffmpeg DCT/IDCT test\n");
491 idct248_error("SIMPLE-C", ff_simple_idct248_put, speed);
493 const int cpu_flags = av_get_cpu_flags();
495 for (i = 0; i < FF_ARRAY_ELEMS(idct_tab); i++)
496 err |= dct_error(&idct_tab[i], test, test_idct, speed, bits);
498 for (i = 0; idct_tab_arch[i].name; i++)
499 if (!(~cpu_flags & idct_tab_arch[i].cpu_flag))
500 err |= dct_error(&idct_tab_arch[i], test, test_idct, speed, bits);
504 for (i = 0; i < FF_ARRAY_ELEMS(fdct_tab); i++)
505 err |= dct_error(&fdct_tab[i], test, test_idct, speed, bits);
507 for (i = 0; fdct_tab_arch[i].name; i++)
508 if (!(~cpu_flags & fdct_tab_arch[i].cpu_flag))
509 err |= dct_error(&fdct_tab_arch[i], test, test_idct, speed, bits);
511 #endif /* CONFIG_FDCTDSP */
515 printf("Error: %d.\n", err);