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/internal.h"
40 #include "libavutil/lfg.h"
41 #include "libavutil/mem_internal.h"
42 #include "libavutil/time.h"
44 #include "libavcodec/dct.h"
45 #include "libavcodec/idctdsp.h"
46 #include "libavcodec/simple_idct.h"
47 #include "libavcodec/xvididct.h"
48 #include "libavcodec/aandcttab.h"
49 #include "libavcodec/faandct.h"
50 #include "libavcodec/faanidct.h"
51 #include "libavcodec/dctref.h"
55 void (*func)(int16_t *block);
56 enum idct_permutation_type perm_type;
61 static const struct algo fdct_tab[] = {
62 { "REF-DBL", ff_ref_fdct, FF_IDCT_PERM_NONE },
63 { "IJG-AAN-INT", ff_fdct_ifast, FF_IDCT_PERM_NONE },
64 { "IJG-LLM-INT", ff_jpeg_fdct_islow_8, FF_IDCT_PERM_NONE },
66 { "FAAN", ff_faandct, FF_IDCT_PERM_NONE },
67 #endif /* CONFIG_FAANDCT */
70 static void ff_prores_idct_wrap(int16_t *dst){
71 LOCAL_ALIGNED(16, int16_t, qmat, [64]);
77 ff_prores_idct_10(dst, qmat);
83 static const struct algo idct_tab[] = {
84 { "REF-DBL", ff_ref_idct, FF_IDCT_PERM_NONE },
85 { "INT", ff_j_rev_dct, FF_IDCT_PERM_LIBMPEG2 },
86 { "SIMPLE-C", ff_simple_idct_int16_8bit, FF_IDCT_PERM_NONE },
87 { "SIMPLE-C10", ff_simple_idct_int16_10bit, FF_IDCT_PERM_NONE },
88 { "SIMPLE-C12", ff_simple_idct_int16_12bit, FF_IDCT_PERM_NONE, 0, 1 },
89 { "PR-C", ff_prores_idct_wrap, FF_IDCT_PERM_NONE, 0, 1 },
91 { "FAANI", ff_faanidct, FF_IDCT_PERM_NONE },
92 #endif /* CONFIG_FAANIDCT */
93 #if CONFIG_MPEG4_DECODER
94 { "XVID", ff_xvid_idct, FF_IDCT_PERM_NONE, 0, 1 },
95 #endif /* CONFIG_MPEG4_DECODER */
99 #include "aarch64/dct.c"
107 static const struct algo fdct_tab_arch[] = { { 0 } };
108 static const struct algo idct_tab_arch[] = { { 0 } };
111 #define AANSCALE_BITS 12
114 #define NB_ITS_SPEED 50000
116 DECLARE_ALIGNED(16, static int16_t, block)[64];
117 DECLARE_ALIGNED(8, static int16_t, block1)[64];
119 static void init_block(int16_t block[64], int test, int is_idct, AVLFG *prng, int vals)
123 memset(block, 0, 64 * sizeof(*block));
127 for (i = 0; i < 64; i++)
128 block[i] = (av_lfg_get(prng) % (2*vals)) -vals;
131 for (i = 0; i < 64; i++)
136 j = av_lfg_get(prng) % 10 + 1;
137 for (i = 0; i < j; i++) {
138 int idx = av_lfg_get(prng) % 64;
139 block[idx] = av_lfg_get(prng) % (2*vals) -vals;
143 block[ 0] = av_lfg_get(prng) % (16*vals) - (8*vals);
144 block[63] = (block[0] & 1) ^ 1;
149 static void permute(int16_t dst[64], const int16_t src[64],
150 enum idct_permutation_type perm_type)
155 if (permute_x86(dst, src, perm_type))
160 case FF_IDCT_PERM_LIBMPEG2:
161 for (i = 0; i < 64; i++)
162 dst[(i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2)] = src[i];
164 case FF_IDCT_PERM_PARTTRANS:
165 for (i = 0; i < 64; i++)
166 dst[(i & 0x24) | ((i & 3) << 3) | ((i >> 3) & 3)] = src[i];
168 case FF_IDCT_PERM_TRANSPOSE:
169 for (i = 0; i < 64; i++)
170 dst[(i>>3) | ((i<<3)&0x38)] = src[i];
173 for (i = 0; i < 64; i++)
179 static int dct_error(const struct algo *dct, int test, int is_idct, int speed, const int bits)
181 void (*ref)(int16_t *block) = is_idct ? ff_ref_idct : ff_ref_fdct;
184 int64_t err2, ti, ti1, it1, err_sum = 0;
185 int64_t sysErr[64], sysErrMax = 0;
186 int64_t err2_matrix[64], err2_max = 0;
188 int blockSumErrMax = 0, blockSumErr;
190 const int vals=1<<bits;
194 av_lfg_init(&prng, 1);
198 for (i = 0; i < 64; i++)
199 err2_matrix[i] = sysErr[i] = 0;
200 for (it = 0; it < NB_ITS; it++) {
201 init_block(block1, test, is_idct, &prng, vals);
202 permute(block, block1, dct->perm_type);
207 if (!strcmp(dct->name, "IJG-AAN-INT")) {
208 for (i = 0; i < 64; i++) {
209 scale = 8 * (1 << (AANSCALE_BITS + 11)) / ff_aanscales[i];
210 block[i] = (block[i] * scale) >> AANSCALE_BITS;
215 if (!strcmp(dct->name, "PR-SSE2"))
216 for (i = 0; i < 64; i++)
217 block1[i] = av_clip(block1[i], 4-512, 1019-512);
220 for (i = 0; i < 64; i++) {
221 int err = block[i] - block1[i];
226 err2_matrix[i] += v * v;
228 sysErr[i] += block[i] - block1[i];
230 if (abs(block[i]) > maxout)
231 maxout = abs(block[i]);
233 if (blockSumErrMax < blockSumErr)
234 blockSumErrMax = blockSumErr;
236 for (i = 0; i < 64; i++) {
237 sysErrMax = FFMAX(sysErrMax, FFABS(sysErr[i]));
238 err2_max = FFMAX(err2_max , FFABS(err2_matrix[i]));
241 for (i = 0; i < 64; i++) {
244 printf("%7d ", (int) sysErr[i]);
248 omse = (double) err2 / NB_ITS / 64;
249 ome = (double) err_sum / NB_ITS / 64;
251 spec_err = is_idct && (err_inf > 1 || omse > 0.02 || fabs(ome) > 0.0015);
253 spec_err = is_idct && ((double) err2_max / NB_ITS > 0.06 || (double) sysErrMax / NB_ITS > 0.015);
255 printf("%s %s: max_err=%d omse=%0.8f ome=%0.8f syserr=%0.8f maxout=%d blockSumErr=%d\n",
256 is_idct ? "IDCT" : "DCT", dct->name, err_inf,
257 omse, ome, (double) sysErrMax / NB_ITS,
258 maxout, blockSumErrMax);
260 if (spec_err && !dct->nonspec) {
270 init_block(block, test, is_idct, &prng, vals);
271 permute(block1, block, dct->perm_type);
273 ti = av_gettime_relative();
276 for (it = 0; it < NB_ITS_SPEED; it++) {
277 memcpy(block, block1, sizeof(block));
282 ti1 = av_gettime_relative() - ti;
283 } while (ti1 < 1000000);
285 printf("%s %s: %0.1f kdct/s\n", is_idct ? "IDCT" : "DCT", dct->name,
286 (double) it1 * 1000.0 / (double) ti1);
291 DECLARE_ALIGNED(8, static uint8_t, img_dest)[64];
292 DECLARE_ALIGNED(8, static uint8_t, img_dest1)[64];
294 static void idct248_ref(uint8_t *dest, ptrdiff_t linesize, int16_t *block)
297 static double c8[8][8];
298 static double c4[4][4];
299 double block1[64], block2[64], block3[64];
306 for (i = 0; i < 8; i++) {
308 for (j = 0; j < 8; j++) {
309 s = (i == 0) ? sqrt(1.0 / 8.0) : sqrt(1.0 / 4.0);
310 c8[i][j] = s * cos(M_PI * i * (j + 0.5) / 8.0);
311 sum += c8[i][j] * c8[i][j];
315 for (i = 0; i < 4; i++) {
317 for (j = 0; j < 4; j++) {
318 s = (i == 0) ? sqrt(1.0 / 4.0) : sqrt(1.0 / 2.0);
319 c4[i][j] = s * cos(M_PI * i * (j + 0.5) / 4.0);
320 sum += c4[i][j] * c4[i][j];
327 for (i = 0; i < 4; i++) {
328 for (j = 0; j < 8; j++) {
329 block1[8 * (2 * i) + j] =
330 (block[8 * (2 * i) + j] + block[8 * (2 * i + 1) + j]) * s;
331 block1[8 * (2 * i + 1) + j] =
332 (block[8 * (2 * i) + j] - block[8 * (2 * i + 1) + j]) * s;
337 for (i = 0; i < 8; i++) {
338 for (j = 0; j < 8; j++) {
340 for (k = 0; k < 8; k++)
341 sum += c8[k][j] * block1[8 * i + k];
342 block2[8 * i + j] = sum;
347 for (i = 0; i < 8; i++) {
348 for (j = 0; j < 4; j++) {
351 for (k = 0; k < 4; k++)
352 sum += c4[k][j] * block2[8 * (2 * k) + i];
353 block3[8 * (2 * j) + i] = sum;
357 for (k = 0; k < 4; k++)
358 sum += c4[k][j] * block2[8 * (2 * k + 1) + i];
359 block3[8 * (2 * j + 1) + i] = sum;
363 /* clamp and store the result */
364 for (i = 0; i < 8; i++) {
365 for (j = 0; j < 8; j++) {
366 v = block3[8 * i + j];
368 else if (v > 255) v = 255;
369 dest[i * linesize + j] = (int) rint(v);
374 static void idct248_error(const char *name,
375 void (*idct248_put)(uint8_t *dest,
380 int it, i, it1, ti, ti1, err_max, v;
383 av_lfg_init(&prng, 1);
385 /* just one test to see if code is correct (precision is less
388 for (it = 0; it < NB_ITS; it++) {
389 /* XXX: use forward transform to generate values */
390 for (i = 0; i < 64; i++)
391 block1[i] = av_lfg_get(&prng) % 256 - 128;
394 for (i = 0; i < 64; i++)
395 block[i] = block1[i];
396 idct248_ref(img_dest1, 8, block);
398 for (i = 0; i < 64; i++)
399 block[i] = block1[i];
400 idct248_put(img_dest, 8, block);
402 for (i = 0; i < 64; i++) {
403 v = abs((int) img_dest[i] - (int) img_dest1[i]);
405 printf("%d %d\n", img_dest[i], img_dest1[i]);
414 printf(" %3d", img_dest1[i*8+j]);
423 printf(" %3d", img_dest[i*8+j]);
429 printf("%s %s: err_inf=%d\n", 1 ? "IDCT248" : "DCT248", name, err_max);
434 ti = av_gettime_relative();
437 for (it = 0; it < NB_ITS_SPEED; it++) {
438 for (i = 0; i < 64; i++)
439 block[i] = block1[i];
440 idct248_put(img_dest, 8, block);
444 ti1 = av_gettime_relative() - ti;
445 } while (ti1 < 1000000);
447 printf("%s %s: %0.1f kdct/s\n", 1 ? "IDCT248" : "DCT248", name,
448 (double) it1 * 1000.0 / (double) ti1);
451 static void help(void)
453 printf("dct-test [-i] [<test-number>] [<bits>]\n"
454 "test-number 0 -> test with random matrixes\n"
455 " 1 -> test with random sparse matrixes\n"
456 " 2 -> do 3. test from MPEG-4 std\n"
457 "bits Number of time domain bits to use, 8 is default\n"
458 "-i test IDCT implementations\n"
459 "-4 test IDCT248 implementations\n"
464 #include "compat/getopt.c"
467 int main(int argc, char **argv)
469 int test_idct = 0, test_248_dct = 0;
479 c = getopt(argc, argv, "ih4t");
500 test = atoi(argv[optind]);
501 if(optind+1 < argc) bits= atoi(argv[optind+1]);
503 printf("ffmpeg DCT/IDCT test\n");
506 idct248_error("SIMPLE-C", ff_simple_idct248_put, speed);
508 const int cpu_flags = av_get_cpu_flags();
510 for (i = 0; i < FF_ARRAY_ELEMS(idct_tab); i++)
511 err |= dct_error(&idct_tab[i], test, test_idct, speed, bits);
513 for (i = 0; idct_tab_arch[i].name; i++)
514 if (!(~cpu_flags & idct_tab_arch[i].cpu_flag))
515 err |= dct_error(&idct_tab_arch[i], test, test_idct, speed, bits);
519 for (i = 0; i < FF_ARRAY_ELEMS(fdct_tab); i++)
520 err |= dct_error(&fdct_tab[i], test, test_idct, speed, bits);
522 for (i = 0; fdct_tab_arch[i].name; i++)
523 if (!(~cpu_flags & fdct_tab_arch[i].cpu_flag))
524 err |= dct_error(&fdct_tab_arch[i], test, test_idct, speed, bits);
526 #endif /* CONFIG_FDCTDSP */
530 printf("Error: %d.\n", err);