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/time.h"
43 #include "libavcodec/dct.h"
44 #include "libavcodec/idctdsp.h"
45 #include "libavcodec/simple_idct.h"
46 #include "libavcodec/xvididct.h"
47 #include "libavcodec/aandcttab.h"
48 #include "libavcodec/faandct.h"
49 #include "libavcodec/faanidct.h"
50 #include "libavcodec/dctref.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 void ff_prores_idct_wrap(int16_t *dst){
70 LOCAL_ALIGNED(16, int16_t, qmat, [64]);
76 ff_prores_idct(dst, qmat);
82 static const struct algo idct_tab[] = {
83 { "REF-DBL", ff_ref_idct, FF_IDCT_PERM_NONE },
84 { "INT", ff_j_rev_dct, FF_IDCT_PERM_LIBMPEG2 },
85 { "SIMPLE-C", ff_simple_idct_8, FF_IDCT_PERM_NONE },
86 { "SIMPLE-C10", ff_simple_idct_10, FF_IDCT_PERM_NONE },
87 { "SIMPLE-C12", ff_simple_idct_12, FF_IDCT_PERM_NONE, 0, 1 },
88 { "PR-C", ff_prores_idct_wrap, FF_IDCT_PERM_NONE, 0, 1 },
90 { "FAANI", ff_faanidct, FF_IDCT_PERM_NONE },
91 #endif /* CONFIG_FAANIDCT */
92 #if CONFIG_MPEG4_DECODER
93 { "XVID", ff_xvid_idct, FF_IDCT_PERM_NONE, 0, 1 },
94 #endif /* CONFIG_MPEG4_DECODER */
98 #include "aarch64/dct.c"
106 static const struct algo fdct_tab_arch[] = { { 0 } };
107 static const struct algo idct_tab_arch[] = { { 0 } };
110 #define AANSCALE_BITS 12
113 #define NB_ITS_SPEED 50000
115 DECLARE_ALIGNED(16, static int16_t, block)[64];
116 DECLARE_ALIGNED(8, static int16_t, block1)[64];
118 static void init_block(int16_t block[64], int test, int is_idct, AVLFG *prng, int vals)
122 memset(block, 0, 64 * sizeof(*block));
126 for (i = 0; i < 64; i++)
127 block[i] = (av_lfg_get(prng) % (2*vals)) -vals;
130 for (i = 0; i < 64; i++)
135 j = av_lfg_get(prng) % 10 + 1;
136 for (i = 0; i < j; i++) {
137 int idx = av_lfg_get(prng) % 64;
138 block[idx] = av_lfg_get(prng) % (2*vals) -vals;
142 block[ 0] = av_lfg_get(prng) % (16*vals) - (8*vals);
143 block[63] = (block[0] & 1) ^ 1;
148 static void permute(int16_t dst[64], const int16_t src[64],
149 enum idct_permutation_type perm_type)
154 if (permute_x86(dst, src, perm_type))
159 case FF_IDCT_PERM_LIBMPEG2:
160 for (i = 0; i < 64; i++)
161 dst[(i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2)] = src[i];
163 case FF_IDCT_PERM_PARTTRANS:
164 for (i = 0; i < 64; i++)
165 dst[(i & 0x24) | ((i & 3) << 3) | ((i >> 3) & 3)] = src[i];
167 case FF_IDCT_PERM_TRANSPOSE:
168 for (i = 0; i < 64; i++)
169 dst[(i>>3) | ((i<<3)&0x38)] = src[i];
172 for (i = 0; i < 64; i++)
178 static int dct_error(const struct algo *dct, int test, int is_idct, int speed, const int bits)
180 void (*ref)(int16_t *block) = is_idct ? ff_ref_idct : ff_ref_fdct;
183 int64_t err2, ti, ti1, it1, err_sum = 0;
184 int64_t sysErr[64], sysErrMax = 0;
185 int64_t err2_matrix[64], err2_max = 0;
187 int blockSumErrMax = 0, blockSumErr;
189 const int vals=1<<bits;
193 av_lfg_init(&prng, 1);
197 for (i = 0; i < 64; i++)
198 err2_matrix[i] = sysErr[i] = 0;
199 for (it = 0; it < NB_ITS; it++) {
200 init_block(block1, test, is_idct, &prng, vals);
201 permute(block, block1, dct->perm_type);
206 if (!strcmp(dct->name, "IJG-AAN-INT")) {
207 for (i = 0; i < 64; i++) {
208 scale = 8 * (1 << (AANSCALE_BITS + 11)) / ff_aanscales[i];
209 block[i] = (block[i] * scale) >> AANSCALE_BITS;
214 if (!strcmp(dct->name, "PR-SSE2"))
215 for (i = 0; i < 64; i++)
216 block1[i] = av_clip(block1[i], 4-512, 1019-512);
219 for (i = 0; i < 64; i++) {
220 int err = block[i] - block1[i];
225 err2_matrix[i] += v * v;
227 sysErr[i] += block[i] - block1[i];
229 if (abs(block[i]) > maxout)
230 maxout = abs(block[i]);
232 if (blockSumErrMax < blockSumErr)
233 blockSumErrMax = blockSumErr;
235 for (i = 0; i < 64; i++) {
236 sysErrMax = FFMAX(sysErrMax, FFABS(sysErr[i]));
237 err2_max = FFMAX(err2_max , FFABS(err2_matrix[i]));
240 for (i = 0; i < 64; i++) {
243 printf("%7d ", (int) sysErr[i]);
247 omse = (double) err2 / NB_ITS / 64;
248 ome = (double) err_sum / NB_ITS / 64;
250 spec_err = is_idct && (err_inf > 1 || omse > 0.02 || fabs(ome) > 0.0015);
252 spec_err = is_idct && ((double) err2_max / NB_ITS > 0.06 || (double) sysErrMax / NB_ITS > 0.015);
254 printf("%s %s: max_err=%d omse=%0.8f ome=%0.8f syserr=%0.8f maxout=%d blockSumErr=%d\n",
255 is_idct ? "IDCT" : "DCT", dct->name, err_inf,
256 omse, ome, (double) sysErrMax / NB_ITS,
257 maxout, blockSumErrMax);
259 if (spec_err && !dct->nonspec) {
269 init_block(block, test, is_idct, &prng, vals);
270 permute(block1, block, dct->perm_type);
272 ti = av_gettime_relative();
275 for (it = 0; it < NB_ITS_SPEED; it++) {
276 memcpy(block, block1, sizeof(block));
281 ti1 = av_gettime_relative() - ti;
282 } while (ti1 < 1000000);
284 printf("%s %s: %0.1f kdct/s\n", is_idct ? "IDCT" : "DCT", dct->name,
285 (double) it1 * 1000.0 / (double) ti1);
290 DECLARE_ALIGNED(8, static uint8_t, img_dest)[64];
291 DECLARE_ALIGNED(8, static uint8_t, img_dest1)[64];
293 static void idct248_ref(uint8_t *dest, ptrdiff_t linesize, int16_t *block)
296 static double c8[8][8];
297 static double c4[4][4];
298 double block1[64], block2[64], block3[64];
305 for (i = 0; i < 8; i++) {
307 for (j = 0; j < 8; j++) {
308 s = (i == 0) ? sqrt(1.0 / 8.0) : sqrt(1.0 / 4.0);
309 c8[i][j] = s * cos(M_PI * i * (j + 0.5) / 8.0);
310 sum += c8[i][j] * c8[i][j];
314 for (i = 0; i < 4; i++) {
316 for (j = 0; j < 4; j++) {
317 s = (i == 0) ? sqrt(1.0 / 4.0) : sqrt(1.0 / 2.0);
318 c4[i][j] = s * cos(M_PI * i * (j + 0.5) / 4.0);
319 sum += c4[i][j] * c4[i][j];
326 for (i = 0; i < 4; i++) {
327 for (j = 0; j < 8; j++) {
328 block1[8 * (2 * i) + j] =
329 (block[8 * (2 * i) + j] + block[8 * (2 * i + 1) + j]) * s;
330 block1[8 * (2 * i + 1) + j] =
331 (block[8 * (2 * i) + j] - block[8 * (2 * i + 1) + j]) * s;
336 for (i = 0; i < 8; i++) {
337 for (j = 0; j < 8; j++) {
339 for (k = 0; k < 8; k++)
340 sum += c8[k][j] * block1[8 * i + k];
341 block2[8 * i + j] = sum;
346 for (i = 0; i < 8; i++) {
347 for (j = 0; j < 4; j++) {
350 for (k = 0; k < 4; k++)
351 sum += c4[k][j] * block2[8 * (2 * k) + i];
352 block3[8 * (2 * j) + i] = sum;
356 for (k = 0; k < 4; k++)
357 sum += c4[k][j] * block2[8 * (2 * k + 1) + i];
358 block3[8 * (2 * j + 1) + i] = sum;
362 /* clamp and store the result */
363 for (i = 0; i < 8; i++) {
364 for (j = 0; j < 8; j++) {
365 v = block3[8 * i + j];
367 else if (v > 255) v = 255;
368 dest[i * linesize + j] = (int) rint(v);
373 static void idct248_error(const char *name,
374 void (*idct248_put)(uint8_t *dest,
379 int it, i, it1, ti, ti1, err_max, v;
382 av_lfg_init(&prng, 1);
384 /* just one test to see if code is correct (precision is less
387 for (it = 0; it < NB_ITS; it++) {
388 /* XXX: use forward transform to generate values */
389 for (i = 0; i < 64; i++)
390 block1[i] = av_lfg_get(&prng) % 256 - 128;
393 for (i = 0; i < 64; i++)
394 block[i] = block1[i];
395 idct248_ref(img_dest1, 8, block);
397 for (i = 0; i < 64; i++)
398 block[i] = block1[i];
399 idct248_put(img_dest, 8, block);
401 for (i = 0; i < 64; i++) {
402 v = abs((int) img_dest[i] - (int) img_dest1[i]);
404 printf("%d %d\n", img_dest[i], img_dest1[i]);
413 printf(" %3d", img_dest1[i*8+j]);
422 printf(" %3d", img_dest[i*8+j]);
428 printf("%s %s: err_inf=%d\n", 1 ? "IDCT248" : "DCT248", name, err_max);
433 ti = av_gettime_relative();
436 for (it = 0; it < NB_ITS_SPEED; it++) {
437 for (i = 0; i < 64; i++)
438 block[i] = block1[i];
439 idct248_put(img_dest, 8, block);
443 ti1 = av_gettime_relative() - ti;
444 } while (ti1 < 1000000);
446 printf("%s %s: %0.1f kdct/s\n", 1 ? "IDCT248" : "DCT248", name,
447 (double) it1 * 1000.0 / (double) ti1);
450 static void help(void)
452 printf("dct-test [-i] [<test-number>] [<bits>]\n"
453 "test-number 0 -> test with random matrixes\n"
454 " 1 -> test with random sparse matrixes\n"
455 " 2 -> do 3. test from MPEG-4 std\n"
456 "bits Number of time domain bits to use, 8 is default\n"
457 "-i test IDCT implementations\n"
458 "-4 test IDCT248 implementations\n"
463 #include "compat/getopt.c"
466 int main(int argc, char **argv)
468 int test_idct = 0, test_248_dct = 0;
478 c = getopt(argc, argv, "ih4t");
499 test = atoi(argv[optind]);
500 if(optind+1 < argc) bits= atoi(argv[optind+1]);
502 printf("ffmpeg DCT/IDCT test\n");
505 idct248_error("SIMPLE-C", ff_simple_idct248_put, speed);
507 const int cpu_flags = av_get_cpu_flags();
509 for (i = 0; i < FF_ARRAY_ELEMS(idct_tab); i++)
510 err |= dct_error(&idct_tab[i], test, test_idct, speed, bits);
512 for (i = 0; idct_tab_arch[i].name; i++)
513 if (!(~cpu_flags & idct_tab_arch[i].cpu_flag))
514 err |= dct_error(&idct_tab_arch[i], test, test_idct, speed, bits);
518 for (i = 0; i < FF_ARRAY_ELEMS(fdct_tab); i++)
519 err |= dct_error(&fdct_tab[i], test, test_idct, speed, bits);
521 for (i = 0; fdct_tab_arch[i].name; i++)
522 if (!(~cpu_flags & fdct_tab_arch[i].cpu_flag))
523 err |= dct_error(&fdct_tab_arch[i], test, test_idct, speed, bits);
525 #endif /* CONFIG_FDCTDSP */
529 printf("Error: %d.\n", err);