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 { "FAAN", ff_faandct, 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 },
66 static void ff_prores_idct_wrap(int16_t *dst){
67 DECLARE_ALIGNED(16, static int16_t, qmat)[64];
73 ff_prores_idct(dst, qmat);
79 static const struct algo idct_tab[] = {
80 { "FAANI", ff_faanidct, FF_IDCT_PERM_NONE },
81 { "REF-DBL", ff_ref_idct, FF_IDCT_PERM_NONE },
82 { "INT", ff_j_rev_dct, FF_IDCT_PERM_LIBMPEG2 },
83 { "SIMPLE-C", ff_simple_idct_8, FF_IDCT_PERM_NONE },
84 { "PR-C", ff_prores_idct_wrap, FF_IDCT_PERM_NONE, 0, 1 },
85 #if CONFIG_MPEG4_DECODER
86 { "XVID", ff_xvid_idct, FF_IDCT_PERM_NONE, 0, 1 },
87 #endif /* CONFIG_MPEG4_DECODER */
91 #include "arm/dct-test.c"
93 #include "ppc/dct-test.c"
95 #include "x86/dct-test.c"
97 static const struct algo fdct_tab_arch[] = { 0 };
98 static const struct algo idct_tab_arch[] = { 0 };
101 #define AANSCALE_BITS 12
104 #define NB_ITS_SPEED 50000
106 DECLARE_ALIGNED(16, static int16_t, block)[64];
107 DECLARE_ALIGNED(8, static int16_t, block1)[64];
109 static void init_block(int16_t block[64], int test, int is_idct, AVLFG *prng, int vals)
113 memset(block, 0, 64 * sizeof(*block));
117 for (i = 0; i < 64; i++)
118 block[i] = (av_lfg_get(prng) % (2*vals)) -vals;
121 for (i = 0; i < 64; i++)
126 j = av_lfg_get(prng) % 10 + 1;
127 for (i = 0; i < j; i++) {
128 int idx = av_lfg_get(prng) % 64;
129 block[idx] = av_lfg_get(prng) % (2*vals) -vals;
133 block[ 0] = av_lfg_get(prng) % (16*vals) - (8*vals);
134 block[63] = (block[0] & 1) ^ 1;
139 static void permute(int16_t dst[64], const int16_t src[64],
140 enum idct_permutation_type perm_type)
145 if (permute_x86(dst, src, perm_type))
150 case FF_IDCT_PERM_LIBMPEG2:
151 for (i = 0; i < 64; i++)
152 dst[(i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2)] = src[i];
154 case FF_IDCT_PERM_PARTTRANS:
155 for (i = 0; i < 64; i++)
156 dst[(i & 0x24) | ((i & 3) << 3) | ((i >> 3) & 3)] = src[i];
158 case FF_IDCT_PERM_TRANSPOSE:
159 for (i = 0; i < 64; i++)
160 dst[(i>>3) | ((i<<3)&0x38)] = src[i];
163 for (i = 0; i < 64; i++)
169 static int dct_error(const struct algo *dct, int test, int is_idct, int speed, const int bits)
171 void (*ref)(int16_t *block) = is_idct ? ff_ref_idct : ff_ref_fdct;
174 int64_t err2, ti, ti1, it1, err_sum = 0;
175 int64_t sysErr[64], sysErrMax = 0;
177 int blockSumErrMax = 0, blockSumErr;
179 const int vals=1<<bits;
183 av_lfg_init(&prng, 1);
187 for (i = 0; i < 64; i++)
189 for (it = 0; it < NB_ITS; it++) {
190 init_block(block1, test, is_idct, &prng, vals);
191 permute(block, block1, dct->perm_type);
196 if (!strcmp(dct->name, "IJG-AAN-INT")) {
197 for (i = 0; i < 64; i++) {
198 scale = 8 * (1 << (AANSCALE_BITS + 11)) / ff_aanscales[i];
199 block[i] = (block[i] * scale) >> AANSCALE_BITS;
204 if (!strcmp(dct->name, "PR-SSE2"))
205 for (i = 0; i < 64; i++)
206 block1[i] = av_clip(block1[i], 4-512, 1019-512);
209 for (i = 0; i < 64; i++) {
210 int err = block[i] - block1[i];
216 sysErr[i] += block[i] - block1[i];
218 if (abs(block[i]) > maxout)
219 maxout = abs(block[i]);
221 if (blockSumErrMax < blockSumErr)
222 blockSumErrMax = blockSumErr;
224 for (i = 0; i < 64; i++)
225 sysErrMax = FFMAX(sysErrMax, FFABS(sysErr[i]));
227 for (i = 0; i < 64; i++) {
230 printf("%7d ", (int) sysErr[i]);
234 omse = (double) err2 / NB_ITS / 64;
235 ome = (double) err_sum / NB_ITS / 64;
237 spec_err = is_idct && (err_inf > 1 || omse > 0.02 || fabs(ome) > 0.0015);
239 printf("%s %s: max_err=%d omse=%0.8f ome=%0.8f syserr=%0.8f maxout=%d blockSumErr=%d\n",
240 is_idct ? "IDCT" : "DCT", dct->name, err_inf,
241 omse, ome, (double) sysErrMax / NB_ITS,
242 maxout, blockSumErrMax);
244 if (spec_err && !dct->nonspec)
252 init_block(block, test, is_idct, &prng, vals);
253 permute(block1, block, dct->perm_type);
255 ti = av_gettime_relative();
258 for (it = 0; it < NB_ITS_SPEED; it++) {
259 memcpy(block, block1, sizeof(block));
264 ti1 = av_gettime_relative() - ti;
265 } while (ti1 < 1000000);
267 printf("%s %s: %0.1f kdct/s\n", is_idct ? "IDCT" : "DCT", dct->name,
268 (double) it1 * 1000.0 / (double) ti1);
273 DECLARE_ALIGNED(8, static uint8_t, img_dest)[64];
274 DECLARE_ALIGNED(8, static uint8_t, img_dest1)[64];
276 static void idct248_ref(uint8_t *dest, int linesize, int16_t *block)
279 static double c8[8][8];
280 static double c4[4][4];
281 double block1[64], block2[64], block3[64];
288 for (i = 0; i < 8; i++) {
290 for (j = 0; j < 8; j++) {
291 s = (i == 0) ? sqrt(1.0 / 8.0) : sqrt(1.0 / 4.0);
292 c8[i][j] = s * cos(M_PI * i * (j + 0.5) / 8.0);
293 sum += c8[i][j] * c8[i][j];
297 for (i = 0; i < 4; i++) {
299 for (j = 0; j < 4; j++) {
300 s = (i == 0) ? sqrt(1.0 / 4.0) : sqrt(1.0 / 2.0);
301 c4[i][j] = s * cos(M_PI * i * (j + 0.5) / 4.0);
302 sum += c4[i][j] * c4[i][j];
309 for (i = 0; i < 4; i++) {
310 for (j = 0; j < 8; j++) {
311 block1[8 * (2 * i) + j] =
312 (block[8 * (2 * i) + j] + block[8 * (2 * i + 1) + j]) * s;
313 block1[8 * (2 * i + 1) + j] =
314 (block[8 * (2 * i) + j] - block[8 * (2 * i + 1) + j]) * s;
319 for (i = 0; i < 8; i++) {
320 for (j = 0; j < 8; j++) {
322 for (k = 0; k < 8; k++)
323 sum += c8[k][j] * block1[8 * i + k];
324 block2[8 * i + j] = sum;
329 for (i = 0; i < 8; i++) {
330 for (j = 0; j < 4; j++) {
333 for (k = 0; k < 4; k++)
334 sum += c4[k][j] * block2[8 * (2 * k) + i];
335 block3[8 * (2 * j) + i] = sum;
339 for (k = 0; k < 4; k++)
340 sum += c4[k][j] * block2[8 * (2 * k + 1) + i];
341 block3[8 * (2 * j + 1) + i] = sum;
345 /* clamp and store the result */
346 for (i = 0; i < 8; i++) {
347 for (j = 0; j < 8; j++) {
348 v = block3[8 * i + j];
350 else if (v > 255) v = 255;
351 dest[i * linesize + j] = (int) rint(v);
356 static void idct248_error(const char *name,
357 void (*idct248_put)(uint8_t *dest, int line_size,
361 int it, i, it1, ti, ti1, err_max, v;
364 av_lfg_init(&prng, 1);
366 /* just one test to see if code is correct (precision is less
369 for (it = 0; it < NB_ITS; it++) {
370 /* XXX: use forward transform to generate values */
371 for (i = 0; i < 64; i++)
372 block1[i] = av_lfg_get(&prng) % 256 - 128;
375 for (i = 0; i < 64; i++)
376 block[i] = block1[i];
377 idct248_ref(img_dest1, 8, block);
379 for (i = 0; i < 64; i++)
380 block[i] = block1[i];
381 idct248_put(img_dest, 8, block);
383 for (i = 0; i < 64; i++) {
384 v = abs((int) img_dest[i] - (int) img_dest1[i]);
386 printf("%d %d\n", img_dest[i], img_dest1[i]);
395 printf(" %3d", img_dest1[i*8+j]);
404 printf(" %3d", img_dest[i*8+j]);
410 printf("%s %s: err_inf=%d\n", 1 ? "IDCT248" : "DCT248", name, err_max);
415 ti = av_gettime_relative();
418 for (it = 0; it < NB_ITS_SPEED; it++) {
419 for (i = 0; i < 64; i++)
420 block[i] = block1[i];
421 idct248_put(img_dest, 8, block);
425 ti1 = av_gettime_relative() - ti;
426 } while (ti1 < 1000000);
428 printf("%s %s: %0.1f kdct/s\n", 1 ? "IDCT248" : "DCT248", name,
429 (double) it1 * 1000.0 / (double) ti1);
432 static void help(void)
434 printf("dct-test [-i] [<test-number>] [<bits>]\n"
435 "test-number 0 -> test with random matrixes\n"
436 " 1 -> test with random sparse matrixes\n"
437 " 2 -> do 3. test from mpeg4 std\n"
438 "bits Number of time domain bits to use, 8 is default\n"
439 "-i test IDCT implementations\n"
440 "-4 test IDCT248 implementations\n"
445 #include "compat/getopt.c"
448 int main(int argc, char **argv)
450 int test_idct = 0, test_248_dct = 0;
460 c = getopt(argc, argv, "ih4t");
481 test = atoi(argv[optind]);
482 if(optind+1 < argc) bits= atoi(argv[optind+1]);
484 printf("ffmpeg DCT/IDCT test\n");
487 idct248_error("SIMPLE-C", ff_simple_idct248_put, speed);
489 const int cpu_flags = av_get_cpu_flags();
491 for (i = 0; i < FF_ARRAY_ELEMS(idct_tab); i++)
492 err |= dct_error(&idct_tab[i], test, test_idct, speed, bits);
494 for (i = 0; idct_tab_arch[i].name; i++)
495 if (!(~cpu_flags & idct_tab_arch[i].cpu_flag))
496 err |= dct_error(&idct_tab_arch[i], test, test_idct, speed, bits);
500 for (i = 0; i < FF_ARRAY_ELEMS(fdct_tab); i++)
501 err |= dct_error(&fdct_tab[i], test, test_idct, speed, bits);
503 for (i = 0; fdct_tab_arch[i].name; i++)
504 if (!(~cpu_flags & fdct_tab_arch[i].cpu_flag))
505 err |= dct_error(&fdct_tab_arch[i], test, test_idct, speed, bits);
507 #endif /* CONFIG_FDCTDSP */
511 printf("Error: %d.\n", err);