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 { "SIMPLE-C10", ff_simple_idct_10, FF_IDCT_PERM_NONE },
86 { "SIMPLE-C12", ff_simple_idct_12, FF_IDCT_PERM_NONE, 0, 1 },
87 { "PR-C", ff_prores_idct_wrap, FF_IDCT_PERM_NONE, 0, 1 },
89 { "FAANI", ff_faanidct, FF_IDCT_PERM_NONE },
90 #endif /* CONFIG_FAANIDCT */
91 #if CONFIG_MPEG4_DECODER
92 { "XVID", ff_xvid_idct, FF_IDCT_PERM_NONE, 0, 1 },
93 #endif /* CONFIG_MPEG4_DECODER */
97 #include "arm/dct-test.c"
99 #include "ppc/dct-test.c"
101 #include "x86/dct-test.c"
103 static const struct algo fdct_tab_arch[] = { { 0 } };
104 static const struct algo idct_tab_arch[] = { { 0 } };
107 #define AANSCALE_BITS 12
110 #define NB_ITS_SPEED 50000
112 DECLARE_ALIGNED(16, static int16_t, block)[64];
113 DECLARE_ALIGNED(8, static int16_t, block1)[64];
115 static void init_block(int16_t block[64], int test, int is_idct, AVLFG *prng, int vals)
119 memset(block, 0, 64 * sizeof(*block));
123 for (i = 0; i < 64; i++)
124 block[i] = (av_lfg_get(prng) % (2*vals)) -vals;
127 for (i = 0; i < 64; i++)
132 j = av_lfg_get(prng) % 10 + 1;
133 for (i = 0; i < j; i++) {
134 int idx = av_lfg_get(prng) % 64;
135 block[idx] = av_lfg_get(prng) % (2*vals) -vals;
139 block[ 0] = av_lfg_get(prng) % (16*vals) - (8*vals);
140 block[63] = (block[0] & 1) ^ 1;
145 static void permute(int16_t dst[64], const int16_t src[64],
146 enum idct_permutation_type perm_type)
151 if (permute_x86(dst, src, perm_type))
156 case FF_IDCT_PERM_LIBMPEG2:
157 for (i = 0; i < 64; i++)
158 dst[(i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2)] = src[i];
160 case FF_IDCT_PERM_PARTTRANS:
161 for (i = 0; i < 64; i++)
162 dst[(i & 0x24) | ((i & 3) << 3) | ((i >> 3) & 3)] = src[i];
164 case FF_IDCT_PERM_TRANSPOSE:
165 for (i = 0; i < 64; i++)
166 dst[(i>>3) | ((i<<3)&0x38)] = src[i];
169 for (i = 0; i < 64; i++)
175 static int dct_error(const struct algo *dct, int test, int is_idct, int speed, const int bits)
177 void (*ref)(int16_t *block) = is_idct ? ff_ref_idct : ff_ref_fdct;
180 int64_t err2, ti, ti1, it1, err_sum = 0;
181 int64_t sysErr[64], sysErrMax = 0;
183 int blockSumErrMax = 0, blockSumErr;
185 const int vals=1<<bits;
189 av_lfg_init(&prng, 1);
193 for (i = 0; i < 64; i++)
195 for (it = 0; it < NB_ITS; it++) {
196 init_block(block1, test, is_idct, &prng, vals);
197 permute(block, block1, dct->perm_type);
202 if (!strcmp(dct->name, "IJG-AAN-INT")) {
203 for (i = 0; i < 64; i++) {
204 scale = 8 * (1 << (AANSCALE_BITS + 11)) / ff_aanscales[i];
205 block[i] = (block[i] * scale) >> AANSCALE_BITS;
210 if (!strcmp(dct->name, "PR-SSE2"))
211 for (i = 0; i < 64; i++)
212 block1[i] = av_clip(block1[i], 4-512, 1019-512);
215 for (i = 0; i < 64; i++) {
216 int err = block[i] - block1[i];
222 sysErr[i] += block[i] - block1[i];
224 if (abs(block[i]) > maxout)
225 maxout = abs(block[i]);
227 if (blockSumErrMax < blockSumErr)
228 blockSumErrMax = blockSumErr;
230 for (i = 0; i < 64; i++)
231 sysErrMax = FFMAX(sysErrMax, FFABS(sysErr[i]));
233 for (i = 0; i < 64; i++) {
236 printf("%7d ", (int) sysErr[i]);
240 omse = (double) err2 / NB_ITS / 64;
241 ome = (double) err_sum / NB_ITS / 64;
243 spec_err = is_idct && (err_inf > 1 || omse > 0.02 || fabs(ome) > 0.0015);
245 printf("%s %s: max_err=%d omse=%0.8f ome=%0.8f syserr=%0.8f maxout=%d blockSumErr=%d\n",
246 is_idct ? "IDCT" : "DCT", dct->name, err_inf,
247 omse, ome, (double) sysErrMax / NB_ITS,
248 maxout, blockSumErrMax);
250 if (spec_err && !dct->nonspec) {
260 init_block(block, test, is_idct, &prng, vals);
261 permute(block1, block, dct->perm_type);
263 ti = av_gettime_relative();
266 for (it = 0; it < NB_ITS_SPEED; it++) {
267 memcpy(block, block1, sizeof(block));
272 ti1 = av_gettime_relative() - ti;
273 } while (ti1 < 1000000);
275 printf("%s %s: %0.1f kdct/s\n", is_idct ? "IDCT" : "DCT", dct->name,
276 (double) it1 * 1000.0 / (double) ti1);
281 DECLARE_ALIGNED(8, static uint8_t, img_dest)[64];
282 DECLARE_ALIGNED(8, static uint8_t, img_dest1)[64];
284 static void idct248_ref(uint8_t *dest, int linesize, int16_t *block)
287 static double c8[8][8];
288 static double c4[4][4];
289 double block1[64], block2[64], block3[64];
296 for (i = 0; i < 8; i++) {
298 for (j = 0; j < 8; j++) {
299 s = (i == 0) ? sqrt(1.0 / 8.0) : sqrt(1.0 / 4.0);
300 c8[i][j] = s * cos(M_PI * i * (j + 0.5) / 8.0);
301 sum += c8[i][j] * c8[i][j];
305 for (i = 0; i < 4; i++) {
307 for (j = 0; j < 4; j++) {
308 s = (i == 0) ? sqrt(1.0 / 4.0) : sqrt(1.0 / 2.0);
309 c4[i][j] = s * cos(M_PI * i * (j + 0.5) / 4.0);
310 sum += c4[i][j] * c4[i][j];
317 for (i = 0; i < 4; i++) {
318 for (j = 0; j < 8; j++) {
319 block1[8 * (2 * i) + j] =
320 (block[8 * (2 * i) + j] + block[8 * (2 * i + 1) + j]) * s;
321 block1[8 * (2 * i + 1) + j] =
322 (block[8 * (2 * i) + j] - block[8 * (2 * i + 1) + j]) * s;
327 for (i = 0; i < 8; i++) {
328 for (j = 0; j < 8; j++) {
330 for (k = 0; k < 8; k++)
331 sum += c8[k][j] * block1[8 * i + k];
332 block2[8 * i + j] = sum;
337 for (i = 0; i < 8; i++) {
338 for (j = 0; j < 4; j++) {
341 for (k = 0; k < 4; k++)
342 sum += c4[k][j] * block2[8 * (2 * k) + i];
343 block3[8 * (2 * j) + i] = sum;
347 for (k = 0; k < 4; k++)
348 sum += c4[k][j] * block2[8 * (2 * k + 1) + i];
349 block3[8 * (2 * j + 1) + i] = sum;
353 /* clamp and store the result */
354 for (i = 0; i < 8; i++) {
355 for (j = 0; j < 8; j++) {
356 v = block3[8 * i + j];
358 else if (v > 255) v = 255;
359 dest[i * linesize + j] = (int) rint(v);
364 static void idct248_error(const char *name,
365 void (*idct248_put)(uint8_t *dest, int line_size,
369 int it, i, it1, ti, ti1, err_max, v;
372 av_lfg_init(&prng, 1);
374 /* just one test to see if code is correct (precision is less
377 for (it = 0; it < NB_ITS; it++) {
378 /* XXX: use forward transform to generate values */
379 for (i = 0; i < 64; i++)
380 block1[i] = av_lfg_get(&prng) % 256 - 128;
383 for (i = 0; i < 64; i++)
384 block[i] = block1[i];
385 idct248_ref(img_dest1, 8, block);
387 for (i = 0; i < 64; i++)
388 block[i] = block1[i];
389 idct248_put(img_dest, 8, block);
391 for (i = 0; i < 64; i++) {
392 v = abs((int) img_dest[i] - (int) img_dest1[i]);
394 printf("%d %d\n", img_dest[i], img_dest1[i]);
403 printf(" %3d", img_dest1[i*8+j]);
412 printf(" %3d", img_dest[i*8+j]);
418 printf("%s %s: err_inf=%d\n", 1 ? "IDCT248" : "DCT248", name, err_max);
423 ti = av_gettime_relative();
426 for (it = 0; it < NB_ITS_SPEED; it++) {
427 for (i = 0; i < 64; i++)
428 block[i] = block1[i];
429 idct248_put(img_dest, 8, block);
433 ti1 = av_gettime_relative() - ti;
434 } while (ti1 < 1000000);
436 printf("%s %s: %0.1f kdct/s\n", 1 ? "IDCT248" : "DCT248", name,
437 (double) it1 * 1000.0 / (double) ti1);
440 static void help(void)
442 printf("dct-test [-i] [<test-number>] [<bits>]\n"
443 "test-number 0 -> test with random matrixes\n"
444 " 1 -> test with random sparse matrixes\n"
445 " 2 -> do 3. test from mpeg4 std\n"
446 "bits Number of time domain bits to use, 8 is default\n"
447 "-i test IDCT implementations\n"
448 "-4 test IDCT248 implementations\n"
453 #include "compat/getopt.c"
456 int main(int argc, char **argv)
458 int test_idct = 0, test_248_dct = 0;
468 c = getopt(argc, argv, "ih4t");
489 test = atoi(argv[optind]);
490 if(optind+1 < argc) bits= atoi(argv[optind+1]);
492 printf("ffmpeg DCT/IDCT test\n");
495 idct248_error("SIMPLE-C", ff_simple_idct248_put, speed);
497 const int cpu_flags = av_get_cpu_flags();
499 for (i = 0; i < FF_ARRAY_ELEMS(idct_tab); i++)
500 err |= dct_error(&idct_tab[i], test, test_idct, speed, bits);
502 for (i = 0; idct_tab_arch[i].name; i++)
503 if (!(~cpu_flags & idct_tab_arch[i].cpu_flag))
504 err |= dct_error(&idct_tab_arch[i], test, test_idct, speed, bits);
508 for (i = 0; i < FF_ARRAY_ELEMS(fdct_tab); i++)
509 err |= dct_error(&fdct_tab[i], test, test_idct, speed, bits);
511 for (i = 0; fdct_tab_arch[i].name; i++)
512 if (!(~cpu_flags & fdct_tab_arch[i].cpu_flag))
513 err |= dct_error(&fdct_tab_arch[i], test, test_idct, speed, bits);
515 #endif /* CONFIG_FDCTDSP */
519 printf("Error: %d.\n", err);