dct.c 12.8 KB
Newer Older
1 2
/*
 * (c) 2001 Fabrice Bellard
3
 *     2007 Marc Hoffman <marc.hoffman@analog.com>
4
 *
5
 * This file is part of Libav.
6
 *
7
 * Libav is free software; you can redistribute it and/or
8 9
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
10
 * version 2.1 of the License, or (at your option) any later version.
11
 *
12
 * Libav is distributed in the hope that it will be useful,
13 14 15 16 17
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
18
 * License along with Libav; if not, write to the Free Software
19 20 21
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

Michael Niedermayer's avatar
Michael Niedermayer committed
22
/**
23
 * @file
24
 * DCT test (c) 2001 Fabrice Bellard
Michael Niedermayer's avatar
Michael Niedermayer committed
25 26 27
 * Started from sample code by Juan J. Sierralta P.
 */

28
#include "config.h"
Fabrice Bellard's avatar
Fabrice Bellard committed
29 30 31
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
32
#if HAVE_UNISTD_H
Fabrice Bellard's avatar
Fabrice Bellard committed
33
#include <unistd.h>
34
#endif
35
#include <math.h>
Fabrice Bellard's avatar
Fabrice Bellard committed
36

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"
Fabrice Bellard's avatar
Fabrice Bellard committed
42

43 44 45 46 47 48 49 50
#include "libavcodec/aandcttab.h"
#include "libavcodec/dct.h"
#include "libavcodec/dctref.h"
#include "libavcodec/faandct.h"
#include "libavcodec/faanidct.h"
#include "libavcodec/idctdsp.h"
#include "libavcodec/simple_idct.h"
#include "libavcodec/xvididct.h"
51

52
struct algo {
53
    const char *name;
Diego Biurrun's avatar
Diego Biurrun committed
54
    void (*func)(int16_t *block);
55
    enum idct_permutation_type perm_type;
56
    int cpu_flag;
57
    int nonspec;
58 59
};

60
static const struct algo fdct_tab[] = {
61 62 63
    { "REF-DBL",     ff_ref_fdct,          FF_IDCT_PERM_NONE },
    { "IJG-AAN-INT", ff_fdct_ifast,        FF_IDCT_PERM_NONE },
    { "IJG-LLM-INT", ff_jpeg_fdct_islow_8, FF_IDCT_PERM_NONE },
64 65 66
#if CONFIG_FAANDCT
    { "FAAN",        ff_faandct,           FF_IDCT_PERM_NONE },
#endif /* CONFIG_FAANDCT */
67 68
};

Pascal Massimino's avatar
Pascal Massimino committed
69
static const struct algo idct_tab[] = {
70 71 72
    { "REF-DBL",     ff_ref_idct,          FF_IDCT_PERM_NONE },
    { "INT",         ff_j_rev_dct,         FF_IDCT_PERM_LIBMPEG2 },
    { "SIMPLE-C",    ff_simple_idct_8,     FF_IDCT_PERM_NONE },
73 74 75
#if CONFIG_FAANIDCT
    { "FAANI",       ff_faanidct,          FF_IDCT_PERM_NONE },
#endif /* CONFIG_FAANIDCT */
Pascal Massimino's avatar
Pascal Massimino committed
76 77 78
#if CONFIG_MPEG4_DECODER
    { "XVID",        ff_xvid_idct,         FF_IDCT_PERM_NONE, 0, 1 },
#endif /* CONFIG_MPEG4_DECODER */
79
};
80

81
#if ARCH_ARM
82
#include "arm/dct.c"
83
#elif ARCH_PPC
84
#include "ppc/dct.c"
85
#elif ARCH_X86
86
#include "x86/dct.c"
87
#else
88 89
static const struct algo fdct_tab_arch[] = { { 0 } };
static const struct algo idct_tab_arch[] = { { 0 } };
90 91
#endif

Fabrice Bellard's avatar
Fabrice Bellard committed
92 93 94 95 96
#define AANSCALE_BITS 12

#define NB_ITS 20000
#define NB_ITS_SPEED 50000

Diego Biurrun's avatar
Diego Biurrun committed
97 98
DECLARE_ALIGNED(16, static int16_t, block)[64];
DECLARE_ALIGNED(8,  static int16_t, block1)[64];
99

Diego Biurrun's avatar
Diego Biurrun committed
100
static void init_block(int16_t block[64], int test, int is_idct, AVLFG *prng)
101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127
{
    int i, j;

    memset(block, 0, 64 * sizeof(*block));

    switch (test) {
    case 0:
        for (i = 0; i < 64; i++)
            block[i] = (av_lfg_get(prng) % 512) - 256;
        if (is_idct) {
            ff_ref_fdct(block);
            for (i = 0; i < 64; i++)
                block[i] >>= 3;
        }
        break;
    case 1:
        j = av_lfg_get(prng) % 10 + 1;
        for (i = 0; i < j; i++)
            block[av_lfg_get(prng) % 64] = av_lfg_get(prng) % 512 - 256;
        break;
    case 2:
        block[ 0] = av_lfg_get(prng) % 4096 - 2048;
        block[63] = (block[0] & 1) ^ 1;
        break;
    }
}

128 129
static void permute(int16_t dst[64], const int16_t src[64],
                    enum idct_permutation_type perm_type)
130 131 132
{
    int i;

133 134 135 136 137
#if ARCH_X86
    if (permute_x86(dst, src, perm_type))
        return;
#endif

138 139
    switch (perm_type) {
    case FF_IDCT_PERM_LIBMPEG2:
140
        for (i = 0; i < 64; i++)
141
            dst[(i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2)] = src[i];
142 143
        break;
    case FF_IDCT_PERM_PARTTRANS:
144 145
        for (i = 0; i < 64; i++)
            dst[(i & 0x24) | ((i & 3) << 3) | ((i >> 3) & 3)] = src[i];
146 147
        break;
    default:
148 149
        for (i = 0; i < 64; i++)
            dst[i] = src[i];
150
        break;
151 152 153
    }
}

154
static int dct_error(const struct algo *dct, int test, int is_idct, int speed)
Fabrice Bellard's avatar
Fabrice Bellard committed
155
{
Diego Biurrun's avatar
Diego Biurrun committed
156
    void (*ref)(int16_t *block) = is_idct ? ff_ref_idct : ff_ref_fdct;
Fabrice Bellard's avatar
Fabrice Bellard committed
157 158
    int it, i, scale;
    int err_inf, v;
159
    int64_t err2, ti, ti1, it1, err_sum = 0;
160 161 162
    int64_t sysErr[64], sysErrMax = 0;
    int maxout = 0;
    int blockSumErrMax = 0, blockSumErr;
163
    AVLFG prng;
164 165
    double omse, ome;
    int spec_err;
Fabrice Bellard's avatar
Fabrice Bellard committed
166

167
    av_lfg_init(&prng, 1);
Fabrice Bellard's avatar
Fabrice Bellard committed
168 169 170

    err_inf = 0;
    err2 = 0;
171 172 173
    for (i = 0; i < 64; i++)
        sysErr[i] = 0;
    for (it = 0; it < NB_ITS; it++) {
174
        init_block(block1, test, is_idct, &prng);
175
        permute(block, block1, dct->perm_type);
176

177
        dct->func(block);
178
        emms_c();
179

180
        if (!strcmp(dct->name, "IJG-AAN-INT")) {
181 182 183
            for (i = 0; i < 64; i++) {
                scale = 8 * (1 << (AANSCALE_BITS + 11)) / ff_aanscales[i];
                block[i] = (block[i] * scale) >> AANSCALE_BITS;
Michael Niedermayer's avatar
Michael Niedermayer committed
184 185 186
            }
        }

187
        ref(block1);
Fabrice Bellard's avatar
Fabrice Bellard committed
188

189 190
        blockSumErr = 0;
        for (i = 0; i < 64; i++) {
191 192 193
            int err = block[i] - block1[i];
            err_sum += err;
            v = abs(err);
Fabrice Bellard's avatar
Fabrice Bellard committed
194 195 196
            if (v > err_inf)
                err_inf = v;
            err2 += v * v;
197 198
            sysErr[i] += block[i] - block1[i];
            blockSumErr += v;
199 200
            if (abs(block[i]) > maxout)
                maxout = abs(block[i]);
Fabrice Bellard's avatar
Fabrice Bellard committed
201
        }
202 203
        if (blockSumErrMax < blockSumErr)
            blockSumErrMax = blockSumErr;
Michael Niedermayer's avatar
Michael Niedermayer committed
204
    }
205 206
    for (i = 0; i < 64; i++)
        sysErrMax = FFMAX(sysErrMax, FFABS(sysErr[i]));
207

208 209 210 211
    for (i = 0; i < 64; i++) {
        if (i % 8 == 0)
            printf("\n");
        printf("%7d ", (int) sysErr[i]);
Fabrice Bellard's avatar
Fabrice Bellard committed
212
    }
Michael Niedermayer's avatar
Michael Niedermayer committed
213
    printf("\n");
214

215 216 217 218 219 220
    omse = (double) err2 / NB_ITS / 64;
    ome  = (double) err_sum / NB_ITS / 64;

    spec_err = is_idct && (err_inf > 1 || omse > 0.02 || fabs(ome) > 0.0015);

    printf("%s %s: ppe=%d omse=%0.8f ome=%0.8f syserr=%0.8f maxout=%d blockSumErr=%d\n",
221
           is_idct ? "IDCT" : "DCT", dct->name, err_inf,
222
           omse, ome, (double) sysErrMax / NB_ITS,
223
           maxout, blockSumErrMax);
224

225 226 227
    if (spec_err && !dct->nonspec)
        return 1;

228
    if (!speed)
229
        return 0;
230

Fabrice Bellard's avatar
Fabrice Bellard committed
231
    /* speed test */
232
    init_block(block, test, is_idct, &prng);
233
    permute(block1, block, dct->perm_type);
234

235
    ti = av_gettime_relative();
Fabrice Bellard's avatar
Fabrice Bellard committed
236 237
    it1 = 0;
    do {
238
        for (it = 0; it < NB_ITS_SPEED; it++) {
239
            memcpy(block, block1, sizeof(block));
240
            dct->func(block);
Fabrice Bellard's avatar
Fabrice Bellard committed
241 242
        }
        it1 += NB_ITS_SPEED;
243
        ti1 = av_gettime_relative() - ti;
Fabrice Bellard's avatar
Fabrice Bellard committed
244
    } while (ti1 < 1000000);
245
    emms_c();
Fabrice Bellard's avatar
Fabrice Bellard committed
246

247
    printf("%s %s: %0.1f kdct/s\n", is_idct ? "IDCT" : "DCT", dct->name,
248
           (double) it1 * 1000.0 / (double) ti1);
249 250

    return 0;
Fabrice Bellard's avatar
Fabrice Bellard committed
251 252
}

253 254
DECLARE_ALIGNED(8, static uint8_t, img_dest)[64];
DECLARE_ALIGNED(8, static uint8_t, img_dest1)[64];
Fabrice Bellard's avatar
Fabrice Bellard committed
255

256
static void idct248_ref(uint8_t *dest, int linesize, int16_t *block)
Fabrice Bellard's avatar
Fabrice Bellard committed
257 258 259 260 261 262 263 264 265 266 267
{
    static int init;
    static double c8[8][8];
    static double c4[4][4];
    double block1[64], block2[64], block3[64];
    double s, sum, v;
    int i, j, k;

    if (!init) {
        init = 1;

268
        for (i = 0; i < 8; i++) {
Fabrice Bellard's avatar
Fabrice Bellard committed
269
            sum = 0;
270 271
            for (j = 0; j < 8; j++) {
                s = (i == 0) ? sqrt(1.0 / 8.0) : sqrt(1.0 / 4.0);
Fabrice Bellard's avatar
Fabrice Bellard committed
272 273 274 275
                c8[i][j] = s * cos(M_PI * i * (j + 0.5) / 8.0);
                sum += c8[i][j] * c8[i][j];
            }
        }
276

277
        for (i = 0; i < 4; i++) {
Fabrice Bellard's avatar
Fabrice Bellard committed
278
            sum = 0;
279 280
            for (j = 0; j < 4; j++) {
                s = (i == 0) ? sqrt(1.0 / 4.0) : sqrt(1.0 / 2.0);
Fabrice Bellard's avatar
Fabrice Bellard committed
281 282 283 284 285 286 287
                c4[i][j] = s * cos(M_PI * i * (j + 0.5) / 4.0);
                sum += c4[i][j] * c4[i][j];
            }
        }
    }

    /* butterfly */
288
    s = 0.5 * sqrt(2.0);
289 290 291 292 293 294
    for (i = 0; i < 4; i++) {
        for (j = 0; j < 8; j++) {
            block1[8 * (2 * i) + j] =
                (block[8 * (2 * i) + j] + block[8 * (2 * i + 1) + j]) * s;
            block1[8 * (2 * i + 1) + j] =
                (block[8 * (2 * i) + j] - block[8 * (2 * i + 1) + j]) * s;
Fabrice Bellard's avatar
Fabrice Bellard committed
295 296 297 298
        }
    }

    /* idct8 on lines */
299 300
    for (i = 0; i < 8; i++) {
        for (j = 0; j < 8; j++) {
Fabrice Bellard's avatar
Fabrice Bellard committed
301
            sum = 0;
302 303 304
            for (k = 0; k < 8; k++)
                sum += c8[k][j] * block1[8 * i + k];
            block2[8 * i + j] = sum;
Fabrice Bellard's avatar
Fabrice Bellard committed
305 306 307 308
        }
    }

    /* idct4 */
309 310
    for (i = 0; i < 8; i++) {
        for (j = 0; j < 4; j++) {
Fabrice Bellard's avatar
Fabrice Bellard committed
311 312
            /* top */
            sum = 0;
313 314 315
            for (k = 0; k < 4; k++)
                sum += c4[k][j] * block2[8 * (2 * k) + i];
            block3[8 * (2 * j) + i] = sum;
Fabrice Bellard's avatar
Fabrice Bellard committed
316 317 318

            /* bottom */
            sum = 0;
319 320 321
            for (k = 0; k < 4; k++)
                sum += c4[k][j] * block2[8 * (2 * k + 1) + i];
            block3[8 * (2 * j + 1) + i] = sum;
Fabrice Bellard's avatar
Fabrice Bellard committed
322 323 324 325
        }
    }

    /* clamp and store the result */
326 327 328 329 330 331
    for (i = 0; i < 8; i++) {
        for (j = 0; j < 8; j++) {
            v = block3[8 * i + j];
            if      (v < 0)   v = 0;
            else if (v > 255) v = 255;
            dest[i * linesize + j] = (int) rint(v);
Fabrice Bellard's avatar
Fabrice Bellard committed
332 333 334 335
        }
    }
}

336
static void idct248_error(const char *name,
337
                          void (*idct248_put)(uint8_t *dest, int line_size,
338 339
                                              int16_t *block),
                          int speed)
Fabrice Bellard's avatar
Fabrice Bellard committed
340 341
{
    int it, i, it1, ti, ti1, err_max, v;
342
    AVLFG prng;
343

344
    av_lfg_init(&prng, 1);
345

Fabrice Bellard's avatar
Fabrice Bellard committed
346 347 348
    /* just one test to see if code is correct (precision is less
       important here) */
    err_max = 0;
349
    for (it = 0; it < NB_ITS; it++) {
350
        /* XXX: use forward transform to generate values */
351
        for (i = 0; i < 64; i++)
352
            block1[i] = av_lfg_get(&prng) % 256 - 128;
353 354
        block1[0] += 1024;

355 356
        for (i = 0; i < 64; i++)
            block[i] = block1[i];
Fabrice Bellard's avatar
Fabrice Bellard committed
357
        idct248_ref(img_dest1, 8, block);
358

359 360
        for (i = 0; i < 64; i++)
            block[i] = block1[i];
361
        idct248_put(img_dest, 8, block);
362

363 364
        for (i = 0; i < 64; i++) {
            v = abs((int) img_dest[i] - (int) img_dest1[i]);
365 366 367 368 369
            if (v == 255)
                printf("%d %d\n", img_dest[i], img_dest1[i]);
            if (v > err_max)
                err_max = v;
        }
Fabrice Bellard's avatar
Fabrice Bellard committed
370
    }
371
    printf("%s %s: err_inf=%d\n", 1 ? "IDCT248" : "DCT248", name, err_max);
Fabrice Bellard's avatar
Fabrice Bellard committed
372

373 374 375
    if (!speed)
        return;

376
    ti = av_gettime_relative();
Fabrice Bellard's avatar
Fabrice Bellard committed
377 378
    it1 = 0;
    do {
379 380 381
        for (it = 0; it < NB_ITS_SPEED; it++) {
            for (i = 0; i < 64; i++)
                block[i] = block1[i];
Fabrice Bellard's avatar
Fabrice Bellard committed
382 383 384
            idct248_put(img_dest, 8, block);
        }
        it1 += NB_ITS_SPEED;
385
        ti1 = av_gettime_relative() - ti;
Fabrice Bellard's avatar
Fabrice Bellard committed
386
    } while (ti1 < 1000000);
387
    emms_c();
Fabrice Bellard's avatar
Fabrice Bellard committed
388

389 390
    printf("%s %s: %0.1f kdct/s\n", 1 ? "IDCT248" : "DCT248", name,
           (double) it1 * 1000.0 / (double) ti1);
Fabrice Bellard's avatar
Fabrice Bellard committed
391 392
}

393
static void help(void)
394
{
Michael Niedermayer's avatar
Michael Niedermayer committed
395 396 397
    printf("dct-test [-i] [<test-number>]\n"
           "test-number 0 -> test with random matrixes\n"
           "            1 -> test with random sparse matrixes\n"
398
           "            2 -> do 3. test from MPEG-4 std\n"
Fabrice Bellard's avatar
Fabrice Bellard committed
399
           "-i          test IDCT implementations\n"
400 401
           "-4          test IDCT248 implementations\n"
           "-t          speed test\n");
402 403
}

404 405 406 407
#if !HAVE_GETOPT
#include "compat/getopt.c"
#endif

Fabrice Bellard's avatar
Fabrice Bellard committed
408 409
int main(int argc, char **argv)
{
Fabrice Bellard's avatar
Fabrice Bellard committed
410
    int test_idct = 0, test_248_dct = 0;
411 412
    int c, i;
    int test = 1;
413
    int speed = 0;
414
    int err = 0;
415

416
    ff_ref_dct_init();
417

418
    for (;;) {
419
        c = getopt(argc, argv, "ih4t");
420 421
        if (c == -1)
            break;
422
        switch (c) {
423 424 425
        case 'i':
            test_idct = 1;
            break;
Fabrice Bellard's avatar
Fabrice Bellard committed
426 427 428
        case '4':
            test_248_dct = 1;
            break;
429 430 431
        case 't':
            speed = 1;
            break;
432
        default:
433 434
        case 'h':
            help();
435
            return 0;
436 437
        }
    }
438

439 440
    if (optind < argc)
        test = atoi(argv[optind]);
441

442
    printf("Libav DCT/IDCT test\n");
443

Fabrice Bellard's avatar
Fabrice Bellard committed
444
    if (test_248_dct) {
445
        idct248_error("SIMPLE-C", ff_simple_idct248_put, speed);
446
    } else {
447
        const int cpu_flags = av_get_cpu_flags();
448 449 450 451 452 453 454
        if (test_idct) {
            for (i = 0; i < FF_ARRAY_ELEMS(idct_tab); i++)
                err |= dct_error(&idct_tab[i], test, test_idct, speed);

            for (i = 0; idct_tab_arch[i].name; i++)
                if (!(~cpu_flags & idct_tab_arch[i].cpu_flag))
                    err |= dct_error(&idct_tab_arch[i], test, test_idct, speed);
455 456 457
        }
#if CONFIG_FDCTDSP
        else {
458 459 460 461 462 463 464
            for (i = 0; i < FF_ARRAY_ELEMS(fdct_tab); i++)
                err |= dct_error(&fdct_tab[i], test, test_idct, speed);

            for (i = 0; fdct_tab_arch[i].name; i++)
                if (!(~cpu_flags & fdct_tab_arch[i].cpu_flag))
                    err |= dct_error(&fdct_tab_arch[i], test, test_idct, speed);
        }
465
#endif /* CONFIG_FDCTDSP */
466
    }
467

468 469 470 471
    if (err)
        printf("Error: %d.\n", err);

    return !!err;
Fabrice Bellard's avatar
Fabrice Bellard committed
472
}