dct-test.c 16.1 KB
Newer Older
1 2
/*
 * (c) 2001 Fabrice Bellard
3
 *     2007 Marc Hoffman <marc.hoffman@analog.com>
4
 *
5 6 7
 * This file is part of FFmpeg.
 *
 * FFmpeg 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
 * FFmpeg 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 FFmpeg; 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.
 */

Fabrice Bellard's avatar
Fabrice Bellard committed
28 29 30 31 32
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <sys/time.h>
#include <unistd.h>
33
#include <math.h>
Fabrice Bellard's avatar
Fabrice Bellard committed
34

35
#include "libavutil/cpu.h"
36
#include "libavutil/common.h"
37
#include "libavutil/lfg.h"
Fabrice Bellard's avatar
Fabrice Bellard committed
38

Michael Niedermayer's avatar
Michael Niedermayer committed
39
#include "simple_idct.h"
40
#include "aandcttab.h"
41
#include "faandct.h"
42
#include "faanidct.h"
43
#include "x86/idct_xvid.h"
44
#include "dctref.h"
45

46 47
#undef printf

48 49
void ff_mmx_idct(DCTELEM *data);
void ff_mmxext_idct(DCTELEM *data);
50

51
void odivx_idct_c(short *block);
Michael Niedermayer's avatar
Michael Niedermayer committed
52

53
// BFIN
54 55
void ff_bfin_idct(DCTELEM *block);
void ff_bfin_fdct(DCTELEM *block);
56 57

// ALTIVEC
58 59
void fdct_altivec(DCTELEM *block);
//void idct_altivec(DCTELEM *block);?? no routine
60

61
// ARM
Måns Rullgård's avatar
Måns Rullgård committed
62 63 64
void ff_j_rev_dct_arm(DCTELEM *data);
void ff_simple_idct_arm(DCTELEM *data);
void ff_simple_idct_armv5te(DCTELEM *data);
65 66
void ff_simple_idct_armv6(DCTELEM *data);
void ff_simple_idct_neon(DCTELEM *data);
67

68 69
void ff_simple_idct_axp(DCTELEM *data);

70
struct algo {
71 72 73
    const char *name;
    void (*func)(DCTELEM *block);
    enum formattag { NO_PERM, MMX_PERM, MMX_SIMPLE_PERM, SCALE_PERM,
74
                     SSE2_PERM, PARTTRANS_PERM, TRANSPOSE_PERM } format;
75
    int mm_support;
76
    int nonspec;
77 78 79 80 81 82 83 84
};

#ifndef FAAN_POSTSCALE
#define FAAN_SCALE SCALE_PERM
#else
#define FAAN_SCALE NO_PERM
#endif

85 86
static int cpu_flags;

87
static const struct algo fdct_tab[] = {
88 89 90
    { "REF-DBL",        ff_ref_fdct,           NO_PERM    },
    { "FAAN",           ff_faandct,            FAAN_SCALE },
    { "IJG-AAN-INT",    fdct_ifast,            SCALE_PERM },
91
    { "IJG-LLM-INT",    ff_jpeg_fdct_islow_8,  NO_PERM    },
92

93
#if HAVE_MMX
94 95 96
    { "MMX",            ff_fdct_mmx,           NO_PERM,   AV_CPU_FLAG_MMX     },
    { "MMX2",           ff_fdct_mmx2,          NO_PERM,   AV_CPU_FLAG_MMX2    },
    { "SSE2",           ff_fdct_sse2,          NO_PERM,   AV_CPU_FLAG_SSE2    },
97
#endif
98

99
#if HAVE_ALTIVEC
100
    { "altivecfdct",    fdct_altivec,          NO_PERM,   AV_CPU_FLAG_ALTIVEC },
101
#endif
102 103

#if ARCH_BFIN
104
    { "BFINfdct",       ff_bfin_fdct,          NO_PERM  },
105 106
#endif

107 108 109
    { 0 }
};

110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125
#if HAVE_MMX
void ff_prores_idct_put_10_sse2(uint16_t *dst, int linesize,
                                DCTELEM *block, int16_t *qmat);

static void ff_prores_idct_put_10_sse2_wrap(uint16_t *dst){
    int16_t qmat[64]; int i;
    int16_t tmp[64];

    for(i=0; i<64; i++){
        qmat[i]=4;
        tmp[i]= dst[i];
    }
    ff_prores_idct_put_10_sse2(dst, 16, tmp, qmat);
}
#endif

126
static const struct algo idct_tab[] = {
127 128 129
    { "FAANI",          ff_faanidct,           NO_PERM  },
    { "REF-DBL",        ff_ref_idct,           NO_PERM  },
    { "INT",            j_rev_dct,             MMX_PERM },
130
    { "SIMPLE-C",       ff_simple_idct_8,      NO_PERM  },
131 132

#if HAVE_MMX
133
#if CONFIG_GPL
134 135
    { "LIBMPEG2-MMX",   ff_mmx_idct,           MMX_PERM,  AV_CPU_FLAG_MMX,  1 },
    { "LIBMPEG2-MMX2",  ff_mmxext_idct,        MMX_PERM,  AV_CPU_FLAG_MMX2, 1 },
136
#endif
137 138 139 140
    { "SIMPLE-MMX",     ff_simple_idct_mmx,  MMX_SIMPLE_PERM, AV_CPU_FLAG_MMX },
    { "XVID-MMX",       ff_idct_xvid_mmx,      NO_PERM,   AV_CPU_FLAG_MMX,  1 },
    { "XVID-MMX2",      ff_idct_xvid_mmx2,     NO_PERM,   AV_CPU_FLAG_MMX2, 1 },
    { "XVID-SSE2",      ff_idct_xvid_sse2,     SSE2_PERM, AV_CPU_FLAG_SSE2, 1 },
141
#if ARCH_X86_64
142
    { "PR-SSE2",        ff_prores_idct_put_10_sse2_wrap,     TRANSPOSE_PERM, AV_CPU_FLAG_SSE2, 1 },
143
#endif
144
#endif
145

146
#if ARCH_BFIN
147
    { "BFINidct",       ff_bfin_idct,          NO_PERM  },
148 149
#endif

150
#if ARCH_ARM
151 152
    { "SIMPLE-ARM",     ff_simple_idct_arm,    NO_PERM  },
    { "INT-ARM",        ff_j_rev_dct_arm,      MMX_PERM },
153
#endif
154
#if HAVE_ARMV5TE
155
    { "SIMPLE-ARMV5TE", ff_simple_idct_armv5te,NO_PERM  },
156
#endif
157
#if HAVE_ARMV6
158
    { "SIMPLE-ARMV6",   ff_simple_idct_armv6,  MMX_PERM },
159
#endif
160
#if HAVE_NEON
161
    { "SIMPLE-NEON",    ff_simple_idct_neon,   PARTTRANS_PERM },
162 163
#endif

164
#if ARCH_ALPHA
165
    { "SIMPLE-ALPHA",   ff_simple_idct_axp,    NO_PERM },
166 167
#endif

168
    { 0 }
169 170
};

Fabrice Bellard's avatar
Fabrice Bellard committed
171 172
#define AANSCALE_BITS 12

173
static int64_t gettime(void)
Fabrice Bellard's avatar
Fabrice Bellard committed
174 175
{
    struct timeval tv;
176
    gettimeofday(&tv, NULL);
177
    return (int64_t)tv.tv_sec * 1000000 + tv.tv_usec;
Fabrice Bellard's avatar
Fabrice Bellard committed
178 179 180 181 182
}

#define NB_ITS 20000
#define NB_ITS_SPEED 50000

183 184
static short idct_mmx_perm[64];

185 186 187 188 189 190 191 192 193
static short idct_simple_mmx_perm[64] = {
    0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D,
    0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D,
    0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D,
    0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F,
    0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F,
    0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D,
    0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F,
    0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F,
Michael Niedermayer's avatar
Michael Niedermayer committed
194 195
};

196
static const uint8_t idct_sse2_row_perm[8] = { 0, 4, 1, 5, 2, 6, 3, 7 };
197

198
static void idct_mmx_init(void)
199 200 201 202 203
{
    int i;

    /* the mmx/mmxext idct uses a reordered input, so we patch scan tables */
    for (i = 0; i < 64; i++) {
204
        idct_mmx_perm[i] = (i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2);
205 206 207
    }
}

208
DECLARE_ALIGNED(16, static DCTELEM, block)[64];
209
DECLARE_ALIGNED(8,  static DCTELEM, block1)[64];
210

211 212
static inline void mmx_emms(void)
{
213
#if HAVE_MMX
214
    if (cpu_flags & AV_CPU_FLAG_MMX)
215
        __asm__ volatile ("emms\n\t");
216 217 218
#endif
}

219
static void init_block(DCTELEM block[64], int test, int is_idct, AVLFG *prng, int vals)
220 221 222 223 224 225 226 227
{
    int i, j;

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

    switch (test) {
    case 0:
        for (i = 0; i < 64; i++)
228
            block[i] = (av_lfg_get(prng) % (2*vals)) -vals;
229 230 231 232 233 234 235 236 237
        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++)
238
            block[av_lfg_get(prng) % 64] = av_lfg_get(prng) % (2*vals) -vals;
239 240
        break;
    case 2:
241
        block[ 0] = av_lfg_get(prng) % (16*vals) - (8*vals);
242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262
        block[63] = (block[0] & 1) ^ 1;
        break;
    }
}

static void permute(DCTELEM dst[64], const DCTELEM src[64], int perm)
{
    int i;

    if (perm == MMX_PERM) {
        for (i = 0; i < 64; i++)
            dst[idct_mmx_perm[i]] = src[i];
    } else if (perm == MMX_SIMPLE_PERM) {
        for (i = 0; i < 64; i++)
            dst[idct_simple_mmx_perm[i]] = src[i];
    } else if (perm == SSE2_PERM) {
        for (i = 0; i < 64; i++)
            dst[(i & 0x38) | idct_sse2_row_perm[i & 7]] = src[i];
    } else if (perm == PARTTRANS_PERM) {
        for (i = 0; i < 64; i++)
            dst[(i & 0x24) | ((i & 3) << 3) | ((i >> 3) & 3)] = src[i];
263 264 265
    } else if (perm == TRANSPOSE_PERM) {
        for (i = 0; i < 64; i++)
            dst[(i>>3) | ((i<<3)&0x38)] = src[i];
266 267 268 269 270
    } else {
        for (i = 0; i < 64; i++)
            dst[i] = src[i];
    }
}
271 272

static int dct_error(const struct algo *dct, int test, int is_idct, int speed, const int bits)
Fabrice Bellard's avatar
Fabrice Bellard committed
273
{
274
    void (*ref)(DCTELEM *block) = is_idct ? ff_ref_idct : ff_ref_fdct;
Fabrice Bellard's avatar
Fabrice Bellard committed
275 276
    int it, i, scale;
    int err_inf, v;
277
    int64_t err2, ti, ti1, it1, err_sum = 0;
278 279 280
    int64_t sysErr[64], sysErrMax = 0;
    int maxout = 0;
    int blockSumErrMax = 0, blockSumErr;
281
    AVLFG prng;
282
    const int vals=1<<bits;
283 284
    double omse, ome;
    int spec_err;
Fabrice Bellard's avatar
Fabrice Bellard committed
285

286
    av_lfg_init(&prng, 1);
Fabrice Bellard's avatar
Fabrice Bellard committed
287 288 289

    err_inf = 0;
    err2 = 0;
290 291 292
    for (i = 0; i < 64; i++)
        sysErr[i] = 0;
    for (it = 0; it < NB_ITS; it++) {
293
        init_block(block1, test, is_idct, &prng, vals);
294
        permute(block, block1, dct->format);
295

296
        dct->func(block);
297
        mmx_emms();
298

299
        if (dct->format == SCALE_PERM) {
300 301 302
            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
303 304 305
            }
        }

306
        ref(block1);
Fabrice Bellard's avatar
Fabrice Bellard committed
307

308 309
        blockSumErr = 0;
        for (i = 0; i < 64; i++) {
310 311 312
            int err = block[i] - block1[i];
            err_sum += err;
            v = abs(err);
Fabrice Bellard's avatar
Fabrice Bellard committed
313 314 315
            if (v > err_inf)
                err_inf = v;
            err2 += v * v;
316 317
            sysErr[i] += block[i] - block1[i];
            blockSumErr += v;
318 319
            if (abs(block[i]) > maxout)
                maxout = abs(block[i]);
Fabrice Bellard's avatar
Fabrice Bellard committed
320
        }
321 322
        if (blockSumErrMax < blockSumErr)
            blockSumErrMax = blockSumErr;
Michael Niedermayer's avatar
Michael Niedermayer committed
323
    }
324 325
    for (i = 0; i < 64; i++)
        sysErrMax = FFMAX(sysErrMax, FFABS(sysErr[i]));
326

327 328 329 330
    for (i = 0; i < 64; i++) {
        if (i % 8 == 0)
            printf("\n");
        printf("%7d ", (int) sysErr[i]);
Fabrice Bellard's avatar
Fabrice Bellard committed
331
    }
Michael Niedermayer's avatar
Michael Niedermayer committed
332
    printf("\n");
333

334 335 336 337 338
    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);

339
    printf("%s %s: max_err=%d omse=%0.8f ome=%0.8f syserr=%0.8f maxout=%d blockSumErr=%d\n",
340
           is_idct ? "IDCT" : "DCT", dct->name, err_inf,
341
           omse, ome, (double) sysErrMax / NB_ITS,
342
           maxout, blockSumErrMax);
343

344 345 346
    if (spec_err && !dct->nonspec)
        return 1;

347
    if (!speed)
348
        return 0;
349

Fabrice Bellard's avatar
Fabrice Bellard committed
350
    /* speed test */
351

352
    init_block(block, test, is_idct, &prng, vals);
353
    permute(block1, block, dct->format);
354

Fabrice Bellard's avatar
Fabrice Bellard committed
355 356 357
    ti = gettime();
    it1 = 0;
    do {
358
        for (it = 0; it < NB_ITS_SPEED; it++) {
359
            memcpy(block, block1, sizeof(block));
360
            dct->func(block);
Fabrice Bellard's avatar
Fabrice Bellard committed
361 362 363 364
        }
        it1 += NB_ITS_SPEED;
        ti1 = gettime() - ti;
    } while (ti1 < 1000000);
365
    mmx_emms();
Fabrice Bellard's avatar
Fabrice Bellard committed
366

367
    printf("%s %s: %0.1f kdct/s\n", is_idct ? "IDCT" : "DCT", dct->name,
368
           (double) it1 * 1000.0 / (double) ti1);
369 370

    return 0;
Fabrice Bellard's avatar
Fabrice Bellard committed
371 372
}

373 374
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
375

376
static void idct248_ref(uint8_t *dest, int linesize, int16_t *block)
Fabrice Bellard's avatar
Fabrice Bellard committed
377 378 379 380 381 382 383 384 385 386 387
{
    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;

388
        for (i = 0; i < 8; i++) {
Fabrice Bellard's avatar
Fabrice Bellard committed
389
            sum = 0;
390 391
            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
392 393 394 395
                c8[i][j] = s * cos(M_PI * i * (j + 0.5) / 8.0);
                sum += c8[i][j] * c8[i][j];
            }
        }
396

397
        for (i = 0; i < 4; i++) {
Fabrice Bellard's avatar
Fabrice Bellard committed
398
            sum = 0;
399 400
            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
401 402 403 404 405 406 407
                c4[i][j] = s * cos(M_PI * i * (j + 0.5) / 4.0);
                sum += c4[i][j] * c4[i][j];
            }
        }
    }

    /* butterfly */
408
    s = 0.5 * sqrt(2.0);
409 410 411 412 413 414
    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
415 416 417 418
        }
    }

    /* idct8 on lines */
419 420
    for (i = 0; i < 8; i++) {
        for (j = 0; j < 8; j++) {
Fabrice Bellard's avatar
Fabrice Bellard committed
421
            sum = 0;
422 423 424
            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
425 426 427 428
        }
    }

    /* idct4 */
429 430
    for (i = 0; i < 8; i++) {
        for (j = 0; j < 4; j++) {
Fabrice Bellard's avatar
Fabrice Bellard committed
431 432
            /* top */
            sum = 0;
433 434 435
            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
436 437 438

            /* bottom */
            sum = 0;
439 440 441
            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
442 443 444 445
        }
    }

    /* clamp and store the result */
446 447 448 449 450 451
    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
452 453 454 455
        }
    }
}

456
static void idct248_error(const char *name,
457
                          void (*idct248_put)(uint8_t *dest, int line_size,
458 459
                                              int16_t *block),
                          int speed)
Fabrice Bellard's avatar
Fabrice Bellard committed
460 461
{
    int it, i, it1, ti, ti1, err_max, v;
462
    AVLFG prng;
463

464
    av_lfg_init(&prng, 1);
465

Fabrice Bellard's avatar
Fabrice Bellard committed
466 467 468
    /* just one test to see if code is correct (precision is less
       important here) */
    err_max = 0;
469
    for (it = 0; it < NB_ITS; it++) {
470
        /* XXX: use forward transform to generate values */
471
        for (i = 0; i < 64; i++)
472
            block1[i] = av_lfg_get(&prng) % 256 - 128;
473 474
        block1[0] += 1024;

475 476
        for (i = 0; i < 64; i++)
            block[i] = block1[i];
Fabrice Bellard's avatar
Fabrice Bellard committed
477
        idct248_ref(img_dest1, 8, block);
478

479 480
        for (i = 0; i < 64; i++)
            block[i] = block1[i];
481
        idct248_put(img_dest, 8, block);
482

483 484
        for (i = 0; i < 64; i++) {
            v = abs((int) img_dest[i] - (int) img_dest1[i]);
485 486 487 488 489
            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
490 491 492 493 494 495 496 497 498
#if 0
        printf("ref=\n");
        for(i=0;i<8;i++) {
            int j;
            for(j=0;j<8;j++) {
                printf(" %3d", img_dest1[i*8+j]);
            }
            printf("\n");
        }
499

Fabrice Bellard's avatar
Fabrice Bellard committed
500 501 502 503 504 505 506 507 508 509
        printf("out=\n");
        for(i=0;i<8;i++) {
            int j;
            for(j=0;j<8;j++) {
                printf(" %3d", img_dest[i*8+j]);
            }
            printf("\n");
        }
#endif
    }
510
    printf("%s %s: err_inf=%d\n", 1 ? "IDCT248" : "DCT248", name, err_max);
Fabrice Bellard's avatar
Fabrice Bellard committed
511

512 513
    if (!speed)
        return;
Fabrice Bellard's avatar
Fabrice Bellard committed
514 515 516 517

    ti = gettime();
    it1 = 0;
    do {
518 519 520
        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
521 522 523 524 525
            idct248_put(img_dest, 8, block);
        }
        it1 += NB_ITS_SPEED;
        ti1 = gettime() - ti;
    } while (ti1 < 1000000);
526
    mmx_emms();
Fabrice Bellard's avatar
Fabrice Bellard committed
527

528 529
    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
530 531
}

532
static void help(void)
533
{
534
    printf("dct-test [-i] [<test-number>] [<bits>]\n"
Michael Niedermayer's avatar
Michael Niedermayer committed
535 536 537
           "test-number 0 -> test with random matrixes\n"
           "            1 -> test with random sparse matrixes\n"
           "            2 -> do 3. test from mpeg4 std\n"
538
           "bits        Number of time domain bits to use, 8 is default\n"
Fabrice Bellard's avatar
Fabrice Bellard committed
539
           "-i          test IDCT implementations\n"
540 541
           "-4          test IDCT248 implementations\n"
           "-t          speed test\n");
542 543
}

Fabrice Bellard's avatar
Fabrice Bellard committed
544 545
int main(int argc, char **argv)
{
Fabrice Bellard's avatar
Fabrice Bellard committed
546
    int test_idct = 0, test_248_dct = 0;
547 548
    int c, i;
    int test = 1;
549
    int speed = 0;
550
    int err = 0;
551
    int bits=8;
552

553
    cpu_flags = av_get_cpu_flags();
554

555
    ff_ref_dct_init();
556
    idct_mmx_init();
557

558
    for (;;) {
559
        c = getopt(argc, argv, "ih4t");
560 561
        if (c == -1)
            break;
562
        switch (c) {
563 564 565
        case 'i':
            test_idct = 1;
            break;
Fabrice Bellard's avatar
Fabrice Bellard committed
566 567 568
        case '4':
            test_248_dct = 1;
            break;
569 570 571
        case 't':
            speed = 1;
            break;
572
        default:
573 574
        case 'h':
            help();
575
            return 0;
576 577
        }
    }
578

579 580
    if (optind < argc)
        test = atoi(argv[optind]);
581
    if(optind+1 < argc) bits= atoi(argv[optind+1]);
582

583 584
    printf("ffmpeg DCT/IDCT test\n");

Fabrice Bellard's avatar
Fabrice Bellard committed
585
    if (test_248_dct) {
586
        idct248_error("SIMPLE-C", ff_simple_idct248_put, speed);
587
    } else {
588
        const struct algo *algos = test_idct ? idct_tab : fdct_tab;
589
        for (i = 0; algos[i].name; i++)
590
            if (!(~cpu_flags & algos[i].mm_support)) {
591
                err |= dct_error(&algos[i], test, test_idct, speed, bits);
592
            }
593
    }
594 595

    return err;
Fabrice Bellard's avatar
Fabrice Bellard committed
596
}