mpegaudiodec.c 85.5 KB
Newer Older
F
Fabrice Bellard 已提交
1 2
/*
 * MPEG Audio decoder
F
Fabrice Bellard 已提交
3
 * Copyright (c) 2001, 2002 Fabrice Bellard.
F
Fabrice Bellard 已提交
4
 *
F
Fabrice Bellard 已提交
5 6 7 8
 * This library is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2 of the License, or (at your option) any later version.
F
Fabrice Bellard 已提交
9
 *
F
Fabrice Bellard 已提交
10
 * This library is distributed in the hope that it will be useful,
F
Fabrice Bellard 已提交
11
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
F
Fabrice Bellard 已提交
12 13
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
F
Fabrice Bellard 已提交
14
 *
F
Fabrice Bellard 已提交
15 16
 * You should have received a copy of the GNU Lesser General Public
 * License along with this library; if not, write to the Free Software
17
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
F
Fabrice Bellard 已提交
18
 */
M
Michael Niedermayer 已提交
19 20 21 22

/**
 * @file mpegaudiodec.c
 * MPEG Audio decoder.
23
 */
M
Michael Niedermayer 已提交
24

25
//#define DEBUG
F
Fabrice Bellard 已提交
26
#include "avcodec.h"
27
#include "bitstream.h"
28
#include "dsputil.h"
F
Fabrice Bellard 已提交
29 30

/*
31 32 33
 * TODO:
 *  - in low precision mode, use more 16 bit multiplies in synth filter
 *  - test lsf / mpeg25 extensively.
F
Fabrice Bellard 已提交
34 35
 */

36 37
/* define USE_HIGHPRECISION to have a bit exact (but slower) mpeg
   audio decoder */
38 39 40
#ifdef CONFIG_MPEGAUDIO_HP
#define USE_HIGHPRECISION
#endif
41

R
Roberto Togni 已提交
42
#include "mpegaudio.h"
43

44 45
#define FRAC_ONE    (1 << FRAC_BITS)

46 47
#define MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> FRAC_BITS)
#define MUL64(a,b) ((int64_t)(a) * (int64_t)(b))
48 49 50 51 52
#define FIX(a)   ((int)((a) * FRAC_ONE))
/* WARNING: only correct for posititive numbers */
#define FIXR(a)   ((int)((a) * FRAC_ONE + 0.5))
#define FRAC_RND(a) (((a) + (FRAC_ONE/2)) >> FRAC_BITS)

M
Michael Niedermayer 已提交
53 54 55 56 57 58
#define FIXHR(a) ((int)((a) * (1LL<<32) + 0.5))
//#define MULH(a,b) (((int64_t)(a) * (int64_t)(b))>>32) //gcc 3.4 creates an incredibly bloated mess out of this
static always_inline int MULH(int a, int b){
    return ((int64_t)(a) * (int64_t)(b))>>32;
}

59 60
/****************/

F
Fabrice Bellard 已提交
61 62 63
#define HEADER_SIZE 4
#define BACKSTEP_SIZE 512

64 65
struct GranuleDef;

F
Fabrice Bellard 已提交
66
typedef struct MPADecodeContext {
67
    uint8_t inbuf1[2][MPA_MAX_CODED_FRAME_SIZE + BACKSTEP_SIZE];        /* input buffer */
F
Fabrice Bellard 已提交
68
    int inbuf_index;
69
    uint8_t *inbuf_ptr, *inbuf;
F
Fabrice Bellard 已提交
70
    int frame_size;
71 72 73
    int free_format_frame_size; /* frame size in case of free format
                                   (zero if currently unknown) */
    /* next header (used in free format parsing) */
74
    uint32_t free_format_next_header;
F
Fabrice Bellard 已提交
75 76 77
    int error_protection;
    int layer;
    int sample_rate;
78
    int sample_rate_index; /* between 0 and 8 */
F
Fabrice Bellard 已提交
79 80 81
    int bit_rate;
    int old_frame_size;
    GetBitContext gb;
82 83 84 85
    int nb_channels;
    int mode;
    int mode_ext;
    int lsf;
86
    MPA_INT synth_buf[MPA_MAX_CHANNELS][512 * 2] __attribute__((aligned(16)));
87
    int synth_buf_offset[MPA_MAX_CHANNELS];
88
    int32_t sb_samples[MPA_MAX_CHANNELS][36][SBLIMIT] __attribute__((aligned(16)));
89
    int32_t mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */
90 91 92
#ifdef DEBUG
    int frame_count;
#endif
93
    void (*compute_antialias)(struct MPADecodeContext *s, struct GranuleDef *g);
R
Roberto Togni 已提交
94
    int adu_mode; ///< 0 for standard mp3, 1 for adu formatted mp3
95
    unsigned int dither_state;
F
Fabrice Bellard 已提交
96 97
} MPADecodeContext;

98 99 100 101 102 103 104 105 106
/**
 * Context for MP3On4 decoder
 */
typedef struct MP3On4DecodeContext {
    int frames;   ///< number of mp3 frames per block (number of mp3 decoder instances)
    int chan_cfg; ///< channel config number
    MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance
} MP3On4DecodeContext;

107 108
/* layer 3 "granule" */
typedef struct GranuleDef {
109
    uint8_t scfsi;
110 111 112 113
    int part2_3_length;
    int big_values;
    int global_gain;
    int scalefac_compress;
114 115
    uint8_t block_type;
    uint8_t switch_point;
116 117
    int table_select[3];
    int subblock_gain[3];
118 119
    uint8_t scalefac_scale;
    uint8_t count1table_select;
120 121 122
    int region_size[3]; /* number of huffman codes in each region */
    int preflag;
    int short_start, long_end; /* long/short band indexes */
123 124
    uint8_t scale_factors[40];
    int32_t sb_hybrid[SBLIMIT * 18]; /* 576 samples */
125
} GranuleDef;
F
Fabrice Bellard 已提交
126

127 128 129 130 131 132
#define MODE_EXT_MS_STEREO 2
#define MODE_EXT_I_STEREO  1

/* layer 3 huffman tables */
typedef struct HuffTable {
    int xsize;
133 134
    const uint8_t *bits;
    const uint16_t *codes;
135 136 137 138
} HuffTable;

#include "mpegaudiodectab.h"

139 140 141
static void compute_antialias_integer(MPADecodeContext *s, GranuleDef *g);
static void compute_antialias_float(MPADecodeContext *s, GranuleDef *g);

142
/* vlc structure for decoding layer 3 huffman tables */
143
static VLC huff_vlc[16];
144
static uint8_t *huff_code_table[16];
145 146
static VLC huff_quad_vlc[2];
/* computed from band_size_long */
147
static uint16_t band_index_long[9][23];
148
/* XXX: free when all decoders are closed */
149
#define TABLE_4_3_SIZE (8191 + 16)*4
150 151
static int8_t  *table_4_3_exp;
static uint32_t *table_4_3_value;
152
/* intensity stereo coef table */
153 154
static int32_t is_table[2][16];
static int32_t is_table_lsf[2][2][16];
155 156
static int32_t csa_table[8][4];
static float csa_table_float[8][4];
157
static int32_t mdct_win[8][36];
158 159

/* lower 2 bits: modulo 3, higher bits: shift */
160
static uint16_t scale_factor_modshift[64];
161
/* [i][j]:  2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
162
static int32_t scale_factor_mult[15][3];
163 164 165 166 167
/* mult table for layer 2 group quantization */

#define SCALE_GEN(v) \
{ FIXR(1.0 * (v)), FIXR(0.7937005259 * (v)), FIXR(0.6299605249 * (v)) }

M
Michael Niedermayer 已提交
168
static const int32_t scale_factor_mult2[3][3] = {
169 170 171
    SCALE_GEN(4.0 / 3.0), /* 3 steps */
    SCALE_GEN(4.0 / 5.0), /* 5 steps */
    SCALE_GEN(4.0 / 9.0), /* 9 steps */
172 173
};

174
void ff_mpa_synth_init(MPA_INT *window);
175
static MPA_INT window[512] __attribute__((aligned(16)));
176

177 178 179 180 181
/* layer 1 unscaling */
/* n = number of bits of the mantissa minus 1 */
static inline int l1_unscale(int n, int mant, int scale_factor)
{
    int shift, mod;
182
    int64_t val;
183 184 185 186 187 188

    shift = scale_factor_modshift[scale_factor];
    mod = shift & 3;
    shift >>= 2;
    val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);
    shift += n;
189 190
    /* NOTE: at this point, 1 <= shift >= 21 + 15 */
    return (int)((val + (1LL << (shift - 1))) >> shift);
191 192 193 194 195 196 197 198 199
}

static inline int l2_unscale_group(int steps, int mant, int scale_factor)
{
    int shift, mod, val;

    shift = scale_factor_modshift[scale_factor];
    mod = shift & 3;
    shift >>= 2;
200 201 202 203 204 205

    val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
    /* NOTE: at this point, 0 <= shift <= 21 */
    if (shift > 0)
        val = (val + (1 << (shift - 1))) >> shift;
    return val;
206 207 208 209 210 211 212 213
}

/* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
static inline int l3_unscale(int value, int exponent)
{
    unsigned int m;
    int e;

214 215 216 217
    e = table_4_3_exp  [4*value + (exponent&3)];
    m = table_4_3_value[4*value + (exponent&3)];
    e -= (exponent >> 2);
    assert(e>=1);
218
    if (e > 31)
219
        return 0;
220
    m = (m + (1 << (e-1))) >> e;
221

222 223 224
    return m;
}

225 226 227 228 229 230
/* all integer n^(4/3) computation code */
#define DEV_ORDER 13

#define POW_FRAC_BITS 24
#define POW_FRAC_ONE    (1 << POW_FRAC_BITS)
#define POW_FIX(a)   ((int)((a) * POW_FRAC_ONE))
231
#define POW_MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> POW_FRAC_BITS)
232 233 234

static int dev_4_3_coefs[DEV_ORDER];

235
#if 0 /* unused */
236 237 238 239 240
static int pow_mult3[3] = {
    POW_FIX(1.0),
    POW_FIX(1.25992104989487316476),
    POW_FIX(1.58740105196819947474),
};
241
#endif
242 243 244 245 246 247 248 249 250 251 252 253

static void int_pow_init(void)
{
    int i, a;

    a = POW_FIX(1.0);
    for(i=0;i<DEV_ORDER;i++) {
        a = POW_MULL(a, POW_FIX(4.0 / 3.0) - i * POW_FIX(1.0)) / (i + 1);
        dev_4_3_coefs[i] = a;
    }
}

254
#if 0 /* unused, remove? */
255 256 257 258 259
/* return the mantissa and the binary exponent */
static int int_pow(int i, int *exp_ptr)
{
    int e, er, eq, j;
    int a, a1;
260

261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286
    /* renormalize */
    a = i;
    e = POW_FRAC_BITS;
    while (a < (1 << (POW_FRAC_BITS - 1))) {
        a = a << 1;
        e--;
    }
    a -= (1 << POW_FRAC_BITS);
    a1 = 0;
    for(j = DEV_ORDER - 1; j >= 0; j--)
        a1 = POW_MULL(a, dev_4_3_coefs[j] + a1);
    a = (1 << POW_FRAC_BITS) + a1;
    /* exponent compute (exact) */
    e = e * 4;
    er = e % 3;
    eq = e / 3;
    a = POW_MULL(a, pow_mult3[er]);
    while (a >= 2 * POW_FRAC_ONE) {
        a = a >> 1;
        eq++;
    }
    /* convert to float */
    while (a < POW_FRAC_ONE) {
        a = a << 1;
        eq--;
    }
287
    /* now POW_FRAC_ONE <= a < 2 * POW_FRAC_ONE */
288
#if POW_FRAC_BITS > FRAC_BITS
289 290 291 292 293 294 295
    a = (a + (1 << (POW_FRAC_BITS - FRAC_BITS - 1))) >> (POW_FRAC_BITS - FRAC_BITS);
    /* correct overflow */
    if (a >= 2 * (1 << FRAC_BITS)) {
        a = a >> 1;
        eq++;
    }
#endif
296 297 298
    *exp_ptr = eq;
    return a;
}
299
#endif
F
Fabrice Bellard 已提交
300 301 302 303

static int decode_init(AVCodecContext * avctx)
{
    MPADecodeContext *s = avctx->priv_data;
304
    static int init=0;
305
    int i, j, k;
F
Fabrice Bellard 已提交
306

307
#if defined(USE_HIGHPRECISION) && defined(CONFIG_AUDIO_NONSHORT)
308 309 310
    avctx->sample_fmt= SAMPLE_FMT_S32;
#else
    avctx->sample_fmt= SAMPLE_FMT_S16;
311 312
#endif

M
Michael Niedermayer 已提交
313
    if(avctx->antialias_algo != FF_AA_FLOAT)
M
10l  
Michael Niedermayer 已提交
314 315 316 317
        s->compute_antialias= compute_antialias_integer;
    else
        s->compute_antialias= compute_antialias_float;

318
    if (!init && !avctx->parse_only) {
319 320 321 322
        /* scale factors table for layer 1/2 */
        for(i=0;i<64;i++) {
            int shift, mod;
            /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */
323
            shift = (i / 3);
324 325 326 327 328 329 330 331
            mod = i % 3;
            scale_factor_modshift[i] = mod | (shift << 2);
        }

        /* scale factor multiply for layer 1 */
        for(i=0;i<15;i++) {
            int n, norm;
            n = i + 2;
332
            norm = ((int64_t_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
333 334 335
            scale_factor_mult[i][0] = MULL(FIXR(1.0 * 2.0), norm);
            scale_factor_mult[i][1] = MULL(FIXR(0.7937005259 * 2.0), norm);
            scale_factor_mult[i][2] = MULL(FIXR(0.6299605249 * 2.0), norm);
336
            dprintf("%d: norm=%x s=%x %x %x\n",
337
                    i, norm,
338 339 340 341
                    scale_factor_mult[i][0],
                    scale_factor_mult[i][1],
                    scale_factor_mult[i][2]);
        }
342

343
        ff_mpa_synth_init(window);
344

345 346 347 348
        /* huffman decode tables */
        huff_code_table[0] = NULL;
        for(i=1;i<16;i++) {
            const HuffTable *h = &mpa_huff_tables[i];
349
            int xsize, x, y;
350
            unsigned int n;
351
            uint8_t *code_table;
352 353 354 355

            xsize = h->xsize;
            n = xsize * xsize;
            /* XXX: fail test */
356
            init_vlc(&huff_vlc[i], 8, n,
357
                     h->bits, 1, 1, h->codes, 2, 2, 1);
358

359 360 361 362 363 364 365 366 367
            code_table = av_mallocz(n);
            j = 0;
            for(x=0;x<xsize;x++) {
                for(y=0;y<xsize;y++)
                    code_table[j++] = (x << 4) | y;
            }
            huff_code_table[i] = code_table;
        }
        for(i=0;i<2;i++) {
368
            init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
369
                     mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1, 1);
370 371 372 373 374 375 376 377 378 379 380
        }

        for(i=0;i<9;i++) {
            k = 0;
            for(j=0;j<22;j++) {
                band_index_long[i][j] = k;
                k += band_size_long[i][j];
            }
            band_index_long[i][22] = k;
        }

381 382
        /* compute n ^ (4/3) and store it in mantissa/exp format */
        table_4_3_exp= av_mallocz_static(TABLE_4_3_SIZE * sizeof(table_4_3_exp[0]));
383
        if(!table_4_3_exp)
384 385
            return -1;
        table_4_3_value= av_mallocz_static(TABLE_4_3_SIZE * sizeof(table_4_3_value[0]));
386
        if(!table_4_3_value)
387
            return -1;
388

389
        int_pow_init();
390
        for(i=1;i<TABLE_4_3_SIZE;i++) {
391
            double f, fm;
392
            int e, m;
393 394
            f = pow((double)(i/4), 4.0 / 3.0) * pow(2, (i&3)*0.25);
            fm = frexp(f, &e);
395
            m = (uint32_t)(fm*(1LL<<31) + 0.5);
396
            e+= FRAC_BITS - 31 + 5;
397

398 399
            /* normalized to FRAC_BITS */
            table_4_3_value[i] = m;
400 401
//            av_log(NULL, AV_LOG_DEBUG, "%d %d %f\n", i, m, pow((double)i, 4.0 / 3.0));
            table_4_3_exp[i] = -e;
402
        }
403

404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429
        for(i=0;i<7;i++) {
            float f;
            int v;
            if (i != 6) {
                f = tan((double)i * M_PI / 12.0);
                v = FIXR(f / (1.0 + f));
            } else {
                v = FIXR(1.0);
            }
            is_table[0][i] = v;
            is_table[1][6 - i] = v;
        }
        /* invalid values */
        for(i=7;i<16;i++)
            is_table[0][i] = is_table[1][i] = 0.0;

        for(i=0;i<16;i++) {
            double f;
            int e, k;

            for(j=0;j<2;j++) {
                e = -(j + 1) * ((i + 1) >> 1);
                f = pow(2.0, e / 4.0);
                k = i & 1;
                is_table_lsf[j][k ^ 1][i] = FIXR(f);
                is_table_lsf[j][k][i] = FIXR(1.0);
430
                dprintf("is_table_lsf %d %d: %x %x\n",
431 432 433 434 435 436 437 438 439
                        i, j, is_table_lsf[j][0][i], is_table_lsf[j][1][i]);
            }
        }

        for(i=0;i<8;i++) {
            float ci, cs, ca;
            ci = ci_table[i];
            cs = 1.0 / sqrt(1.0 + ci * ci);
            ca = cs * ci;
M
Michael Niedermayer 已提交
440 441 442
            csa_table[i][0] = FIXHR(cs/4);
            csa_table[i][1] = FIXHR(ca/4);
            csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);
443
            csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4);
444 445 446
            csa_table_float[i][0] = cs;
            csa_table_float[i][1] = ca;
            csa_table_float[i][2] = ca + cs;
447
            csa_table_float[i][3] = ca - cs;
448
//            printf("%d %d %d %d\n", FIX(cs), FIX(cs-1), FIX(ca), FIX(cs)-FIX(ca));
M
Michael Niedermayer 已提交
449
//            av_log(NULL, AV_LOG_DEBUG,"%f %f %f %f\n", cs, ca, ca+cs, ca-cs);
450 451 452 453
        }

        /* compute mdct windows */
        for(i=0;i<36;i++) {
M
Michael Niedermayer 已提交
454 455
            for(j=0; j<4; j++){
                double d;
456

M
Michael Niedermayer 已提交
457 458
                if(j==2 && i%3 != 1)
                    continue;
459

M
Michael Niedermayer 已提交
460 461 462 463 464 465 466 467 468 469 470
                d= sin(M_PI * (i + 0.5) / 36.0);
                if(j==1){
                    if     (i>=30) d= 0;
                    else if(i>=24) d= sin(M_PI * (i - 18 + 0.5) / 12.0);
                    else if(i>=18) d= 1;
                }else if(j==3){
                    if     (i<  6) d= 0;
                    else if(i< 12) d= sin(M_PI * (i -  6 + 0.5) / 12.0);
                    else if(i< 18) d= 1;
                }
                //merge last stage of imdct into the window coefficients
M
Michael Niedermayer 已提交
471 472 473 474 475 476
                d*= 0.5 / cos(M_PI*(2*i + 19)/72);

                if(j==2)
                    mdct_win[j][i/3] = FIXHR((d / (1<<5)));
                else
                    mdct_win[j][i  ] = FIXHR((d / (1<<5)));
M
Michael Niedermayer 已提交
477 478
//                av_log(NULL, AV_LOG_DEBUG, "%2d %d %f\n", i,j,d / (1<<5));
            }
479 480 481 482 483 484 485 486 487 488 489 490 491
        }

        /* NOTE: we do frequency inversion adter the MDCT by changing
           the sign of the right window coefs */
        for(j=0;j<4;j++) {
            for(i=0;i<36;i+=2) {
                mdct_win[j + 4][i] = mdct_win[j][i];
                mdct_win[j + 4][i + 1] = -mdct_win[j][i + 1];
            }
        }

#if defined(DEBUG)
        for(j=0;j<8;j++) {
S
Steve L'Homme 已提交
492
            av_log(avctx, AV_LOG_DEBUG, "win%d=\n", j);
493
            for(i=0;i<36;i++)
S
Steve L'Homme 已提交
494 495
                av_log(avctx, AV_LOG_DEBUG, "%f, ", (double)mdct_win[j][i] / FRAC_ONE);
            av_log(avctx, AV_LOG_DEBUG, "\n");
496 497
        }
#endif
F
Fabrice Bellard 已提交
498 499 500 501 502 503
        init = 1;
    }

    s->inbuf_index = 0;
    s->inbuf = &s->inbuf1[s->inbuf_index][BACKSTEP_SIZE];
    s->inbuf_ptr = s->inbuf;
504 505 506
#ifdef DEBUG
    s->frame_count = 0;
#endif
R
Roberto Togni 已提交
507 508
    if (avctx->codec_id == CODEC_ID_MP3ADU)
        s->adu_mode = 1;
F
Fabrice Bellard 已提交
509 510 511
    return 0;
}

512
/* tab[i][j] = 1.0 / (2.0 * cos(pi*(2*k+1) / 2^(6 - j))) */
513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580

/* cos(i*pi/64) */

#define COS0_0  FIXR(0.50060299823519630134)
#define COS0_1  FIXR(0.50547095989754365998)
#define COS0_2  FIXR(0.51544730992262454697)
#define COS0_3  FIXR(0.53104259108978417447)
#define COS0_4  FIXR(0.55310389603444452782)
#define COS0_5  FIXR(0.58293496820613387367)
#define COS0_6  FIXR(0.62250412303566481615)
#define COS0_7  FIXR(0.67480834145500574602)
#define COS0_8  FIXR(0.74453627100229844977)
#define COS0_9  FIXR(0.83934964541552703873)
#define COS0_10 FIXR(0.97256823786196069369)
#define COS0_11 FIXR(1.16943993343288495515)
#define COS0_12 FIXR(1.48416461631416627724)
#define COS0_13 FIXR(2.05778100995341155085)
#define COS0_14 FIXR(3.40760841846871878570)
#define COS0_15 FIXR(10.19000812354805681150)

#define COS1_0 FIXR(0.50241928618815570551)
#define COS1_1 FIXR(0.52249861493968888062)
#define COS1_2 FIXR(0.56694403481635770368)
#define COS1_3 FIXR(0.64682178335999012954)
#define COS1_4 FIXR(0.78815462345125022473)
#define COS1_5 FIXR(1.06067768599034747134)
#define COS1_6 FIXR(1.72244709823833392782)
#define COS1_7 FIXR(5.10114861868916385802)

#define COS2_0 FIXR(0.50979557910415916894)
#define COS2_1 FIXR(0.60134488693504528054)
#define COS2_2 FIXR(0.89997622313641570463)
#define COS2_3 FIXR(2.56291544774150617881)

#define COS3_0 FIXR(0.54119610014619698439)
#define COS3_1 FIXR(1.30656296487637652785)

#define COS4_0 FIXR(0.70710678118654752439)

/* butterfly operator */
#define BF(a, b, c)\
{\
    tmp0 = tab[a] + tab[b];\
    tmp1 = tab[a] - tab[b];\
    tab[a] = tmp0;\
    tab[b] = MULL(tmp1, c);\
}

#define BF1(a, b, c, d)\
{\
    BF(a, b, COS4_0);\
    BF(c, d, -COS4_0);\
    tab[c] += tab[d];\
}

#define BF2(a, b, c, d)\
{\
    BF(a, b, COS4_0);\
    BF(c, d, -COS4_0);\
    tab[c] += tab[d];\
    tab[a] += tab[c];\
    tab[c] += tab[b];\
    tab[b] += tab[d];\
}

#define ADD(a, b) tab[a] += tab[b]

/* DCT32 without 1/sqrt(2) coef zero scaling. */
581
static void dct32(int32_t *out, int32_t *tab)
582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611
{
    int tmp0, tmp1;

    /* pass 1 */
    BF(0, 31, COS0_0);
    BF(1, 30, COS0_1);
    BF(2, 29, COS0_2);
    BF(3, 28, COS0_3);
    BF(4, 27, COS0_4);
    BF(5, 26, COS0_5);
    BF(6, 25, COS0_6);
    BF(7, 24, COS0_7);
    BF(8, 23, COS0_8);
    BF(9, 22, COS0_9);
    BF(10, 21, COS0_10);
    BF(11, 20, COS0_11);
    BF(12, 19, COS0_12);
    BF(13, 18, COS0_13);
    BF(14, 17, COS0_14);
    BF(15, 16, COS0_15);

    /* pass 2 */
    BF(0, 15, COS1_0);
    BF(1, 14, COS1_1);
    BF(2, 13, COS1_2);
    BF(3, 12, COS1_3);
    BF(4, 11, COS1_4);
    BF(5, 10, COS1_5);
    BF(6,  9, COS1_6);
    BF(7,  8, COS1_7);
612

613 614 615 616 617 618 619 620
    BF(16, 31, -COS1_0);
    BF(17, 30, -COS1_1);
    BF(18, 29, -COS1_2);
    BF(19, 28, -COS1_3);
    BF(20, 27, -COS1_4);
    BF(21, 26, -COS1_5);
    BF(22, 25, -COS1_6);
    BF(23, 24, -COS1_7);
621

622 623 624 625 626
    /* pass 3 */
    BF(0, 7, COS2_0);
    BF(1, 6, COS2_1);
    BF(2, 5, COS2_2);
    BF(3, 4, COS2_3);
627

628 629 630 631
    BF(8, 15, -COS2_0);
    BF(9, 14, -COS2_1);
    BF(10, 13, -COS2_2);
    BF(11, 12, -COS2_3);
632

633 634 635 636
    BF(16, 23, COS2_0);
    BF(17, 22, COS2_1);
    BF(18, 21, COS2_2);
    BF(19, 20, COS2_3);
637

638 639 640 641 642 643 644 645
    BF(24, 31, -COS2_0);
    BF(25, 30, -COS2_1);
    BF(26, 29, -COS2_2);
    BF(27, 28, -COS2_3);

    /* pass 4 */
    BF(0, 3, COS3_0);
    BF(1, 2, COS3_1);
646

647 648
    BF(4, 7, -COS3_0);
    BF(5, 6, -COS3_1);
649

650 651
    BF(8, 11, COS3_0);
    BF(9, 10, COS3_1);
652

653 654
    BF(12, 15, -COS3_0);
    BF(13, 14, -COS3_1);
655

656 657
    BF(16, 19, COS3_0);
    BF(17, 18, COS3_1);
658

659 660
    BF(20, 23, -COS3_0);
    BF(21, 22, -COS3_1);
661

662 663
    BF(24, 27, COS3_0);
    BF(25, 26, COS3_1);
664

665 666
    BF(28, 31, -COS3_0);
    BF(29, 30, -COS3_1);
667

668 669 670 671 672 673 674 675 676
    /* pass 5 */
    BF1(0, 1, 2, 3);
    BF2(4, 5, 6, 7);
    BF1(8, 9, 10, 11);
    BF2(12, 13, 14, 15);
    BF1(16, 17, 18, 19);
    BF2(20, 21, 22, 23);
    BF1(24, 25, 26, 27);
    BF2(28, 29, 30, 31);
677

678
    /* pass 6 */
679

680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703
    ADD( 8, 12);
    ADD(12, 10);
    ADD(10, 14);
    ADD(14,  9);
    ADD( 9, 13);
    ADD(13, 11);
    ADD(11, 15);

    out[ 0] = tab[0];
    out[16] = tab[1];
    out[ 8] = tab[2];
    out[24] = tab[3];
    out[ 4] = tab[4];
    out[20] = tab[5];
    out[12] = tab[6];
    out[28] = tab[7];
    out[ 2] = tab[8];
    out[18] = tab[9];
    out[10] = tab[10];
    out[26] = tab[11];
    out[ 6] = tab[12];
    out[22] = tab[13];
    out[14] = tab[14];
    out[30] = tab[15];
704

705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732
    ADD(24, 28);
    ADD(28, 26);
    ADD(26, 30);
    ADD(30, 25);
    ADD(25, 29);
    ADD(29, 27);
    ADD(27, 31);

    out[ 1] = tab[16] + tab[24];
    out[17] = tab[17] + tab[25];
    out[ 9] = tab[18] + tab[26];
    out[25] = tab[19] + tab[27];
    out[ 5] = tab[20] + tab[28];
    out[21] = tab[21] + tab[29];
    out[13] = tab[22] + tab[30];
    out[29] = tab[23] + tab[31];
    out[ 3] = tab[24] + tab[20];
    out[19] = tab[25] + tab[21];
    out[11] = tab[26] + tab[22];
    out[27] = tab[27] + tab[23];
    out[ 7] = tab[28] + tab[18];
    out[23] = tab[29] + tab[19];
    out[15] = tab[30] + tab[17];
    out[31] = tab[31];
}

#if FRAC_BITS <= 15

733
static inline int round_sample(int *sum)
734 735
{
    int sum1;
736 737
    sum1 = (*sum) >> OUT_SHIFT;
    *sum &= (1<<OUT_SHIFT)-1;
738 739 740 741
    if (sum1 < OUT_MIN)
        sum1 = OUT_MIN;
    else if (sum1 > OUT_MAX)
        sum1 = OUT_MAX;
742
    return sum1;
743 744
}

745 746 747 748 749 750 751 752 753
#if defined(ARCH_POWERPC_405)

/* signed 16x16 -> 32 multiply add accumulate */
#define MACS(rt, ra, rb) \
    asm ("maclhw %0, %2, %3" : "=r" (rt) : "0" (rt), "r" (ra), "r" (rb));

/* signed 16x16 -> 32 multiply */
#define MULS(ra, rb) \
    ({ int __rt; asm ("mullhw %0, %1, %2" : "=r" (__rt) : "r" (ra), "r" (rb)); __rt; })
754 755 756

#else

757 758 759 760 761 762 763 764 765 766
/* signed 16x16 -> 32 multiply add accumulate */
#define MACS(rt, ra, rb) rt += (ra) * (rb)

/* signed 16x16 -> 32 multiply */
#define MULS(ra, rb) ((ra) * (rb))

#endif

#else

767
static inline int round_sample(int64_t *sum)
768 769
{
    int sum1;
770 771
    sum1 = (int)((*sum) >> OUT_SHIFT);
    *sum &= (1<<OUT_SHIFT)-1;
772 773 774 775
    if (sum1 < OUT_MIN)
        sum1 = OUT_MIN;
    else if (sum1 > OUT_MAX)
        sum1 = OUT_MAX;
776
    return sum1;
777 778
}

779 780 781 782 783
#define MULS(ra, rb) MUL64(ra, rb)

#endif

#define SUM8(sum, op, w, p) \
784
{                                               \
785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821
    sum op MULS((w)[0 * 64], p[0 * 64]);\
    sum op MULS((w)[1 * 64], p[1 * 64]);\
    sum op MULS((w)[2 * 64], p[2 * 64]);\
    sum op MULS((w)[3 * 64], p[3 * 64]);\
    sum op MULS((w)[4 * 64], p[4 * 64]);\
    sum op MULS((w)[5 * 64], p[5 * 64]);\
    sum op MULS((w)[6 * 64], p[6 * 64]);\
    sum op MULS((w)[7 * 64], p[7 * 64]);\
}

#define SUM8P2(sum1, op1, sum2, op2, w1, w2, p) \
{                                               \
    int tmp;\
    tmp = p[0 * 64];\
    sum1 op1 MULS((w1)[0 * 64], tmp);\
    sum2 op2 MULS((w2)[0 * 64], tmp);\
    tmp = p[1 * 64];\
    sum1 op1 MULS((w1)[1 * 64], tmp);\
    sum2 op2 MULS((w2)[1 * 64], tmp);\
    tmp = p[2 * 64];\
    sum1 op1 MULS((w1)[2 * 64], tmp);\
    sum2 op2 MULS((w2)[2 * 64], tmp);\
    tmp = p[3 * 64];\
    sum1 op1 MULS((w1)[3 * 64], tmp);\
    sum2 op2 MULS((w2)[3 * 64], tmp);\
    tmp = p[4 * 64];\
    sum1 op1 MULS((w1)[4 * 64], tmp);\
    sum2 op2 MULS((w2)[4 * 64], tmp);\
    tmp = p[5 * 64];\
    sum1 op1 MULS((w1)[5 * 64], tmp);\
    sum2 op2 MULS((w2)[5 * 64], tmp);\
    tmp = p[6 * 64];\
    sum1 op1 MULS((w1)[6 * 64], tmp);\
    sum2 op2 MULS((w2)[6 * 64], tmp);\
    tmp = p[7 * 64];\
    sum1 op1 MULS((w1)[7 * 64], tmp);\
    sum2 op2 MULS((w2)[7 * 64], tmp);\
822 823
}

824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839
void ff_mpa_synth_init(MPA_INT *window)
{
    int i;

    /* max = 18760, max sum over all 16 coefs : 44736 */
    for(i=0;i<257;i++) {
        int v;
        v = mpa_enwindow[i];
#if WFRAC_BITS < 16
        v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS);
#endif
        window[i] = v;
        if ((i & 63) != 0)
            v = -v;
        if (i != 0)
            window[512 - i] = v;
840
    }
841
}
842 843 844 845

/* 32 sub band synthesis filter. Input: 32 sub band samples, Output:
   32 samples. */
/* XXX: optimize by avoiding ring buffer usage */
846
void ff_mpa_synth_filter(MPA_INT *synth_buf_ptr, int *synth_buf_offset,
847
                         MPA_INT *window, int *dither_state,
848
                         OUT_INT *samples, int incr,
849
                         int32_t sb_samples[SBLIMIT])
850
{
851
    int32_t tmp[32];
852
    register MPA_INT *synth_buf;
A
Alex Beregszaszi 已提交
853
    register const MPA_INT *w, *w2, *p;
854
    int j, offset, v;
855
    OUT_INT *samples2;
856
#if FRAC_BITS <= 15
857
    int sum, sum2;
858
#else
859
    int64_t sum, sum2;
860
#endif
861

862
    dct32(tmp, sb_samples);
863

864 865
    offset = *synth_buf_offset;
    synth_buf = synth_buf_ptr + offset;
866 867 868 869

    for(j=0;j<32;j++) {
        v = tmp[j];
#if FRAC_BITS <= 15
870 871
        /* NOTE: can cause a loss in precision if very high amplitude
           sound */
872 873 874 875 876 877 878 879 880 881
        if (v > 32767)
            v = 32767;
        else if (v < -32768)
            v = -32768;
#endif
        synth_buf[j] = v;
    }
    /* copy to avoid wrap */
    memcpy(synth_buf + 512, synth_buf, 32 * sizeof(MPA_INT));

882
    samples2 = samples + 31 * incr;
883
    w = window;
884 885
    w2 = window + 31;

886
    sum = *dither_state;
887 888 889 890
    p = synth_buf + 16;
    SUM8(sum, +=, w, p);
    p = synth_buf + 48;
    SUM8(sum, -=, w + 32, p);
891
    *samples = round_sample(&sum);
892
    samples += incr;
893 894
    w++;

895 896 897 898 899 900 901 902 903
    /* we calculate two samples at the same time to avoid one memory
       access per two sample */
    for(j=1;j<16;j++) {
        sum2 = 0;
        p = synth_buf + 16 + j;
        SUM8P2(sum, +=, sum2, -=, w, w2, p);
        p = synth_buf + 48 - j;
        SUM8P2(sum, -=, sum2, -=, w + 32, w2 + 32, p);

904
        *samples = round_sample(&sum);
905
        samples += incr;
906 907
        sum += sum2;
        *samples2 = round_sample(&sum);
908
        samples2 -= incr;
909
        w++;
910
        w2--;
911
    }
912

913 914
    p = synth_buf + 32;
    SUM8(sum, -=, w + 32, p);
915
    *samples = round_sample(&sum);
916
    *dither_state= sum;
917

918
    offset = (offset - 32) & 511;
919
    *synth_buf_offset = offset;
920 921
}

M
Michael Niedermayer 已提交
922 923 924 925 926 927 928 929 930 931 932 933 934 935
#define C3 FIXHR(0.86602540378443864676/2)

/* 0.5 / cos(pi*(2*i+1)/36) */
static const int icos36[9] = {
    FIXR(0.50190991877167369479),
    FIXR(0.51763809020504152469), //0
    FIXR(0.55168895948124587824),
    FIXR(0.61038729438072803416),
    FIXR(0.70710678118654752439), //1
    FIXR(0.87172339781054900991),
    FIXR(1.18310079157624925896),
    FIXR(1.93185165257813657349), //2
    FIXR(5.73685662283492756461),
};
936 937 938 939 940

/* 12 points IMDCT. We compute it "by hand" by factorizing obvious
   cases. */
static void imdct12(int *out, int *in)
{
M
Michael Niedermayer 已提交
941
    int in0, in1, in2, in3, in4, in5, t1, t2;
942 943 944 945 946 947 948

    in0= in[0*3];
    in1= in[1*3] + in[0*3];
    in2= in[2*3] + in[1*3];
    in3= in[3*3] + in[2*3];
    in4= in[4*3] + in[3*3];
    in5= in[5*3] + in[4*3];
M
Michael Niedermayer 已提交
949 950 951 952 953
    in5 += in3;
    in3 += in1;

    in2= MULH(2*in2, C3);
    in3= MULH(2*in3, C3);
954

M
Michael Niedermayer 已提交
955 956 957
    t1 = in0 - in4;
    t2 = MULL(in1 - in5, icos36[4]);

958
    out[ 7]=
M
Michael Niedermayer 已提交
959 960 961 962 963 964 965
    out[10]= t1 + t2;
    out[ 1]=
    out[ 4]= t1 - t2;

    in0 += in4>>1;
    in4 = in0 + in2;
    in1 += in5>>1;
966 967
    in5 = MULL(in1 + in3, icos36[1]);
    out[ 8]=
M
Michael Niedermayer 已提交
968 969 970
    out[ 9]= in4 + in5;
    out[ 2]=
    out[ 3]= in4 - in5;
971

M
Michael Niedermayer 已提交
972 973 974 975 976
    in0 -= in2;
    in1 = MULL(in1 - in3, icos36[7]);
    out[ 0]=
    out[ 5]= in0 - in1;
    out[ 6]=
977
    out[11]= in0 + in1;
978 979 980
}

/* cos(pi*i/18) */
M
Michael Niedermayer 已提交
981 982 983 984 985 986 987 988 989
#define C1 FIXHR(0.98480775301220805936/2)
#define C2 FIXHR(0.93969262078590838405/2)
#define C3 FIXHR(0.86602540378443864676/2)
#define C4 FIXHR(0.76604444311897803520/2)
#define C5 FIXHR(0.64278760968653932632/2)
#define C6 FIXHR(0.5/2)
#define C7 FIXHR(0.34202014332566873304/2)
#define C8 FIXHR(0.17364817766693034885/2)

990 991

/* using Lee like decomposition followed by hand coded 9 points DCT */
M
Michael Niedermayer 已提交
992
static void imdct36(int *out, int *buf, int *in, int *win)
993 994 995 996 997 998 999 1000 1001 1002 1003 1004
{
    int i, j, t0, t1, t2, t3, s0, s1, s2, s3;
    int tmp[18], *tmp1, *in1;

    for(i=17;i>=1;i--)
        in[i] += in[i-1];
    for(i=17;i>=3;i-=2)
        in[i] += in[i-2];

    for(j=0;j<2;j++) {
        tmp1 = tmp + j;
        in1 = in + j;
M
Michael Niedermayer 已提交
1005 1006 1007 1008
#if 0
//more accurate but slower
        int64_t t0, t1, t2, t3;
        t2 = in1[2*4] + in1[2*8] - in1[2*2];
1009

M
Michael Niedermayer 已提交
1010 1011 1012 1013 1014 1015 1016 1017
        t3 = (in1[2*0] + (int64_t)(in1[2*6]>>1))<<32;
        t1 = in1[2*0] - in1[2*6];
        tmp1[ 6] = t1 - (t2>>1);
        tmp1[16] = t1 + t2;

        t0 = MUL64(2*(in1[2*2] + in1[2*4]),    C2);
        t1 = MUL64(   in1[2*4] - in1[2*8] , -2*C8);
        t2 = MUL64(2*(in1[2*2] + in1[2*8]),   -C4);
1018

M
Michael Niedermayer 已提交
1019 1020 1021
        tmp1[10] = (t3 - t0 - t2) >> 32;
        tmp1[ 2] = (t3 + t0 + t1) >> 32;
        tmp1[14] = (t3 + t2 - t1) >> 32;
1022

M
Michael Niedermayer 已提交
1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034
        tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
        t2 = MUL64(2*(in1[2*1] + in1[2*5]),    C1);
        t3 = MUL64(   in1[2*5] - in1[2*7] , -2*C7);
        t0 = MUL64(2*in1[2*3], C3);

        t1 = MUL64(2*(in1[2*1] + in1[2*7]),   -C5);

        tmp1[ 0] = (t2 + t3 + t0) >> 32;
        tmp1[12] = (t2 + t1 - t0) >> 32;
        tmp1[ 8] = (t3 - t1 - t0) >> 32;
#else
        t2 = in1[2*4] + in1[2*8] - in1[2*2];
1035

M
Michael Niedermayer 已提交
1036 1037 1038 1039 1040 1041 1042 1043
        t3 = in1[2*0] + (in1[2*6]>>1);
        t1 = in1[2*0] - in1[2*6];
        tmp1[ 6] = t1 - (t2>>1);
        tmp1[16] = t1 + t2;

        t0 = MULH(2*(in1[2*2] + in1[2*4]),    C2);
        t1 = MULH(   in1[2*4] - in1[2*8] , -2*C8);
        t2 = MULH(2*(in1[2*2] + in1[2*8]),   -C4);
1044

M
Michael Niedermayer 已提交
1045 1046 1047
        tmp1[10] = t3 - t0 - t2;
        tmp1[ 2] = t3 + t0 + t1;
        tmp1[14] = t3 + t2 - t1;
1048

M
Michael Niedermayer 已提交
1049 1050 1051 1052
        tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
        t2 = MULH(2*(in1[2*1] + in1[2*5]),    C1);
        t3 = MULH(   in1[2*5] - in1[2*7] , -2*C7);
        t0 = MULH(2*in1[2*3], C3);
1053

M
Michael Niedermayer 已提交
1054 1055 1056 1057 1058 1059
        t1 = MULH(2*(in1[2*1] + in1[2*7]),   -C5);

        tmp1[ 0] = t2 + t3 + t0;
        tmp1[12] = t2 + t1 - t0;
        tmp1[ 8] = t3 - t1 - t0;
#endif
1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072
    }

    i = 0;
    for(j=0;j<4;j++) {
        t0 = tmp[i];
        t1 = tmp[i + 2];
        s0 = t1 + t0;
        s2 = t1 - t0;

        t2 = tmp[i + 1];
        t3 = tmp[i + 3];
        s1 = MULL(t3 + t2, icos36[j]);
        s3 = MULL(t3 - t2, icos36[8 - j]);
1073

1074 1075
        t0 = s0 + s1;
        t1 = s0 - s1;
M
Michael Niedermayer 已提交
1076
        out[(9 + j)*SBLIMIT] =  MULH(t1, win[9 + j]) + buf[9 + j];
M
Michael Niedermayer 已提交
1077 1078 1079
        out[(8 - j)*SBLIMIT] =  MULH(t1, win[8 - j]) + buf[8 - j];
        buf[9 + j] = MULH(t0, win[18 + 9 + j]);
        buf[8 - j] = MULH(t0, win[18 + 8 - j]);
1080

1081 1082
        t0 = s2 + s3;
        t1 = s2 - s3;
M
Michael Niedermayer 已提交
1083
        out[(9 + 8 - j)*SBLIMIT] =  MULH(t1, win[9 + 8 - j]) + buf[9 + 8 - j];
M
Michael Niedermayer 已提交
1084 1085 1086
        out[(        j)*SBLIMIT] =  MULH(t1, win[        j]) + buf[        j];
        buf[9 + 8 - j] = MULH(t0, win[18 + 9 + 8 - j]);
        buf[      + j] = MULH(t0, win[18         + j]);
1087 1088 1089 1090 1091
        i += 4;
    }

    s0 = tmp[16];
    s1 = MULL(tmp[17], icos36[4]);
1092 1093
    t0 = s0 + s1;
    t1 = s0 - s1;
M
Michael Niedermayer 已提交
1094
    out[(9 + 4)*SBLIMIT] =  MULH(t1, win[9 + 4]) + buf[9 + 4];
M
Michael Niedermayer 已提交
1095 1096 1097
    out[(8 - 4)*SBLIMIT] =  MULH(t1, win[8 - 4]) + buf[8 - 4];
    buf[9 + 4] = MULH(t0, win[18 + 9 + 4]);
    buf[8 - 4] = MULH(t0, win[18 + 8 - 4]);
1098 1099
}

F
Fabrice Bellard 已提交
1100
/* header decoding. MUST check the header before because no
1101 1102
   consistency check is done there. Return 1 if free format found and
   that the frame size must be computed externally */
1103
static int decode_header(MPADecodeContext *s, uint32_t header)
F
Fabrice Bellard 已提交
1104
{
1105 1106
    int sample_rate, frame_size, mpeg25, padding;
    int sample_rate_index, bitrate_index;
F
Fabrice Bellard 已提交
1107
    if (header & (1<<20)) {
1108 1109
        s->lsf = (header & (1<<19)) ? 0 : 1;
        mpeg25 = 0;
F
Fabrice Bellard 已提交
1110
    } else {
1111 1112
        s->lsf = 1;
        mpeg25 = 1;
F
Fabrice Bellard 已提交
1113
    }
1114

F
Fabrice Bellard 已提交
1115 1116
    s->layer = 4 - ((header >> 17) & 3);
    /* extract frequency */
1117 1118 1119 1120 1121
    sample_rate_index = (header >> 10) & 3;
    sample_rate = mpa_freq_tab[sample_rate_index] >> (s->lsf + mpeg25);
    sample_rate_index += 3 * (s->lsf + mpeg25);
    s->sample_rate_index = sample_rate_index;
    s->error_protection = ((header >> 16) & 1) ^ 1;
1122
    s->sample_rate = sample_rate;
F
Fabrice Bellard 已提交
1123

1124 1125 1126 1127 1128 1129 1130 1131
    bitrate_index = (header >> 12) & 0xf;
    padding = (header >> 9) & 1;
    //extension = (header >> 8) & 1;
    s->mode = (header >> 6) & 3;
    s->mode_ext = (header >> 4) & 3;
    //copyright = (header >> 3) & 1;
    //original = (header >> 2) & 1;
    //emphasis = header & 3;
F
Fabrice Bellard 已提交
1132

1133 1134 1135 1136
    if (s->mode == MPA_MONO)
        s->nb_channels = 1;
    else
        s->nb_channels = 2;
1137

1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178
    if (bitrate_index != 0) {
        frame_size = mpa_bitrate_tab[s->lsf][s->layer - 1][bitrate_index];
        s->bit_rate = frame_size * 1000;
        switch(s->layer) {
        case 1:
            frame_size = (frame_size * 12000) / sample_rate;
            frame_size = (frame_size + padding) * 4;
            break;
        case 2:
            frame_size = (frame_size * 144000) / sample_rate;
            frame_size += padding;
            break;
        default:
        case 3:
            frame_size = (frame_size * 144000) / (sample_rate << s->lsf);
            frame_size += padding;
            break;
        }
        s->frame_size = frame_size;
    } else {
        /* if no frame size computed, signal it */
        if (!s->free_format_frame_size)
            return 1;
        /* free format: compute bitrate and real frame size from the
           frame size we extracted by reading the bitstream */
        s->frame_size = s->free_format_frame_size;
        switch(s->layer) {
        case 1:
            s->frame_size += padding  * 4;
            s->bit_rate = (s->frame_size * sample_rate) / 48000;
            break;
        case 2:
            s->frame_size += padding;
            s->bit_rate = (s->frame_size * sample_rate) / 144000;
            break;
        default:
        case 3:
            s->frame_size += padding;
            s->bit_rate = (s->frame_size * (sample_rate << s->lsf)) / 144000;
            break;
        }
F
Fabrice Bellard 已提交
1179
    }
1180

1181
#if defined(DEBUG)
S
Steve L'Homme 已提交
1182
    dprintf("layer%d, %d Hz, %d kbits/s, ",
1183 1184 1185 1186
           s->layer, s->sample_rate, s->bit_rate);
    if (s->nb_channels == 2) {
        if (s->layer == 3) {
            if (s->mode_ext & MODE_EXT_MS_STEREO)
S
Steve L'Homme 已提交
1187
                dprintf("ms-");
1188
            if (s->mode_ext & MODE_EXT_I_STEREO)
S
Steve L'Homme 已提交
1189
                dprintf("i-");
1190
        }
S
Steve L'Homme 已提交
1191
        dprintf("stereo");
1192
    } else {
S
Steve L'Homme 已提交
1193
        dprintf("mono");
1194
    }
S
Steve L'Homme 已提交
1195
    dprintf("\n");
F
Fabrice Bellard 已提交
1196
#endif
1197
    return 0;
F
Fabrice Bellard 已提交
1198 1199
}

1200
/* useful helper to get mpeg audio stream infos. Return -1 if error in
1201 1202
   header, otherwise the coded frame size in bytes */
int mpa_decode_header(AVCodecContext *avctx, uint32_t head)
1203 1204
{
    MPADecodeContext s1, *s = &s1;
1205
    memset( s, 0, sizeof(MPADecodeContext) );
1206

1207
    if (ff_mpa_check_header(head) != 0)
1208 1209 1210 1211 1212 1213 1214 1215
        return -1;

    if (decode_header(s, head) != 0) {
        return -1;
    }

    switch(s->layer) {
    case 1:
1216
        avctx->frame_size = 384;
1217 1218
        break;
    case 2:
1219
        avctx->frame_size = 1152;
1220 1221 1222 1223
        break;
    default:
    case 3:
        if (s->lsf)
1224
            avctx->frame_size = 576;
1225
        else
1226
            avctx->frame_size = 1152;
1227 1228 1229
        break;
    }

1230 1231 1232 1233 1234
    avctx->sample_rate = s->sample_rate;
    avctx->channels = s->nb_channels;
    avctx->bit_rate = s->bit_rate;
    avctx->sub_id = s->layer;
    return s->frame_size;
1235 1236
}

1237 1238
/* return the number of decoded frames */
static int mp_decode_layer1(MPADecodeContext *s)
F
Fabrice Bellard 已提交
1239
{
1240
    int bound, i, v, n, ch, j, mant;
1241 1242
    uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
    uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
1243

1244
    if (s->mode == MPA_JSTEREO)
1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271
        bound = (s->mode_ext + 1) * 4;
    else
        bound = SBLIMIT;

    /* allocation bits */
    for(i=0;i<bound;i++) {
        for(ch=0;ch<s->nb_channels;ch++) {
            allocation[ch][i] = get_bits(&s->gb, 4);
        }
    }
    for(i=bound;i<SBLIMIT;i++) {
        allocation[0][i] = get_bits(&s->gb, 4);
    }

    /* scale factors */
    for(i=0;i<bound;i++) {
        for(ch=0;ch<s->nb_channels;ch++) {
            if (allocation[ch][i])
                scale_factors[ch][i] = get_bits(&s->gb, 6);
        }
    }
    for(i=bound;i<SBLIMIT;i++) {
        if (allocation[0][i]) {
            scale_factors[0][i] = get_bits(&s->gb, 6);
            scale_factors[1][i] = get_bits(&s->gb, 6);
        }
    }
1272

1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307
    /* compute samples */
    for(j=0;j<12;j++) {
        for(i=0;i<bound;i++) {
            for(ch=0;ch<s->nb_channels;ch++) {
                n = allocation[ch][i];
                if (n) {
                    mant = get_bits(&s->gb, n + 1);
                    v = l1_unscale(n, mant, scale_factors[ch][i]);
                } else {
                    v = 0;
                }
                s->sb_samples[ch][j][i] = v;
            }
        }
        for(i=bound;i<SBLIMIT;i++) {
            n = allocation[0][i];
            if (n) {
                mant = get_bits(&s->gb, n + 1);
                v = l1_unscale(n, mant, scale_factors[0][i]);
                s->sb_samples[0][j][i] = v;
                v = l1_unscale(n, mant, scale_factors[1][i]);
                s->sb_samples[1][j][i] = v;
            } else {
                s->sb_samples[0][j][i] = 0;
                s->sb_samples[1][j][i] = 0;
            }
        }
    }
    return 12;
}

/* bitrate is in kb/s */
int l2_select_table(int bitrate, int nb_channels, int freq, int lsf)
{
    int ch_bitrate, table;
1308

1309 1310 1311
    ch_bitrate = bitrate / nb_channels;
    if (!lsf) {
        if ((freq == 48000 && ch_bitrate >= 56) ||
1312
            (ch_bitrate >= 56 && ch_bitrate <= 80))
1313
            table = 0;
1314
        else if (freq != 48000 && ch_bitrate >= 96)
1315
            table = 1;
1316
        else if (freq != 32000 && ch_bitrate <= 48)
1317
            table = 2;
1318
        else
1319 1320 1321 1322 1323 1324
            table = 3;
    } else {
        table = 4;
    }
    return table;
}
F
Fabrice Bellard 已提交
1325

1326 1327 1328 1329 1330 1331 1332 1333 1334
static int mp_decode_layer2(MPADecodeContext *s)
{
    int sblimit; /* number of used subbands */
    const unsigned char *alloc_table;
    int table, bit_alloc_bits, i, j, ch, bound, v;
    unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
    unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
    unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
    int scale, qindex, bits, steps, k, l, m, b;
F
Fabrice Bellard 已提交
1335

1336
    /* select decoding table */
1337
    table = l2_select_table(s->bit_rate / 1000, s->nb_channels,
1338 1339 1340 1341
                            s->sample_rate, s->lsf);
    sblimit = sblimit_table[table];
    alloc_table = alloc_tables[table];

1342
    if (s->mode == MPA_JSTEREO)
1343 1344 1345 1346 1347
        bound = (s->mode_ext + 1) * 4;
    else
        bound = sblimit;

    dprintf("bound=%d sblimit=%d\n", bound, sblimit);
1348 1349 1350 1351

    /* sanity check */
    if( bound > sblimit ) bound = sblimit;

1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366
    /* parse bit allocation */
    j = 0;
    for(i=0;i<bound;i++) {
        bit_alloc_bits = alloc_table[j];
        for(ch=0;ch<s->nb_channels;ch++) {
            bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
        }
        j += 1 << bit_alloc_bits;
    }
    for(i=bound;i<sblimit;i++) {
        bit_alloc_bits = alloc_table[j];
        v = get_bits(&s->gb, bit_alloc_bits);
        bit_alloc[0][i] = v;
        bit_alloc[1][i] = v;
        j += 1 << bit_alloc_bits;
F
Fabrice Bellard 已提交
1367
    }
1368 1369 1370 1371 1372

#ifdef DEBUG
    {
        for(ch=0;ch<s->nb_channels;ch++) {
            for(i=0;i<sblimit;i++)
S
Steve L'Homme 已提交
1373 1374
                dprintf(" %d", bit_alloc[ch][i]);
            dprintf("\n");
1375 1376 1377 1378 1379 1380 1381
        }
    }
#endif

    /* scale codes */
    for(i=0;i<sblimit;i++) {
        for(ch=0;ch<s->nb_channels;ch++) {
1382
            if (bit_alloc[ch][i])
1383 1384 1385
                scale_code[ch][i] = get_bits(&s->gb, 2);
        }
    }
1386

1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423
    /* scale factors */
    for(i=0;i<sblimit;i++) {
        for(ch=0;ch<s->nb_channels;ch++) {
            if (bit_alloc[ch][i]) {
                sf = scale_factors[ch][i];
                switch(scale_code[ch][i]) {
                default:
                case 0:
                    sf[0] = get_bits(&s->gb, 6);
                    sf[1] = get_bits(&s->gb, 6);
                    sf[2] = get_bits(&s->gb, 6);
                    break;
                case 2:
                    sf[0] = get_bits(&s->gb, 6);
                    sf[1] = sf[0];
                    sf[2] = sf[0];
                    break;
                case 1:
                    sf[0] = get_bits(&s->gb, 6);
                    sf[2] = get_bits(&s->gb, 6);
                    sf[1] = sf[0];
                    break;
                case 3:
                    sf[0] = get_bits(&s->gb, 6);
                    sf[2] = get_bits(&s->gb, 6);
                    sf[1] = sf[2];
                    break;
                }
            }
        }
    }

#ifdef DEBUG
    for(ch=0;ch<s->nb_channels;ch++) {
        for(i=0;i<sblimit;i++) {
            if (bit_alloc[ch][i]) {
                sf = scale_factors[ch][i];
S
Steve L'Homme 已提交
1424
                dprintf(" %d %d %d", sf[0], sf[1], sf[2]);
1425
            } else {
S
Steve L'Homme 已提交
1426
                dprintf(" -");
1427 1428
            }
        }
S
Steve L'Homme 已提交
1429
        dprintf("\n");
1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448
    }
#endif

    /* samples */
    for(k=0;k<3;k++) {
        for(l=0;l<12;l+=3) {
            j = 0;
            for(i=0;i<bound;i++) {
                bit_alloc_bits = alloc_table[j];
                for(ch=0;ch<s->nb_channels;ch++) {
                    b = bit_alloc[ch][i];
                    if (b) {
                        scale = scale_factors[ch][i][k];
                        qindex = alloc_table[j+b];
                        bits = quant_bits[qindex];
                        if (bits < 0) {
                            /* 3 values at the same time */
                            v = get_bits(&s->gb, -bits);
                            steps = quant_steps[qindex];
1449
                            s->sb_samples[ch][k * 12 + l + 0][i] =
1450 1451
                                l2_unscale_group(steps, v % steps, scale);
                            v = v / steps;
1452
                            s->sb_samples[ch][k * 12 + l + 1][i] =
1453 1454
                                l2_unscale_group(steps, v % steps, scale);
                            v = v / steps;
1455
                            s->sb_samples[ch][k * 12 + l + 2][i] =
1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470
                                l2_unscale_group(steps, v, scale);
                        } else {
                            for(m=0;m<3;m++) {
                                v = get_bits(&s->gb, bits);
                                v = l1_unscale(bits - 1, v, scale);
                                s->sb_samples[ch][k * 12 + l + m][i] = v;
                            }
                        }
                    } else {
                        s->sb_samples[ch][k * 12 + l + 0][i] = 0;
                        s->sb_samples[ch][k * 12 + l + 1][i] = 0;
                        s->sb_samples[ch][k * 12 + l + 2][i] = 0;
                    }
                }
                /* next subband in alloc table */
1471
                j += 1 << bit_alloc_bits;
1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488
            }
            /* XXX: find a way to avoid this duplication of code */
            for(i=bound;i<sblimit;i++) {
                bit_alloc_bits = alloc_table[j];
                b = bit_alloc[0][i];
                if (b) {
                    int mant, scale0, scale1;
                    scale0 = scale_factors[0][i][k];
                    scale1 = scale_factors[1][i][k];
                    qindex = alloc_table[j+b];
                    bits = quant_bits[qindex];
                    if (bits < 0) {
                        /* 3 values at the same time */
                        v = get_bits(&s->gb, -bits);
                        steps = quant_steps[qindex];
                        mant = v % steps;
                        v = v / steps;
1489
                        s->sb_samples[0][k * 12 + l + 0][i] =
1490
                            l2_unscale_group(steps, mant, scale0);
1491
                        s->sb_samples[1][k * 12 + l + 0][i] =
1492 1493 1494
                            l2_unscale_group(steps, mant, scale1);
                        mant = v % steps;
                        v = v / steps;
1495
                        s->sb_samples[0][k * 12 + l + 1][i] =
1496
                            l2_unscale_group(steps, mant, scale0);
1497
                        s->sb_samples[1][k * 12 + l + 1][i] =
1498
                            l2_unscale_group(steps, mant, scale1);
1499
                        s->sb_samples[0][k * 12 + l + 2][i] =
1500
                            l2_unscale_group(steps, v, scale0);
1501
                        s->sb_samples[1][k * 12 + l + 2][i] =
1502 1503 1504 1505
                            l2_unscale_group(steps, v, scale1);
                    } else {
                        for(m=0;m<3;m++) {
                            mant = get_bits(&s->gb, bits);
1506
                            s->sb_samples[0][k * 12 + l + m][i] =
1507
                                l1_unscale(bits - 1, mant, scale0);
1508
                            s->sb_samples[1][k * 12 + l + m][i] =
1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520
                                l1_unscale(bits - 1, mant, scale1);
                        }
                    }
                } else {
                    s->sb_samples[0][k * 12 + l + 0][i] = 0;
                    s->sb_samples[0][k * 12 + l + 1][i] = 0;
                    s->sb_samples[0][k * 12 + l + 2][i] = 0;
                    s->sb_samples[1][k * 12 + l + 0][i] = 0;
                    s->sb_samples[1][k * 12 + l + 1][i] = 0;
                    s->sb_samples[1][k * 12 + l + 2][i] = 0;
                }
                /* next subband in alloc table */
1521
                j += 1 << bit_alloc_bits;
1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533
            }
            /* fill remaining samples to zero */
            for(i=sblimit;i<SBLIMIT;i++) {
                for(ch=0;ch<s->nb_channels;ch++) {
                    s->sb_samples[ch][k * 12 + l + 0][i] = 0;
                    s->sb_samples[ch][k * 12 + l + 1][i] = 0;
                    s->sb_samples[ch][k * 12 + l + 2][i] = 0;
                }
            }
        }
    }
    return 3 * 12;
F
Fabrice Bellard 已提交
1534 1535 1536
}

/*
1537
 * Seek back in the stream for backstep bytes (at most 511 bytes)
F
Fabrice Bellard 已提交
1538
 */
1539
static void seek_to_maindata(MPADecodeContext *s, unsigned int backstep)
F
Fabrice Bellard 已提交
1540
{
1541
    uint8_t *ptr;
F
Fabrice Bellard 已提交
1542 1543

    /* compute current position in stream */
F
Fabrice Bellard 已提交
1544
    ptr = (uint8_t *)(s->gb.buffer + (get_bits_count(&s->gb)>>3));
1545

F
Fabrice Bellard 已提交
1546 1547
    /* copy old data before current one */
    ptr -= backstep;
1548
    memcpy(ptr, s->inbuf1[s->inbuf_index ^ 1] +
1549
           BACKSTEP_SIZE + s->old_frame_size - backstep, backstep);
F
Fabrice Bellard 已提交
1550
    /* init get bits again */
1551
    init_get_bits(&s->gb, ptr, (s->frame_size + backstep)*8);
F
Fabrice Bellard 已提交
1552

1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578
    /* prepare next buffer */
    s->inbuf_index ^= 1;
    s->inbuf = &s->inbuf1[s->inbuf_index][BACKSTEP_SIZE];
    s->old_frame_size = s->frame_size;
}

static inline void lsf_sf_expand(int *slen,
                                 int sf, int n1, int n2, int n3)
{
    if (n3) {
        slen[3] = sf % n3;
        sf /= n3;
    } else {
        slen[3] = 0;
    }
    if (n2) {
        slen[2] = sf % n2;
        sf /= n2;
    } else {
        slen[2] = 0;
    }
    slen[1] = sf % n1;
    sf /= n1;
    slen[0] = sf;
}

1579
static void exponents_from_scale_factors(MPADecodeContext *s,
1580
                                         GranuleDef *g,
1581
                                         int16_t *exponents)
1582
{
1583
    const uint8_t *bstab, *pretab;
1584
    int len, i, j, k, l, v0, shift, gain, gains[3];
1585
    int16_t *exp_ptr;
1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626

    exp_ptr = exponents;
    gain = g->global_gain - 210;
    shift = g->scalefac_scale + 1;

    bstab = band_size_long[s->sample_rate_index];
    pretab = mpa_pretab[g->preflag];
    for(i=0;i<g->long_end;i++) {
        v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift);
        len = bstab[i];
        for(j=len;j>0;j--)
            *exp_ptr++ = v0;
    }

    if (g->short_start < 13) {
        bstab = band_size_short[s->sample_rate_index];
        gains[0] = gain - (g->subblock_gain[0] << 3);
        gains[1] = gain - (g->subblock_gain[1] << 3);
        gains[2] = gain - (g->subblock_gain[2] << 3);
        k = g->long_end;
        for(i=g->short_start;i<13;i++) {
            len = bstab[i];
            for(l=0;l<3;l++) {
                v0 = gains[l] - (g->scale_factors[k++] << shift);
                for(j=len;j>0;j--)
                *exp_ptr++ = v0;
            }
        }
    }
}

/* handle n = 0 too */
static inline int get_bitsz(GetBitContext *s, int n)
{
    if (n == 0)
        return 0;
    else
        return get_bits(s, n);
}

static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
1627
                          int16_t *exponents, int end_pos)
1628 1629 1630
{
    int s_index;
    int linbits, code, x, y, l, v, i, j, k, pos;
1631
    GetBitContext last_gb;
1632
    VLC *vlc;
1633
    uint8_t *code_table;
1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652

    /* low frequencies (called big values) */
    s_index = 0;
    for(i=0;i<3;i++) {
        j = g->region_size[i];
        if (j == 0)
            continue;
        /* select vlc table */
        k = g->table_select[i];
        l = mpa_huff_data[k][0];
        linbits = mpa_huff_data[k][1];
        vlc = &huff_vlc[l];
        code_table = huff_code_table[l];

        /* read huffcode and compute each couple */
        for(;j>0;j--) {
            if (get_bits_count(&s->gb) >= end_pos)
                break;
            if (code_table) {
1653
                code = get_vlc2(&s->gb, vlc->table, 8, 3);
1654 1655 1656 1657 1658 1659 1660 1661 1662
                if (code < 0)
                    return -1;
                y = code_table[code];
                x = y >> 4;
                y = y & 0x0f;
            } else {
                x = 0;
                y = 0;
            }
1663
            dprintf("region=%d n=%d x=%d y=%d exp=%d\n",
1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686
                    i, g->region_size[i] - j, x, y, exponents[s_index]);
            if (x) {
                if (x == 15)
                    x += get_bitsz(&s->gb, linbits);
                v = l3_unscale(x, exponents[s_index]);
                if (get_bits1(&s->gb))
                    v = -v;
            } else {
                v = 0;
            }
            g->sb_hybrid[s_index++] = v;
            if (y) {
                if (y == 15)
                    y += get_bitsz(&s->gb, linbits);
                v = l3_unscale(y, exponents[s_index]);
                if (get_bits1(&s->gb))
                    v = -v;
            } else {
                v = 0;
            }
            g->sb_hybrid[s_index++] = v;
        }
    }
1687

1688 1689
    /* high frequencies */
    vlc = &huff_quad_vlc[g->count1table_select];
1690
    last_gb.buffer = NULL;
1691 1692 1693
    while (s_index <= 572) {
        pos = get_bits_count(&s->gb);
        if (pos >= end_pos) {
1694
            if (pos > end_pos && last_gb.buffer != NULL) {
1695 1696 1697
                /* some encoders generate an incorrect size for this
                   part. We must go back into the data */
                s_index -= 4;
1698
                s->gb = last_gb;
1699 1700 1701
            }
            break;
        }
1702 1703
        last_gb= s->gb;

1704
        code = get_vlc2(&s->gb, vlc->table, vlc->bits, 2);
1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722
        dprintf("t=%d code=%d\n", g->count1table_select, code);
        if (code < 0)
            return -1;
        for(i=0;i<4;i++) {
            if (code & (8 >> i)) {
                /* non zero value. Could use a hand coded function for
                   'one' value */
                v = l3_unscale(1, exponents[s_index]);
                if(get_bits1(&s->gb))
                    v = -v;
            } else {
                v = 0;
            }
            g->sb_hybrid[s_index++] = v;
        }
    }
    while (s_index < 576)
        g->sb_hybrid[s_index++] = 0;
F
Fabrice Bellard 已提交
1723 1724 1725
    return 0;
}

1726 1727 1728 1729 1730 1731
/* Reorder short blocks from bitstream order to interleaved order. It
   would be faster to do it in parsing, but the code would be far more
   complicated */
static void reorder_block(MPADecodeContext *s, GranuleDef *g)
{
    int i, j, k, len;
1732 1733
    int32_t *ptr, *dst, *ptr1;
    int32_t tmp[576];
1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746

    if (g->block_type != 2)
        return;

    if (g->switch_point) {
        if (s->sample_rate_index != 8) {
            ptr = g->sb_hybrid + 36;
        } else {
            ptr = g->sb_hybrid + 48;
        }
    } else {
        ptr = g->sb_hybrid;
    }
1747

1748 1749 1750 1751 1752 1753 1754 1755 1756 1757
    for(i=g->short_start;i<13;i++) {
        len = band_size_short[s->sample_rate_index][i];
        ptr1 = ptr;
        for(k=0;k<3;k++) {
            dst = tmp + k;
            for(j=len;j>0;j--) {
                *dst = *ptr++;
                dst += 3;
            }
        }
1758
        memcpy(ptr1, tmp, len * 3 * sizeof(int32_t));
1759 1760 1761 1762 1763 1764 1765 1766 1767
    }
}

#define ISQRT2 FIXR(0.70710678118654752440)

static void compute_stereo(MPADecodeContext *s,
                           GranuleDef *g0, GranuleDef *g1)
{
    int i, j, k, l;
1768
    int32_t v1, v2;
1769
    int sf_max, tmp0, tmp1, sf, len, non_zero_found;
1770 1771
    int32_t (*is_tab)[16];
    int32_t *tab0, *tab1;
1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782
    int non_zero_found_short[3];

    /* intensity stereo */
    if (s->mode_ext & MODE_EXT_I_STEREO) {
        if (!s->lsf) {
            is_tab = is_table;
            sf_max = 7;
        } else {
            is_tab = is_table_lsf[g1->scalefac_compress & 1];
            sf_max = 16;
        }
1783

1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833
        tab0 = g0->sb_hybrid + 576;
        tab1 = g1->sb_hybrid + 576;

        non_zero_found_short[0] = 0;
        non_zero_found_short[1] = 0;
        non_zero_found_short[2] = 0;
        k = (13 - g1->short_start) * 3 + g1->long_end - 3;
        for(i = 12;i >= g1->short_start;i--) {
            /* for last band, use previous scale factor */
            if (i != 11)
                k -= 3;
            len = band_size_short[s->sample_rate_index][i];
            for(l=2;l>=0;l--) {
                tab0 -= len;
                tab1 -= len;
                if (!non_zero_found_short[l]) {
                    /* test if non zero band. if so, stop doing i-stereo */
                    for(j=0;j<len;j++) {
                        if (tab1[j] != 0) {
                            non_zero_found_short[l] = 1;
                            goto found1;
                        }
                    }
                    sf = g1->scale_factors[k + l];
                    if (sf >= sf_max)
                        goto found1;

                    v1 = is_tab[0][sf];
                    v2 = is_tab[1][sf];
                    for(j=0;j<len;j++) {
                        tmp0 = tab0[j];
                        tab0[j] = MULL(tmp0, v1);
                        tab1[j] = MULL(tmp0, v2);
                    }
                } else {
                found1:
                    if (s->mode_ext & MODE_EXT_MS_STEREO) {
                        /* lower part of the spectrum : do ms stereo
                           if enabled */
                        for(j=0;j<len;j++) {
                            tmp0 = tab0[j];
                            tmp1 = tab1[j];
                            tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
                            tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
                        }
                    }
                }
            }
        }

1834 1835
        non_zero_found = non_zero_found_short[0] |
            non_zero_found_short[1] |
1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890
            non_zero_found_short[2];

        for(i = g1->long_end - 1;i >= 0;i--) {
            len = band_size_long[s->sample_rate_index][i];
            tab0 -= len;
            tab1 -= len;
            /* test if non zero band. if so, stop doing i-stereo */
            if (!non_zero_found) {
                for(j=0;j<len;j++) {
                    if (tab1[j] != 0) {
                        non_zero_found = 1;
                        goto found2;
                    }
                }
                /* for last band, use previous scale factor */
                k = (i == 21) ? 20 : i;
                sf = g1->scale_factors[k];
                if (sf >= sf_max)
                    goto found2;
                v1 = is_tab[0][sf];
                v2 = is_tab[1][sf];
                for(j=0;j<len;j++) {
                    tmp0 = tab0[j];
                    tab0[j] = MULL(tmp0, v1);
                    tab1[j] = MULL(tmp0, v2);
                }
            } else {
            found2:
                if (s->mode_ext & MODE_EXT_MS_STEREO) {
                    /* lower part of the spectrum : do ms stereo
                       if enabled */
                    for(j=0;j<len;j++) {
                        tmp0 = tab0[j];
                        tmp1 = tab1[j];
                        tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
                        tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
                    }
                }
            }
        }
    } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
        /* ms stereo ONLY */
        /* NOTE: the 1/sqrt(2) normalization factor is included in the
           global gain */
        tab0 = g0->sb_hybrid;
        tab1 = g1->sb_hybrid;
        for(i=0;i<576;i++) {
            tmp0 = tab0[i];
            tmp1 = tab1[i];
            tab0[i] = tmp0 + tmp1;
            tab1[i] = tmp0 - tmp1;
        }
    }
}

1891
static void compute_antialias_integer(MPADecodeContext *s,
1892 1893
                              GranuleDef *g)
{
M
Michael Niedermayer 已提交
1894 1895
    int32_t *ptr, *csa;
    int n, i;
1896 1897 1898 1899 1900 1901 1902 1903 1904 1905

    /* we antialias only "long" bands */
    if (g->block_type == 2) {
        if (!g->switch_point)
            return;
        /* XXX: check this for 8000Hz case */
        n = 1;
    } else {
        n = SBLIMIT - 1;
    }
1906

1907 1908
    ptr = g->sb_hybrid + 18;
    for(i = n;i > 0;i--) {
M
Michael Niedermayer 已提交
1909 1910 1911
        int tmp0, tmp1, tmp2;
        csa = &csa_table[0][0];
#define INT_AA(j) \
1912 1913
            tmp0 = ptr[-1-j];\
            tmp1 = ptr[   j];\
M
Michael Niedermayer 已提交
1914
            tmp2= MULH(tmp0 + tmp1, csa[0+4*j]);\
1915 1916
            ptr[-1-j] = 4*(tmp2 - MULH(tmp1, csa[2+4*j]));\
            ptr[   j] = 4*(tmp2 + MULH(tmp0, csa[3+4*j]));
M
Michael Niedermayer 已提交
1917 1918 1919 1920 1921 1922 1923 1924 1925

        INT_AA(0)
        INT_AA(1)
        INT_AA(2)
        INT_AA(3)
        INT_AA(4)
        INT_AA(5)
        INT_AA(6)
        INT_AA(7)
1926 1927

        ptr += 18;
1928 1929 1930 1931 1932 1933
    }
}

static void compute_antialias_float(MPADecodeContext *s,
                              GranuleDef *g)
{
M
Michael Niedermayer 已提交
1934 1935
    int32_t *ptr;
    int n, i;
1936 1937 1938 1939 1940 1941 1942 1943 1944 1945

    /* we antialias only "long" bands */
    if (g->block_type == 2) {
        if (!g->switch_point)
            return;
        /* XXX: check this for 8000Hz case */
        n = 1;
    } else {
        n = SBLIMIT - 1;
    }
1946

1947 1948
    ptr = g->sb_hybrid + 18;
    for(i = n;i > 0;i--) {
M
Michael Niedermayer 已提交
1949
        float tmp0, tmp1;
1950
        float *csa = &csa_table_float[0][0];
M
Michael Niedermayer 已提交
1951 1952 1953 1954 1955
#define FLOAT_AA(j)\
        tmp0= ptr[-1-j];\
        tmp1= ptr[   j];\
        ptr[-1-j] = lrintf(tmp0 * csa[0+4*j] - tmp1 * csa[1+4*j]);\
        ptr[   j] = lrintf(tmp0 * csa[1+4*j] + tmp1 * csa[0+4*j]);
1956

M
Michael Niedermayer 已提交
1957 1958 1959 1960 1961 1962 1963 1964 1965
        FLOAT_AA(0)
        FLOAT_AA(1)
        FLOAT_AA(2)
        FLOAT_AA(3)
        FLOAT_AA(4)
        FLOAT_AA(5)
        FLOAT_AA(6)
        FLOAT_AA(7)

1966
        ptr += 18;
1967 1968 1969 1970
    }
}

static void compute_imdct(MPADecodeContext *s,
1971
                          GranuleDef *g,
1972 1973
                          int32_t *sb_samples,
                          int32_t *mdct_buf)
1974
{
M
Michael Niedermayer 已提交
1975
    int32_t *ptr, *win, *win1, *buf, *out_ptr, *ptr1;
1976
    int32_t out2[12];
M
Michael Niedermayer 已提交
1977
    int i, j, mdct_long_end, v, sblimit;
1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

    /* find last non zero block */
    ptr = g->sb_hybrid + 576;
    ptr1 = g->sb_hybrid + 2 * 18;
    while (ptr >= ptr1) {
        ptr -= 6;
        v = ptr[0] | ptr[1] | ptr[2] | ptr[3] | ptr[4] | ptr[5];
        if (v != 0)
            break;
    }
    sblimit = ((ptr - g->sb_hybrid) / 18) + 1;

    if (g->block_type == 2) {
        /* XXX: check for 8000 Hz */
        if (g->switch_point)
            mdct_long_end = 2;
        else
            mdct_long_end = 0;
    } else {
        mdct_long_end = sblimit;
    }

    buf = mdct_buf;
    ptr = g->sb_hybrid;
    for(j=0;j<mdct_long_end;j++) {
        /* apply window & overlap with previous buffer */
        out_ptr = sb_samples + j;
        /* select window */
        if (g->switch_point && j < 2)
            win1 = mdct_win[0];
        else
            win1 = mdct_win[g->block_type];
        /* select frequency inversion */
        win = win1 + ((4 * 36) & -(j & 1));
M
Michael Niedermayer 已提交
2012 2013
        imdct36(out_ptr, buf, ptr, win);
        out_ptr += 18*SBLIMIT;
2014 2015 2016 2017 2018 2019 2020
        ptr += 18;
        buf += 18;
    }
    for(j=mdct_long_end;j<sblimit;j++) {
        /* select frequency inversion */
        win = mdct_win[2] + ((4 * 36) & -(j & 1));
        out_ptr = sb_samples + j;
2021

M
Michael Niedermayer 已提交
2022 2023 2024 2025 2026 2027 2028 2029
        for(i=0; i<6; i++){
            *out_ptr = buf[i];
            out_ptr += SBLIMIT;
        }
        imdct12(out2, ptr + 0);
        for(i=0;i<6;i++) {
            *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*1];
            buf[i + 6*2] = MULH(out2[i + 6], win[i + 6]);
2030 2031
            out_ptr += SBLIMIT;
        }
M
Michael Niedermayer 已提交
2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043
        imdct12(out2, ptr + 1);
        for(i=0;i<6;i++) {
            *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*2];
            buf[i + 6*0] = MULH(out2[i + 6], win[i + 6]);
            out_ptr += SBLIMIT;
        }
        imdct12(out2, ptr + 2);
        for(i=0;i<6;i++) {
            buf[i + 6*0] = MULH(out2[i], win[i]) + buf[i + 6*0];
            buf[i + 6*1] = MULH(out2[i + 6], win[i + 6]);
            buf[i + 6*2] = 0;
        }
2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059
        ptr += 18;
        buf += 18;
    }
    /* zero bands */
    for(j=sblimit;j<SBLIMIT;j++) {
        /* overlap */
        out_ptr = sb_samples + j;
        for(i=0;i<18;i++) {
            *out_ptr = buf[i];
            buf[i] = 0;
            out_ptr += SBLIMIT;
        }
        buf += 18;
    }
}

F
Fabrice Bellard 已提交
2060
#if defined(DEBUG)
2061
void sample_dump(int fnum, int32_t *tab, int n)
2062 2063 2064
{
    static FILE *files[16], *f;
    char buf[512];
2065
    int i;
2066
    int32_t v;
2067

2068 2069
    f = files[fnum];
    if (!f) {
2070 2071
        snprintf(buf, sizeof(buf), "/tmp/out%d.%s.pcm",
                fnum,
2072 2073 2074 2075 2076 2077
#ifdef USE_HIGHPRECISION
                "hp"
#else
                "lp"
#endif
                );
2078 2079 2080 2081 2082
        f = fopen(buf, "w");
        if (!f)
            return;
        files[fnum] = f;
    }
2083

2084 2085
    if (fnum == 0) {
        static int pos = 0;
2086
        av_log(NULL, AV_LOG_DEBUG, "pos=%d\n", pos);
2087
        for(i=0;i<n;i++) {
2088
            av_log(NULL, AV_LOG_DEBUG, " %0.4f", (double)tab[i] / FRAC_ONE);
2089
            if ((i % 18) == 17)
2090
                av_log(NULL, AV_LOG_DEBUG, "\n");
2091 2092 2093
        }
        pos += n;
    }
2094 2095 2096
    for(i=0;i<n;i++) {
        /* normalize to 23 frac bits */
        v = tab[i] << (23 - FRAC_BITS);
2097
        fwrite(&v, 1, sizeof(int32_t), f);
2098
    }
2099 2100 2101 2102 2103 2104 2105 2106 2107 2108
}
#endif


/* main layer3 decoding function */
static int mp_decode_layer3(MPADecodeContext *s)
{
    int nb_granules, main_data_begin, private_bits;
    int gr, ch, blocksplit_flag, i, j, k, n, bits_pos, bits_left;
    GranuleDef granules[2][2], *g;
2109
    int16_t exponents[576];
2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130

    /* read side info */
    if (s->lsf) {
        main_data_begin = get_bits(&s->gb, 8);
        if (s->nb_channels == 2)
            private_bits = get_bits(&s->gb, 2);
        else
            private_bits = get_bits(&s->gb, 1);
        nb_granules = 1;
    } else {
        main_data_begin = get_bits(&s->gb, 9);
        if (s->nb_channels == 2)
            private_bits = get_bits(&s->gb, 3);
        else
            private_bits = get_bits(&s->gb, 5);
        nb_granules = 2;
        for(ch=0;ch<s->nb_channels;ch++) {
            granules[ch][0].scfsi = 0; /* all scale factors are transmitted */
            granules[ch][1].scfsi = get_bits(&s->gb, 4);
        }
    }
2131

2132 2133 2134 2135 2136 2137 2138 2139 2140
    for(gr=0;gr<nb_granules;gr++) {
        for(ch=0;ch<s->nb_channels;ch++) {
            dprintf("gr=%d ch=%d: side_info\n", gr, ch);
            g = &granules[ch][gr];
            g->part2_3_length = get_bits(&s->gb, 12);
            g->big_values = get_bits(&s->gb, 9);
            g->global_gain = get_bits(&s->gb, 8);
            /* if MS stereo only is selected, we precompute the
               1/sqrt(2) renormalization factor */
2141
            if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155
                MODE_EXT_MS_STEREO)
                g->global_gain -= 2;
            if (s->lsf)
                g->scalefac_compress = get_bits(&s->gb, 9);
            else
                g->scalefac_compress = get_bits(&s->gb, 4);
            blocksplit_flag = get_bits(&s->gb, 1);
            if (blocksplit_flag) {
                g->block_type = get_bits(&s->gb, 2);
                if (g->block_type == 0)
                    return -1;
                g->switch_point = get_bits(&s->gb, 1);
                for(i=0;i<2;i++)
                    g->table_select[i] = get_bits(&s->gb, 5);
2156
                for(i=0;i<3;i++)
2157 2158 2159 2160 2161
                    g->subblock_gain[i] = get_bits(&s->gb, 3);
                /* compute huffman coded region sizes */
                if (g->block_type == 2)
                    g->region_size[0] = (36 / 2);
                else {
2162
                    if (s->sample_rate_index <= 2)
2163
                        g->region_size[0] = (36 / 2);
2164
                    else if (s->sample_rate_index != 8)
2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178
                        g->region_size[0] = (54 / 2);
                    else
                        g->region_size[0] = (108 / 2);
                }
                g->region_size[1] = (576 / 2);
            } else {
                int region_address1, region_address2, l;
                g->block_type = 0;
                g->switch_point = 0;
                for(i=0;i<3;i++)
                    g->table_select[i] = get_bits(&s->gb, 5);
                /* compute huffman coded region sizes */
                region_address1 = get_bits(&s->gb, 4);
                region_address2 = get_bits(&s->gb, 3);
2179
                dprintf("region1=%d region2=%d\n",
2180
                        region_address1, region_address2);
2181
                g->region_size[0] =
2182 2183 2184 2185 2186
                    band_index_long[s->sample_rate_index][region_address1 + 1] >> 1;
                l = region_address1 + region_address2 + 2;
                /* should not overflow */
                if (l > 22)
                    l = 22;
2187
                g->region_size[1] =
2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213
                    band_index_long[s->sample_rate_index][l] >> 1;
            }
            /* convert region offsets to region sizes and truncate
               size to big_values */
            g->region_size[2] = (576 / 2);
            j = 0;
            for(i=0;i<3;i++) {
                k = g->region_size[i];
                if (k > g->big_values)
                    k = g->big_values;
                g->region_size[i] = k - j;
                j = k;
            }

            /* compute band indexes */
            if (g->block_type == 2) {
                if (g->switch_point) {
                    /* if switched mode, we handle the 36 first samples as
                       long blocks.  For 8000Hz, we handle the 48 first
                       exponents as long blocks (XXX: check this!) */
                    if (s->sample_rate_index <= 2)
                        g->long_end = 8;
                    else if (s->sample_rate_index != 8)
                        g->long_end = 6;
                    else
                        g->long_end = 4; /* 8000 Hz */
2214

2215 2216 2217
                    if (s->sample_rate_index != 8)
                        g->short_start = 3;
                    else
2218
                        g->short_start = 2;
2219 2220 2221 2222 2223 2224 2225 2226
                } else {
                    g->long_end = 0;
                    g->short_start = 0;
                }
            } else {
                g->short_start = 13;
                g->long_end = 22;
            }
2227

2228 2229 2230 2231 2232 2233 2234 2235 2236 2237
            g->preflag = 0;
            if (!s->lsf)
                g->preflag = get_bits(&s->gb, 1);
            g->scalefac_scale = get_bits(&s->gb, 1);
            g->count1table_select = get_bits(&s->gb, 1);
            dprintf("block_type=%d switch_point=%d\n",
                    g->block_type, g->switch_point);
        }
    }

R
Roberto Togni 已提交
2238
  if (!s->adu_mode) {
2239 2240 2241
    /* now we get bits from the main_data_begin offset */
    dprintf("seekback: %d\n", main_data_begin);
    seek_to_maindata(s, main_data_begin);
R
Roberto Togni 已提交
2242
  }
2243 2244 2245 2246

    for(gr=0;gr<nb_granules;gr++) {
        for(ch=0;ch<s->nb_channels;ch++) {
            g = &granules[ch][gr];
2247

2248
            bits_pos = get_bits_count(&s->gb);
2249

2250
            if (!s->lsf) {
2251
                uint8_t *sc;
2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285
                int slen, slen1, slen2;

                /* MPEG1 scale factors */
                slen1 = slen_table[0][g->scalefac_compress];
                slen2 = slen_table[1][g->scalefac_compress];
                dprintf("slen1=%d slen2=%d\n", slen1, slen2);
                if (g->block_type == 2) {
                    n = g->switch_point ? 17 : 18;
                    j = 0;
                    for(i=0;i<n;i++)
                        g->scale_factors[j++] = get_bitsz(&s->gb, slen1);
                    for(i=0;i<18;i++)
                        g->scale_factors[j++] = get_bitsz(&s->gb, slen2);
                    for(i=0;i<3;i++)
                        g->scale_factors[j++] = 0;
                } else {
                    sc = granules[ch][0].scale_factors;
                    j = 0;
                    for(k=0;k<4;k++) {
                        n = (k == 0 ? 6 : 5);
                        if ((g->scfsi & (0x8 >> k)) == 0) {
                            slen = (k < 2) ? slen1 : slen2;
                            for(i=0;i<n;i++)
                                g->scale_factors[j++] = get_bitsz(&s->gb, slen);
                        } else {
                            /* simply copy from last granule */
                            for(i=0;i<n;i++) {
                                g->scale_factors[j] = sc[j];
                                j++;
                            }
                        }
                    }
                    g->scale_factors[j++] = 0;
                }
F
Fabrice Bellard 已提交
2286
#if defined(DEBUG)
2287
                {
S
Steve L'Homme 已提交
2288
                    dprintf("scfsi=%x gr=%d ch=%d scale_factors:\n",
2289 2290
                           g->scfsi, gr, ch);
                    for(i=0;i<j;i++)
S
Steve L'Homme 已提交
2291 2292
                        dprintf(" %d", g->scale_factors[i]);
                    dprintf("\n");
2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342
                }
#endif
            } else {
                int tindex, tindex2, slen[4], sl, sf;

                /* LSF scale factors */
                if (g->block_type == 2) {
                    tindex = g->switch_point ? 2 : 1;
                } else {
                    tindex = 0;
                }
                sf = g->scalefac_compress;
                if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
                    /* intensity stereo case */
                    sf >>= 1;
                    if (sf < 180) {
                        lsf_sf_expand(slen, sf, 6, 6, 0);
                        tindex2 = 3;
                    } else if (sf < 244) {
                        lsf_sf_expand(slen, sf - 180, 4, 4, 0);
                        tindex2 = 4;
                    } else {
                        lsf_sf_expand(slen, sf - 244, 3, 0, 0);
                        tindex2 = 5;
                    }
                } else {
                    /* normal case */
                    if (sf < 400) {
                        lsf_sf_expand(slen, sf, 5, 4, 4);
                        tindex2 = 0;
                    } else if (sf < 500) {
                        lsf_sf_expand(slen, sf - 400, 5, 4, 0);
                        tindex2 = 1;
                    } else {
                        lsf_sf_expand(slen, sf - 500, 3, 0, 0);
                        tindex2 = 2;
                        g->preflag = 1;
                    }
                }

                j = 0;
                for(k=0;k<4;k++) {
                    n = lsf_nsf_table[tindex2][tindex][k];
                    sl = slen[k];
                    for(i=0;i<n;i++)
                        g->scale_factors[j++] = get_bitsz(&s->gb, sl);
                }
                /* XXX: should compute exact size */
                for(;j<40;j++)
                    g->scale_factors[j] = 0;
F
Fabrice Bellard 已提交
2343
#if defined(DEBUG)
2344
                {
S
Steve L'Homme 已提交
2345
                    dprintf("gr=%d ch=%d scale_factors:\n",
2346 2347
                           gr, ch);
                    for(i=0;i<40;i++)
S
Steve L'Homme 已提交
2348 2349
                        dprintf(" %d", g->scale_factors[i]);
                    dprintf("\n");
2350 2351 2352 2353 2354 2355 2356 2357 2358 2359
                }
#endif
            }

            exponents_from_scale_factors(s, g, exponents);

            /* read Huffman coded residue */
            if (huffman_decode(s, g, exponents,
                               bits_pos + g->part2_3_length) < 0)
                return -1;
F
Fabrice Bellard 已提交
2360 2361
#if defined(DEBUG)
            sample_dump(0, g->sb_hybrid, 576);
2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384
#endif

            /* skip extension bits */
            bits_left = g->part2_3_length - (get_bits_count(&s->gb) - bits_pos);
            if (bits_left < 0) {
                dprintf("bits_left=%d\n", bits_left);
                return -1;
            }
            while (bits_left >= 16) {
                skip_bits(&s->gb, 16);
                bits_left -= 16;
            }
            if (bits_left > 0)
                skip_bits(&s->gb, bits_left);
        } /* ch */

        if (s->nb_channels == 2)
            compute_stereo(s, &granules[0][gr], &granules[1][gr]);

        for(ch=0;ch<s->nb_channels;ch++) {
            g = &granules[ch][gr];

            reorder_block(s, g);
F
Fabrice Bellard 已提交
2385
#if defined(DEBUG)
2386 2387
            sample_dump(0, g->sb_hybrid, 576);
#endif
2388
            s->compute_antialias(s, g);
2389
#if defined(DEBUG)
2390 2391
            sample_dump(1, g->sb_hybrid, 576);
#endif
2392
            compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
2393
#if defined(DEBUG)
2394 2395 2396 2397 2398 2399 2400
            sample_dump(2, &s->sb_samples[ch][18 * gr][0], 576);
#endif
        }
    } /* gr */
    return nb_granules * 18;
}

2401
static int mp_decode_frame(MPADecodeContext *s,
2402
                           OUT_INT *samples)
2403 2404
{
    int i, nb_frames, ch;
2405
    OUT_INT *samples_ptr;
2406

2407
    init_get_bits(&s->gb, s->inbuf + HEADER_SIZE,
2408
                  (s->inbuf_ptr - s->inbuf - HEADER_SIZE)*8);
2409

2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430
    /* skip error protection field */
    if (s->error_protection)
        get_bits(&s->gb, 16);

    dprintf("frame %d:\n", s->frame_count);
    switch(s->layer) {
    case 1:
        nb_frames = mp_decode_layer1(s);
        break;
    case 2:
        nb_frames = mp_decode_layer2(s);
        break;
    case 3:
    default:
        nb_frames = mp_decode_layer3(s);
        break;
    }
#if defined(DEBUG)
    for(i=0;i<nb_frames;i++) {
        for(ch=0;ch<s->nb_channels;ch++) {
            int j;
S
Steve L'Homme 已提交
2431
            dprintf("%d-%d:", i, ch);
2432
            for(j=0;j<SBLIMIT;j++)
S
Steve L'Homme 已提交
2433 2434
                dprintf(" %0.6f", (double)s->sb_samples[ch][i][j] / FRAC_ONE);
            dprintf("\n");
2435 2436 2437 2438 2439 2440 2441
        }
    }
#endif
    /* apply the synthesis filter */
    for(ch=0;ch<s->nb_channels;ch++) {
        samples_ptr = samples + ch;
        for(i=0;i<nb_frames;i++) {
2442
            ff_mpa_synth_filter(s->synth_buf[ch], &(s->synth_buf_offset[ch]),
2443 2444
                         window, &s->dither_state,
                         samples_ptr, s->nb_channels,
2445 2446 2447 2448 2449
                         s->sb_samples[ch][i]);
            samples_ptr += 32 * s->nb_channels;
        }
    }
#ifdef DEBUG
2450
    s->frame_count++;
2451
#endif
2452
    return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
2453 2454
}

F
Fabrice Bellard 已提交
2455
static int decode_frame(AVCodecContext * avctx,
2456 2457
                        void *data, int *data_size,
                        uint8_t * buf, int buf_size)
F
Fabrice Bellard 已提交
2458 2459
{
    MPADecodeContext *s = avctx->priv_data;
2460 2461
    uint32_t header;
    uint8_t *buf_ptr;
F
Fabrice Bellard 已提交
2462
    int len, out_size;
2463
    OUT_INT *out_samples = data;
F
Fabrice Bellard 已提交
2464 2465 2466

    buf_ptr = buf;
    while (buf_size > 0) {
2467 2468
        len = s->inbuf_ptr - s->inbuf;
        if (s->frame_size == 0) {
2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479
            /* special case for next header for first frame in free
               format case (XXX: find a simpler method) */
            if (s->free_format_next_header != 0) {
                s->inbuf[0] = s->free_format_next_header >> 24;
                s->inbuf[1] = s->free_format_next_header >> 16;
                s->inbuf[2] = s->free_format_next_header >> 8;
                s->inbuf[3] = s->free_format_next_header;
                s->inbuf_ptr = s->inbuf + 4;
                s->free_format_next_header = 0;
                goto got_header;
            }
2480
            /* no header seen : find one. We need at least HEADER_SIZE
2481
               bytes to parse it */
2482 2483 2484 2485 2486 2487 2488 2489 2490 2491
            len = HEADER_SIZE - len;
            if (len > buf_size)
                len = buf_size;
            if (len > 0) {
                memcpy(s->inbuf_ptr, buf_ptr, len);
                buf_ptr += len;
                buf_size -= len;
                s->inbuf_ptr += len;
            }
            if ((s->inbuf_ptr - s->inbuf) >= HEADER_SIZE) {
2492
            got_header:
2493 2494
                header = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |
                    (s->inbuf[2] << 8) | s->inbuf[3];
Z
Zdenek Kabelac 已提交
2495

2496 2497 2498 2499
                if (ff_mpa_check_header(header) < 0) {
                    /* no sync found : move by one byte (inefficient, but simple!) */
                    memmove(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1);
                    s->inbuf_ptr--;
2500 2501 2502 2503
                    dprintf("skip %x\n", header);
                    /* reset free format frame size to give a chance
                       to get a new bitrate */
                    s->free_format_frame_size = 0;
2504 2505
                } else {
                    if (decode_header(s, header) == 1) {
2506
                        /* free format: prepare to compute frame size */
2507
                        s->frame_size = -1;
2508
                    }
2509 2510 2511 2512
                    /* update codec info */
                    avctx->sample_rate = s->sample_rate;
                    avctx->channels = s->nb_channels;
                    avctx->bit_rate = s->bit_rate;
2513
                    avctx->sub_id = s->layer;
2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527
                    switch(s->layer) {
                    case 1:
                        avctx->frame_size = 384;
                        break;
                    case 2:
                        avctx->frame_size = 1152;
                        break;
                    case 3:
                        if (s->lsf)
                            avctx->frame_size = 576;
                        else
                            avctx->frame_size = 1152;
                        break;
                    }
2528 2529
                }
            }
2530 2531
        } else if (s->frame_size == -1) {
            /* free format : find next sync to compute frame size */
2532 2533 2534
            len = MPA_MAX_CODED_FRAME_SIZE - len;
            if (len > buf_size)
                len = buf_size;
2535
            if (len == 0) {
2536
                /* frame too long: resync */
2537
                s->frame_size = 0;
2538 2539
                memmove(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1);
                s->inbuf_ptr--;
2540
            } else {
2541 2542
                uint8_t *p, *pend;
                uint32_t header1;
2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570
                int padding;

                memcpy(s->inbuf_ptr, buf_ptr, len);
                /* check for header */
                p = s->inbuf_ptr - 3;
                pend = s->inbuf_ptr + len - 4;
                while (p <= pend) {
                    header = (p[0] << 24) | (p[1] << 16) |
                        (p[2] << 8) | p[3];
                    header1 = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |
                        (s->inbuf[2] << 8) | s->inbuf[3];
                    /* check with high probability that we have a
                       valid header */
                    if ((header & SAME_HEADER_MASK) ==
                        (header1 & SAME_HEADER_MASK)) {
                        /* header found: update pointers */
                        len = (p + 4) - s->inbuf_ptr;
                        buf_ptr += len;
                        buf_size -= len;
                        s->inbuf_ptr = p;
                        /* compute frame size */
                        s->free_format_next_header = header;
                        s->free_format_frame_size = s->inbuf_ptr - s->inbuf;
                        padding = (header1 >> 9) & 1;
                        if (s->layer == 1)
                            s->free_format_frame_size -= padding * 4;
                        else
                            s->free_format_frame_size -= padding;
2571
                        dprintf("free frame size=%d padding=%d\n",
2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582
                                s->free_format_frame_size, padding);
                        decode_header(s, header1);
                        goto next_data;
                    }
                    p++;
                }
                /* not found: simply increase pointers */
                buf_ptr += len;
                s->inbuf_ptr += len;
                buf_size -= len;
            }
2583
        } else if (len < s->frame_size) {
2584 2585
            if (s->frame_size > MPA_MAX_CODED_FRAME_SIZE)
                s->frame_size = MPA_MAX_CODED_FRAME_SIZE;
2586 2587 2588 2589 2590 2591 2592 2593
            len = s->frame_size - len;
            if (len > buf_size)
                len = buf_size;
            memcpy(s->inbuf_ptr, buf_ptr, len);
            buf_ptr += len;
            s->inbuf_ptr += len;
            buf_size -= len;
        }
2594
    next_data:
2595
        if (s->frame_size > 0 &&
2596 2597 2598 2599 2600 2601 2602 2603
            (s->inbuf_ptr - s->inbuf) >= s->frame_size) {
            if (avctx->parse_only) {
                /* simply return the frame data */
                *(uint8_t **)data = s->inbuf;
                out_size = s->inbuf_ptr - s->inbuf;
            } else {
                out_size = mp_decode_frame(s, out_samples);
            }
2604 2605
            s->inbuf_ptr = s->inbuf;
            s->frame_size = 0;
M
Michael Niedermayer 已提交
2606
            if(out_size>=0)
2607
                *data_size = out_size;
M
Michael Niedermayer 已提交
2608 2609
            else
                av_log(avctx, AV_LOG_DEBUG, "Error while decoding mpeg audio frame\n"); //FIXME return -1 / but also return the number of bytes consumed
2610 2611
            break;
        }
F
Fabrice Bellard 已提交
2612 2613 2614 2615
    }
    return buf_ptr - buf;
}

R
Roberto Togni 已提交
2616 2617

static int decode_frame_adu(AVCodecContext * avctx,
2618 2619
                        void *data, int *data_size,
                        uint8_t * buf, int buf_size)
R
Roberto Togni 已提交
2620 2621 2622 2623
{
    MPADecodeContext *s = avctx->priv_data;
    uint32_t header;
    int len, out_size;
2624
    OUT_INT *out_samples = data;
R
Roberto Togni 已提交
2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644

    len = buf_size;

    // Discard too short frames
    if (buf_size < HEADER_SIZE) {
        *data_size = 0;
        return buf_size;
    }


    if (len > MPA_MAX_CODED_FRAME_SIZE)
        len = MPA_MAX_CODED_FRAME_SIZE;

    memcpy(s->inbuf, buf, len);
    s->inbuf_ptr = s->inbuf + len;

    // Get header and restore sync word
    header = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |
              (s->inbuf[2] << 8) | s->inbuf[3] | 0xffe00000;

2645
    if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame
R
Roberto Togni 已提交
2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671
        *data_size = 0;
        return buf_size;
    }

    decode_header(s, header);
    /* update codec info */
    avctx->sample_rate = s->sample_rate;
    avctx->channels = s->nb_channels;
    avctx->bit_rate = s->bit_rate;
    avctx->sub_id = s->layer;

    avctx->frame_size=s->frame_size = len;

    if (avctx->parse_only) {
        /* simply return the frame data */
        *(uint8_t **)data = s->inbuf;
        out_size = s->inbuf_ptr - s->inbuf;
    } else {
        out_size = mp_decode_frame(s, out_samples);
    }

    *data_size = out_size;
    return buf_size;
}


2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749
/* Next 3 arrays are indexed by channel config number (passed via codecdata) */
static int mp3Frames[16] = {0,1,1,2,3,3,4,5,2};   /* number of mp3 decoder instances */
static int mp3Channels[16] = {0,1,2,3,4,5,6,8,4}; /* total output channels */
/* offsets into output buffer, assume output order is FL FR BL BR C LFE */
static int chan_offset[9][5] = {
    {0},
    {0},            // C
    {0},            // FLR
    {2,0},          // C FLR
    {2,0,3},        // C FLR BS
    {4,0,2},        // C FLR BLRS
    {4,0,2,5},      // C FLR BLRS LFE
    {4,0,2,6,5},    // C FLR BLRS BLR LFE
    {0,2}           // FLR BLRS
};


static int decode_init_mp3on4(AVCodecContext * avctx)
{
    MP3On4DecodeContext *s = avctx->priv_data;
    int i;

    if ((avctx->extradata_size < 2) || (avctx->extradata == NULL)) {
        av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
        return -1;
    }

    s->chan_cfg = (((unsigned char *)avctx->extradata)[1] >> 3) & 0x0f;
    s->frames = mp3Frames[s->chan_cfg];
    if(!s->frames) {
        av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
        return -1;
    }
    avctx->channels = mp3Channels[s->chan_cfg];

    /* Init the first mp3 decoder in standard way, so that all tables get builded
     * We replace avctx->priv_data with the context of the first decoder so that
     * decode_init() does not have to be changed.
     * Other decoders will be inited here copying data from the first context
     */
    // Allocate zeroed memory for the first decoder context
    s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
    // Put decoder context in place to make init_decode() happy
    avctx->priv_data = s->mp3decctx[0];
    decode_init(avctx);
    // Restore mp3on4 context pointer
    avctx->priv_data = s;
    s->mp3decctx[0]->adu_mode = 1; // Set adu mode

    /* Create a separate codec/context for each frame (first is already ok).
     * Each frame is 1 or 2 channels - up to 5 frames allowed
     */
    for (i = 1; i < s->frames; i++) {
        s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
        s->mp3decctx[i]->compute_antialias = s->mp3decctx[0]->compute_antialias;
        s->mp3decctx[i]->inbuf = &s->mp3decctx[i]->inbuf1[0][BACKSTEP_SIZE];
        s->mp3decctx[i]->inbuf_ptr = s->mp3decctx[i]->inbuf;
        s->mp3decctx[i]->adu_mode = 1;
    }

    return 0;
}


static int decode_close_mp3on4(AVCodecContext * avctx)
{
    MP3On4DecodeContext *s = avctx->priv_data;
    int i;

    for (i = 0; i < s->frames; i++)
        if (s->mp3decctx[i])
            av_free(s->mp3decctx[i]);

    return 0;
}


static int decode_frame_mp3on4(AVCodecContext * avctx,
2750 2751
                        void *data, int *data_size,
                        uint8_t * buf, int buf_size)
2752 2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835
{
    MP3On4DecodeContext *s = avctx->priv_data;
    MPADecodeContext *m;
    int len, out_size = 0;
    uint32_t header;
    OUT_INT *out_samples = data;
    OUT_INT decoded_buf[MPA_FRAME_SIZE * MPA_MAX_CHANNELS];
    OUT_INT *outptr, *bp;
    int fsize;
    unsigned char *start2 = buf, *start;
    int fr, i, j, n;
    int off = avctx->channels;
    int *coff = chan_offset[s->chan_cfg];

    len = buf_size;

    // Discard too short frames
    if (buf_size < HEADER_SIZE) {
        *data_size = 0;
        return buf_size;
    }

    // If only one decoder interleave is not needed
    outptr = s->frames == 1 ? out_samples : decoded_buf;

    for (fr = 0; fr < s->frames; fr++) {
        start = start2;
        fsize = (start[0] << 4) | (start[1] >> 4);
        start2 += fsize;
        if (fsize > len)
            fsize = len;
        len -= fsize;
        if (fsize > MPA_MAX_CODED_FRAME_SIZE)
            fsize = MPA_MAX_CODED_FRAME_SIZE;
        m = s->mp3decctx[fr];
        assert (m != NULL);
        /* copy original to new */
        m->inbuf_ptr = m->inbuf + fsize;
        memcpy(m->inbuf, start, fsize);

        // Get header
        header = (m->inbuf[0] << 24) | (m->inbuf[1] << 16) |
                  (m->inbuf[2] << 8) | m->inbuf[3] | 0xfff00000;

        if (ff_mpa_check_header(header) < 0) { // Bad header, discard block
            *data_size = 0;
            return buf_size;
        }

        decode_header(m, header);
        mp_decode_frame(m, decoded_buf);

        n = MPA_FRAME_SIZE * m->nb_channels;
        out_size += n * sizeof(OUT_INT);
        if(s->frames > 1) {
            /* interleave output data */
            bp = out_samples + coff[fr];
            if(m->nb_channels == 1) {
                for(j = 0; j < n; j++) {
                    *bp = decoded_buf[j];
                    bp += off;
                }
            } else {
                for(j = 0; j < n; j++) {
                    bp[0] = decoded_buf[j++];
                    bp[1] = decoded_buf[j];
                    bp += off;
                }
            }
        }
    }

    /* update codec info */
    avctx->sample_rate = s->mp3decctx[0]->sample_rate;
    avctx->frame_size= buf_size;
    avctx->bit_rate = 0;
    for (i = 0; i < s->frames; i++)
        avctx->bit_rate += s->mp3decctx[i]->bit_rate;

    *data_size = out_size;
    return buf_size;
}


2836
AVCodec mp2_decoder =
F
Fabrice Bellard 已提交
2837
{
2838
    "mp2",
F
Fabrice Bellard 已提交
2839 2840 2841 2842 2843 2844 2845
    CODEC_TYPE_AUDIO,
    CODEC_ID_MP2,
    sizeof(MPADecodeContext),
    decode_init,
    NULL,
    NULL,
    decode_frame,
2846
    CODEC_CAP_PARSE_ONLY,
F
Fabrice Bellard 已提交
2847
};
2848 2849 2850 2851 2852

AVCodec mp3_decoder =
{
    "mp3",
    CODEC_TYPE_AUDIO,
F
Fabrice Bellard 已提交
2853
    CODEC_ID_MP3,
2854 2855 2856 2857 2858
    sizeof(MPADecodeContext),
    decode_init,
    NULL,
    NULL,
    decode_frame,
2859
    CODEC_CAP_PARSE_ONLY,
2860
};
R
Roberto Togni 已提交
2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873

AVCodec mp3adu_decoder =
{
    "mp3adu",
    CODEC_TYPE_AUDIO,
    CODEC_ID_MP3ADU,
    sizeof(MPADecodeContext),
    decode_init,
    NULL,
    NULL,
    decode_frame_adu,
    CODEC_CAP_PARSE_ONLY,
};
2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886

AVCodec mp3on4_decoder =
{
    "mp3on4",
    CODEC_TYPE_AUDIO,
    CODEC_ID_MP3ON4,
    sizeof(MP3On4DecodeContext),
    decode_init_mp3on4,
    NULL,
    decode_close_mp3on4,
    decode_frame_mp3on4,
    0
};