Statistics
| Branch: | Revision:

ffmpeg / libavcodec / mpegaudiodec.c @ fe037229

History | View | Annotate | Download (85.4 KB)

1
/*
2
 * MPEG Audio decoder
3
 * Copyright (c) 2001, 2002 Fabrice Bellard.
4
 *
5
 * This file is part of FFmpeg.
6
 *
7
 * FFmpeg is free software; you can redistribute it and/or
8
 * modify it under the terms of the GNU Lesser General Public
9
 * License as published by the Free Software Foundation; either
10
 * version 2.1 of the License, or (at your option) any later version.
11
 *
12
 * FFmpeg is distributed in the hope that it will be useful,
13
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15
 * Lesser General Public License for more details.
16
 *
17
 * You should have received a copy of the GNU Lesser General Public
18
 * License along with FFmpeg; if not, write to the Free Software
19
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20
 */
21

    
22
/**
23
 * @file mpegaudiodec.c
24
 * MPEG Audio decoder.
25
 */
26

    
27
//#define DEBUG
28
#include "avcodec.h"
29
#include "bitstream.h"
30
#include "dsputil.h"
31

    
32
/*
33
 * TODO:
34
 *  - in low precision mode, use more 16 bit multiplies in synth filter
35
 *  - test lsf / mpeg25 extensively.
36
 */
37

    
38
/* define USE_HIGHPRECISION to have a bit exact (but slower) mpeg
39
   audio decoder */
40
#ifdef CONFIG_MPEGAUDIO_HP
41
#   define USE_HIGHPRECISION
42
#endif
43

    
44
#include "mpegaudio.h"
45

    
46
#include "mathops.h"
47

    
48
#define FRAC_ONE    (1 << FRAC_BITS)
49

    
50
#define FIX(a)   ((int)((a) * FRAC_ONE))
51
/* WARNING: only correct for posititive numbers */
52
#define FIXR(a)   ((int)((a) * FRAC_ONE + 0.5))
53
#define FRAC_RND(a) (((a) + (FRAC_ONE/2)) >> FRAC_BITS)
54

    
55
#define FIXHR(a) ((int)((a) * (1LL<<32) + 0.5))
56

    
57
/****************/
58

    
59
#define HEADER_SIZE 4
60
#define BACKSTEP_SIZE 512
61
#define EXTRABYTES 24
62

    
63
struct GranuleDef;
64

    
65
typedef struct MPADecodeContext {
66
    DECLARE_ALIGNED_8(uint8_t, last_buf[2*BACKSTEP_SIZE + EXTRABYTES]);
67
    int last_buf_size;
68
    int frame_size;
69
    /* next header (used in free format parsing) */
70
    uint32_t free_format_next_header;
71
    int error_protection;
72
    int layer;
73
    int sample_rate;
74
    int sample_rate_index; /* between 0 and 8 */
75
    int bit_rate;
76
    GetBitContext gb;
77
    GetBitContext in_gb;
78
    int nb_channels;
79
    int mode;
80
    int mode_ext;
81
    int lsf;
82
    MPA_INT synth_buf[MPA_MAX_CHANNELS][512 * 2] __attribute__((aligned(16)));
83
    int synth_buf_offset[MPA_MAX_CHANNELS];
84
    int32_t sb_samples[MPA_MAX_CHANNELS][36][SBLIMIT] __attribute__((aligned(16)));
85
    int32_t mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */
86
#ifdef DEBUG
87
    int frame_count;
88
#endif
89
    void (*compute_antialias)(struct MPADecodeContext *s, struct GranuleDef *g);
90
    int adu_mode; ///< 0 for standard mp3, 1 for adu formatted mp3
91
    int dither_state;
92
    int error_resilience;
93
    AVCodecContext* avctx;
94
} MPADecodeContext;
95

    
96
/**
97
 * Context for MP3On4 decoder
98
 */
99
typedef struct MP3On4DecodeContext {
100
    int frames;   ///< number of mp3 frames per block (number of mp3 decoder instances)
101
    int chan_cfg; ///< channel config number
102
    MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance
103
} MP3On4DecodeContext;
104

    
105
/* layer 3 "granule" */
106
typedef struct GranuleDef {
107
    uint8_t scfsi;
108
    int part2_3_length;
109
    int big_values;
110
    int global_gain;
111
    int scalefac_compress;
112
    uint8_t block_type;
113
    uint8_t switch_point;
114
    int table_select[3];
115
    int subblock_gain[3];
116
    uint8_t scalefac_scale;
117
    uint8_t count1table_select;
118
    int region_size[3]; /* number of huffman codes in each region */
119
    int preflag;
120
    int short_start, long_end; /* long/short band indexes */
121
    uint8_t scale_factors[40];
122
    int32_t sb_hybrid[SBLIMIT * 18]; /* 576 samples */
123
} GranuleDef;
124

    
125
#define MODE_EXT_MS_STEREO 2
126
#define MODE_EXT_I_STEREO  1
127

    
128
/* layer 3 huffman tables */
129
typedef struct HuffTable {
130
    int xsize;
131
    const uint8_t *bits;
132
    const uint16_t *codes;
133
} HuffTable;
134

    
135
#include "mpegaudiodectab.h"
136

    
137
static void compute_antialias_integer(MPADecodeContext *s, GranuleDef *g);
138
static void compute_antialias_float(MPADecodeContext *s, GranuleDef *g);
139

    
140
/* vlc structure for decoding layer 3 huffman tables */
141
static VLC huff_vlc[16];
142
static VLC huff_quad_vlc[2];
143
/* computed from band_size_long */
144
static uint16_t band_index_long[9][23];
145
/* XXX: free when all decoders are closed */
146
#define TABLE_4_3_SIZE (8191 + 16)*4
147
static int8_t  table_4_3_exp[TABLE_4_3_SIZE];
148
static uint32_t table_4_3_value[TABLE_4_3_SIZE];
149
static uint32_t exp_table[512];
150
static uint32_t expval_table[512][16];
151
/* intensity stereo coef table */
152
static int32_t is_table[2][16];
153
static int32_t is_table_lsf[2][2][16];
154
static int32_t csa_table[8][4];
155
static float csa_table_float[8][4];
156
static int32_t mdct_win[8][36];
157

    
158
/* lower 2 bits: modulo 3, higher bits: shift */
159
static uint16_t scale_factor_modshift[64];
160
/* [i][j]:  2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
161
static int32_t scale_factor_mult[15][3];
162
/* mult table for layer 2 group quantization */
163

    
164
#define SCALE_GEN(v) \
165
{ FIXR(1.0 * (v)), FIXR(0.7937005259 * (v)), FIXR(0.6299605249 * (v)) }
166

    
167
static const int32_t scale_factor_mult2[3][3] = {
168
    SCALE_GEN(4.0 / 3.0), /* 3 steps */
169
    SCALE_GEN(4.0 / 5.0), /* 5 steps */
170
    SCALE_GEN(4.0 / 9.0), /* 9 steps */
171
};
172

    
173
static MPA_INT window[512] __attribute__((aligned(16)));
174

    
175
/* layer 1 unscaling */
176
/* n = number of bits of the mantissa minus 1 */
177
static inline int l1_unscale(int n, int mant, int scale_factor)
178
{
179
    int shift, mod;
180
    int64_t val;
181

    
182
    shift = scale_factor_modshift[scale_factor];
183
    mod = shift & 3;
184
    shift >>= 2;
185
    val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);
186
    shift += n;
187
    /* NOTE: at this point, 1 <= shift >= 21 + 15 */
188
    return (int)((val + (1LL << (shift - 1))) >> shift);
189
}
190

    
191
static inline int l2_unscale_group(int steps, int mant, int scale_factor)
192
{
193
    int shift, mod, val;
194

    
195
    shift = scale_factor_modshift[scale_factor];
196
    mod = shift & 3;
197
    shift >>= 2;
198

    
199
    val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
200
    /* NOTE: at this point, 0 <= shift <= 21 */
201
    if (shift > 0)
202
        val = (val + (1 << (shift - 1))) >> shift;
203
    return val;
204
}
205

    
206
/* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
207
static inline int l3_unscale(int value, int exponent)
208
{
209
    unsigned int m;
210
    int e;
211

    
212
    e = table_4_3_exp  [4*value + (exponent&3)];
213
    m = table_4_3_value[4*value + (exponent&3)];
214
    e -= (exponent >> 2);
215
    assert(e>=1);
216
    if (e > 31)
217
        return 0;
218
    m = (m + (1 << (e-1))) >> e;
219

    
220
    return m;
221
}
222

    
223
/* all integer n^(4/3) computation code */
224
#define DEV_ORDER 13
225

    
226
#define POW_FRAC_BITS 24
227
#define POW_FRAC_ONE    (1 << POW_FRAC_BITS)
228
#define POW_FIX(a)   ((int)((a) * POW_FRAC_ONE))
229
#define POW_MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> POW_FRAC_BITS)
230

    
231
static int dev_4_3_coefs[DEV_ORDER];
232

    
233
#if 0 /* unused */
234
static int pow_mult3[3] = {
235
    POW_FIX(1.0),
236
    POW_FIX(1.25992104989487316476),
237
    POW_FIX(1.58740105196819947474),
238
};
239
#endif
240

    
241
static void int_pow_init(void)
242
{
243
    int i, a;
244

    
245
    a = POW_FIX(1.0);
246
    for(i=0;i<DEV_ORDER;i++) {
247
        a = POW_MULL(a, POW_FIX(4.0 / 3.0) - i * POW_FIX(1.0)) / (i + 1);
248
        dev_4_3_coefs[i] = a;
249
    }
250
}
251

    
252
#if 0 /* unused, remove? */
253
/* return the mantissa and the binary exponent */
254
static int int_pow(int i, int *exp_ptr)
255
{
256
    int e, er, eq, j;
257
    int a, a1;
258

259
    /* renormalize */
260
    a = i;
261
    e = POW_FRAC_BITS;
262
    while (a < (1 << (POW_FRAC_BITS - 1))) {
263
        a = a << 1;
264
        e--;
265
    }
266
    a -= (1 << POW_FRAC_BITS);
267
    a1 = 0;
268
    for(j = DEV_ORDER - 1; j >= 0; j--)
269
        a1 = POW_MULL(a, dev_4_3_coefs[j] + a1);
270
    a = (1 << POW_FRAC_BITS) + a1;
271
    /* exponent compute (exact) */
272
    e = e * 4;
273
    er = e % 3;
274
    eq = e / 3;
275
    a = POW_MULL(a, pow_mult3[er]);
276
    while (a >= 2 * POW_FRAC_ONE) {
277
        a = a >> 1;
278
        eq++;
279
    }
280
    /* convert to float */
281
    while (a < POW_FRAC_ONE) {
282
        a = a << 1;
283
        eq--;
284
    }
285
    /* now POW_FRAC_ONE <= a < 2 * POW_FRAC_ONE */
286
#if POW_FRAC_BITS > FRAC_BITS
287
    a = (a + (1 << (POW_FRAC_BITS - FRAC_BITS - 1))) >> (POW_FRAC_BITS - FRAC_BITS);
288
    /* correct overflow */
289
    if (a >= 2 * (1 << FRAC_BITS)) {
290
        a = a >> 1;
291
        eq++;
292
    }
293
#endif
294
    *exp_ptr = eq;
295
    return a;
296
}
297
#endif
298

    
299
static int decode_init(AVCodecContext * avctx)
300
{
301
    MPADecodeContext *s = avctx->priv_data;
302
    static int init=0;
303
    int i, j, k;
304

    
305
    s->avctx = avctx;
306

    
307
#if defined(USE_HIGHPRECISION) && defined(CONFIG_AUDIO_NONSHORT)
308
    avctx->sample_fmt= SAMPLE_FMT_S32;
309
#else
310
    avctx->sample_fmt= SAMPLE_FMT_S16;
311
#endif
312
    s->error_resilience= avctx->error_resilience;
313

    
314
    if(avctx->antialias_algo != FF_AA_FLOAT)
315
        s->compute_antialias= compute_antialias_integer;
316
    else
317
        s->compute_antialias= compute_antialias_float;
318

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

    
329
        /* scale factor multiply for layer 1 */
330
        for(i=0;i<15;i++) {
331
            int n, norm;
332
            n = i + 2;
333
            norm = ((INT64_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
334
            scale_factor_mult[i][0] = MULL(FIXR(1.0 * 2.0), norm);
335
            scale_factor_mult[i][1] = MULL(FIXR(0.7937005259 * 2.0), norm);
336
            scale_factor_mult[i][2] = MULL(FIXR(0.6299605249 * 2.0), norm);
337
            dprintf(avctx, "%d: norm=%x s=%x %x %x\n",
338
                    i, norm,
339
                    scale_factor_mult[i][0],
340
                    scale_factor_mult[i][1],
341
                    scale_factor_mult[i][2]);
342
        }
343

    
344
        ff_mpa_synth_init(window);
345

    
346
        /* huffman decode tables */
347
        for(i=1;i<16;i++) {
348
            const HuffTable *h = &mpa_huff_tables[i];
349
            int xsize, x, y;
350
            unsigned int n;
351
            uint8_t  tmp_bits [512];
352
            uint16_t tmp_codes[512];
353

    
354
            memset(tmp_bits , 0, sizeof(tmp_bits ));
355
            memset(tmp_codes, 0, sizeof(tmp_codes));
356

    
357
            xsize = h->xsize;
358
            n = xsize * xsize;
359

    
360
            j = 0;
361
            for(x=0;x<xsize;x++) {
362
                for(y=0;y<xsize;y++){
363
                    tmp_bits [(x << 5) | y | ((x&&y)<<4)]= h->bits [j  ];
364
                    tmp_codes[(x << 5) | y | ((x&&y)<<4)]= h->codes[j++];
365
                }
366
            }
367

    
368
            /* XXX: fail test */
369
            init_vlc(&huff_vlc[i], 7, 512,
370
                     tmp_bits, 1, 1, tmp_codes, 2, 2, 1);
371
        }
372
        for(i=0;i<2;i++) {
373
            init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
374
                     mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1, 1);
375
        }
376

    
377
        for(i=0;i<9;i++) {
378
            k = 0;
379
            for(j=0;j<22;j++) {
380
                band_index_long[i][j] = k;
381
                k += band_size_long[i][j];
382
            }
383
            band_index_long[i][22] = k;
384
        }
385

    
386
        /* compute n ^ (4/3) and store it in mantissa/exp format */
387

    
388
        int_pow_init();
389
        for(i=1;i<TABLE_4_3_SIZE;i++) {
390
            double f, fm;
391
            int e, m;
392
            f = pow((double)(i/4), 4.0 / 3.0) * pow(2, (i&3)*0.25);
393
            fm = frexp(f, &e);
394
            m = (uint32_t)(fm*(1LL<<31) + 0.5);
395
            e+= FRAC_BITS - 31 + 5 - 100;
396

    
397
            /* normalized to FRAC_BITS */
398
            table_4_3_value[i] = m;
399
//            av_log(NULL, AV_LOG_DEBUG, "%d %d %f\n", i, m, pow((double)i, 4.0 / 3.0));
400
            table_4_3_exp[i] = -e;
401
        }
402
        for(i=0; i<512*16; i++){
403
            int exponent= (i>>4);
404
            double f= pow(i&15, 4.0 / 3.0) * pow(2, (exponent-400)*0.25 + FRAC_BITS + 5);
405
            expval_table[exponent][i&15]= llrint(f);
406
            if((i&15)==1)
407
                exp_table[exponent]= llrint(f);
408
        }
409

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

    
426
        for(i=0;i<16;i++) {
427
            double f;
428
            int e, k;
429

    
430
            for(j=0;j<2;j++) {
431
                e = -(j + 1) * ((i + 1) >> 1);
432
                f = pow(2.0, e / 4.0);
433
                k = i & 1;
434
                is_table_lsf[j][k ^ 1][i] = FIXR(f);
435
                is_table_lsf[j][k][i] = FIXR(1.0);
436
                dprintf(avctx, "is_table_lsf %d %d: %x %x\n",
437
                        i, j, is_table_lsf[j][0][i], is_table_lsf[j][1][i]);
438
            }
439
        }
440

    
441
        for(i=0;i<8;i++) {
442
            float ci, cs, ca;
443
            ci = ci_table[i];
444
            cs = 1.0 / sqrt(1.0 + ci * ci);
445
            ca = cs * ci;
446
            csa_table[i][0] = FIXHR(cs/4);
447
            csa_table[i][1] = FIXHR(ca/4);
448
            csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);
449
            csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4);
450
            csa_table_float[i][0] = cs;
451
            csa_table_float[i][1] = ca;
452
            csa_table_float[i][2] = ca + cs;
453
            csa_table_float[i][3] = ca - cs;
454
//            printf("%d %d %d %d\n", FIX(cs), FIX(cs-1), FIX(ca), FIX(cs)-FIX(ca));
455
//            av_log(NULL, AV_LOG_DEBUG,"%f %f %f %f\n", cs, ca, ca+cs, ca-cs);
456
        }
457

    
458
        /* compute mdct windows */
459
        for(i=0;i<36;i++) {
460
            for(j=0; j<4; j++){
461
                double d;
462

    
463
                if(j==2 && i%3 != 1)
464
                    continue;
465

    
466
                d= sin(M_PI * (i + 0.5) / 36.0);
467
                if(j==1){
468
                    if     (i>=30) d= 0;
469
                    else if(i>=24) d= sin(M_PI * (i - 18 + 0.5) / 12.0);
470
                    else if(i>=18) d= 1;
471
                }else if(j==3){
472
                    if     (i<  6) d= 0;
473
                    else if(i< 12) d= sin(M_PI * (i -  6 + 0.5) / 12.0);
474
                    else if(i< 18) d= 1;
475
                }
476
                //merge last stage of imdct into the window coefficients
477
                d*= 0.5 / cos(M_PI*(2*i + 19)/72);
478

    
479
                if(j==2)
480
                    mdct_win[j][i/3] = FIXHR((d / (1<<5)));
481
                else
482
                    mdct_win[j][i  ] = FIXHR((d / (1<<5)));
483
//                av_log(NULL, AV_LOG_DEBUG, "%2d %d %f\n", i,j,d / (1<<5));
484
            }
485
        }
486

    
487
        /* NOTE: we do frequency inversion adter the MDCT by changing
488
           the sign of the right window coefs */
489
        for(j=0;j<4;j++) {
490
            for(i=0;i<36;i+=2) {
491
                mdct_win[j + 4][i] = mdct_win[j][i];
492
                mdct_win[j + 4][i + 1] = -mdct_win[j][i + 1];
493
            }
494
        }
495

    
496
#if defined(DEBUG)
497
        for(j=0;j<8;j++) {
498
            av_log(avctx, AV_LOG_DEBUG, "win%d=\n", j);
499
            for(i=0;i<36;i++)
500
                av_log(avctx, AV_LOG_DEBUG, "%f, ", (double)mdct_win[j][i] / FRAC_ONE);
501
            av_log(avctx, AV_LOG_DEBUG, "\n");
502
        }
503
#endif
504
        init = 1;
505
    }
506

    
507
#ifdef DEBUG
508
    s->frame_count = 0;
509
#endif
510
    if (avctx->codec_id == CODEC_ID_MP3ADU)
511
        s->adu_mode = 1;
512
    return 0;
513
}
514

    
515
/* tab[i][j] = 1.0 / (2.0 * cos(pi*(2*k+1) / 2^(6 - j))) */
516

    
517
/* cos(i*pi/64) */
518

    
519
#define COS0_0  FIXHR(0.50060299823519630134/2)
520
#define COS0_1  FIXHR(0.50547095989754365998/2)
521
#define COS0_2  FIXHR(0.51544730992262454697/2)
522
#define COS0_3  FIXHR(0.53104259108978417447/2)
523
#define COS0_4  FIXHR(0.55310389603444452782/2)
524
#define COS0_5  FIXHR(0.58293496820613387367/2)
525
#define COS0_6  FIXHR(0.62250412303566481615/2)
526
#define COS0_7  FIXHR(0.67480834145500574602/2)
527
#define COS0_8  FIXHR(0.74453627100229844977/2)
528
#define COS0_9  FIXHR(0.83934964541552703873/2)
529
#define COS0_10 FIXHR(0.97256823786196069369/2)
530
#define COS0_11 FIXHR(1.16943993343288495515/4)
531
#define COS0_12 FIXHR(1.48416461631416627724/4)
532
#define COS0_13 FIXHR(2.05778100995341155085/8)
533
#define COS0_14 FIXHR(3.40760841846871878570/8)
534
#define COS0_15 FIXHR(10.19000812354805681150/32)
535

    
536
#define COS1_0 FIXHR(0.50241928618815570551/2)
537
#define COS1_1 FIXHR(0.52249861493968888062/2)
538
#define COS1_2 FIXHR(0.56694403481635770368/2)
539
#define COS1_3 FIXHR(0.64682178335999012954/2)
540
#define COS1_4 FIXHR(0.78815462345125022473/2)
541
#define COS1_5 FIXHR(1.06067768599034747134/4)
542
#define COS1_6 FIXHR(1.72244709823833392782/4)
543
#define COS1_7 FIXHR(5.10114861868916385802/16)
544

    
545
#define COS2_0 FIXHR(0.50979557910415916894/2)
546
#define COS2_1 FIXHR(0.60134488693504528054/2)
547
#define COS2_2 FIXHR(0.89997622313641570463/2)
548
#define COS2_3 FIXHR(2.56291544774150617881/8)
549

    
550
#define COS3_0 FIXHR(0.54119610014619698439/2)
551
#define COS3_1 FIXHR(1.30656296487637652785/4)
552

    
553
#define COS4_0 FIXHR(0.70710678118654752439/2)
554

    
555
/* butterfly operator */
556
#define BF(a, b, c, s)\
557
{\
558
    tmp0 = tab[a] + tab[b];\
559
    tmp1 = tab[a] - tab[b];\
560
    tab[a] = tmp0;\
561
    tab[b] = MULH(tmp1<<(s), c);\
562
}
563

    
564
#define BF1(a, b, c, d)\
565
{\
566
    BF(a, b, COS4_0, 1);\
567
    BF(c, d,-COS4_0, 1);\
568
    tab[c] += tab[d];\
569
}
570

    
571
#define BF2(a, b, c, d)\
572
{\
573
    BF(a, b, COS4_0, 1);\
574
    BF(c, d,-COS4_0, 1);\
575
    tab[c] += tab[d];\
576
    tab[a] += tab[c];\
577
    tab[c] += tab[b];\
578
    tab[b] += tab[d];\
579
}
580

    
581
#define ADD(a, b) tab[a] += tab[b]
582

    
583
/* DCT32 without 1/sqrt(2) coef zero scaling. */
584
static void dct32(int32_t *out, int32_t *tab)
585
{
586
    int tmp0, tmp1;
587

    
588
    /* pass 1 */
589
    BF( 0, 31, COS0_0 , 1);
590
    BF(15, 16, COS0_15, 5);
591
    /* pass 2 */
592
    BF( 0, 15, COS1_0 , 1);
593
    BF(16, 31,-COS1_0 , 1);
594
    /* pass 1 */
595
    BF( 7, 24, COS0_7 , 1);
596
    BF( 8, 23, COS0_8 , 1);
597
    /* pass 2 */
598
    BF( 7,  8, COS1_7 , 4);
599
    BF(23, 24,-COS1_7 , 4);
600
    /* pass 3 */
601
    BF( 0,  7, COS2_0 , 1);
602
    BF( 8, 15,-COS2_0 , 1);
603
    BF(16, 23, COS2_0 , 1);
604
    BF(24, 31,-COS2_0 , 1);
605
    /* pass 1 */
606
    BF( 3, 28, COS0_3 , 1);
607
    BF(12, 19, COS0_12, 2);
608
    /* pass 2 */
609
    BF( 3, 12, COS1_3 , 1);
610
    BF(19, 28,-COS1_3 , 1);
611
    /* pass 1 */
612
    BF( 4, 27, COS0_4 , 1);
613
    BF(11, 20, COS0_11, 2);
614
    /* pass 2 */
615
    BF( 4, 11, COS1_4 , 1);
616
    BF(20, 27,-COS1_4 , 1);
617
    /* pass 3 */
618
    BF( 3,  4, COS2_3 , 3);
619
    BF(11, 12,-COS2_3 , 3);
620
    BF(19, 20, COS2_3 , 3);
621
    BF(27, 28,-COS2_3 , 3);
622
    /* pass 4 */
623
    BF( 0,  3, COS3_0 , 1);
624
    BF( 4,  7,-COS3_0 , 1);
625
    BF( 8, 11, COS3_0 , 1);
626
    BF(12, 15,-COS3_0 , 1);
627
    BF(16, 19, COS3_0 , 1);
628
    BF(20, 23,-COS3_0 , 1);
629
    BF(24, 27, COS3_0 , 1);
630
    BF(28, 31,-COS3_0 , 1);
631

    
632

    
633

    
634
    /* pass 1 */
635
    BF( 1, 30, COS0_1 , 1);
636
    BF(14, 17, COS0_14, 3);
637
    /* pass 2 */
638
    BF( 1, 14, COS1_1 , 1);
639
    BF(17, 30,-COS1_1 , 1);
640
    /* pass 1 */
641
    BF( 6, 25, COS0_6 , 1);
642
    BF( 9, 22, COS0_9 , 1);
643
    /* pass 2 */
644
    BF( 6,  9, COS1_6 , 2);
645
    BF(22, 25,-COS1_6 , 2);
646
    /* pass 3 */
647
    BF( 1,  6, COS2_1 , 1);
648
    BF( 9, 14,-COS2_1 , 1);
649
    BF(17, 22, COS2_1 , 1);
650
    BF(25, 30,-COS2_1 , 1);
651

    
652
    /* pass 1 */
653
    BF( 2, 29, COS0_2 , 1);
654
    BF(13, 18, COS0_13, 3);
655
    /* pass 2 */
656
    BF( 2, 13, COS1_2 , 1);
657
    BF(18, 29,-COS1_2 , 1);
658
    /* pass 1 */
659
    BF( 5, 26, COS0_5 , 1);
660
    BF(10, 21, COS0_10, 1);
661
    /* pass 2 */
662
    BF( 5, 10, COS1_5 , 2);
663
    BF(21, 26,-COS1_5 , 2);
664
    /* pass 3 */
665
    BF( 2,  5, COS2_2 , 1);
666
    BF(10, 13,-COS2_2 , 1);
667
    BF(18, 21, COS2_2 , 1);
668
    BF(26, 29,-COS2_2 , 1);
669
    /* pass 4 */
670
    BF( 1,  2, COS3_1 , 2);
671
    BF( 5,  6,-COS3_1 , 2);
672
    BF( 9, 10, COS3_1 , 2);
673
    BF(13, 14,-COS3_1 , 2);
674
    BF(17, 18, COS3_1 , 2);
675
    BF(21, 22,-COS3_1 , 2);
676
    BF(25, 26, COS3_1 , 2);
677
    BF(29, 30,-COS3_1 , 2);
678

    
679
    /* pass 5 */
680
    BF1( 0,  1,  2,  3);
681
    BF2( 4,  5,  6,  7);
682
    BF1( 8,  9, 10, 11);
683
    BF2(12, 13, 14, 15);
684
    BF1(16, 17, 18, 19);
685
    BF2(20, 21, 22, 23);
686
    BF1(24, 25, 26, 27);
687
    BF2(28, 29, 30, 31);
688

    
689
    /* pass 6 */
690

    
691
    ADD( 8, 12);
692
    ADD(12, 10);
693
    ADD(10, 14);
694
    ADD(14,  9);
695
    ADD( 9, 13);
696
    ADD(13, 11);
697
    ADD(11, 15);
698

    
699
    out[ 0] = tab[0];
700
    out[16] = tab[1];
701
    out[ 8] = tab[2];
702
    out[24] = tab[3];
703
    out[ 4] = tab[4];
704
    out[20] = tab[5];
705
    out[12] = tab[6];
706
    out[28] = tab[7];
707
    out[ 2] = tab[8];
708
    out[18] = tab[9];
709
    out[10] = tab[10];
710
    out[26] = tab[11];
711
    out[ 6] = tab[12];
712
    out[22] = tab[13];
713
    out[14] = tab[14];
714
    out[30] = tab[15];
715

    
716
    ADD(24, 28);
717
    ADD(28, 26);
718
    ADD(26, 30);
719
    ADD(30, 25);
720
    ADD(25, 29);
721
    ADD(29, 27);
722
    ADD(27, 31);
723

    
724
    out[ 1] = tab[16] + tab[24];
725
    out[17] = tab[17] + tab[25];
726
    out[ 9] = tab[18] + tab[26];
727
    out[25] = tab[19] + tab[27];
728
    out[ 5] = tab[20] + tab[28];
729
    out[21] = tab[21] + tab[29];
730
    out[13] = tab[22] + tab[30];
731
    out[29] = tab[23] + tab[31];
732
    out[ 3] = tab[24] + tab[20];
733
    out[19] = tab[25] + tab[21];
734
    out[11] = tab[26] + tab[22];
735
    out[27] = tab[27] + tab[23];
736
    out[ 7] = tab[28] + tab[18];
737
    out[23] = tab[29] + tab[19];
738
    out[15] = tab[30] + tab[17];
739
    out[31] = tab[31];
740
}
741

    
742
#if FRAC_BITS <= 15
743

    
744
static inline int round_sample(int *sum)
745
{
746
    int sum1;
747
    sum1 = (*sum) >> OUT_SHIFT;
748
    *sum &= (1<<OUT_SHIFT)-1;
749
    if (sum1 < OUT_MIN)
750
        sum1 = OUT_MIN;
751
    else if (sum1 > OUT_MAX)
752
        sum1 = OUT_MAX;
753
    return sum1;
754
}
755

    
756
/* signed 16x16 -> 32 multiply add accumulate */
757
#define MACS(rt, ra, rb) MAC16(rt, ra, rb)
758

    
759
/* signed 16x16 -> 32 multiply */
760
#define MULS(ra, rb) MUL16(ra, rb)
761

    
762
#else
763

    
764
static inline int round_sample(int64_t *sum)
765
{
766
    int sum1;
767
    sum1 = (int)((*sum) >> OUT_SHIFT);
768
    *sum &= (1<<OUT_SHIFT)-1;
769
    if (sum1 < OUT_MIN)
770
        sum1 = OUT_MIN;
771
    else if (sum1 > OUT_MAX)
772
        sum1 = OUT_MAX;
773
    return sum1;
774
}
775

    
776
#   define MULS(ra, rb) MUL64(ra, rb)
777
#endif
778

    
779
#define SUM8(sum, op, w, p) \
780
{                                               \
781
    sum op MULS((w)[0 * 64], p[0 * 64]);\
782
    sum op MULS((w)[1 * 64], p[1 * 64]);\
783
    sum op MULS((w)[2 * 64], p[2 * 64]);\
784
    sum op MULS((w)[3 * 64], p[3 * 64]);\
785
    sum op MULS((w)[4 * 64], p[4 * 64]);\
786
    sum op MULS((w)[5 * 64], p[5 * 64]);\
787
    sum op MULS((w)[6 * 64], p[6 * 64]);\
788
    sum op MULS((w)[7 * 64], p[7 * 64]);\
789
}
790

    
791
#define SUM8P2(sum1, op1, sum2, op2, w1, w2, p) \
792
{                                               \
793
    int tmp;\
794
    tmp = p[0 * 64];\
795
    sum1 op1 MULS((w1)[0 * 64], tmp);\
796
    sum2 op2 MULS((w2)[0 * 64], tmp);\
797
    tmp = p[1 * 64];\
798
    sum1 op1 MULS((w1)[1 * 64], tmp);\
799
    sum2 op2 MULS((w2)[1 * 64], tmp);\
800
    tmp = p[2 * 64];\
801
    sum1 op1 MULS((w1)[2 * 64], tmp);\
802
    sum2 op2 MULS((w2)[2 * 64], tmp);\
803
    tmp = p[3 * 64];\
804
    sum1 op1 MULS((w1)[3 * 64], tmp);\
805
    sum2 op2 MULS((w2)[3 * 64], tmp);\
806
    tmp = p[4 * 64];\
807
    sum1 op1 MULS((w1)[4 * 64], tmp);\
808
    sum2 op2 MULS((w2)[4 * 64], tmp);\
809
    tmp = p[5 * 64];\
810
    sum1 op1 MULS((w1)[5 * 64], tmp);\
811
    sum2 op2 MULS((w2)[5 * 64], tmp);\
812
    tmp = p[6 * 64];\
813
    sum1 op1 MULS((w1)[6 * 64], tmp);\
814
    sum2 op2 MULS((w2)[6 * 64], tmp);\
815
    tmp = p[7 * 64];\
816
    sum1 op1 MULS((w1)[7 * 64], tmp);\
817
    sum2 op2 MULS((w2)[7 * 64], tmp);\
818
}
819

    
820
void ff_mpa_synth_init(MPA_INT *window)
821
{
822
    int i;
823

    
824
    /* max = 18760, max sum over all 16 coefs : 44736 */
825
    for(i=0;i<257;i++) {
826
        int v;
827
        v = mpa_enwindow[i];
828
#if WFRAC_BITS < 16
829
        v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS);
830
#endif
831
        window[i] = v;
832
        if ((i & 63) != 0)
833
            v = -v;
834
        if (i != 0)
835
            window[512 - i] = v;
836
    }
837
}
838

    
839
/* 32 sub band synthesis filter. Input: 32 sub band samples, Output:
840
   32 samples. */
841
/* XXX: optimize by avoiding ring buffer usage */
842
void ff_mpa_synth_filter(MPA_INT *synth_buf_ptr, int *synth_buf_offset,
843
                         MPA_INT *window, int *dither_state,
844
                         OUT_INT *samples, int incr,
845
                         int32_t sb_samples[SBLIMIT])
846
{
847
    int32_t tmp[32];
848
    register MPA_INT *synth_buf;
849
    register const MPA_INT *w, *w2, *p;
850
    int j, offset, v;
851
    OUT_INT *samples2;
852
#if FRAC_BITS <= 15
853
    int sum, sum2;
854
#else
855
    int64_t sum, sum2;
856
#endif
857

    
858
    dct32(tmp, sb_samples);
859

    
860
    offset = *synth_buf_offset;
861
    synth_buf = synth_buf_ptr + offset;
862

    
863
    for(j=0;j<32;j++) {
864
        v = tmp[j];
865
#if FRAC_BITS <= 15
866
        /* NOTE: can cause a loss in precision if very high amplitude
867
           sound */
868
        if (v > 32767)
869
            v = 32767;
870
        else if (v < -32768)
871
            v = -32768;
872
#endif
873
        synth_buf[j] = v;
874
    }
875
    /* copy to avoid wrap */
876
    memcpy(synth_buf + 512, synth_buf, 32 * sizeof(MPA_INT));
877

    
878
    samples2 = samples + 31 * incr;
879
    w = window;
880
    w2 = window + 31;
881

    
882
    sum = *dither_state;
883
    p = synth_buf + 16;
884
    SUM8(sum, +=, w, p);
885
    p = synth_buf + 48;
886
    SUM8(sum, -=, w + 32, p);
887
    *samples = round_sample(&sum);
888
    samples += incr;
889
    w++;
890

    
891
    /* we calculate two samples at the same time to avoid one memory
892
       access per two sample */
893
    for(j=1;j<16;j++) {
894
        sum2 = 0;
895
        p = synth_buf + 16 + j;
896
        SUM8P2(sum, +=, sum2, -=, w, w2, p);
897
        p = synth_buf + 48 - j;
898
        SUM8P2(sum, -=, sum2, -=, w + 32, w2 + 32, p);
899

    
900
        *samples = round_sample(&sum);
901
        samples += incr;
902
        sum += sum2;
903
        *samples2 = round_sample(&sum);
904
        samples2 -= incr;
905
        w++;
906
        w2--;
907
    }
908

    
909
    p = synth_buf + 32;
910
    SUM8(sum, -=, w + 32, p);
911
    *samples = round_sample(&sum);
912
    *dither_state= sum;
913

    
914
    offset = (offset - 32) & 511;
915
    *synth_buf_offset = offset;
916
}
917

    
918
#define C3 FIXHR(0.86602540378443864676/2)
919

    
920
/* 0.5 / cos(pi*(2*i+1)/36) */
921
static const int icos36[9] = {
922
    FIXR(0.50190991877167369479),
923
    FIXR(0.51763809020504152469), //0
924
    FIXR(0.55168895948124587824),
925
    FIXR(0.61038729438072803416),
926
    FIXR(0.70710678118654752439), //1
927
    FIXR(0.87172339781054900991),
928
    FIXR(1.18310079157624925896),
929
    FIXR(1.93185165257813657349), //2
930
    FIXR(5.73685662283492756461),
931
};
932

    
933
/* 0.5 / cos(pi*(2*i+1)/36) */
934
static const int icos36h[9] = {
935
    FIXHR(0.50190991877167369479/2),
936
    FIXHR(0.51763809020504152469/2), //0
937
    FIXHR(0.55168895948124587824/2),
938
    FIXHR(0.61038729438072803416/2),
939
    FIXHR(0.70710678118654752439/2), //1
940
    FIXHR(0.87172339781054900991/2),
941
    FIXHR(1.18310079157624925896/4),
942
    FIXHR(1.93185165257813657349/4), //2
943
//    FIXHR(5.73685662283492756461),
944
};
945

    
946
/* 12 points IMDCT. We compute it "by hand" by factorizing obvious
947
   cases. */
948
static void imdct12(int *out, int *in)
949
{
950
    int in0, in1, in2, in3, in4, in5, t1, t2;
951

    
952
    in0= in[0*3];
953
    in1= in[1*3] + in[0*3];
954
    in2= in[2*3] + in[1*3];
955
    in3= in[3*3] + in[2*3];
956
    in4= in[4*3] + in[3*3];
957
    in5= in[5*3] + in[4*3];
958
    in5 += in3;
959
    in3 += in1;
960

    
961
    in2= MULH(2*in2, C3);
962
    in3= MULH(4*in3, C3);
963

    
964
    t1 = in0 - in4;
965
    t2 = MULH(2*(in1 - in5), icos36h[4]);
966

    
967
    out[ 7]=
968
    out[10]= t1 + t2;
969
    out[ 1]=
970
    out[ 4]= t1 - t2;
971

    
972
    in0 += in4>>1;
973
    in4 = in0 + in2;
974
    in5 += 2*in1;
975
    in1 = MULH(in5 + in3, icos36h[1]);
976
    out[ 8]=
977
    out[ 9]= in4 + in1;
978
    out[ 2]=
979
    out[ 3]= in4 - in1;
980

    
981
    in0 -= in2;
982
    in5 = MULH(2*(in5 - in3), icos36h[7]);
983
    out[ 0]=
984
    out[ 5]= in0 - in5;
985
    out[ 6]=
986
    out[11]= in0 + in5;
987
}
988

    
989
/* cos(pi*i/18) */
990
#define C1 FIXHR(0.98480775301220805936/2)
991
#define C2 FIXHR(0.93969262078590838405/2)
992
#define C3 FIXHR(0.86602540378443864676/2)
993
#define C4 FIXHR(0.76604444311897803520/2)
994
#define C5 FIXHR(0.64278760968653932632/2)
995
#define C6 FIXHR(0.5/2)
996
#define C7 FIXHR(0.34202014332566873304/2)
997
#define C8 FIXHR(0.17364817766693034885/2)
998

    
999

    
1000
/* using Lee like decomposition followed by hand coded 9 points DCT */
1001
static void imdct36(int *out, int *buf, int *in, int *win)
1002
{
1003
    int i, j, t0, t1, t2, t3, s0, s1, s2, s3;
1004
    int tmp[18], *tmp1, *in1;
1005

    
1006
    for(i=17;i>=1;i--)
1007
        in[i] += in[i-1];
1008
    for(i=17;i>=3;i-=2)
1009
        in[i] += in[i-2];
1010

    
1011
    for(j=0;j<2;j++) {
1012
        tmp1 = tmp + j;
1013
        in1 = in + j;
1014
#if 0
1015
//more accurate but slower
1016
        int64_t t0, t1, t2, t3;
1017
        t2 = in1[2*4] + in1[2*8] - in1[2*2];
1018

1019
        t3 = (in1[2*0] + (int64_t)(in1[2*6]>>1))<<32;
1020
        t1 = in1[2*0] - in1[2*6];
1021
        tmp1[ 6] = t1 - (t2>>1);
1022
        tmp1[16] = t1 + t2;
1023

1024
        t0 = MUL64(2*(in1[2*2] + in1[2*4]),    C2);
1025
        t1 = MUL64(   in1[2*4] - in1[2*8] , -2*C8);
1026
        t2 = MUL64(2*(in1[2*2] + in1[2*8]),   -C4);
1027

1028
        tmp1[10] = (t3 - t0 - t2) >> 32;
1029
        tmp1[ 2] = (t3 + t0 + t1) >> 32;
1030
        tmp1[14] = (t3 + t2 - t1) >> 32;
1031

1032
        tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
1033
        t2 = MUL64(2*(in1[2*1] + in1[2*5]),    C1);
1034
        t3 = MUL64(   in1[2*5] - in1[2*7] , -2*C7);
1035
        t0 = MUL64(2*in1[2*3], C3);
1036

1037
        t1 = MUL64(2*(in1[2*1] + in1[2*7]),   -C5);
1038

1039
        tmp1[ 0] = (t2 + t3 + t0) >> 32;
1040
        tmp1[12] = (t2 + t1 - t0) >> 32;
1041
        tmp1[ 8] = (t3 - t1 - t0) >> 32;
1042
#else
1043
        t2 = in1[2*4] + in1[2*8] - in1[2*2];
1044

    
1045
        t3 = in1[2*0] + (in1[2*6]>>1);
1046
        t1 = in1[2*0] - in1[2*6];
1047
        tmp1[ 6] = t1 - (t2>>1);
1048
        tmp1[16] = t1 + t2;
1049

    
1050
        t0 = MULH(2*(in1[2*2] + in1[2*4]),    C2);
1051
        t1 = MULH(   in1[2*4] - in1[2*8] , -2*C8);
1052
        t2 = MULH(2*(in1[2*2] + in1[2*8]),   -C4);
1053

    
1054
        tmp1[10] = t3 - t0 - t2;
1055
        tmp1[ 2] = t3 + t0 + t1;
1056
        tmp1[14] = t3 + t2 - t1;
1057

    
1058
        tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
1059
        t2 = MULH(2*(in1[2*1] + in1[2*5]),    C1);
1060
        t3 = MULH(   in1[2*5] - in1[2*7] , -2*C7);
1061
        t0 = MULH(2*in1[2*3], C3);
1062

    
1063
        t1 = MULH(2*(in1[2*1] + in1[2*7]),   -C5);
1064

    
1065
        tmp1[ 0] = t2 + t3 + t0;
1066
        tmp1[12] = t2 + t1 - t0;
1067
        tmp1[ 8] = t3 - t1 - t0;
1068
#endif
1069
    }
1070

    
1071
    i = 0;
1072
    for(j=0;j<4;j++) {
1073
        t0 = tmp[i];
1074
        t1 = tmp[i + 2];
1075
        s0 = t1 + t0;
1076
        s2 = t1 - t0;
1077

    
1078
        t2 = tmp[i + 1];
1079
        t3 = tmp[i + 3];
1080
        s1 = MULH(2*(t3 + t2), icos36h[j]);
1081
        s3 = MULL(t3 - t2, icos36[8 - j]);
1082

    
1083
        t0 = s0 + s1;
1084
        t1 = s0 - s1;
1085
        out[(9 + j)*SBLIMIT] =  MULH(t1, win[9 + j]) + buf[9 + j];
1086
        out[(8 - j)*SBLIMIT] =  MULH(t1, win[8 - j]) + buf[8 - j];
1087
        buf[9 + j] = MULH(t0, win[18 + 9 + j]);
1088
        buf[8 - j] = MULH(t0, win[18 + 8 - j]);
1089

    
1090
        t0 = s2 + s3;
1091
        t1 = s2 - s3;
1092
        out[(9 + 8 - j)*SBLIMIT] =  MULH(t1, win[9 + 8 - j]) + buf[9 + 8 - j];
1093
        out[(        j)*SBLIMIT] =  MULH(t1, win[        j]) + buf[        j];
1094
        buf[9 + 8 - j] = MULH(t0, win[18 + 9 + 8 - j]);
1095
        buf[      + j] = MULH(t0, win[18         + j]);
1096
        i += 4;
1097
    }
1098

    
1099
    s0 = tmp[16];
1100
    s1 = MULH(2*tmp[17], icos36h[4]);
1101
    t0 = s0 + s1;
1102
    t1 = s0 - s1;
1103
    out[(9 + 4)*SBLIMIT] =  MULH(t1, win[9 + 4]) + buf[9 + 4];
1104
    out[(8 - 4)*SBLIMIT] =  MULH(t1, win[8 - 4]) + buf[8 - 4];
1105
    buf[9 + 4] = MULH(t0, win[18 + 9 + 4]);
1106
    buf[8 - 4] = MULH(t0, win[18 + 8 - 4]);
1107
}
1108

    
1109
/* header decoding. MUST check the header before because no
1110
   consistency check is done there. Return 1 if free format found and
1111
   that the frame size must be computed externally */
1112
static int decode_header(MPADecodeContext *s, uint32_t header)
1113
{
1114
    int sample_rate, frame_size, mpeg25, padding;
1115
    int sample_rate_index, bitrate_index;
1116
    if (header & (1<<20)) {
1117
        s->lsf = (header & (1<<19)) ? 0 : 1;
1118
        mpeg25 = 0;
1119
    } else {
1120
        s->lsf = 1;
1121
        mpeg25 = 1;
1122
    }
1123

    
1124
    s->layer = 4 - ((header >> 17) & 3);
1125
    /* extract frequency */
1126
    sample_rate_index = (header >> 10) & 3;
1127
    sample_rate = mpa_freq_tab[sample_rate_index] >> (s->lsf + mpeg25);
1128
    sample_rate_index += 3 * (s->lsf + mpeg25);
1129
    s->sample_rate_index = sample_rate_index;
1130
    s->error_protection = ((header >> 16) & 1) ^ 1;
1131
    s->sample_rate = sample_rate;
1132

    
1133
    bitrate_index = (header >> 12) & 0xf;
1134
    padding = (header >> 9) & 1;
1135
    //extension = (header >> 8) & 1;
1136
    s->mode = (header >> 6) & 3;
1137
    s->mode_ext = (header >> 4) & 3;
1138
    //copyright = (header >> 3) & 1;
1139
    //original = (header >> 2) & 1;
1140
    //emphasis = header & 3;
1141

    
1142
    if (s->mode == MPA_MONO)
1143
        s->nb_channels = 1;
1144
    else
1145
        s->nb_channels = 2;
1146

    
1147
    if (bitrate_index != 0) {
1148
        frame_size = mpa_bitrate_tab[s->lsf][s->layer - 1][bitrate_index];
1149
        s->bit_rate = frame_size * 1000;
1150
        switch(s->layer) {
1151
        case 1:
1152
            frame_size = (frame_size * 12000) / sample_rate;
1153
            frame_size = (frame_size + padding) * 4;
1154
            break;
1155
        case 2:
1156
            frame_size = (frame_size * 144000) / sample_rate;
1157
            frame_size += padding;
1158
            break;
1159
        default:
1160
        case 3:
1161
            frame_size = (frame_size * 144000) / (sample_rate << s->lsf);
1162
            frame_size += padding;
1163
            break;
1164
        }
1165
        s->frame_size = frame_size;
1166
    } else {
1167
        /* if no frame size computed, signal it */
1168
        return 1;
1169
    }
1170

    
1171
#if defined(DEBUG)
1172
    dprintf(s->avctx, "layer%d, %d Hz, %d kbits/s, ",
1173
           s->layer, s->sample_rate, s->bit_rate);
1174
    if (s->nb_channels == 2) {
1175
        if (s->layer == 3) {
1176
            if (s->mode_ext & MODE_EXT_MS_STEREO)
1177
                dprintf(s->avctx, "ms-");
1178
            if (s->mode_ext & MODE_EXT_I_STEREO)
1179
                dprintf(s->avctx, "i-");
1180
        }
1181
        dprintf(s->avctx, "stereo");
1182
    } else {
1183
        dprintf(s->avctx, "mono");
1184
    }
1185
    dprintf(s->avctx, "\n");
1186
#endif
1187
    return 0;
1188
}
1189

    
1190
/* useful helper to get mpeg audio stream infos. Return -1 if error in
1191
   header, otherwise the coded frame size in bytes */
1192
int mpa_decode_header(AVCodecContext *avctx, uint32_t head, int *sample_rate)
1193
{
1194
    MPADecodeContext s1, *s = &s1;
1195
    s1.avctx = avctx;
1196

    
1197
    if (ff_mpa_check_header(head) != 0)
1198
        return -1;
1199

    
1200
    if (decode_header(s, head) != 0) {
1201
        return -1;
1202
    }
1203

    
1204
    switch(s->layer) {
1205
    case 1:
1206
        avctx->frame_size = 384;
1207
        break;
1208
    case 2:
1209
        avctx->frame_size = 1152;
1210
        break;
1211
    default:
1212
    case 3:
1213
        if (s->lsf)
1214
            avctx->frame_size = 576;
1215
        else
1216
            avctx->frame_size = 1152;
1217
        break;
1218
    }
1219

    
1220
    *sample_rate = s->sample_rate;
1221
    avctx->channels = s->nb_channels;
1222
    avctx->bit_rate = s->bit_rate;
1223
    avctx->sub_id = s->layer;
1224
    return s->frame_size;
1225
}
1226

    
1227
/* return the number of decoded frames */
1228
static int mp_decode_layer1(MPADecodeContext *s)
1229
{
1230
    int bound, i, v, n, ch, j, mant;
1231
    uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
1232
    uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
1233

    
1234
    if (s->mode == MPA_JSTEREO)
1235
        bound = (s->mode_ext + 1) * 4;
1236
    else
1237
        bound = SBLIMIT;
1238

    
1239
    /* allocation bits */
1240
    for(i=0;i<bound;i++) {
1241
        for(ch=0;ch<s->nb_channels;ch++) {
1242
            allocation[ch][i] = get_bits(&s->gb, 4);
1243
        }
1244
    }
1245
    for(i=bound;i<SBLIMIT;i++) {
1246
        allocation[0][i] = get_bits(&s->gb, 4);
1247
    }
1248

    
1249
    /* scale factors */
1250
    for(i=0;i<bound;i++) {
1251
        for(ch=0;ch<s->nb_channels;ch++) {
1252
            if (allocation[ch][i])
1253
                scale_factors[ch][i] = get_bits(&s->gb, 6);
1254
        }
1255
    }
1256
    for(i=bound;i<SBLIMIT;i++) {
1257
        if (allocation[0][i]) {
1258
            scale_factors[0][i] = get_bits(&s->gb, 6);
1259
            scale_factors[1][i] = get_bits(&s->gb, 6);
1260
        }
1261
    }
1262

    
1263
    /* compute samples */
1264
    for(j=0;j<12;j++) {
1265
        for(i=0;i<bound;i++) {
1266
            for(ch=0;ch<s->nb_channels;ch++) {
1267
                n = allocation[ch][i];
1268
                if (n) {
1269
                    mant = get_bits(&s->gb, n + 1);
1270
                    v = l1_unscale(n, mant, scale_factors[ch][i]);
1271
                } else {
1272
                    v = 0;
1273
                }
1274
                s->sb_samples[ch][j][i] = v;
1275
            }
1276
        }
1277
        for(i=bound;i<SBLIMIT;i++) {
1278
            n = allocation[0][i];
1279
            if (n) {
1280
                mant = get_bits(&s->gb, n + 1);
1281
                v = l1_unscale(n, mant, scale_factors[0][i]);
1282
                s->sb_samples[0][j][i] = v;
1283
                v = l1_unscale(n, mant, scale_factors[1][i]);
1284
                s->sb_samples[1][j][i] = v;
1285
            } else {
1286
                s->sb_samples[0][j][i] = 0;
1287
                s->sb_samples[1][j][i] = 0;
1288
            }
1289
        }
1290
    }
1291
    return 12;
1292
}
1293

    
1294
/* bitrate is in kb/s */
1295
int l2_select_table(int bitrate, int nb_channels, int freq, int lsf)
1296
{
1297
    int ch_bitrate, table;
1298

    
1299
    ch_bitrate = bitrate / nb_channels;
1300
    if (!lsf) {
1301
        if ((freq == 48000 && ch_bitrate >= 56) ||
1302
            (ch_bitrate >= 56 && ch_bitrate <= 80))
1303
            table = 0;
1304
        else if (freq != 48000 && ch_bitrate >= 96)
1305
            table = 1;
1306
        else if (freq != 32000 && ch_bitrate <= 48)
1307
            table = 2;
1308
        else
1309
            table = 3;
1310
    } else {
1311
        table = 4;
1312
    }
1313
    return table;
1314
}
1315

    
1316
static int mp_decode_layer2(MPADecodeContext *s)
1317
{
1318
    int sblimit; /* number of used subbands */
1319
    const unsigned char *alloc_table;
1320
    int table, bit_alloc_bits, i, j, ch, bound, v;
1321
    unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
1322
    unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
1323
    unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
1324
    int scale, qindex, bits, steps, k, l, m, b;
1325

    
1326
    /* select decoding table */
1327
    table = l2_select_table(s->bit_rate / 1000, s->nb_channels,
1328
                            s->sample_rate, s->lsf);
1329
    sblimit = sblimit_table[table];
1330
    alloc_table = alloc_tables[table];
1331

    
1332
    if (s->mode == MPA_JSTEREO)
1333
        bound = (s->mode_ext + 1) * 4;
1334
    else
1335
        bound = sblimit;
1336

    
1337
    dprintf(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit);
1338

    
1339
    /* sanity check */
1340
    if( bound > sblimit ) bound = sblimit;
1341

    
1342
    /* parse bit allocation */
1343
    j = 0;
1344
    for(i=0;i<bound;i++) {
1345
        bit_alloc_bits = alloc_table[j];
1346
        for(ch=0;ch<s->nb_channels;ch++) {
1347
            bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
1348
        }
1349
        j += 1 << bit_alloc_bits;
1350
    }
1351
    for(i=bound;i<sblimit;i++) {
1352
        bit_alloc_bits = alloc_table[j];
1353
        v = get_bits(&s->gb, bit_alloc_bits);
1354
        bit_alloc[0][i] = v;
1355
        bit_alloc[1][i] = v;
1356
        j += 1 << bit_alloc_bits;
1357
    }
1358

    
1359
#ifdef DEBUG
1360
    {
1361
        for(ch=0;ch<s->nb_channels;ch++) {
1362
            for(i=0;i<sblimit;i++)
1363
                dprintf(s->avctx, " %d", bit_alloc[ch][i]);
1364
            dprintf(s->avctx, "\n");
1365
        }
1366
    }
1367
#endif
1368

    
1369
    /* scale codes */
1370
    for(i=0;i<sblimit;i++) {
1371
        for(ch=0;ch<s->nb_channels;ch++) {
1372
            if (bit_alloc[ch][i])
1373
                scale_code[ch][i] = get_bits(&s->gb, 2);
1374
        }
1375
    }
1376

    
1377
    /* scale factors */
1378
    for(i=0;i<sblimit;i++) {
1379
        for(ch=0;ch<s->nb_channels;ch++) {
1380
            if (bit_alloc[ch][i]) {
1381
                sf = scale_factors[ch][i];
1382
                switch(scale_code[ch][i]) {
1383
                default:
1384
                case 0:
1385
                    sf[0] = get_bits(&s->gb, 6);
1386
                    sf[1] = get_bits(&s->gb, 6);
1387
                    sf[2] = get_bits(&s->gb, 6);
1388
                    break;
1389
                case 2:
1390
                    sf[0] = get_bits(&s->gb, 6);
1391
                    sf[1] = sf[0];
1392
                    sf[2] = sf[0];
1393
                    break;
1394
                case 1:
1395
                    sf[0] = get_bits(&s->gb, 6);
1396
                    sf[2] = get_bits(&s->gb, 6);
1397
                    sf[1] = sf[0];
1398
                    break;
1399
                case 3:
1400
                    sf[0] = get_bits(&s->gb, 6);
1401
                    sf[2] = get_bits(&s->gb, 6);
1402
                    sf[1] = sf[2];
1403
                    break;
1404
                }
1405
            }
1406
        }
1407
    }
1408

    
1409
#ifdef DEBUG
1410
    for(ch=0;ch<s->nb_channels;ch++) {
1411
        for(i=0;i<sblimit;i++) {
1412
            if (bit_alloc[ch][i]) {
1413
                sf = scale_factors[ch][i];
1414
                dprintf(s->avctx, " %d %d %d", sf[0], sf[1], sf[2]);
1415
            } else {
1416
                dprintf(s->avctx, " -");
1417
            }
1418
        }
1419
        dprintf(s->avctx, "\n");
1420
    }
1421
#endif
1422

    
1423
    /* samples */
1424
    for(k=0;k<3;k++) {
1425
        for(l=0;l<12;l+=3) {
1426
            j = 0;
1427
            for(i=0;i<bound;i++) {
1428
                bit_alloc_bits = alloc_table[j];
1429
                for(ch=0;ch<s->nb_channels;ch++) {
1430
                    b = bit_alloc[ch][i];
1431
                    if (b) {
1432
                        scale = scale_factors[ch][i][k];
1433
                        qindex = alloc_table[j+b];
1434
                        bits = quant_bits[qindex];
1435
                        if (bits < 0) {
1436
                            /* 3 values at the same time */
1437
                            v = get_bits(&s->gb, -bits);
1438
                            steps = quant_steps[qindex];
1439
                            s->sb_samples[ch][k * 12 + l + 0][i] =
1440
                                l2_unscale_group(steps, v % steps, scale);
1441
                            v = v / steps;
1442
                            s->sb_samples[ch][k * 12 + l + 1][i] =
1443
                                l2_unscale_group(steps, v % steps, scale);
1444
                            v = v / steps;
1445
                            s->sb_samples[ch][k * 12 + l + 2][i] =
1446
                                l2_unscale_group(steps, v, scale);
1447
                        } else {
1448
                            for(m=0;m<3;m++) {
1449
                                v = get_bits(&s->gb, bits);
1450
                                v = l1_unscale(bits - 1, v, scale);
1451
                                s->sb_samples[ch][k * 12 + l + m][i] = v;
1452
                            }
1453
                        }
1454
                    } else {
1455
                        s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1456
                        s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1457
                        s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1458
                    }
1459
                }
1460
                /* next subband in alloc table */
1461
                j += 1 << bit_alloc_bits;
1462
            }
1463
            /* XXX: find a way to avoid this duplication of code */
1464
            for(i=bound;i<sblimit;i++) {
1465
                bit_alloc_bits = alloc_table[j];
1466
                b = bit_alloc[0][i];
1467
                if (b) {
1468
                    int mant, scale0, scale1;
1469
                    scale0 = scale_factors[0][i][k];
1470
                    scale1 = scale_factors[1][i][k];
1471
                    qindex = alloc_table[j+b];
1472
                    bits = quant_bits[qindex];
1473
                    if (bits < 0) {
1474
                        /* 3 values at the same time */
1475
                        v = get_bits(&s->gb, -bits);
1476
                        steps = quant_steps[qindex];
1477
                        mant = v % steps;
1478
                        v = v / steps;
1479
                        s->sb_samples[0][k * 12 + l + 0][i] =
1480
                            l2_unscale_group(steps, mant, scale0);
1481
                        s->sb_samples[1][k * 12 + l + 0][i] =
1482
                            l2_unscale_group(steps, mant, scale1);
1483
                        mant = v % steps;
1484
                        v = v / steps;
1485
                        s->sb_samples[0][k * 12 + l + 1][i] =
1486
                            l2_unscale_group(steps, mant, scale0);
1487
                        s->sb_samples[1][k * 12 + l + 1][i] =
1488
                            l2_unscale_group(steps, mant, scale1);
1489
                        s->sb_samples[0][k * 12 + l + 2][i] =
1490
                            l2_unscale_group(steps, v, scale0);
1491
                        s->sb_samples[1][k * 12 + l + 2][i] =
1492
                            l2_unscale_group(steps, v, scale1);
1493
                    } else {
1494
                        for(m=0;m<3;m++) {
1495
                            mant = get_bits(&s->gb, bits);
1496
                            s->sb_samples[0][k * 12 + l + m][i] =
1497
                                l1_unscale(bits - 1, mant, scale0);
1498
                            s->sb_samples[1][k * 12 + l + m][i] =
1499
                                l1_unscale(bits - 1, mant, scale1);
1500
                        }
1501
                    }
1502
                } else {
1503
                    s->sb_samples[0][k * 12 + l + 0][i] = 0;
1504
                    s->sb_samples[0][k * 12 + l + 1][i] = 0;
1505
                    s->sb_samples[0][k * 12 + l + 2][i] = 0;
1506
                    s->sb_samples[1][k * 12 + l + 0][i] = 0;
1507
                    s->sb_samples[1][k * 12 + l + 1][i] = 0;
1508
                    s->sb_samples[1][k * 12 + l + 2][i] = 0;
1509
                }
1510
                /* next subband in alloc table */
1511
                j += 1 << bit_alloc_bits;
1512
            }
1513
            /* fill remaining samples to zero */
1514
            for(i=sblimit;i<SBLIMIT;i++) {
1515
                for(ch=0;ch<s->nb_channels;ch++) {
1516
                    s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1517
                    s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1518
                    s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1519
                }
1520
            }
1521
        }
1522
    }
1523
    return 3 * 12;
1524
}
1525

    
1526
static inline void lsf_sf_expand(int *slen,
1527
                                 int sf, int n1, int n2, int n3)
1528
{
1529
    if (n3) {
1530
        slen[3] = sf % n3;
1531
        sf /= n3;
1532
    } else {
1533
        slen[3] = 0;
1534
    }
1535
    if (n2) {
1536
        slen[2] = sf % n2;
1537
        sf /= n2;
1538
    } else {
1539
        slen[2] = 0;
1540
    }
1541
    slen[1] = sf % n1;
1542
    sf /= n1;
1543
    slen[0] = sf;
1544
}
1545

    
1546
static void exponents_from_scale_factors(MPADecodeContext *s,
1547
                                         GranuleDef *g,
1548
                                         int16_t *exponents)
1549
{
1550
    const uint8_t *bstab, *pretab;
1551
    int len, i, j, k, l, v0, shift, gain, gains[3];
1552
    int16_t *exp_ptr;
1553

    
1554
    exp_ptr = exponents;
1555
    gain = g->global_gain - 210;
1556
    shift = g->scalefac_scale + 1;
1557

    
1558
    bstab = band_size_long[s->sample_rate_index];
1559
    pretab = mpa_pretab[g->preflag];
1560
    for(i=0;i<g->long_end;i++) {
1561
        v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400;
1562
        len = bstab[i];
1563
        for(j=len;j>0;j--)
1564
            *exp_ptr++ = v0;
1565
    }
1566

    
1567
    if (g->short_start < 13) {
1568
        bstab = band_size_short[s->sample_rate_index];
1569
        gains[0] = gain - (g->subblock_gain[0] << 3);
1570
        gains[1] = gain - (g->subblock_gain[1] << 3);
1571
        gains[2] = gain - (g->subblock_gain[2] << 3);
1572
        k = g->long_end;
1573
        for(i=g->short_start;i<13;i++) {
1574
            len = bstab[i];
1575
            for(l=0;l<3;l++) {
1576
                v0 = gains[l] - (g->scale_factors[k++] << shift) + 400;
1577
                for(j=len;j>0;j--)
1578
                *exp_ptr++ = v0;
1579
            }
1580
        }
1581
    }
1582
}
1583

    
1584
/* handle n = 0 too */
1585
static inline int get_bitsz(GetBitContext *s, int n)
1586
{
1587
    if (n == 0)
1588
        return 0;
1589
    else
1590
        return get_bits(s, n);
1591
}
1592

    
1593

    
1594
static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos, int *end_pos2){
1595
    if(s->in_gb.buffer && *pos >= s->gb.size_in_bits){
1596
        s->gb= s->in_gb;
1597
        s->in_gb.buffer=NULL;
1598
        assert((get_bits_count(&s->gb) & 7) == 0);
1599
        skip_bits_long(&s->gb, *pos - *end_pos);
1600
        *end_pos2=
1601
        *end_pos= *end_pos2 + get_bits_count(&s->gb) - *pos;
1602
        *pos= get_bits_count(&s->gb);
1603
    }
1604
}
1605

    
1606
static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
1607
                          int16_t *exponents, int end_pos2)
1608
{
1609
    int s_index;
1610
    int i;
1611
    int last_pos, bits_left;
1612
    VLC *vlc;
1613
    int end_pos= FFMIN(end_pos2, s->gb.size_in_bits);
1614

    
1615
    /* low frequencies (called big values) */
1616
    s_index = 0;
1617
    for(i=0;i<3;i++) {
1618
        int j, k, l, linbits;
1619
        j = g->region_size[i];
1620
        if (j == 0)
1621
            continue;
1622
        /* select vlc table */
1623
        k = g->table_select[i];
1624
        l = mpa_huff_data[k][0];
1625
        linbits = mpa_huff_data[k][1];
1626
        vlc = &huff_vlc[l];
1627

    
1628
        if(!l){
1629
            memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*2*j);
1630
            s_index += 2*j;
1631
            continue;
1632
        }
1633

    
1634
        /* read huffcode and compute each couple */
1635
        for(;j>0;j--) {
1636
            int exponent, x, y, v;
1637
            int pos= get_bits_count(&s->gb);
1638

    
1639
            if (pos >= end_pos){
1640
//                av_log(NULL, AV_LOG_ERROR, "pos: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);
1641
                switch_buffer(s, &pos, &end_pos, &end_pos2);
1642
//                av_log(NULL, AV_LOG_ERROR, "new pos: %d %d\n", pos, end_pos);
1643
                if(pos >= end_pos)
1644
                    break;
1645
            }
1646
            y = get_vlc2(&s->gb, vlc->table, 7, 3);
1647

    
1648
            if(!y){
1649
                g->sb_hybrid[s_index  ] =
1650
                g->sb_hybrid[s_index+1] = 0;
1651
                s_index += 2;
1652
                continue;
1653
            }
1654

    
1655
            exponent= exponents[s_index];
1656

    
1657
            dprintf(s->avctx, "region=%d n=%d x=%d y=%d exp=%d\n",
1658
                    i, g->region_size[i] - j, x, y, exponent);
1659
            if(y&16){
1660
                x = y >> 5;
1661
                y = y & 0x0f;
1662
                if (x < 15){
1663
                    v = expval_table[ exponent ][ x ];
1664
//                      v = expval_table[ (exponent&3) ][ x ] >> FFMIN(0 - (exponent>>2), 31);
1665
                }else{
1666
                    x += get_bitsz(&s->gb, linbits);
1667
                    v = l3_unscale(x, exponent);
1668
                }
1669
                if (get_bits1(&s->gb))
1670
                    v = -v;
1671
                g->sb_hybrid[s_index] = v;
1672
                if (y < 15){
1673
                    v = expval_table[ exponent ][ y ];
1674
                }else{
1675
                    y += get_bitsz(&s->gb, linbits);
1676
                    v = l3_unscale(y, exponent);
1677
                }
1678
                if (get_bits1(&s->gb))
1679
                    v = -v;
1680
                g->sb_hybrid[s_index+1] = v;
1681
            }else{
1682
                x = y >> 5;
1683
                y = y & 0x0f;
1684
                x += y;
1685
                if (x < 15){
1686
                    v = expval_table[ exponent ][ x ];
1687
                }else{
1688
                    x += get_bitsz(&s->gb, linbits);
1689
                    v = l3_unscale(x, exponent);
1690
                }
1691
                if (get_bits1(&s->gb))
1692
                    v = -v;
1693
                g->sb_hybrid[s_index+!!y] = v;
1694
                g->sb_hybrid[s_index+ !y] = 0;
1695
            }
1696
            s_index+=2;
1697
        }
1698
    }
1699

    
1700
    /* high frequencies */
1701
    vlc = &huff_quad_vlc[g->count1table_select];
1702
    last_pos=0;
1703
    while (s_index <= 572) {
1704
        int pos, code;
1705
        pos = get_bits_count(&s->gb);
1706
        if (pos >= end_pos) {
1707
            if (pos > end_pos2 && last_pos){
1708
                /* some encoders generate an incorrect size for this
1709
                   part. We must go back into the data */
1710
                s_index -= 4;
1711
                skip_bits_long(&s->gb, last_pos - pos);
1712
                av_log(NULL, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos);
1713
                if(s->error_resilience >= FF_ER_COMPLIANT)
1714
                    s_index=0;
1715
                break;
1716
            }
1717
//                av_log(NULL, AV_LOG_ERROR, "pos2: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);
1718
            switch_buffer(s, &pos, &end_pos, &end_pos2);
1719
//                av_log(NULL, AV_LOG_ERROR, "new pos2: %d %d %d\n", pos, end_pos, s_index);
1720
            if(pos >= end_pos)
1721
                break;
1722
        }
1723
        last_pos= pos;
1724

    
1725
        code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);
1726
        dprintf(s->avctx, "t=%d code=%d\n", g->count1table_select, code);
1727
        g->sb_hybrid[s_index+0]=
1728
        g->sb_hybrid[s_index+1]=
1729
        g->sb_hybrid[s_index+2]=
1730
        g->sb_hybrid[s_index+3]= 0;
1731
        while(code){
1732
            const static int idxtab[16]={3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0};
1733
            int v;
1734
            int pos= s_index+idxtab[code];
1735
            code ^= 8>>idxtab[code];
1736
            v = exp_table[ exponents[pos] ];
1737
//            v = exp_table[ (exponents[pos]&3) ] >> FFMIN(0 - (exponents[pos]>>2), 31);
1738
            if(get_bits1(&s->gb))
1739
                v = -v;
1740
            g->sb_hybrid[pos] = v;
1741
        }
1742
        s_index+=4;
1743
    }
1744
    /* skip extension bits */
1745
    bits_left = end_pos2 - get_bits_count(&s->gb);
1746
//av_log(NULL, AV_LOG_ERROR, "left:%d buf:%p\n", bits_left, s->in_gb.buffer);
1747
    if (bits_left < 0/* || bits_left > 500*/) {
1748
        av_log(NULL, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
1749
        s_index=0;
1750
    }else if(bits_left > 0 && s->error_resilience >= FF_ER_AGGRESSIVE){
1751
        av_log(NULL, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
1752
        s_index=0;
1753
    }
1754
    memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*(576 - s_index));
1755
    skip_bits_long(&s->gb, bits_left);
1756

    
1757
    i= get_bits_count(&s->gb);
1758
    switch_buffer(s, &i, &end_pos, &end_pos2);
1759

    
1760
    return 0;
1761
}
1762

    
1763
/* Reorder short blocks from bitstream order to interleaved order. It
1764
   would be faster to do it in parsing, but the code would be far more
1765
   complicated */
1766
static void reorder_block(MPADecodeContext *s, GranuleDef *g)
1767
{
1768
    int i, j, len;
1769
    int32_t *ptr, *dst, *ptr1;
1770
    int32_t tmp[576];
1771

    
1772
    if (g->block_type != 2)
1773
        return;
1774

    
1775
    if (g->switch_point) {
1776
        if (s->sample_rate_index != 8) {
1777
            ptr = g->sb_hybrid + 36;
1778
        } else {
1779
            ptr = g->sb_hybrid + 48;
1780
        }
1781
    } else {
1782
        ptr = g->sb_hybrid;
1783
    }
1784

    
1785
    for(i=g->short_start;i<13;i++) {
1786
        len = band_size_short[s->sample_rate_index][i];
1787
        ptr1 = ptr;
1788
        dst = tmp;
1789
        for(j=len;j>0;j--) {
1790
            *dst++ = ptr[0*len];
1791
            *dst++ = ptr[1*len];
1792
            *dst++ = ptr[2*len];
1793
            ptr++;
1794
        }
1795
        ptr+=2*len;
1796
        memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));
1797
    }
1798
}
1799

    
1800
#define ISQRT2 FIXR(0.70710678118654752440)
1801

    
1802
static void compute_stereo(MPADecodeContext *s,
1803
                           GranuleDef *g0, GranuleDef *g1)
1804
{
1805
    int i, j, k, l;
1806
    int32_t v1, v2;
1807
    int sf_max, tmp0, tmp1, sf, len, non_zero_found;
1808
    int32_t (*is_tab)[16];
1809
    int32_t *tab0, *tab1;
1810
    int non_zero_found_short[3];
1811

    
1812
    /* intensity stereo */
1813
    if (s->mode_ext & MODE_EXT_I_STEREO) {
1814
        if (!s->lsf) {
1815
            is_tab = is_table;
1816
            sf_max = 7;
1817
        } else {
1818
            is_tab = is_table_lsf[g1->scalefac_compress & 1];
1819
            sf_max = 16;
1820
        }
1821

    
1822
        tab0 = g0->sb_hybrid + 576;
1823
        tab1 = g1->sb_hybrid + 576;
1824

    
1825
        non_zero_found_short[0] = 0;
1826
        non_zero_found_short[1] = 0;
1827
        non_zero_found_short[2] = 0;
1828
        k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1829
        for(i = 12;i >= g1->short_start;i--) {
1830
            /* for last band, use previous scale factor */
1831
            if (i != 11)
1832
                k -= 3;
1833
            len = band_size_short[s->sample_rate_index][i];
1834
            for(l=2;l>=0;l--) {
1835
                tab0 -= len;
1836
                tab1 -= len;
1837
                if (!non_zero_found_short[l]) {
1838
                    /* test if non zero band. if so, stop doing i-stereo */
1839
                    for(j=0;j<len;j++) {
1840
                        if (tab1[j] != 0) {
1841
                            non_zero_found_short[l] = 1;
1842
                            goto found1;
1843
                        }
1844
                    }
1845
                    sf = g1->scale_factors[k + l];
1846
                    if (sf >= sf_max)
1847
                        goto found1;
1848

    
1849
                    v1 = is_tab[0][sf];
1850
                    v2 = is_tab[1][sf];
1851
                    for(j=0;j<len;j++) {
1852
                        tmp0 = tab0[j];
1853
                        tab0[j] = MULL(tmp0, v1);
1854
                        tab1[j] = MULL(tmp0, v2);
1855
                    }
1856
                } else {
1857
                found1:
1858
                    if (s->mode_ext & MODE_EXT_MS_STEREO) {
1859
                        /* lower part of the spectrum : do ms stereo
1860
                           if enabled */
1861
                        for(j=0;j<len;j++) {
1862
                            tmp0 = tab0[j];
1863
                            tmp1 = tab1[j];
1864
                            tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
1865
                            tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
1866
                        }
1867
                    }
1868
                }
1869
            }
1870
        }
1871

    
1872
        non_zero_found = non_zero_found_short[0] |
1873
            non_zero_found_short[1] |
1874
            non_zero_found_short[2];
1875

    
1876
        for(i = g1->long_end - 1;i >= 0;i--) {
1877
            len = band_size_long[s->sample_rate_index][i];
1878
            tab0 -= len;
1879
            tab1 -= len;
1880
            /* test if non zero band. if so, stop doing i-stereo */
1881
            if (!non_zero_found) {
1882
                for(j=0;j<len;j++) {
1883
                    if (tab1[j] != 0) {
1884
                        non_zero_found = 1;
1885
                        goto found2;
1886
                    }
1887
                }
1888
                /* for last band, use previous scale factor */
1889
                k = (i == 21) ? 20 : i;
1890
                sf = g1->scale_factors[k];
1891
                if (sf >= sf_max)
1892
                    goto found2;
1893
                v1 = is_tab[0][sf];
1894
                v2 = is_tab[1][sf];
1895
                for(j=0;j<len;j++) {
1896
                    tmp0 = tab0[j];
1897
                    tab0[j] = MULL(tmp0, v1);
1898
                    tab1[j] = MULL(tmp0, v2);
1899
                }
1900
            } else {
1901
            found2:
1902
                if (s->mode_ext & MODE_EXT_MS_STEREO) {
1903
                    /* lower part of the spectrum : do ms stereo
1904
                       if enabled */
1905
                    for(j=0;j<len;j++) {
1906
                        tmp0 = tab0[j];
1907
                        tmp1 = tab1[j];
1908
                        tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
1909
                        tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
1910
                    }
1911
                }
1912
            }
1913
        }
1914
    } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1915
        /* ms stereo ONLY */
1916
        /* NOTE: the 1/sqrt(2) normalization factor is included in the
1917
           global gain */
1918
        tab0 = g0->sb_hybrid;
1919
        tab1 = g1->sb_hybrid;
1920
        for(i=0;i<576;i++) {
1921
            tmp0 = tab0[i];
1922
            tmp1 = tab1[i];
1923
            tab0[i] = tmp0 + tmp1;
1924
            tab1[i] = tmp0 - tmp1;
1925
        }
1926
    }
1927
}
1928

    
1929
static void compute_antialias_integer(MPADecodeContext *s,
1930
                              GranuleDef *g)
1931
{
1932
    int32_t *ptr, *csa;
1933
    int n, i;
1934

    
1935
    /* we antialias only "long" bands */
1936
    if (g->block_type == 2) {
1937
        if (!g->switch_point)
1938
            return;
1939
        /* XXX: check this for 8000Hz case */
1940
        n = 1;
1941
    } else {
1942
        n = SBLIMIT - 1;
1943
    }
1944

    
1945
    ptr = g->sb_hybrid + 18;
1946
    for(i = n;i > 0;i--) {
1947
        int tmp0, tmp1, tmp2;
1948
        csa = &csa_table[0][0];
1949
#define INT_AA(j) \
1950
            tmp0 = ptr[-1-j];\
1951
            tmp1 = ptr[   j];\
1952
            tmp2= MULH(tmp0 + tmp1, csa[0+4*j]);\
1953
            ptr[-1-j] = 4*(tmp2 - MULH(tmp1, csa[2+4*j]));\
1954
            ptr[   j] = 4*(tmp2 + MULH(tmp0, csa[3+4*j]));
1955

    
1956
        INT_AA(0)
1957
        INT_AA(1)
1958
        INT_AA(2)
1959
        INT_AA(3)
1960
        INT_AA(4)
1961
        INT_AA(5)
1962
        INT_AA(6)
1963
        INT_AA(7)
1964

    
1965
        ptr += 18;
1966
    }
1967
}
1968

    
1969
static void compute_antialias_float(MPADecodeContext *s,
1970
                              GranuleDef *g)
1971
{
1972
    int32_t *ptr;
1973
    int n, i;
1974

    
1975
    /* we antialias only "long" bands */
1976
    if (g->block_type == 2) {
1977
        if (!g->switch_point)
1978
            return;
1979
        /* XXX: check this for 8000Hz case */
1980
        n = 1;
1981
    } else {
1982
        n = SBLIMIT - 1;
1983
    }
1984

    
1985
    ptr = g->sb_hybrid + 18;
1986
    for(i = n;i > 0;i--) {
1987
        float tmp0, tmp1;
1988
        float *csa = &csa_table_float[0][0];
1989
#define FLOAT_AA(j)\
1990
        tmp0= ptr[-1-j];\
1991
        tmp1= ptr[   j];\
1992
        ptr[-1-j] = lrintf(tmp0 * csa[0+4*j] - tmp1 * csa[1+4*j]);\
1993
        ptr[   j] = lrintf(tmp0 * csa[1+4*j] + tmp1 * csa[0+4*j]);
1994

    
1995
        FLOAT_AA(0)
1996
        FLOAT_AA(1)
1997
        FLOAT_AA(2)
1998
        FLOAT_AA(3)
1999
        FLOAT_AA(4)
2000
        FLOAT_AA(5)
2001
        FLOAT_AA(6)
2002
        FLOAT_AA(7)
2003

    
2004
        ptr += 18;
2005
    }
2006
}
2007

    
2008
static void compute_imdct(MPADecodeContext *s,
2009
                          GranuleDef *g,
2010
                          int32_t *sb_samples,
2011
                          int32_t *mdct_buf)
2012
{
2013
    int32_t *ptr, *win, *win1, *buf, *out_ptr, *ptr1;
2014
    int32_t out2[12];
2015
    int i, j, mdct_long_end, v, sblimit;
2016

    
2017
    /* find last non zero block */
2018
    ptr = g->sb_hybrid + 576;
2019
    ptr1 = g->sb_hybrid + 2 * 18;
2020
    while (ptr >= ptr1) {
2021
        ptr -= 6;
2022
        v = ptr[0] | ptr[1] | ptr[2] | ptr[3] | ptr[4] | ptr[5];
2023
        if (v != 0)
2024
            break;
2025
    }
2026
    sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
2027

    
2028
    if (g->block_type == 2) {
2029
        /* XXX: check for 8000 Hz */
2030
        if (g->switch_point)
2031
            mdct_long_end = 2;
2032
        else
2033
            mdct_long_end = 0;
2034
    } else {
2035
        mdct_long_end = sblimit;
2036
    }
2037

    
2038
    buf = mdct_buf;
2039
    ptr = g->sb_hybrid;
2040
    for(j=0;j<mdct_long_end;j++) {
2041
        /* apply window & overlap with previous buffer */
2042
        out_ptr = sb_samples + j;
2043
        /* select window */
2044
        if (g->switch_point && j < 2)
2045
            win1 = mdct_win[0];
2046
        else
2047
            win1 = mdct_win[g->block_type];
2048
        /* select frequency inversion */
2049
        win = win1 + ((4 * 36) & -(j & 1));
2050
        imdct36(out_ptr, buf, ptr, win);
2051
        out_ptr += 18*SBLIMIT;
2052
        ptr += 18;
2053
        buf += 18;
2054
    }
2055
    for(j=mdct_long_end;j<sblimit;j++) {
2056
        /* select frequency inversion */
2057
        win = mdct_win[2] + ((4 * 36) & -(j & 1));
2058
        out_ptr = sb_samples + j;
2059

    
2060
        for(i=0; i<6; i++){
2061
            *out_ptr = buf[i];
2062
            out_ptr += SBLIMIT;
2063
        }
2064
        imdct12(out2, ptr + 0);
2065
        for(i=0;i<6;i++) {
2066
            *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*1];
2067
            buf[i + 6*2] = MULH(out2[i + 6], win[i + 6]);
2068
            out_ptr += SBLIMIT;
2069
        }
2070
        imdct12(out2, ptr + 1);
2071
        for(i=0;i<6;i++) {
2072
            *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*2];
2073
            buf[i + 6*0] = MULH(out2[i + 6], win[i + 6]);
2074
            out_ptr += SBLIMIT;
2075
        }
2076
        imdct12(out2, ptr + 2);
2077
        for(i=0;i<6;i++) {
2078
            buf[i + 6*0] = MULH(out2[i], win[i]) + buf[i + 6*0];
2079
            buf[i + 6*1] = MULH(out2[i + 6], win[i + 6]);
2080
            buf[i + 6*2] = 0;
2081
        }
2082
        ptr += 18;
2083
        buf += 18;
2084
    }
2085
    /* zero bands */
2086
    for(j=sblimit;j<SBLIMIT;j++) {
2087
        /* overlap */
2088
        out_ptr = sb_samples + j;
2089
        for(i=0;i<18;i++) {
2090
            *out_ptr = buf[i];
2091
            buf[i] = 0;
2092
            out_ptr += SBLIMIT;
2093
        }
2094
        buf += 18;
2095
    }
2096
}
2097

    
2098
#if defined(DEBUG)
2099
void sample_dump(int fnum, int32_t *tab, int n)
2100
{
2101
    static FILE *files[16], *f;
2102
    char buf[512];
2103
    int i;
2104
    int32_t v;
2105

    
2106
    f = files[fnum];
2107
    if (!f) {
2108
        snprintf(buf, sizeof(buf), "/tmp/out%d.%s.pcm",
2109
                fnum,
2110
#ifdef USE_HIGHPRECISION
2111
                "hp"
2112
#else
2113
                "lp"
2114
#endif
2115
                );
2116
        f = fopen(buf, "w");
2117
        if (!f)
2118
            return;
2119
        files[fnum] = f;
2120
    }
2121

    
2122
    if (fnum == 0) {
2123
        static int pos = 0;
2124
        av_log(NULL, AV_LOG_DEBUG, "pos=%d\n", pos);
2125
        for(i=0;i<n;i++) {
2126
            av_log(NULL, AV_LOG_DEBUG, " %0.4f", (double)tab[i] / FRAC_ONE);
2127
            if ((i % 18) == 17)
2128
                av_log(NULL, AV_LOG_DEBUG, "\n");
2129
        }
2130
        pos += n;
2131
    }
2132
    for(i=0;i<n;i++) {
2133
        /* normalize to 23 frac bits */
2134
        v = tab[i] << (23 - FRAC_BITS);
2135
        fwrite(&v, 1, sizeof(int32_t), f);
2136
    }
2137
}
2138
#endif
2139

    
2140

    
2141
/* main layer3 decoding function */
2142
static int mp_decode_layer3(MPADecodeContext *s)
2143
{
2144
    int nb_granules, main_data_begin, private_bits;
2145
    int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;
2146
    GranuleDef granules[2][2], *g;
2147
    int16_t exponents[576];
2148

    
2149
    /* read side info */
2150
    if (s->lsf) {
2151
        main_data_begin = get_bits(&s->gb, 8);
2152
        private_bits = get_bits(&s->gb, s->nb_channels);
2153
        nb_granules = 1;
2154
    } else {
2155
        main_data_begin = get_bits(&s->gb, 9);
2156
        if (s->nb_channels == 2)
2157
            private_bits = get_bits(&s->gb, 3);
2158
        else
2159
            private_bits = get_bits(&s->gb, 5);
2160
        nb_granules = 2;
2161
        for(ch=0;ch<s->nb_channels;ch++) {
2162
            granules[ch][0].scfsi = 0; /* all scale factors are transmitted */
2163
            granules[ch][1].scfsi = get_bits(&s->gb, 4);
2164
        }
2165
    }
2166

    
2167
    for(gr=0;gr<nb_granules;gr++) {
2168
        for(ch=0;ch<s->nb_channels;ch++) {
2169
            dprintf(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);
2170
            g = &granules[ch][gr];
2171
            g->part2_3_length = get_bits(&s->gb, 12);
2172
            g->big_values = get_bits(&s->gb, 9);
2173
            if(g->big_values > 288){
2174
                av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");
2175
                return -1;
2176
            }
2177

    
2178
            g->global_gain = get_bits(&s->gb, 8);
2179
            /* if MS stereo only is selected, we precompute the
2180
               1/sqrt(2) renormalization factor */
2181
            if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
2182
                MODE_EXT_MS_STEREO)
2183
                g->global_gain -= 2;
2184
            if (s->lsf)
2185
                g->scalefac_compress = get_bits(&s->gb, 9);
2186
            else
2187
                g->scalefac_compress = get_bits(&s->gb, 4);
2188
            blocksplit_flag = get_bits(&s->gb, 1);
2189
            if (blocksplit_flag) {
2190
                g->block_type = get_bits(&s->gb, 2);
2191
                if (g->block_type == 0){
2192
                    av_log(NULL, AV_LOG_ERROR, "invalid block type\n");
2193
                    return -1;
2194
                }
2195
                g->switch_point = get_bits(&s->gb, 1);
2196
                for(i=0;i<2;i++)
2197
                    g->table_select[i] = get_bits(&s->gb, 5);
2198
                for(i=0;i<3;i++)
2199
                    g->subblock_gain[i] = get_bits(&s->gb, 3);
2200
                /* compute huffman coded region sizes */
2201
                if (g->block_type == 2)
2202
                    g->region_size[0] = (36 / 2);
2203
                else {
2204
                    if (s->sample_rate_index <= 2)
2205
                        g->region_size[0] = (36 / 2);
2206
                    else if (s->sample_rate_index != 8)
2207
                        g->region_size[0] = (54 / 2);
2208
                    else
2209
                        g->region_size[0] = (108 / 2);
2210
                }
2211
                g->region_size[1] = (576 / 2);
2212
            } else {
2213
                int region_address1, region_address2, l;
2214
                g->block_type = 0;
2215
                g->switch_point = 0;
2216
                for(i=0;i<3;i++)
2217
                    g->table_select[i] = get_bits(&s->gb, 5);
2218
                /* compute huffman coded region sizes */
2219
                region_address1 = get_bits(&s->gb, 4);
2220
                region_address2 = get_bits(&s->gb, 3);
2221
                dprintf(s->avctx, "region1=%d region2=%d\n",
2222
                        region_address1, region_address2);
2223
                g->region_size[0] =
2224
                    band_index_long[s->sample_rate_index][region_address1 + 1] >> 1;
2225
                l = region_address1 + region_address2 + 2;
2226
                /* should not overflow */
2227
                if (l > 22)
2228
                    l = 22;
2229
                g->region_size[1] =
2230
                    band_index_long[s->sample_rate_index][l] >> 1;
2231
            }
2232
            /* convert region offsets to region sizes and truncate
2233
               size to big_values */
2234
            g->region_size[2] = (576 / 2);
2235
            j = 0;
2236
            for(i=0;i<3;i++) {
2237
                k = FFMIN(g->region_size[i], g->big_values);
2238
                g->region_size[i] = k - j;
2239
                j = k;
2240
            }
2241

    
2242
            /* compute band indexes */
2243
            if (g->block_type == 2) {
2244
                if (g->switch_point) {
2245
                    /* if switched mode, we handle the 36 first samples as
2246
                       long blocks.  For 8000Hz, we handle the 48 first
2247
                       exponents as long blocks (XXX: check this!) */
2248
                    if (s->sample_rate_index <= 2)
2249
                        g->long_end = 8;
2250
                    else if (s->sample_rate_index != 8)
2251
                        g->long_end = 6;
2252
                    else
2253
                        g->long_end = 4; /* 8000 Hz */
2254

    
2255
                    g->short_start = 2 + (s->sample_rate_index != 8);
2256
                } else {
2257
                    g->long_end = 0;
2258
                    g->short_start = 0;
2259
                }
2260
            } else {
2261
                g->short_start = 13;
2262
                g->long_end = 22;
2263
            }
2264

    
2265
            g->preflag = 0;
2266
            if (!s->lsf)
2267
                g->preflag = get_bits(&s->gb, 1);
2268
            g->scalefac_scale = get_bits(&s->gb, 1);
2269
            g->count1table_select = get_bits(&s->gb, 1);
2270
            dprintf(s->avctx, "block_type=%d switch_point=%d\n",
2271
                    g->block_type, g->switch_point);
2272
        }
2273
    }
2274

    
2275
  if (!s->adu_mode) {
2276
    const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);
2277
    assert((get_bits_count(&s->gb) & 7) == 0);
2278
    /* now we get bits from the main_data_begin offset */
2279
    dprintf(s->avctx, "seekback: %d\n", main_data_begin);
2280
//av_log(NULL, AV_LOG_ERROR, "backstep:%d, lastbuf:%d\n", main_data_begin, s->last_buf_size);
2281

    
2282
    memcpy(s->last_buf + s->last_buf_size, ptr, EXTRABYTES);
2283
    s->in_gb= s->gb;
2284
        init_get_bits(&s->gb, s->last_buf, s->last_buf_size*8);
2285
        skip_bits_long(&s->gb, 8*(s->last_buf_size - main_data_begin));
2286
  }
2287

    
2288
    for(gr=0;gr<nb_granules;gr++) {
2289
        for(ch=0;ch<s->nb_channels;ch++) {
2290
            g = &granules[ch][gr];
2291
            if(get_bits_count(&s->gb)<0){
2292
                av_log(NULL, AV_LOG_ERROR, "mdb:%d, lastbuf:%d skipping granule %d\n",
2293
                                            main_data_begin, s->last_buf_size, gr);
2294
                skip_bits_long(&s->gb, g->part2_3_length);
2295
                memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));
2296
                if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->in_gb.buffer){
2297
                    skip_bits_long(&s->in_gb, get_bits_count(&s->gb) - s->gb.size_in_bits);
2298
                    s->gb= s->in_gb;
2299
                    s->in_gb.buffer=NULL;
2300
                }
2301
                continue;
2302
            }
2303

    
2304
            bits_pos = get_bits_count(&s->gb);
2305

    
2306
            if (!s->lsf) {
2307
                uint8_t *sc;
2308
                int slen, slen1, slen2;
2309

    
2310
                /* MPEG1 scale factors */
2311
                slen1 = slen_table[0][g->scalefac_compress];
2312
                slen2 = slen_table[1][g->scalefac_compress];
2313
                dprintf(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);
2314
                if (g->block_type == 2) {
2315
                    n = g->switch_point ? 17 : 18;
2316
                    j = 0;
2317
                    if(slen1){
2318
                        for(i=0;i<n;i++)
2319
                            g->scale_factors[j++] = get_bits(&s->gb, slen1);
2320
                    }else{
2321
                        for(i=0;i<n;i++)
2322
                            g->scale_factors[j++] = 0;
2323
                    }
2324
                    if(slen2){
2325
                        for(i=0;i<18;i++)
2326
                            g->scale_factors[j++] = get_bits(&s->gb, slen2);
2327
                        for(i=0;i<3;i++)
2328
                            g->scale_factors[j++] = 0;
2329
                    }else{
2330
                        for(i=0;i<21;i++)
2331
                            g->scale_factors[j++] = 0;
2332
                    }
2333
                } else {
2334
                    sc = granules[ch][0].scale_factors;
2335
                    j = 0;
2336
                    for(k=0;k<4;k++) {
2337
                        n = (k == 0 ? 6 : 5);
2338
                        if ((g->scfsi & (0x8 >> k)) == 0) {
2339
                            slen = (k < 2) ? slen1 : slen2;
2340
                            if(slen){
2341
                                for(i=0;i<n;i++)
2342
                                    g->scale_factors[j++] = get_bits(&s->gb, slen);
2343
                            }else{
2344
                                for(i=0;i<n;i++)
2345
                                    g->scale_factors[j++] = 0;
2346
                            }
2347
                        } else {
2348
                            /* simply copy from last granule */
2349
                            for(i=0;i<n;i++) {
2350
                                g->scale_factors[j] = sc[j];
2351
                                j++;
2352
                            }
2353
                        }
2354
                    }
2355
                    g->scale_factors[j++] = 0;
2356
                }
2357
#if defined(DEBUG)
2358
                {
2359
                    dprintf(s->avctx, "scfsi=%x gr=%d ch=%d scale_factors:\n",
2360
                           g->scfsi, gr, ch);
2361
                    for(i=0;i<j;i++)
2362
                        dprintf(s->avctx, " %d", g->scale_factors[i]);
2363
                    dprintf(s->avctx, "\n");
2364
                }
2365
#endif
2366
            } else {
2367
                int tindex, tindex2, slen[4], sl, sf;
2368

    
2369
                /* LSF scale factors */
2370
                if (g->block_type == 2) {
2371
                    tindex = g->switch_point ? 2 : 1;
2372
                } else {
2373
                    tindex = 0;
2374
                }
2375
                sf = g->scalefac_compress;
2376
                if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
2377
                    /* intensity stereo case */
2378
                    sf >>= 1;
2379
                    if (sf < 180) {
2380
                        lsf_sf_expand(slen, sf, 6, 6, 0);
2381
                        tindex2 = 3;
2382
                    } else if (sf < 244) {
2383
                        lsf_sf_expand(slen, sf - 180, 4, 4, 0);
2384
                        tindex2 = 4;
2385
                    } else {
2386
                        lsf_sf_expand(slen, sf - 244, 3, 0, 0);
2387
                        tindex2 = 5;
2388
                    }
2389
                } else {
2390
                    /* normal case */
2391
                    if (sf < 400) {
2392
                        lsf_sf_expand(slen, sf, 5, 4, 4);
2393
                        tindex2 = 0;
2394
                    } else if (sf < 500) {
2395
                        lsf_sf_expand(slen, sf - 400, 5, 4, 0);
2396
                        tindex2 = 1;
2397
                    } else {
2398
                        lsf_sf_expand(slen, sf - 500, 3, 0, 0);
2399
                        tindex2 = 2;
2400
                        g->preflag = 1;
2401
                    }
2402
                }
2403

    
2404
                j = 0;
2405
                for(k=0;k<4;k++) {
2406
                    n = lsf_nsf_table[tindex2][tindex][k];
2407
                    sl = slen[k];
2408
                    if(sl){
2409
                        for(i=0;i<n;i++)
2410
                            g->scale_factors[j++] = get_bits(&s->gb, sl);
2411
                    }else{
2412
                        for(i=0;i<n;i++)
2413
                            g->scale_factors[j++] = 0;
2414
                    }
2415
                }
2416
                /* XXX: should compute exact size */
2417
                for(;j<40;j++)
2418
                    g->scale_factors[j] = 0;
2419
#if defined(DEBUG)
2420
                {
2421
                    dprintf(s->avctx, "gr=%d ch=%d scale_factors:\n",
2422
                           gr, ch);
2423
                    for(i=0;i<40;i++)
2424
                        dprintf(s->avctx, " %d", g->scale_factors[i]);
2425
                    dprintf(s->avctx, "\n");
2426
                }
2427
#endif
2428
            }
2429

    
2430
            exponents_from_scale_factors(s, g, exponents);
2431

    
2432
            /* read Huffman coded residue */
2433
            huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);
2434
#if defined(DEBUG)
2435
            sample_dump(0, g->sb_hybrid, 576);
2436
#endif
2437
        } /* ch */
2438

    
2439
        if (s->nb_channels == 2)
2440
            compute_stereo(s, &granules[0][gr], &granules[1][gr]);
2441

    
2442
        for(ch=0;ch<s->nb_channels;ch++) {
2443
            g = &granules[ch][gr];
2444

    
2445
            reorder_block(s, g);
2446
#if defined(DEBUG)
2447
            sample_dump(0, g->sb_hybrid, 576);
2448
#endif
2449
            s->compute_antialias(s, g);
2450
#if defined(DEBUG)
2451
            sample_dump(1, g->sb_hybrid, 576);
2452
#endif
2453
            compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
2454
#if defined(DEBUG)
2455
            sample_dump(2, &s->sb_samples[ch][18 * gr][0], 576);
2456
#endif
2457
        }
2458
    } /* gr */
2459
    if(get_bits_count(&s->gb)<0)
2460
        skip_bits_long(&s->gb, -get_bits_count(&s->gb));
2461
    return nb_granules * 18;
2462
}
2463

    
2464
static int mp_decode_frame(MPADecodeContext *s,
2465
                           OUT_INT *samples, const uint8_t *buf, int buf_size)
2466
{
2467
    int i, nb_frames, ch;
2468
    OUT_INT *samples_ptr;
2469

    
2470
    init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE)*8);
2471

    
2472
    /* skip error protection field */
2473
    if (s->error_protection)
2474
        get_bits(&s->gb, 16);
2475

    
2476
    dprintf(s->avctx, "frame %d:\n", s->frame_count);
2477
    switch(s->layer) {
2478
    case 1:
2479
        nb_frames = mp_decode_layer1(s);
2480
        break;
2481
    case 2:
2482
        nb_frames = mp_decode_layer2(s);
2483
        break;
2484
    case 3:
2485
    default:
2486
        nb_frames = mp_decode_layer3(s);
2487

    
2488
        s->last_buf_size=0;
2489
        if(s->in_gb.buffer){
2490
            align_get_bits(&s->gb);
2491
            i= (s->gb.size_in_bits - get_bits_count(&s->gb))>>3;
2492
            if(i >= 0 && i <= BACKSTEP_SIZE){
2493
                memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);
2494
                s->last_buf_size=i;
2495
            }else
2496
                av_log(NULL, AV_LOG_ERROR, "invalid old backstep %d\n", i);
2497
            s->gb= s->in_gb;
2498
            s->in_gb.buffer= NULL;
2499
        }
2500

    
2501
        align_get_bits(&s->gb);
2502
        assert((get_bits_count(&s->gb) & 7) == 0);
2503
        i= (s->gb.size_in_bits - get_bits_count(&s->gb))>>3;
2504

    
2505
        if(i<0 || i > BACKSTEP_SIZE || nb_frames<0){
2506
            av_log(NULL, AV_LOG_ERROR, "invalid new backstep %d\n", i);
2507
            i= FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);
2508
        }
2509
        assert(i <= buf_size - HEADER_SIZE && i>= 0);
2510
        memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);
2511
        s->last_buf_size += i;
2512

    
2513
        break;
2514
    }
2515
#if defined(DEBUG)
2516
    for(i=0;i<nb_frames;i++) {
2517
        for(ch=0;ch<s->nb_channels;ch++) {
2518
            int j;
2519
            dprintf(s->avctx, "%d-%d:", i, ch);
2520
            for(j=0;j<SBLIMIT;j++)
2521
                dprintf(s->avctx, " %0.6f", (double)s->sb_samples[ch][i][j] / FRAC_ONE);
2522
            dprintf(s->avctx, "\n");
2523
        }
2524
    }
2525
#endif
2526
    /* apply the synthesis filter */
2527
    for(ch=0;ch<s->nb_channels;ch++) {
2528
        samples_ptr = samples + ch;
2529
        for(i=0;i<nb_frames;i++) {
2530
            ff_mpa_synth_filter(s->synth_buf[ch], &(s->synth_buf_offset[ch]),
2531
                         window, &s->dither_state,
2532
                         samples_ptr, s->nb_channels,
2533
                         s->sb_samples[ch][i]);
2534
            samples_ptr += 32 * s->nb_channels;
2535
        }
2536
    }
2537
#ifdef DEBUG
2538
    s->frame_count++;
2539
#endif
2540
    return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
2541
}
2542

    
2543
static int decode_frame(AVCodecContext * avctx,
2544
                        void *data, int *data_size,
2545
                        uint8_t * buf, int buf_size)
2546
{
2547
    MPADecodeContext *s = avctx->priv_data;
2548
    uint32_t header;
2549
    int out_size;
2550
    OUT_INT *out_samples = data;
2551

    
2552
retry:
2553
    if(buf_size < HEADER_SIZE)
2554
        return -1;
2555

    
2556
    header = (buf[0] << 24) | (buf[1] << 16) | (buf[2] << 8) | buf[3];
2557
    if(ff_mpa_check_header(header) < 0){
2558
        buf++;
2559
//        buf_size--;
2560
        av_log(avctx, AV_LOG_ERROR, "Header missing skipping one byte.\n");
2561
        goto retry;
2562
    }
2563

    
2564
    if (decode_header(s, header) == 1) {
2565
        /* free format: prepare to compute frame size */
2566
        s->frame_size = -1;
2567
        return -1;
2568
    }
2569
    /* update codec info */
2570
    avctx->channels = s->nb_channels;
2571
    avctx->bit_rate = s->bit_rate;
2572
    avctx->sub_id = s->layer;
2573
    switch(s->layer) {
2574
    case 1:
2575
        avctx->frame_size = 384;
2576
        break;
2577
    case 2:
2578
        avctx->frame_size = 1152;
2579
        break;
2580
    case 3:
2581
        if (s->lsf)
2582
            avctx->frame_size = 576;
2583
        else
2584
            avctx->frame_size = 1152;
2585
        break;
2586
    }
2587

    
2588
    if(s->frame_size<=0 || s->frame_size > buf_size){
2589
        av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
2590
        return -1;
2591
    }else if(s->frame_size < buf_size){
2592
        av_log(avctx, AV_LOG_ERROR, "incorrect frame size\n");
2593
    }
2594

    
2595
    out_size = mp_decode_frame(s, out_samples, buf, buf_size);
2596
    if(out_size>=0){
2597
        *data_size = out_size;
2598
        avctx->sample_rate = s->sample_rate;
2599
        //FIXME maybe move the other codec info stuff from above here too
2600
    }else
2601
        av_log(avctx, AV_LOG_DEBUG, "Error while decoding MPEG audio frame.\n"); //FIXME return -1 / but also return the number of bytes consumed
2602
    s->frame_size = 0;
2603
    return buf_size;
2604
}
2605

    
2606
static void flush(AVCodecContext *avctx){
2607
    MPADecodeContext *s = avctx->priv_data;
2608
    s->last_buf_size= 0;
2609
}
2610

    
2611
#ifdef CONFIG_MP3ADU_DECODER
2612
static int decode_frame_adu(AVCodecContext * avctx,
2613
                        void *data, int *data_size,
2614
                        uint8_t * buf, int buf_size)
2615
{
2616
    MPADecodeContext *s = avctx->priv_data;
2617
    uint32_t header;
2618
    int len, out_size;
2619
    OUT_INT *out_samples = data;
2620

    
2621
    len = buf_size;
2622

    
2623
    // Discard too short frames
2624
    if (buf_size < HEADER_SIZE) {
2625
        *data_size = 0;
2626
        return buf_size;
2627
    }
2628

    
2629

    
2630
    if (len > MPA_MAX_CODED_FRAME_SIZE)
2631
        len = MPA_MAX_CODED_FRAME_SIZE;
2632

    
2633
    // Get header and restore sync word
2634
    header = (buf[0] << 24) | (buf[1] << 16) | (buf[2] << 8) | buf[3] | 0xffe00000;
2635

    
2636
    if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame
2637
        *data_size = 0;
2638
        return buf_size;
2639
    }
2640

    
2641
    decode_header(s, header);
2642
    /* update codec info */
2643
    avctx->sample_rate = s->sample_rate;
2644
    avctx->channels = s->nb_channels;
2645
    avctx->bit_rate = s->bit_rate;
2646
    avctx->sub_id = s->layer;
2647

    
2648
    avctx->frame_size=s->frame_size = len;
2649

    
2650
    if (avctx->parse_only) {
2651
        out_size = buf_size;
2652
    } else {
2653
        out_size = mp_decode_frame(s, out_samples, buf, buf_size);
2654
    }
2655

    
2656
    *data_size = out_size;
2657
    return buf_size;
2658
}
2659
#endif /* CONFIG_MP3ADU_DECODER */
2660

    
2661
#ifdef CONFIG_MP3ON4_DECODER
2662
/* Next 3 arrays are indexed by channel config number (passed via codecdata) */
2663
static int mp3Frames[16] = {0,1,1,2,3,3,4,5,2};   /* number of mp3 decoder instances */
2664
static int mp3Channels[16] = {0,1,2,3,4,5,6,8,4}; /* total output channels */
2665
/* offsets into output buffer, assume output order is FL FR BL BR C LFE */
2666
static int chan_offset[9][5] = {
2667
    {0},
2668
    {0},            // C
2669
    {0},            // FLR
2670
    {2,0},          // C FLR
2671
    {2,0,3},        // C FLR BS
2672
    {4,0,2},        // C FLR BLRS
2673
    {4,0,2,5},      // C FLR BLRS LFE
2674
    {4,0,2,6,5},    // C FLR BLRS BLR LFE
2675
    {0,2}           // FLR BLRS
2676
};
2677

    
2678

    
2679
static int decode_init_mp3on4(AVCodecContext * avctx)
2680
{
2681
    MP3On4DecodeContext *s = avctx->priv_data;
2682
    int i;
2683

    
2684
    if ((avctx->extradata_size < 2) || (avctx->extradata == NULL)) {
2685
        av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
2686
        return -1;
2687
    }
2688

    
2689
    s->chan_cfg = (((unsigned char *)avctx->extradata)[1] >> 3) & 0x0f;
2690
    s->frames = mp3Frames[s->chan_cfg];
2691
    if(!s->frames) {
2692
        av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
2693
        return -1;
2694
    }
2695
    avctx->channels = mp3Channels[s->chan_cfg];
2696

    
2697
    /* Init the first mp3 decoder in standard way, so that all tables get builded
2698
     * We replace avctx->priv_data with the context of the first decoder so that
2699
     * decode_init() does not have to be changed.
2700
     * Other decoders will be inited here copying data from the first context
2701
     */
2702
    // Allocate zeroed memory for the first decoder context
2703
    s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
2704
    // Put decoder context in place to make init_decode() happy
2705
    avctx->priv_data = s->mp3decctx[0];
2706
    decode_init(avctx);
2707
    // Restore mp3on4 context pointer
2708
    avctx->priv_data = s;
2709
    s->mp3decctx[0]->adu_mode = 1; // Set adu mode
2710

    
2711
    /* Create a separate codec/context for each frame (first is already ok).
2712
     * Each frame is 1 or 2 channels - up to 5 frames allowed
2713
     */
2714
    for (i = 1; i < s->frames; i++) {
2715
        s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
2716
        s->mp3decctx[i]->compute_antialias = s->mp3decctx[0]->compute_antialias;
2717
        s->mp3decctx[i]->adu_mode = 1;
2718
        s->mp3decctx[i]->avctx = avctx;
2719
    }
2720

    
2721
    return 0;
2722
}
2723

    
2724

    
2725
static int decode_close_mp3on4(AVCodecContext * avctx)
2726
{
2727
    MP3On4DecodeContext *s = avctx->priv_data;
2728
    int i;
2729

    
2730
    for (i = 0; i < s->frames; i++)
2731
        if (s->mp3decctx[i])
2732
            av_free(s->mp3decctx[i]);
2733

    
2734
    return 0;
2735
}
2736

    
2737

    
2738
static int decode_frame_mp3on4(AVCodecContext * avctx,
2739
                        void *data, int *data_size,
2740
                        uint8_t * buf, int buf_size)
2741
{
2742
    MP3On4DecodeContext *s = avctx->priv_data;
2743
    MPADecodeContext *m;
2744
    int len, out_size = 0;
2745
    uint32_t header;
2746
    OUT_INT *out_samples = data;
2747
    OUT_INT decoded_buf[MPA_FRAME_SIZE * MPA_MAX_CHANNELS];
2748
    OUT_INT *outptr, *bp;
2749
    int fsize;
2750
    unsigned char *start2 = buf, *start;
2751
    int fr, i, j, n;
2752
    int off = avctx->channels;
2753
    int *coff = chan_offset[s->chan_cfg];
2754

    
2755
    len = buf_size;
2756

    
2757
    // Discard too short frames
2758
    if (buf_size < HEADER_SIZE) {
2759
        *data_size = 0;
2760
        return buf_size;
2761
    }
2762

    
2763
    // If only one decoder interleave is not needed
2764
    outptr = s->frames == 1 ? out_samples : decoded_buf;
2765

    
2766
    for (fr = 0; fr < s->frames; fr++) {
2767
        start = start2;
2768
        fsize = (start[0] << 4) | (start[1] >> 4);
2769
        start2 += fsize;
2770
        if (fsize > len)
2771
            fsize = len;
2772
        len -= fsize;
2773
        if (fsize > MPA_MAX_CODED_FRAME_SIZE)
2774
            fsize = MPA_MAX_CODED_FRAME_SIZE;
2775
        m = s->mp3decctx[fr];
2776
        assert (m != NULL);
2777

    
2778
        // Get header
2779
        header = (start[0] << 24) | (start[1] << 16) | (start[2] << 8) | start[3] | 0xfff00000;
2780

    
2781
        if (ff_mpa_check_header(header) < 0) { // Bad header, discard block
2782
            *data_size = 0;
2783
            return buf_size;
2784
        }
2785

    
2786
        decode_header(m, header);
2787
        mp_decode_frame(m, decoded_buf, start, fsize);
2788

    
2789
        n = MPA_FRAME_SIZE * m->nb_channels;
2790
        out_size += n * sizeof(OUT_INT);
2791
        if(s->frames > 1) {
2792
            /* interleave output data */
2793
            bp = out_samples + coff[fr];
2794
            if(m->nb_channels == 1) {
2795
                for(j = 0; j < n; j++) {
2796
                    *bp = decoded_buf[j];
2797
                    bp += off;
2798
                }
2799
            } else {
2800
                for(j = 0; j < n; j++) {
2801
                    bp[0] = decoded_buf[j++];
2802
                    bp[1] = decoded_buf[j];
2803
                    bp += off;
2804
                }
2805
            }
2806
        }
2807
    }
2808

    
2809
    /* update codec info */
2810
    avctx->sample_rate = s->mp3decctx[0]->sample_rate;
2811
    avctx->frame_size= buf_size;
2812
    avctx->bit_rate = 0;
2813
    for (i = 0; i < s->frames; i++)
2814
        avctx->bit_rate += s->mp3decctx[i]->bit_rate;
2815

    
2816
    *data_size = out_size;
2817
    return buf_size;
2818
}
2819
#endif /* CONFIG_MP3ON4_DECODER */
2820

    
2821
#ifdef CONFIG_MP2_DECODER
2822
AVCodec mp2_decoder =
2823
{
2824
    "mp2",
2825
    CODEC_TYPE_AUDIO,
2826
    CODEC_ID_MP2,
2827
    sizeof(MPADecodeContext),
2828
    decode_init,
2829
    NULL,
2830
    NULL,
2831
    decode_frame,
2832
    CODEC_CAP_PARSE_ONLY,
2833
};
2834
#endif
2835
#ifdef CONFIG_MP3_DECODER
2836
AVCodec mp3_decoder =
2837
{
2838
    "mp3",
2839
    CODEC_TYPE_AUDIO,
2840
    CODEC_ID_MP3,
2841
    sizeof(MPADecodeContext),
2842
    decode_init,
2843
    NULL,
2844
    NULL,
2845
    decode_frame,
2846
    CODEC_CAP_PARSE_ONLY,
2847
    .flush= flush,
2848
};
2849
#endif
2850
#ifdef CONFIG_MP3ADU_DECODER
2851
AVCodec mp3adu_decoder =
2852
{
2853
    "mp3adu",
2854
    CODEC_TYPE_AUDIO,
2855
    CODEC_ID_MP3ADU,
2856
    sizeof(MPADecodeContext),
2857
    decode_init,
2858
    NULL,
2859
    NULL,
2860
    decode_frame_adu,
2861
    CODEC_CAP_PARSE_ONLY,
2862
    .flush= flush,
2863
};
2864
#endif
2865
#ifdef CONFIG_MP3ON4_DECODER
2866
AVCodec mp3on4_decoder =
2867
{
2868
    "mp3on4",
2869
    CODEC_TYPE_AUDIO,
2870
    CODEC_ID_MP3ON4,
2871
    sizeof(MP3On4DecodeContext),
2872
    decode_init_mp3on4,
2873
    NULL,
2874
    decode_close_mp3on4,
2875
    decode_frame_mp3on4,
2876
    .flush= flush,
2877
};
2878
#endif