Statistics
| Branch: | Revision:

ffmpeg / libavcodec / mpegaudiodec.c @ 4bd8e17c

History | View | Annotate | Download (81.1 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
#include "mpegaudiodecheader.h"
46

    
47
#include "mathops.h"
48

    
49
#define FRAC_ONE    (1 << FRAC_BITS)
50

    
51
#define FIX(a)   ((int)((a) * FRAC_ONE))
52
/* WARNING: only correct for posititive numbers */
53
#define FIXR(a)   ((int)((a) * FRAC_ONE + 0.5))
54
#define FRAC_RND(a) (((a) + (FRAC_ONE/2)) >> FRAC_BITS)
55

    
56
#define FIXHR(a) ((int)((a) * (1LL<<32) + 0.5))
57

    
58
/****************/
59

    
60
#define HEADER_SIZE 4
61

    
62
/**
63
 * Context for MP3On4 decoder
64
 */
65
typedef struct MP3On4DecodeContext {
66
    int frames;   ///< number of mp3 frames per block (number of mp3 decoder instances)
67
    int chan_cfg; ///< channel config number
68
    MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance
69
} MP3On4DecodeContext;
70

    
71
/* layer 3 "granule" */
72
typedef struct GranuleDef {
73
    uint8_t scfsi;
74
    int part2_3_length;
75
    int big_values;
76
    int global_gain;
77
    int scalefac_compress;
78
    uint8_t block_type;
79
    uint8_t switch_point;
80
    int table_select[3];
81
    int subblock_gain[3];
82
    uint8_t scalefac_scale;
83
    uint8_t count1table_select;
84
    int region_size[3]; /* number of huffman codes in each region */
85
    int preflag;
86
    int short_start, long_end; /* long/short band indexes */
87
    uint8_t scale_factors[40];
88
    int32_t sb_hybrid[SBLIMIT * 18]; /* 576 samples */
89
} GranuleDef;
90

    
91
#define MODE_EXT_MS_STEREO 2
92
#define MODE_EXT_I_STEREO  1
93

    
94
/* layer 3 huffman tables */
95
typedef struct HuffTable {
96
    int xsize;
97
    const uint8_t *bits;
98
    const uint16_t *codes;
99
} HuffTable;
100

    
101
#include "mpegaudiodata.h"
102
#include "mpegaudiodectab.h"
103

    
104
static void compute_antialias_integer(MPADecodeContext *s, GranuleDef *g);
105
static void compute_antialias_float(MPADecodeContext *s, GranuleDef *g);
106

    
107
/* vlc structure for decoding layer 3 huffman tables */
108
static VLC huff_vlc[16];
109
static VLC huff_quad_vlc[2];
110
/* computed from band_size_long */
111
static uint16_t band_index_long[9][23];
112
/* XXX: free when all decoders are closed */
113
#define TABLE_4_3_SIZE (8191 + 16)*4
114
static int8_t  table_4_3_exp[TABLE_4_3_SIZE];
115
static uint32_t table_4_3_value[TABLE_4_3_SIZE];
116
static uint32_t exp_table[512];
117
static uint32_t expval_table[512][16];
118
/* intensity stereo coef table */
119
static int32_t is_table[2][16];
120
static int32_t is_table_lsf[2][2][16];
121
static int32_t csa_table[8][4];
122
static float csa_table_float[8][4];
123
static int32_t mdct_win[8][36];
124

    
125
/* lower 2 bits: modulo 3, higher bits: shift */
126
static uint16_t scale_factor_modshift[64];
127
/* [i][j]:  2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
128
static int32_t scale_factor_mult[15][3];
129
/* mult table for layer 2 group quantization */
130

    
131
#define SCALE_GEN(v) \
132
{ FIXR(1.0 * (v)), FIXR(0.7937005259 * (v)), FIXR(0.6299605249 * (v)) }
133

    
134
static const int32_t scale_factor_mult2[3][3] = {
135
    SCALE_GEN(4.0 / 3.0), /* 3 steps */
136
    SCALE_GEN(4.0 / 5.0), /* 5 steps */
137
    SCALE_GEN(4.0 / 9.0), /* 9 steps */
138
};
139

    
140
static DECLARE_ALIGNED_16(MPA_INT, window[512]);
141

    
142
/* layer 1 unscaling */
143
/* n = number of bits of the mantissa minus 1 */
144
static inline int l1_unscale(int n, int mant, int scale_factor)
145
{
146
    int shift, mod;
147
    int64_t val;
148

    
149
    shift = scale_factor_modshift[scale_factor];
150
    mod = shift & 3;
151
    shift >>= 2;
152
    val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);
153
    shift += n;
154
    /* NOTE: at this point, 1 <= shift >= 21 + 15 */
155
    return (int)((val + (1LL << (shift - 1))) >> shift);
156
}
157

    
158
static inline int l2_unscale_group(int steps, int mant, int scale_factor)
159
{
160
    int shift, mod, val;
161

    
162
    shift = scale_factor_modshift[scale_factor];
163
    mod = shift & 3;
164
    shift >>= 2;
165

    
166
    val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
167
    /* NOTE: at this point, 0 <= shift <= 21 */
168
    if (shift > 0)
169
        val = (val + (1 << (shift - 1))) >> shift;
170
    return val;
171
}
172

    
173
/* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
174
static inline int l3_unscale(int value, int exponent)
175
{
176
    unsigned int m;
177
    int e;
178

    
179
    e = table_4_3_exp  [4*value + (exponent&3)];
180
    m = table_4_3_value[4*value + (exponent&3)];
181
    e -= (exponent >> 2);
182
    assert(e>=1);
183
    if (e > 31)
184
        return 0;
185
    m = (m + (1 << (e-1))) >> e;
186

    
187
    return m;
188
}
189

    
190
/* all integer n^(4/3) computation code */
191
#define DEV_ORDER 13
192

    
193
#define POW_FRAC_BITS 24
194
#define POW_FRAC_ONE    (1 << POW_FRAC_BITS)
195
#define POW_FIX(a)   ((int)((a) * POW_FRAC_ONE))
196
#define POW_MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> POW_FRAC_BITS)
197

    
198
static int dev_4_3_coefs[DEV_ORDER];
199

    
200
#if 0 /* unused */
201
static int pow_mult3[3] = {
202
    POW_FIX(1.0),
203
    POW_FIX(1.25992104989487316476),
204
    POW_FIX(1.58740105196819947474),
205
};
206
#endif
207

    
208
static void int_pow_init(void)
209
{
210
    int i, a;
211

    
212
    a = POW_FIX(1.0);
213
    for(i=0;i<DEV_ORDER;i++) {
214
        a = POW_MULL(a, POW_FIX(4.0 / 3.0) - i * POW_FIX(1.0)) / (i + 1);
215
        dev_4_3_coefs[i] = a;
216
    }
217
}
218

    
219
#if 0 /* unused, remove? */
220
/* return the mantissa and the binary exponent */
221
static int int_pow(int i, int *exp_ptr)
222
{
223
    int e, er, eq, j;
224
    int a, a1;
225

226
    /* renormalize */
227
    a = i;
228
    e = POW_FRAC_BITS;
229
    while (a < (1 << (POW_FRAC_BITS - 1))) {
230
        a = a << 1;
231
        e--;
232
    }
233
    a -= (1 << POW_FRAC_BITS);
234
    a1 = 0;
235
    for(j = DEV_ORDER - 1; j >= 0; j--)
236
        a1 = POW_MULL(a, dev_4_3_coefs[j] + a1);
237
    a = (1 << POW_FRAC_BITS) + a1;
238
    /* exponent compute (exact) */
239
    e = e * 4;
240
    er = e % 3;
241
    eq = e / 3;
242
    a = POW_MULL(a, pow_mult3[er]);
243
    while (a >= 2 * POW_FRAC_ONE) {
244
        a = a >> 1;
245
        eq++;
246
    }
247
    /* convert to float */
248
    while (a < POW_FRAC_ONE) {
249
        a = a << 1;
250
        eq--;
251
    }
252
    /* now POW_FRAC_ONE <= a < 2 * POW_FRAC_ONE */
253
#if POW_FRAC_BITS > FRAC_BITS
254
    a = (a + (1 << (POW_FRAC_BITS - FRAC_BITS - 1))) >> (POW_FRAC_BITS - FRAC_BITS);
255
    /* correct overflow */
256
    if (a >= 2 * (1 << FRAC_BITS)) {
257
        a = a >> 1;
258
        eq++;
259
    }
260
#endif
261
    *exp_ptr = eq;
262
    return a;
263
}
264
#endif
265

    
266
static int decode_init(AVCodecContext * avctx)
267
{
268
    MPADecodeContext *s = avctx->priv_data;
269
    static int init=0;
270
    int i, j, k;
271

    
272
    s->avctx = avctx;
273

    
274
#if defined(USE_HIGHPRECISION) && defined(CONFIG_AUDIO_NONSHORT)
275
    avctx->sample_fmt= SAMPLE_FMT_S32;
276
#else
277
    avctx->sample_fmt= SAMPLE_FMT_S16;
278
#endif
279
    s->error_resilience= avctx->error_resilience;
280

    
281
    if(avctx->antialias_algo != FF_AA_FLOAT)
282
        s->compute_antialias= compute_antialias_integer;
283
    else
284
        s->compute_antialias= compute_antialias_float;
285

    
286
    if (!init && !avctx->parse_only) {
287
        /* scale factors table for layer 1/2 */
288
        for(i=0;i<64;i++) {
289
            int shift, mod;
290
            /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */
291
            shift = (i / 3);
292
            mod = i % 3;
293
            scale_factor_modshift[i] = mod | (shift << 2);
294
        }
295

    
296
        /* scale factor multiply for layer 1 */
297
        for(i=0;i<15;i++) {
298
            int n, norm;
299
            n = i + 2;
300
            norm = ((INT64_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
301
            scale_factor_mult[i][0] = MULL(FIXR(1.0 * 2.0), norm);
302
            scale_factor_mult[i][1] = MULL(FIXR(0.7937005259 * 2.0), norm);
303
            scale_factor_mult[i][2] = MULL(FIXR(0.6299605249 * 2.0), norm);
304
            dprintf(avctx, "%d: norm=%x s=%x %x %x\n",
305
                    i, norm,
306
                    scale_factor_mult[i][0],
307
                    scale_factor_mult[i][1],
308
                    scale_factor_mult[i][2]);
309
        }
310

    
311
        ff_mpa_synth_init(window);
312

    
313
        /* huffman decode tables */
314
        for(i=1;i<16;i++) {
315
            const HuffTable *h = &mpa_huff_tables[i];
316
            int xsize, x, y;
317
            unsigned int n;
318
            uint8_t  tmp_bits [512];
319
            uint16_t tmp_codes[512];
320

    
321
            memset(tmp_bits , 0, sizeof(tmp_bits ));
322
            memset(tmp_codes, 0, sizeof(tmp_codes));
323

    
324
            xsize = h->xsize;
325
            n = xsize * xsize;
326

    
327
            j = 0;
328
            for(x=0;x<xsize;x++) {
329
                for(y=0;y<xsize;y++){
330
                    tmp_bits [(x << 5) | y | ((x&&y)<<4)]= h->bits [j  ];
331
                    tmp_codes[(x << 5) | y | ((x&&y)<<4)]= h->codes[j++];
332
                }
333
            }
334

    
335
            /* XXX: fail test */
336
            init_vlc(&huff_vlc[i], 7, 512,
337
                     tmp_bits, 1, 1, tmp_codes, 2, 2, 1);
338
        }
339
        for(i=0;i<2;i++) {
340
            init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
341
                     mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1, 1);
342
        }
343

    
344
        for(i=0;i<9;i++) {
345
            k = 0;
346
            for(j=0;j<22;j++) {
347
                band_index_long[i][j] = k;
348
                k += band_size_long[i][j];
349
            }
350
            band_index_long[i][22] = k;
351
        }
352

    
353
        /* compute n ^ (4/3) and store it in mantissa/exp format */
354

    
355
        int_pow_init();
356
        for(i=1;i<TABLE_4_3_SIZE;i++) {
357
            double f, fm;
358
            int e, m;
359
            f = pow((double)(i/4), 4.0 / 3.0) * pow(2, (i&3)*0.25);
360
            fm = frexp(f, &e);
361
            m = (uint32_t)(fm*(1LL<<31) + 0.5);
362
            e+= FRAC_BITS - 31 + 5 - 100;
363

    
364
            /* normalized to FRAC_BITS */
365
            table_4_3_value[i] = m;
366
//            av_log(NULL, AV_LOG_DEBUG, "%d %d %f\n", i, m, pow((double)i, 4.0 / 3.0));
367
            table_4_3_exp[i] = -e;
368
        }
369
        for(i=0; i<512*16; i++){
370
            int exponent= (i>>4);
371
            double f= pow(i&15, 4.0 / 3.0) * pow(2, (exponent-400)*0.25 + FRAC_BITS + 5);
372
            expval_table[exponent][i&15]= llrint(f);
373
            if((i&15)==1)
374
                exp_table[exponent]= llrint(f);
375
        }
376

    
377
        for(i=0;i<7;i++) {
378
            float f;
379
            int v;
380
            if (i != 6) {
381
                f = tan((double)i * M_PI / 12.0);
382
                v = FIXR(f / (1.0 + f));
383
            } else {
384
                v = FIXR(1.0);
385
            }
386
            is_table[0][i] = v;
387
            is_table[1][6 - i] = v;
388
        }
389
        /* invalid values */
390
        for(i=7;i<16;i++)
391
            is_table[0][i] = is_table[1][i] = 0.0;
392

    
393
        for(i=0;i<16;i++) {
394
            double f;
395
            int e, k;
396

    
397
            for(j=0;j<2;j++) {
398
                e = -(j + 1) * ((i + 1) >> 1);
399
                f = pow(2.0, e / 4.0);
400
                k = i & 1;
401
                is_table_lsf[j][k ^ 1][i] = FIXR(f);
402
                is_table_lsf[j][k][i] = FIXR(1.0);
403
                dprintf(avctx, "is_table_lsf %d %d: %x %x\n",
404
                        i, j, is_table_lsf[j][0][i], is_table_lsf[j][1][i]);
405
            }
406
        }
407

    
408
        for(i=0;i<8;i++) {
409
            float ci, cs, ca;
410
            ci = ci_table[i];
411
            cs = 1.0 / sqrt(1.0 + ci * ci);
412
            ca = cs * ci;
413
            csa_table[i][0] = FIXHR(cs/4);
414
            csa_table[i][1] = FIXHR(ca/4);
415
            csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);
416
            csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4);
417
            csa_table_float[i][0] = cs;
418
            csa_table_float[i][1] = ca;
419
            csa_table_float[i][2] = ca + cs;
420
            csa_table_float[i][3] = ca - cs;
421
//            printf("%d %d %d %d\n", FIX(cs), FIX(cs-1), FIX(ca), FIX(cs)-FIX(ca));
422
//            av_log(NULL, AV_LOG_DEBUG,"%f %f %f %f\n", cs, ca, ca+cs, ca-cs);
423
        }
424

    
425
        /* compute mdct windows */
426
        for(i=0;i<36;i++) {
427
            for(j=0; j<4; j++){
428
                double d;
429

    
430
                if(j==2 && i%3 != 1)
431
                    continue;
432

    
433
                d= sin(M_PI * (i + 0.5) / 36.0);
434
                if(j==1){
435
                    if     (i>=30) d= 0;
436
                    else if(i>=24) d= sin(M_PI * (i - 18 + 0.5) / 12.0);
437
                    else if(i>=18) d= 1;
438
                }else if(j==3){
439
                    if     (i<  6) d= 0;
440
                    else if(i< 12) d= sin(M_PI * (i -  6 + 0.5) / 12.0);
441
                    else if(i< 18) d= 1;
442
                }
443
                //merge last stage of imdct into the window coefficients
444
                d*= 0.5 / cos(M_PI*(2*i + 19)/72);
445

    
446
                if(j==2)
447
                    mdct_win[j][i/3] = FIXHR((d / (1<<5)));
448
                else
449
                    mdct_win[j][i  ] = FIXHR((d / (1<<5)));
450
//                av_log(NULL, AV_LOG_DEBUG, "%2d %d %f\n", i,j,d / (1<<5));
451
            }
452
        }
453

    
454
        /* NOTE: we do frequency inversion adter the MDCT by changing
455
           the sign of the right window coefs */
456
        for(j=0;j<4;j++) {
457
            for(i=0;i<36;i+=2) {
458
                mdct_win[j + 4][i] = mdct_win[j][i];
459
                mdct_win[j + 4][i + 1] = -mdct_win[j][i + 1];
460
            }
461
        }
462

    
463
#if defined(DEBUG)
464
        for(j=0;j<8;j++) {
465
            av_log(avctx, AV_LOG_DEBUG, "win%d=\n", j);
466
            for(i=0;i<36;i++)
467
                av_log(avctx, AV_LOG_DEBUG, "%f, ", (double)mdct_win[j][i] / FRAC_ONE);
468
            av_log(avctx, AV_LOG_DEBUG, "\n");
469
        }
470
#endif
471
        init = 1;
472
    }
473

    
474
#ifdef DEBUG
475
    s->frame_count = 0;
476
#endif
477
    if (avctx->codec_id == CODEC_ID_MP3ADU)
478
        s->adu_mode = 1;
479
    return 0;
480
}
481

    
482
/* tab[i][j] = 1.0 / (2.0 * cos(pi*(2*k+1) / 2^(6 - j))) */
483

    
484
/* cos(i*pi/64) */
485

    
486
#define COS0_0  FIXHR(0.50060299823519630134/2)
487
#define COS0_1  FIXHR(0.50547095989754365998/2)
488
#define COS0_2  FIXHR(0.51544730992262454697/2)
489
#define COS0_3  FIXHR(0.53104259108978417447/2)
490
#define COS0_4  FIXHR(0.55310389603444452782/2)
491
#define COS0_5  FIXHR(0.58293496820613387367/2)
492
#define COS0_6  FIXHR(0.62250412303566481615/2)
493
#define COS0_7  FIXHR(0.67480834145500574602/2)
494
#define COS0_8  FIXHR(0.74453627100229844977/2)
495
#define COS0_9  FIXHR(0.83934964541552703873/2)
496
#define COS0_10 FIXHR(0.97256823786196069369/2)
497
#define COS0_11 FIXHR(1.16943993343288495515/4)
498
#define COS0_12 FIXHR(1.48416461631416627724/4)
499
#define COS0_13 FIXHR(2.05778100995341155085/8)
500
#define COS0_14 FIXHR(3.40760841846871878570/8)
501
#define COS0_15 FIXHR(10.19000812354805681150/32)
502

    
503
#define COS1_0 FIXHR(0.50241928618815570551/2)
504
#define COS1_1 FIXHR(0.52249861493968888062/2)
505
#define COS1_2 FIXHR(0.56694403481635770368/2)
506
#define COS1_3 FIXHR(0.64682178335999012954/2)
507
#define COS1_4 FIXHR(0.78815462345125022473/2)
508
#define COS1_5 FIXHR(1.06067768599034747134/4)
509
#define COS1_6 FIXHR(1.72244709823833392782/4)
510
#define COS1_7 FIXHR(5.10114861868916385802/16)
511

    
512
#define COS2_0 FIXHR(0.50979557910415916894/2)
513
#define COS2_1 FIXHR(0.60134488693504528054/2)
514
#define COS2_2 FIXHR(0.89997622313641570463/2)
515
#define COS2_3 FIXHR(2.56291544774150617881/8)
516

    
517
#define COS3_0 FIXHR(0.54119610014619698439/2)
518
#define COS3_1 FIXHR(1.30656296487637652785/4)
519

    
520
#define COS4_0 FIXHR(0.70710678118654752439/2)
521

    
522
/* butterfly operator */
523
#define BF(a, b, c, s)\
524
{\
525
    tmp0 = tab[a] + tab[b];\
526
    tmp1 = tab[a] - tab[b];\
527
    tab[a] = tmp0;\
528
    tab[b] = MULH(tmp1<<(s), c);\
529
}
530

    
531
#define BF1(a, b, c, d)\
532
{\
533
    BF(a, b, COS4_0, 1);\
534
    BF(c, d,-COS4_0, 1);\
535
    tab[c] += tab[d];\
536
}
537

    
538
#define BF2(a, b, c, d)\
539
{\
540
    BF(a, b, COS4_0, 1);\
541
    BF(c, d,-COS4_0, 1);\
542
    tab[c] += tab[d];\
543
    tab[a] += tab[c];\
544
    tab[c] += tab[b];\
545
    tab[b] += tab[d];\
546
}
547

    
548
#define ADD(a, b) tab[a] += tab[b]
549

    
550
/* DCT32 without 1/sqrt(2) coef zero scaling. */
551
static void dct32(int32_t *out, int32_t *tab)
552
{
553
    int tmp0, tmp1;
554

    
555
    /* pass 1 */
556
    BF( 0, 31, COS0_0 , 1);
557
    BF(15, 16, COS0_15, 5);
558
    /* pass 2 */
559
    BF( 0, 15, COS1_0 , 1);
560
    BF(16, 31,-COS1_0 , 1);
561
    /* pass 1 */
562
    BF( 7, 24, COS0_7 , 1);
563
    BF( 8, 23, COS0_8 , 1);
564
    /* pass 2 */
565
    BF( 7,  8, COS1_7 , 4);
566
    BF(23, 24,-COS1_7 , 4);
567
    /* pass 3 */
568
    BF( 0,  7, COS2_0 , 1);
569
    BF( 8, 15,-COS2_0 , 1);
570
    BF(16, 23, COS2_0 , 1);
571
    BF(24, 31,-COS2_0 , 1);
572
    /* pass 1 */
573
    BF( 3, 28, COS0_3 , 1);
574
    BF(12, 19, COS0_12, 2);
575
    /* pass 2 */
576
    BF( 3, 12, COS1_3 , 1);
577
    BF(19, 28,-COS1_3 , 1);
578
    /* pass 1 */
579
    BF( 4, 27, COS0_4 , 1);
580
    BF(11, 20, COS0_11, 2);
581
    /* pass 2 */
582
    BF( 4, 11, COS1_4 , 1);
583
    BF(20, 27,-COS1_4 , 1);
584
    /* pass 3 */
585
    BF( 3,  4, COS2_3 , 3);
586
    BF(11, 12,-COS2_3 , 3);
587
    BF(19, 20, COS2_3 , 3);
588
    BF(27, 28,-COS2_3 , 3);
589
    /* pass 4 */
590
    BF( 0,  3, COS3_0 , 1);
591
    BF( 4,  7,-COS3_0 , 1);
592
    BF( 8, 11, COS3_0 , 1);
593
    BF(12, 15,-COS3_0 , 1);
594
    BF(16, 19, COS3_0 , 1);
595
    BF(20, 23,-COS3_0 , 1);
596
    BF(24, 27, COS3_0 , 1);
597
    BF(28, 31,-COS3_0 , 1);
598

    
599

    
600

    
601
    /* pass 1 */
602
    BF( 1, 30, COS0_1 , 1);
603
    BF(14, 17, COS0_14, 3);
604
    /* pass 2 */
605
    BF( 1, 14, COS1_1 , 1);
606
    BF(17, 30,-COS1_1 , 1);
607
    /* pass 1 */
608
    BF( 6, 25, COS0_6 , 1);
609
    BF( 9, 22, COS0_9 , 1);
610
    /* pass 2 */
611
    BF( 6,  9, COS1_6 , 2);
612
    BF(22, 25,-COS1_6 , 2);
613
    /* pass 3 */
614
    BF( 1,  6, COS2_1 , 1);
615
    BF( 9, 14,-COS2_1 , 1);
616
    BF(17, 22, COS2_1 , 1);
617
    BF(25, 30,-COS2_1 , 1);
618

    
619
    /* pass 1 */
620
    BF( 2, 29, COS0_2 , 1);
621
    BF(13, 18, COS0_13, 3);
622
    /* pass 2 */
623
    BF( 2, 13, COS1_2 , 1);
624
    BF(18, 29,-COS1_2 , 1);
625
    /* pass 1 */
626
    BF( 5, 26, COS0_5 , 1);
627
    BF(10, 21, COS0_10, 1);
628
    /* pass 2 */
629
    BF( 5, 10, COS1_5 , 2);
630
    BF(21, 26,-COS1_5 , 2);
631
    /* pass 3 */
632
    BF( 2,  5, COS2_2 , 1);
633
    BF(10, 13,-COS2_2 , 1);
634
    BF(18, 21, COS2_2 , 1);
635
    BF(26, 29,-COS2_2 , 1);
636
    /* pass 4 */
637
    BF( 1,  2, COS3_1 , 2);
638
    BF( 5,  6,-COS3_1 , 2);
639
    BF( 9, 10, COS3_1 , 2);
640
    BF(13, 14,-COS3_1 , 2);
641
    BF(17, 18, COS3_1 , 2);
642
    BF(21, 22,-COS3_1 , 2);
643
    BF(25, 26, COS3_1 , 2);
644
    BF(29, 30,-COS3_1 , 2);
645

    
646
    /* pass 5 */
647
    BF1( 0,  1,  2,  3);
648
    BF2( 4,  5,  6,  7);
649
    BF1( 8,  9, 10, 11);
650
    BF2(12, 13, 14, 15);
651
    BF1(16, 17, 18, 19);
652
    BF2(20, 21, 22, 23);
653
    BF1(24, 25, 26, 27);
654
    BF2(28, 29, 30, 31);
655

    
656
    /* pass 6 */
657

    
658
    ADD( 8, 12);
659
    ADD(12, 10);
660
    ADD(10, 14);
661
    ADD(14,  9);
662
    ADD( 9, 13);
663
    ADD(13, 11);
664
    ADD(11, 15);
665

    
666
    out[ 0] = tab[0];
667
    out[16] = tab[1];
668
    out[ 8] = tab[2];
669
    out[24] = tab[3];
670
    out[ 4] = tab[4];
671
    out[20] = tab[5];
672
    out[12] = tab[6];
673
    out[28] = tab[7];
674
    out[ 2] = tab[8];
675
    out[18] = tab[9];
676
    out[10] = tab[10];
677
    out[26] = tab[11];
678
    out[ 6] = tab[12];
679
    out[22] = tab[13];
680
    out[14] = tab[14];
681
    out[30] = tab[15];
682

    
683
    ADD(24, 28);
684
    ADD(28, 26);
685
    ADD(26, 30);
686
    ADD(30, 25);
687
    ADD(25, 29);
688
    ADD(29, 27);
689
    ADD(27, 31);
690

    
691
    out[ 1] = tab[16] + tab[24];
692
    out[17] = tab[17] + tab[25];
693
    out[ 9] = tab[18] + tab[26];
694
    out[25] = tab[19] + tab[27];
695
    out[ 5] = tab[20] + tab[28];
696
    out[21] = tab[21] + tab[29];
697
    out[13] = tab[22] + tab[30];
698
    out[29] = tab[23] + tab[31];
699
    out[ 3] = tab[24] + tab[20];
700
    out[19] = tab[25] + tab[21];
701
    out[11] = tab[26] + tab[22];
702
    out[27] = tab[27] + tab[23];
703
    out[ 7] = tab[28] + tab[18];
704
    out[23] = tab[29] + tab[19];
705
    out[15] = tab[30] + tab[17];
706
    out[31] = tab[31];
707
}
708

    
709
#if FRAC_BITS <= 15
710

    
711
static inline int round_sample(int *sum)
712
{
713
    int sum1;
714
    sum1 = (*sum) >> OUT_SHIFT;
715
    *sum &= (1<<OUT_SHIFT)-1;
716
    if (sum1 < OUT_MIN)
717
        sum1 = OUT_MIN;
718
    else if (sum1 > OUT_MAX)
719
        sum1 = OUT_MAX;
720
    return sum1;
721
}
722

    
723
/* signed 16x16 -> 32 multiply add accumulate */
724
#define MACS(rt, ra, rb) MAC16(rt, ra, rb)
725

    
726
/* signed 16x16 -> 32 multiply */
727
#define MULS(ra, rb) MUL16(ra, rb)
728

    
729
#else
730

    
731
static inline int round_sample(int64_t *sum)
732
{
733
    int sum1;
734
    sum1 = (int)((*sum) >> OUT_SHIFT);
735
    *sum &= (1<<OUT_SHIFT)-1;
736
    if (sum1 < OUT_MIN)
737
        sum1 = OUT_MIN;
738
    else if (sum1 > OUT_MAX)
739
        sum1 = OUT_MAX;
740
    return sum1;
741
}
742

    
743
#   define MULS(ra, rb) MUL64(ra, rb)
744
#endif
745

    
746
#define SUM8(sum, op, w, p) \
747
{                                               \
748
    sum op MULS((w)[0 * 64], p[0 * 64]);\
749
    sum op MULS((w)[1 * 64], p[1 * 64]);\
750
    sum op MULS((w)[2 * 64], p[2 * 64]);\
751
    sum op MULS((w)[3 * 64], p[3 * 64]);\
752
    sum op MULS((w)[4 * 64], p[4 * 64]);\
753
    sum op MULS((w)[5 * 64], p[5 * 64]);\
754
    sum op MULS((w)[6 * 64], p[6 * 64]);\
755
    sum op MULS((w)[7 * 64], p[7 * 64]);\
756
}
757

    
758
#define SUM8P2(sum1, op1, sum2, op2, w1, w2, p) \
759
{                                               \
760
    int tmp;\
761
    tmp = p[0 * 64];\
762
    sum1 op1 MULS((w1)[0 * 64], tmp);\
763
    sum2 op2 MULS((w2)[0 * 64], tmp);\
764
    tmp = p[1 * 64];\
765
    sum1 op1 MULS((w1)[1 * 64], tmp);\
766
    sum2 op2 MULS((w2)[1 * 64], tmp);\
767
    tmp = p[2 * 64];\
768
    sum1 op1 MULS((w1)[2 * 64], tmp);\
769
    sum2 op2 MULS((w2)[2 * 64], tmp);\
770
    tmp = p[3 * 64];\
771
    sum1 op1 MULS((w1)[3 * 64], tmp);\
772
    sum2 op2 MULS((w2)[3 * 64], tmp);\
773
    tmp = p[4 * 64];\
774
    sum1 op1 MULS((w1)[4 * 64], tmp);\
775
    sum2 op2 MULS((w2)[4 * 64], tmp);\
776
    tmp = p[5 * 64];\
777
    sum1 op1 MULS((w1)[5 * 64], tmp);\
778
    sum2 op2 MULS((w2)[5 * 64], tmp);\
779
    tmp = p[6 * 64];\
780
    sum1 op1 MULS((w1)[6 * 64], tmp);\
781
    sum2 op2 MULS((w2)[6 * 64], tmp);\
782
    tmp = p[7 * 64];\
783
    sum1 op1 MULS((w1)[7 * 64], tmp);\
784
    sum2 op2 MULS((w2)[7 * 64], tmp);\
785
}
786

    
787
void ff_mpa_synth_init(MPA_INT *window)
788
{
789
    int i;
790

    
791
    /* max = 18760, max sum over all 16 coefs : 44736 */
792
    for(i=0;i<257;i++) {
793
        int v;
794
        v = ff_mpa_enwindow[i];
795
#if WFRAC_BITS < 16
796
        v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS);
797
#endif
798
        window[i] = v;
799
        if ((i & 63) != 0)
800
            v = -v;
801
        if (i != 0)
802
            window[512 - i] = v;
803
    }
804
}
805

    
806
/* 32 sub band synthesis filter. Input: 32 sub band samples, Output:
807
   32 samples. */
808
/* XXX: optimize by avoiding ring buffer usage */
809
void ff_mpa_synth_filter(MPA_INT *synth_buf_ptr, int *synth_buf_offset,
810
                         MPA_INT *window, int *dither_state,
811
                         OUT_INT *samples, int incr,
812
                         int32_t sb_samples[SBLIMIT])
813
{
814
    int32_t tmp[32];
815
    register MPA_INT *synth_buf;
816
    register const MPA_INT *w, *w2, *p;
817
    int j, offset, v;
818
    OUT_INT *samples2;
819
#if FRAC_BITS <= 15
820
    int sum, sum2;
821
#else
822
    int64_t sum, sum2;
823
#endif
824

    
825
    dct32(tmp, sb_samples);
826

    
827
    offset = *synth_buf_offset;
828
    synth_buf = synth_buf_ptr + offset;
829

    
830
    for(j=0;j<32;j++) {
831
        v = tmp[j];
832
#if FRAC_BITS <= 15
833
        /* NOTE: can cause a loss in precision if very high amplitude
834
           sound */
835
        if (v > 32767)
836
            v = 32767;
837
        else if (v < -32768)
838
            v = -32768;
839
#endif
840
        synth_buf[j] = v;
841
    }
842
    /* copy to avoid wrap */
843
    memcpy(synth_buf + 512, synth_buf, 32 * sizeof(MPA_INT));
844

    
845
    samples2 = samples + 31 * incr;
846
    w = window;
847
    w2 = window + 31;
848

    
849
    sum = *dither_state;
850
    p = synth_buf + 16;
851
    SUM8(sum, +=, w, p);
852
    p = synth_buf + 48;
853
    SUM8(sum, -=, w + 32, p);
854
    *samples = round_sample(&sum);
855
    samples += incr;
856
    w++;
857

    
858
    /* we calculate two samples at the same time to avoid one memory
859
       access per two sample */
860
    for(j=1;j<16;j++) {
861
        sum2 = 0;
862
        p = synth_buf + 16 + j;
863
        SUM8P2(sum, +=, sum2, -=, w, w2, p);
864
        p = synth_buf + 48 - j;
865
        SUM8P2(sum, -=, sum2, -=, w + 32, w2 + 32, p);
866

    
867
        *samples = round_sample(&sum);
868
        samples += incr;
869
        sum += sum2;
870
        *samples2 = round_sample(&sum);
871
        samples2 -= incr;
872
        w++;
873
        w2--;
874
    }
875

    
876
    p = synth_buf + 32;
877
    SUM8(sum, -=, w + 32, p);
878
    *samples = round_sample(&sum);
879
    *dither_state= sum;
880

    
881
    offset = (offset - 32) & 511;
882
    *synth_buf_offset = offset;
883
}
884

    
885
#define C3 FIXHR(0.86602540378443864676/2)
886

    
887
/* 0.5 / cos(pi*(2*i+1)/36) */
888
static const int icos36[9] = {
889
    FIXR(0.50190991877167369479),
890
    FIXR(0.51763809020504152469), //0
891
    FIXR(0.55168895948124587824),
892
    FIXR(0.61038729438072803416),
893
    FIXR(0.70710678118654752439), //1
894
    FIXR(0.87172339781054900991),
895
    FIXR(1.18310079157624925896),
896
    FIXR(1.93185165257813657349), //2
897
    FIXR(5.73685662283492756461),
898
};
899

    
900
/* 0.5 / cos(pi*(2*i+1)/36) */
901
static const int icos36h[9] = {
902
    FIXHR(0.50190991877167369479/2),
903
    FIXHR(0.51763809020504152469/2), //0
904
    FIXHR(0.55168895948124587824/2),
905
    FIXHR(0.61038729438072803416/2),
906
    FIXHR(0.70710678118654752439/2), //1
907
    FIXHR(0.87172339781054900991/2),
908
    FIXHR(1.18310079157624925896/4),
909
    FIXHR(1.93185165257813657349/4), //2
910
//    FIXHR(5.73685662283492756461),
911
};
912

    
913
/* 12 points IMDCT. We compute it "by hand" by factorizing obvious
914
   cases. */
915
static void imdct12(int *out, int *in)
916
{
917
    int in0, in1, in2, in3, in4, in5, t1, t2;
918

    
919
    in0= in[0*3];
920
    in1= in[1*3] + in[0*3];
921
    in2= in[2*3] + in[1*3];
922
    in3= in[3*3] + in[2*3];
923
    in4= in[4*3] + in[3*3];
924
    in5= in[5*3] + in[4*3];
925
    in5 += in3;
926
    in3 += in1;
927

    
928
    in2= MULH(2*in2, C3);
929
    in3= MULH(4*in3, C3);
930

    
931
    t1 = in0 - in4;
932
    t2 = MULH(2*(in1 - in5), icos36h[4]);
933

    
934
    out[ 7]=
935
    out[10]= t1 + t2;
936
    out[ 1]=
937
    out[ 4]= t1 - t2;
938

    
939
    in0 += in4>>1;
940
    in4 = in0 + in2;
941
    in5 += 2*in1;
942
    in1 = MULH(in5 + in3, icos36h[1]);
943
    out[ 8]=
944
    out[ 9]= in4 + in1;
945
    out[ 2]=
946
    out[ 3]= in4 - in1;
947

    
948
    in0 -= in2;
949
    in5 = MULH(2*(in5 - in3), icos36h[7]);
950
    out[ 0]=
951
    out[ 5]= in0 - in5;
952
    out[ 6]=
953
    out[11]= in0 + in5;
954
}
955

    
956
/* cos(pi*i/18) */
957
#define C1 FIXHR(0.98480775301220805936/2)
958
#define C2 FIXHR(0.93969262078590838405/2)
959
#define C3 FIXHR(0.86602540378443864676/2)
960
#define C4 FIXHR(0.76604444311897803520/2)
961
#define C5 FIXHR(0.64278760968653932632/2)
962
#define C6 FIXHR(0.5/2)
963
#define C7 FIXHR(0.34202014332566873304/2)
964
#define C8 FIXHR(0.17364817766693034885/2)
965

    
966

    
967
/* using Lee like decomposition followed by hand coded 9 points DCT */
968
static void imdct36(int *out, int *buf, int *in, int *win)
969
{
970
    int i, j, t0, t1, t2, t3, s0, s1, s2, s3;
971
    int tmp[18], *tmp1, *in1;
972

    
973
    for(i=17;i>=1;i--)
974
        in[i] += in[i-1];
975
    for(i=17;i>=3;i-=2)
976
        in[i] += in[i-2];
977

    
978
    for(j=0;j<2;j++) {
979
        tmp1 = tmp + j;
980
        in1 = in + j;
981
#if 0
982
//more accurate but slower
983
        int64_t t0, t1, t2, t3;
984
        t2 = in1[2*4] + in1[2*8] - in1[2*2];
985

986
        t3 = (in1[2*0] + (int64_t)(in1[2*6]>>1))<<32;
987
        t1 = in1[2*0] - in1[2*6];
988
        tmp1[ 6] = t1 - (t2>>1);
989
        tmp1[16] = t1 + t2;
990

991
        t0 = MUL64(2*(in1[2*2] + in1[2*4]),    C2);
992
        t1 = MUL64(   in1[2*4] - in1[2*8] , -2*C8);
993
        t2 = MUL64(2*(in1[2*2] + in1[2*8]),   -C4);
994

995
        tmp1[10] = (t3 - t0 - t2) >> 32;
996
        tmp1[ 2] = (t3 + t0 + t1) >> 32;
997
        tmp1[14] = (t3 + t2 - t1) >> 32;
998

999
        tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
1000
        t2 = MUL64(2*(in1[2*1] + in1[2*5]),    C1);
1001
        t3 = MUL64(   in1[2*5] - in1[2*7] , -2*C7);
1002
        t0 = MUL64(2*in1[2*3], C3);
1003

1004
        t1 = MUL64(2*(in1[2*1] + in1[2*7]),   -C5);
1005

1006
        tmp1[ 0] = (t2 + t3 + t0) >> 32;
1007
        tmp1[12] = (t2 + t1 - t0) >> 32;
1008
        tmp1[ 8] = (t3 - t1 - t0) >> 32;
1009
#else
1010
        t2 = in1[2*4] + in1[2*8] - in1[2*2];
1011

    
1012
        t3 = in1[2*0] + (in1[2*6]>>1);
1013
        t1 = in1[2*0] - in1[2*6];
1014
        tmp1[ 6] = t1 - (t2>>1);
1015
        tmp1[16] = t1 + t2;
1016

    
1017
        t0 = MULH(2*(in1[2*2] + in1[2*4]),    C2);
1018
        t1 = MULH(   in1[2*4] - in1[2*8] , -2*C8);
1019
        t2 = MULH(2*(in1[2*2] + in1[2*8]),   -C4);
1020

    
1021
        tmp1[10] = t3 - t0 - t2;
1022
        tmp1[ 2] = t3 + t0 + t1;
1023
        tmp1[14] = t3 + t2 - t1;
1024

    
1025
        tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
1026
        t2 = MULH(2*(in1[2*1] + in1[2*5]),    C1);
1027
        t3 = MULH(   in1[2*5] - in1[2*7] , -2*C7);
1028
        t0 = MULH(2*in1[2*3], C3);
1029

    
1030
        t1 = MULH(2*(in1[2*1] + in1[2*7]),   -C5);
1031

    
1032
        tmp1[ 0] = t2 + t3 + t0;
1033
        tmp1[12] = t2 + t1 - t0;
1034
        tmp1[ 8] = t3 - t1 - t0;
1035
#endif
1036
    }
1037

    
1038
    i = 0;
1039
    for(j=0;j<4;j++) {
1040
        t0 = tmp[i];
1041
        t1 = tmp[i + 2];
1042
        s0 = t1 + t0;
1043
        s2 = t1 - t0;
1044

    
1045
        t2 = tmp[i + 1];
1046
        t3 = tmp[i + 3];
1047
        s1 = MULH(2*(t3 + t2), icos36h[j]);
1048
        s3 = MULL(t3 - t2, icos36[8 - j]);
1049

    
1050
        t0 = s0 + s1;
1051
        t1 = s0 - s1;
1052
        out[(9 + j)*SBLIMIT] =  MULH(t1, win[9 + j]) + buf[9 + j];
1053
        out[(8 - j)*SBLIMIT] =  MULH(t1, win[8 - j]) + buf[8 - j];
1054
        buf[9 + j] = MULH(t0, win[18 + 9 + j]);
1055
        buf[8 - j] = MULH(t0, win[18 + 8 - j]);
1056

    
1057
        t0 = s2 + s3;
1058
        t1 = s2 - s3;
1059
        out[(9 + 8 - j)*SBLIMIT] =  MULH(t1, win[9 + 8 - j]) + buf[9 + 8 - j];
1060
        out[(        j)*SBLIMIT] =  MULH(t1, win[        j]) + buf[        j];
1061
        buf[9 + 8 - j] = MULH(t0, win[18 + 9 + 8 - j]);
1062
        buf[      + j] = MULH(t0, win[18         + j]);
1063
        i += 4;
1064
    }
1065

    
1066
    s0 = tmp[16];
1067
    s1 = MULH(2*tmp[17], icos36h[4]);
1068
    t0 = s0 + s1;
1069
    t1 = s0 - s1;
1070
    out[(9 + 4)*SBLIMIT] =  MULH(t1, win[9 + 4]) + buf[9 + 4];
1071
    out[(8 - 4)*SBLIMIT] =  MULH(t1, win[8 - 4]) + buf[8 - 4];
1072
    buf[9 + 4] = MULH(t0, win[18 + 9 + 4]);
1073
    buf[8 - 4] = MULH(t0, win[18 + 8 - 4]);
1074
}
1075

    
1076
/* return the number of decoded frames */
1077
static int mp_decode_layer1(MPADecodeContext *s)
1078
{
1079
    int bound, i, v, n, ch, j, mant;
1080
    uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
1081
    uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
1082

    
1083
    if (s->mode == MPA_JSTEREO)
1084
        bound = (s->mode_ext + 1) * 4;
1085
    else
1086
        bound = SBLIMIT;
1087

    
1088
    /* allocation bits */
1089
    for(i=0;i<bound;i++) {
1090
        for(ch=0;ch<s->nb_channels;ch++) {
1091
            allocation[ch][i] = get_bits(&s->gb, 4);
1092
        }
1093
    }
1094
    for(i=bound;i<SBLIMIT;i++) {
1095
        allocation[0][i] = get_bits(&s->gb, 4);
1096
    }
1097

    
1098
    /* scale factors */
1099
    for(i=0;i<bound;i++) {
1100
        for(ch=0;ch<s->nb_channels;ch++) {
1101
            if (allocation[ch][i])
1102
                scale_factors[ch][i] = get_bits(&s->gb, 6);
1103
        }
1104
    }
1105
    for(i=bound;i<SBLIMIT;i++) {
1106
        if (allocation[0][i]) {
1107
            scale_factors[0][i] = get_bits(&s->gb, 6);
1108
            scale_factors[1][i] = get_bits(&s->gb, 6);
1109
        }
1110
    }
1111

    
1112
    /* compute samples */
1113
    for(j=0;j<12;j++) {
1114
        for(i=0;i<bound;i++) {
1115
            for(ch=0;ch<s->nb_channels;ch++) {
1116
                n = allocation[ch][i];
1117
                if (n) {
1118
                    mant = get_bits(&s->gb, n + 1);
1119
                    v = l1_unscale(n, mant, scale_factors[ch][i]);
1120
                } else {
1121
                    v = 0;
1122
                }
1123
                s->sb_samples[ch][j][i] = v;
1124
            }
1125
        }
1126
        for(i=bound;i<SBLIMIT;i++) {
1127
            n = allocation[0][i];
1128
            if (n) {
1129
                mant = get_bits(&s->gb, n + 1);
1130
                v = l1_unscale(n, mant, scale_factors[0][i]);
1131
                s->sb_samples[0][j][i] = v;
1132
                v = l1_unscale(n, mant, scale_factors[1][i]);
1133
                s->sb_samples[1][j][i] = v;
1134
            } else {
1135
                s->sb_samples[0][j][i] = 0;
1136
                s->sb_samples[1][j][i] = 0;
1137
            }
1138
        }
1139
    }
1140
    return 12;
1141
}
1142

    
1143
/* bitrate is in kb/s */
1144
int l2_select_table(int bitrate, int nb_channels, int freq, int lsf)
1145
{
1146
    int ch_bitrate, table;
1147

    
1148
    ch_bitrate = bitrate / nb_channels;
1149
    if (!lsf) {
1150
        if ((freq == 48000 && ch_bitrate >= 56) ||
1151
            (ch_bitrate >= 56 && ch_bitrate <= 80))
1152
            table = 0;
1153
        else if (freq != 48000 && ch_bitrate >= 96)
1154
            table = 1;
1155
        else if (freq != 32000 && ch_bitrate <= 48)
1156
            table = 2;
1157
        else
1158
            table = 3;
1159
    } else {
1160
        table = 4;
1161
    }
1162
    return table;
1163
}
1164

    
1165
static int mp_decode_layer2(MPADecodeContext *s)
1166
{
1167
    int sblimit; /* number of used subbands */
1168
    const unsigned char *alloc_table;
1169
    int table, bit_alloc_bits, i, j, ch, bound, v;
1170
    unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
1171
    unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
1172
    unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
1173
    int scale, qindex, bits, steps, k, l, m, b;
1174

    
1175
    /* select decoding table */
1176
    table = l2_select_table(s->bit_rate / 1000, s->nb_channels,
1177
                            s->sample_rate, s->lsf);
1178
    sblimit = ff_mpa_sblimit_table[table];
1179
    alloc_table = ff_mpa_alloc_tables[table];
1180

    
1181
    if (s->mode == MPA_JSTEREO)
1182
        bound = (s->mode_ext + 1) * 4;
1183
    else
1184
        bound = sblimit;
1185

    
1186
    dprintf(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit);
1187

    
1188
    /* sanity check */
1189
    if( bound > sblimit ) bound = sblimit;
1190

    
1191
    /* parse bit allocation */
1192
    j = 0;
1193
    for(i=0;i<bound;i++) {
1194
        bit_alloc_bits = alloc_table[j];
1195
        for(ch=0;ch<s->nb_channels;ch++) {
1196
            bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
1197
        }
1198
        j += 1 << bit_alloc_bits;
1199
    }
1200
    for(i=bound;i<sblimit;i++) {
1201
        bit_alloc_bits = alloc_table[j];
1202
        v = get_bits(&s->gb, bit_alloc_bits);
1203
        bit_alloc[0][i] = v;
1204
        bit_alloc[1][i] = v;
1205
        j += 1 << bit_alloc_bits;
1206
    }
1207

    
1208
#ifdef DEBUG
1209
    {
1210
        for(ch=0;ch<s->nb_channels;ch++) {
1211
            for(i=0;i<sblimit;i++)
1212
                dprintf(s->avctx, " %d", bit_alloc[ch][i]);
1213
            dprintf(s->avctx, "\n");
1214
        }
1215
    }
1216
#endif
1217

    
1218
    /* scale codes */
1219
    for(i=0;i<sblimit;i++) {
1220
        for(ch=0;ch<s->nb_channels;ch++) {
1221
            if (bit_alloc[ch][i])
1222
                scale_code[ch][i] = get_bits(&s->gb, 2);
1223
        }
1224
    }
1225

    
1226
    /* scale factors */
1227
    for(i=0;i<sblimit;i++) {
1228
        for(ch=0;ch<s->nb_channels;ch++) {
1229
            if (bit_alloc[ch][i]) {
1230
                sf = scale_factors[ch][i];
1231
                switch(scale_code[ch][i]) {
1232
                default:
1233
                case 0:
1234
                    sf[0] = get_bits(&s->gb, 6);
1235
                    sf[1] = get_bits(&s->gb, 6);
1236
                    sf[2] = get_bits(&s->gb, 6);
1237
                    break;
1238
                case 2:
1239
                    sf[0] = get_bits(&s->gb, 6);
1240
                    sf[1] = sf[0];
1241
                    sf[2] = sf[0];
1242
                    break;
1243
                case 1:
1244
                    sf[0] = get_bits(&s->gb, 6);
1245
                    sf[2] = get_bits(&s->gb, 6);
1246
                    sf[1] = sf[0];
1247
                    break;
1248
                case 3:
1249
                    sf[0] = get_bits(&s->gb, 6);
1250
                    sf[2] = get_bits(&s->gb, 6);
1251
                    sf[1] = sf[2];
1252
                    break;
1253
                }
1254
            }
1255
        }
1256
    }
1257

    
1258
#ifdef DEBUG
1259
    for(ch=0;ch<s->nb_channels;ch++) {
1260
        for(i=0;i<sblimit;i++) {
1261
            if (bit_alloc[ch][i]) {
1262
                sf = scale_factors[ch][i];
1263
                dprintf(s->avctx, " %d %d %d", sf[0], sf[1], sf[2]);
1264
            } else {
1265
                dprintf(s->avctx, " -");
1266
            }
1267
        }
1268
        dprintf(s->avctx, "\n");
1269
    }
1270
#endif
1271

    
1272
    /* samples */
1273
    for(k=0;k<3;k++) {
1274
        for(l=0;l<12;l+=3) {
1275
            j = 0;
1276
            for(i=0;i<bound;i++) {
1277
                bit_alloc_bits = alloc_table[j];
1278
                for(ch=0;ch<s->nb_channels;ch++) {
1279
                    b = bit_alloc[ch][i];
1280
                    if (b) {
1281
                        scale = scale_factors[ch][i][k];
1282
                        qindex = alloc_table[j+b];
1283
                        bits = ff_mpa_quant_bits[qindex];
1284
                        if (bits < 0) {
1285
                            /* 3 values at the same time */
1286
                            v = get_bits(&s->gb, -bits);
1287
                            steps = ff_mpa_quant_steps[qindex];
1288
                            s->sb_samples[ch][k * 12 + l + 0][i] =
1289
                                l2_unscale_group(steps, v % steps, scale);
1290
                            v = v / steps;
1291
                            s->sb_samples[ch][k * 12 + l + 1][i] =
1292
                                l2_unscale_group(steps, v % steps, scale);
1293
                            v = v / steps;
1294
                            s->sb_samples[ch][k * 12 + l + 2][i] =
1295
                                l2_unscale_group(steps, v, scale);
1296
                        } else {
1297
                            for(m=0;m<3;m++) {
1298
                                v = get_bits(&s->gb, bits);
1299
                                v = l1_unscale(bits - 1, v, scale);
1300
                                s->sb_samples[ch][k * 12 + l + m][i] = v;
1301
                            }
1302
                        }
1303
                    } else {
1304
                        s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1305
                        s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1306
                        s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1307
                    }
1308
                }
1309
                /* next subband in alloc table */
1310
                j += 1 << bit_alloc_bits;
1311
            }
1312
            /* XXX: find a way to avoid this duplication of code */
1313
            for(i=bound;i<sblimit;i++) {
1314
                bit_alloc_bits = alloc_table[j];
1315
                b = bit_alloc[0][i];
1316
                if (b) {
1317
                    int mant, scale0, scale1;
1318
                    scale0 = scale_factors[0][i][k];
1319
                    scale1 = scale_factors[1][i][k];
1320
                    qindex = alloc_table[j+b];
1321
                    bits = ff_mpa_quant_bits[qindex];
1322
                    if (bits < 0) {
1323
                        /* 3 values at the same time */
1324
                        v = get_bits(&s->gb, -bits);
1325
                        steps = ff_mpa_quant_steps[qindex];
1326
                        mant = v % steps;
1327
                        v = v / steps;
1328
                        s->sb_samples[0][k * 12 + l + 0][i] =
1329
                            l2_unscale_group(steps, mant, scale0);
1330
                        s->sb_samples[1][k * 12 + l + 0][i] =
1331
                            l2_unscale_group(steps, mant, scale1);
1332
                        mant = v % steps;
1333
                        v = v / steps;
1334
                        s->sb_samples[0][k * 12 + l + 1][i] =
1335
                            l2_unscale_group(steps, mant, scale0);
1336
                        s->sb_samples[1][k * 12 + l + 1][i] =
1337
                            l2_unscale_group(steps, mant, scale1);
1338
                        s->sb_samples[0][k * 12 + l + 2][i] =
1339
                            l2_unscale_group(steps, v, scale0);
1340
                        s->sb_samples[1][k * 12 + l + 2][i] =
1341
                            l2_unscale_group(steps, v, scale1);
1342
                    } else {
1343
                        for(m=0;m<3;m++) {
1344
                            mant = get_bits(&s->gb, bits);
1345
                            s->sb_samples[0][k * 12 + l + m][i] =
1346
                                l1_unscale(bits - 1, mant, scale0);
1347
                            s->sb_samples[1][k * 12 + l + m][i] =
1348
                                l1_unscale(bits - 1, mant, scale1);
1349
                        }
1350
                    }
1351
                } else {
1352
                    s->sb_samples[0][k * 12 + l + 0][i] = 0;
1353
                    s->sb_samples[0][k * 12 + l + 1][i] = 0;
1354
                    s->sb_samples[0][k * 12 + l + 2][i] = 0;
1355
                    s->sb_samples[1][k * 12 + l + 0][i] = 0;
1356
                    s->sb_samples[1][k * 12 + l + 1][i] = 0;
1357
                    s->sb_samples[1][k * 12 + l + 2][i] = 0;
1358
                }
1359
                /* next subband in alloc table */
1360
                j += 1 << bit_alloc_bits;
1361
            }
1362
            /* fill remaining samples to zero */
1363
            for(i=sblimit;i<SBLIMIT;i++) {
1364
                for(ch=0;ch<s->nb_channels;ch++) {
1365
                    s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1366
                    s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1367
                    s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1368
                }
1369
            }
1370
        }
1371
    }
1372
    return 3 * 12;
1373
}
1374

    
1375
static inline void lsf_sf_expand(int *slen,
1376
                                 int sf, int n1, int n2, int n3)
1377
{
1378
    if (n3) {
1379
        slen[3] = sf % n3;
1380
        sf /= n3;
1381
    } else {
1382
        slen[3] = 0;
1383
    }
1384
    if (n2) {
1385
        slen[2] = sf % n2;
1386
        sf /= n2;
1387
    } else {
1388
        slen[2] = 0;
1389
    }
1390
    slen[1] = sf % n1;
1391
    sf /= n1;
1392
    slen[0] = sf;
1393
}
1394

    
1395
static void exponents_from_scale_factors(MPADecodeContext *s,
1396
                                         GranuleDef *g,
1397
                                         int16_t *exponents)
1398
{
1399
    const uint8_t *bstab, *pretab;
1400
    int len, i, j, k, l, v0, shift, gain, gains[3];
1401
    int16_t *exp_ptr;
1402

    
1403
    exp_ptr = exponents;
1404
    gain = g->global_gain - 210;
1405
    shift = g->scalefac_scale + 1;
1406

    
1407
    bstab = band_size_long[s->sample_rate_index];
1408
    pretab = mpa_pretab[g->preflag];
1409
    for(i=0;i<g->long_end;i++) {
1410
        v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400;
1411
        len = bstab[i];
1412
        for(j=len;j>0;j--)
1413
            *exp_ptr++ = v0;
1414
    }
1415

    
1416
    if (g->short_start < 13) {
1417
        bstab = band_size_short[s->sample_rate_index];
1418
        gains[0] = gain - (g->subblock_gain[0] << 3);
1419
        gains[1] = gain - (g->subblock_gain[1] << 3);
1420
        gains[2] = gain - (g->subblock_gain[2] << 3);
1421
        k = g->long_end;
1422
        for(i=g->short_start;i<13;i++) {
1423
            len = bstab[i];
1424
            for(l=0;l<3;l++) {
1425
                v0 = gains[l] - (g->scale_factors[k++] << shift) + 400;
1426
                for(j=len;j>0;j--)
1427
                *exp_ptr++ = v0;
1428
            }
1429
        }
1430
    }
1431
}
1432

    
1433
/* handle n = 0 too */
1434
static inline int get_bitsz(GetBitContext *s, int n)
1435
{
1436
    if (n == 0)
1437
        return 0;
1438
    else
1439
        return get_bits(s, n);
1440
}
1441

    
1442

    
1443
static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos, int *end_pos2){
1444
    if(s->in_gb.buffer && *pos >= s->gb.size_in_bits){
1445
        s->gb= s->in_gb;
1446
        s->in_gb.buffer=NULL;
1447
        assert((get_bits_count(&s->gb) & 7) == 0);
1448
        skip_bits_long(&s->gb, *pos - *end_pos);
1449
        *end_pos2=
1450
        *end_pos= *end_pos2 + get_bits_count(&s->gb) - *pos;
1451
        *pos= get_bits_count(&s->gb);
1452
    }
1453
}
1454

    
1455
static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
1456
                          int16_t *exponents, int end_pos2)
1457
{
1458
    int s_index;
1459
    int i;
1460
    int last_pos, bits_left;
1461
    VLC *vlc;
1462
    int end_pos= FFMIN(end_pos2, s->gb.size_in_bits);
1463

    
1464
    /* low frequencies (called big values) */
1465
    s_index = 0;
1466
    for(i=0;i<3;i++) {
1467
        int j, k, l, linbits;
1468
        j = g->region_size[i];
1469
        if (j == 0)
1470
            continue;
1471
        /* select vlc table */
1472
        k = g->table_select[i];
1473
        l = mpa_huff_data[k][0];
1474
        linbits = mpa_huff_data[k][1];
1475
        vlc = &huff_vlc[l];
1476

    
1477
        if(!l){
1478
            memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*2*j);
1479
            s_index += 2*j;
1480
            continue;
1481
        }
1482

    
1483
        /* read huffcode and compute each couple */
1484
        for(;j>0;j--) {
1485
            int exponent, x, y, v;
1486
            int pos= get_bits_count(&s->gb);
1487

    
1488
            if (pos >= end_pos){
1489
//                av_log(NULL, AV_LOG_ERROR, "pos: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);
1490
                switch_buffer(s, &pos, &end_pos, &end_pos2);
1491
//                av_log(NULL, AV_LOG_ERROR, "new pos: %d %d\n", pos, end_pos);
1492
                if(pos >= end_pos)
1493
                    break;
1494
            }
1495
            y = get_vlc2(&s->gb, vlc->table, 7, 3);
1496

    
1497
            if(!y){
1498
                g->sb_hybrid[s_index  ] =
1499
                g->sb_hybrid[s_index+1] = 0;
1500
                s_index += 2;
1501
                continue;
1502
            }
1503

    
1504
            exponent= exponents[s_index];
1505

    
1506
            dprintf(s->avctx, "region=%d n=%d x=%d y=%d exp=%d\n",
1507
                    i, g->region_size[i] - j, x, y, exponent);
1508
            if(y&16){
1509
                x = y >> 5;
1510
                y = y & 0x0f;
1511
                if (x < 15){
1512
                    v = expval_table[ exponent ][ x ];
1513
//                      v = expval_table[ (exponent&3) ][ x ] >> FFMIN(0 - (exponent>>2), 31);
1514
                }else{
1515
                    x += get_bitsz(&s->gb, linbits);
1516
                    v = l3_unscale(x, exponent);
1517
                }
1518
                if (get_bits1(&s->gb))
1519
                    v = -v;
1520
                g->sb_hybrid[s_index] = v;
1521
                if (y < 15){
1522
                    v = expval_table[ exponent ][ y ];
1523
                }else{
1524
                    y += get_bitsz(&s->gb, linbits);
1525
                    v = l3_unscale(y, exponent);
1526
                }
1527
                if (get_bits1(&s->gb))
1528
                    v = -v;
1529
                g->sb_hybrid[s_index+1] = v;
1530
            }else{
1531
                x = y >> 5;
1532
                y = y & 0x0f;
1533
                x += y;
1534
                if (x < 15){
1535
                    v = expval_table[ exponent ][ x ];
1536
                }else{
1537
                    x += get_bitsz(&s->gb, linbits);
1538
                    v = l3_unscale(x, exponent);
1539
                }
1540
                if (get_bits1(&s->gb))
1541
                    v = -v;
1542
                g->sb_hybrid[s_index+!!y] = v;
1543
                g->sb_hybrid[s_index+ !y] = 0;
1544
            }
1545
            s_index+=2;
1546
        }
1547
    }
1548

    
1549
    /* high frequencies */
1550
    vlc = &huff_quad_vlc[g->count1table_select];
1551
    last_pos=0;
1552
    while (s_index <= 572) {
1553
        int pos, code;
1554
        pos = get_bits_count(&s->gb);
1555
        if (pos >= end_pos) {
1556
            if (pos > end_pos2 && last_pos){
1557
                /* some encoders generate an incorrect size for this
1558
                   part. We must go back into the data */
1559
                s_index -= 4;
1560
                skip_bits_long(&s->gb, last_pos - pos);
1561
                av_log(NULL, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos);
1562
                if(s->error_resilience >= FF_ER_COMPLIANT)
1563
                    s_index=0;
1564
                break;
1565
            }
1566
//                av_log(NULL, AV_LOG_ERROR, "pos2: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);
1567
            switch_buffer(s, &pos, &end_pos, &end_pos2);
1568
//                av_log(NULL, AV_LOG_ERROR, "new pos2: %d %d %d\n", pos, end_pos, s_index);
1569
            if(pos >= end_pos)
1570
                break;
1571
        }
1572
        last_pos= pos;
1573

    
1574
        code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);
1575
        dprintf(s->avctx, "t=%d code=%d\n", g->count1table_select, code);
1576
        g->sb_hybrid[s_index+0]=
1577
        g->sb_hybrid[s_index+1]=
1578
        g->sb_hybrid[s_index+2]=
1579
        g->sb_hybrid[s_index+3]= 0;
1580
        while(code){
1581
            static const int idxtab[16]={3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0};
1582
            int v;
1583
            int pos= s_index+idxtab[code];
1584
            code ^= 8>>idxtab[code];
1585
            v = exp_table[ exponents[pos] ];
1586
//            v = exp_table[ (exponents[pos]&3) ] >> FFMIN(0 - (exponents[pos]>>2), 31);
1587
            if(get_bits1(&s->gb))
1588
                v = -v;
1589
            g->sb_hybrid[pos] = v;
1590
        }
1591
        s_index+=4;
1592
    }
1593
    /* skip extension bits */
1594
    bits_left = end_pos2 - get_bits_count(&s->gb);
1595
//av_log(NULL, AV_LOG_ERROR, "left:%d buf:%p\n", bits_left, s->in_gb.buffer);
1596
    if (bits_left < 0/* || bits_left > 500*/) {
1597
        av_log(NULL, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
1598
        s_index=0;
1599
    }else if(bits_left > 0 && s->error_resilience >= FF_ER_AGGRESSIVE){
1600
        av_log(NULL, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
1601
        s_index=0;
1602
    }
1603
    memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*(576 - s_index));
1604
    skip_bits_long(&s->gb, bits_left);
1605

    
1606
    i= get_bits_count(&s->gb);
1607
    switch_buffer(s, &i, &end_pos, &end_pos2);
1608

    
1609
    return 0;
1610
}
1611

    
1612
/* Reorder short blocks from bitstream order to interleaved order. It
1613
   would be faster to do it in parsing, but the code would be far more
1614
   complicated */
1615
static void reorder_block(MPADecodeContext *s, GranuleDef *g)
1616
{
1617
    int i, j, len;
1618
    int32_t *ptr, *dst, *ptr1;
1619
    int32_t tmp[576];
1620

    
1621
    if (g->block_type != 2)
1622
        return;
1623

    
1624
    if (g->switch_point) {
1625
        if (s->sample_rate_index != 8) {
1626
            ptr = g->sb_hybrid + 36;
1627
        } else {
1628
            ptr = g->sb_hybrid + 48;
1629
        }
1630
    } else {
1631
        ptr = g->sb_hybrid;
1632
    }
1633

    
1634
    for(i=g->short_start;i<13;i++) {
1635
        len = band_size_short[s->sample_rate_index][i];
1636
        ptr1 = ptr;
1637
        dst = tmp;
1638
        for(j=len;j>0;j--) {
1639
            *dst++ = ptr[0*len];
1640
            *dst++ = ptr[1*len];
1641
            *dst++ = ptr[2*len];
1642
            ptr++;
1643
        }
1644
        ptr+=2*len;
1645
        memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));
1646
    }
1647
}
1648

    
1649
#define ISQRT2 FIXR(0.70710678118654752440)
1650

    
1651
static void compute_stereo(MPADecodeContext *s,
1652
                           GranuleDef *g0, GranuleDef *g1)
1653
{
1654
    int i, j, k, l;
1655
    int32_t v1, v2;
1656
    int sf_max, tmp0, tmp1, sf, len, non_zero_found;
1657
    int32_t (*is_tab)[16];
1658
    int32_t *tab0, *tab1;
1659
    int non_zero_found_short[3];
1660

    
1661
    /* intensity stereo */
1662
    if (s->mode_ext & MODE_EXT_I_STEREO) {
1663
        if (!s->lsf) {
1664
            is_tab = is_table;
1665
            sf_max = 7;
1666
        } else {
1667
            is_tab = is_table_lsf[g1->scalefac_compress & 1];
1668
            sf_max = 16;
1669
        }
1670

    
1671
        tab0 = g0->sb_hybrid + 576;
1672
        tab1 = g1->sb_hybrid + 576;
1673

    
1674
        non_zero_found_short[0] = 0;
1675
        non_zero_found_short[1] = 0;
1676
        non_zero_found_short[2] = 0;
1677
        k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1678
        for(i = 12;i >= g1->short_start;i--) {
1679
            /* for last band, use previous scale factor */
1680
            if (i != 11)
1681
                k -= 3;
1682
            len = band_size_short[s->sample_rate_index][i];
1683
            for(l=2;l>=0;l--) {
1684
                tab0 -= len;
1685
                tab1 -= len;
1686
                if (!non_zero_found_short[l]) {
1687
                    /* test if non zero band. if so, stop doing i-stereo */
1688
                    for(j=0;j<len;j++) {
1689
                        if (tab1[j] != 0) {
1690
                            non_zero_found_short[l] = 1;
1691
                            goto found1;
1692
                        }
1693
                    }
1694
                    sf = g1->scale_factors[k + l];
1695
                    if (sf >= sf_max)
1696
                        goto found1;
1697

    
1698
                    v1 = is_tab[0][sf];
1699
                    v2 = is_tab[1][sf];
1700
                    for(j=0;j<len;j++) {
1701
                        tmp0 = tab0[j];
1702
                        tab0[j] = MULL(tmp0, v1);
1703
                        tab1[j] = MULL(tmp0, v2);
1704
                    }
1705
                } else {
1706
                found1:
1707
                    if (s->mode_ext & MODE_EXT_MS_STEREO) {
1708
                        /* lower part of the spectrum : do ms stereo
1709
                           if enabled */
1710
                        for(j=0;j<len;j++) {
1711
                            tmp0 = tab0[j];
1712
                            tmp1 = tab1[j];
1713
                            tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
1714
                            tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
1715
                        }
1716
                    }
1717
                }
1718
            }
1719
        }
1720

    
1721
        non_zero_found = non_zero_found_short[0] |
1722
            non_zero_found_short[1] |
1723
            non_zero_found_short[2];
1724

    
1725
        for(i = g1->long_end - 1;i >= 0;i--) {
1726
            len = band_size_long[s->sample_rate_index][i];
1727
            tab0 -= len;
1728
            tab1 -= len;
1729
            /* test if non zero band. if so, stop doing i-stereo */
1730
            if (!non_zero_found) {
1731
                for(j=0;j<len;j++) {
1732
                    if (tab1[j] != 0) {
1733
                        non_zero_found = 1;
1734
                        goto found2;
1735
                    }
1736
                }
1737
                /* for last band, use previous scale factor */
1738
                k = (i == 21) ? 20 : i;
1739
                sf = g1->scale_factors[k];
1740
                if (sf >= sf_max)
1741
                    goto found2;
1742
                v1 = is_tab[0][sf];
1743
                v2 = is_tab[1][sf];
1744
                for(j=0;j<len;j++) {
1745
                    tmp0 = tab0[j];
1746
                    tab0[j] = MULL(tmp0, v1);
1747
                    tab1[j] = MULL(tmp0, v2);
1748
                }
1749
            } else {
1750
            found2:
1751
                if (s->mode_ext & MODE_EXT_MS_STEREO) {
1752
                    /* lower part of the spectrum : do ms stereo
1753
                       if enabled */
1754
                    for(j=0;j<len;j++) {
1755
                        tmp0 = tab0[j];
1756
                        tmp1 = tab1[j];
1757
                        tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
1758
                        tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
1759
                    }
1760
                }
1761
            }
1762
        }
1763
    } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1764
        /* ms stereo ONLY */
1765
        /* NOTE: the 1/sqrt(2) normalization factor is included in the
1766
           global gain */
1767
        tab0 = g0->sb_hybrid;
1768
        tab1 = g1->sb_hybrid;
1769
        for(i=0;i<576;i++) {
1770
            tmp0 = tab0[i];
1771
            tmp1 = tab1[i];
1772
            tab0[i] = tmp0 + tmp1;
1773
            tab1[i] = tmp0 - tmp1;
1774
        }
1775
    }
1776
}
1777

    
1778
static void compute_antialias_integer(MPADecodeContext *s,
1779
                              GranuleDef *g)
1780
{
1781
    int32_t *ptr, *csa;
1782
    int n, i;
1783

    
1784
    /* we antialias only "long" bands */
1785
    if (g->block_type == 2) {
1786
        if (!g->switch_point)
1787
            return;
1788
        /* XXX: check this for 8000Hz case */
1789
        n = 1;
1790
    } else {
1791
        n = SBLIMIT - 1;
1792
    }
1793

    
1794
    ptr = g->sb_hybrid + 18;
1795
    for(i = n;i > 0;i--) {
1796
        int tmp0, tmp1, tmp2;
1797
        csa = &csa_table[0][0];
1798
#define INT_AA(j) \
1799
            tmp0 = ptr[-1-j];\
1800
            tmp1 = ptr[   j];\
1801
            tmp2= MULH(tmp0 + tmp1, csa[0+4*j]);\
1802
            ptr[-1-j] = 4*(tmp2 - MULH(tmp1, csa[2+4*j]));\
1803
            ptr[   j] = 4*(tmp2 + MULH(tmp0, csa[3+4*j]));
1804

    
1805
        INT_AA(0)
1806
        INT_AA(1)
1807
        INT_AA(2)
1808
        INT_AA(3)
1809
        INT_AA(4)
1810
        INT_AA(5)
1811
        INT_AA(6)
1812
        INT_AA(7)
1813

    
1814
        ptr += 18;
1815
    }
1816
}
1817

    
1818
static void compute_antialias_float(MPADecodeContext *s,
1819
                              GranuleDef *g)
1820
{
1821
    int32_t *ptr;
1822
    int n, i;
1823

    
1824
    /* we antialias only "long" bands */
1825
    if (g->block_type == 2) {
1826
        if (!g->switch_point)
1827
            return;
1828
        /* XXX: check this for 8000Hz case */
1829
        n = 1;
1830
    } else {
1831
        n = SBLIMIT - 1;
1832
    }
1833

    
1834
    ptr = g->sb_hybrid + 18;
1835
    for(i = n;i > 0;i--) {
1836
        float tmp0, tmp1;
1837
        float *csa = &csa_table_float[0][0];
1838
#define FLOAT_AA(j)\
1839
        tmp0= ptr[-1-j];\
1840
        tmp1= ptr[   j];\
1841
        ptr[-1-j] = lrintf(tmp0 * csa[0+4*j] - tmp1 * csa[1+4*j]);\
1842
        ptr[   j] = lrintf(tmp0 * csa[1+4*j] + tmp1 * csa[0+4*j]);
1843

    
1844
        FLOAT_AA(0)
1845
        FLOAT_AA(1)
1846
        FLOAT_AA(2)
1847
        FLOAT_AA(3)
1848
        FLOAT_AA(4)
1849
        FLOAT_AA(5)
1850
        FLOAT_AA(6)
1851
        FLOAT_AA(7)
1852

    
1853
        ptr += 18;
1854
    }
1855
}
1856

    
1857
static void compute_imdct(MPADecodeContext *s,
1858
                          GranuleDef *g,
1859
                          int32_t *sb_samples,
1860
                          int32_t *mdct_buf)
1861
{
1862
    int32_t *ptr, *win, *win1, *buf, *out_ptr, *ptr1;
1863
    int32_t out2[12];
1864
    int i, j, mdct_long_end, v, sblimit;
1865

    
1866
    /* find last non zero block */
1867
    ptr = g->sb_hybrid + 576;
1868
    ptr1 = g->sb_hybrid + 2 * 18;
1869
    while (ptr >= ptr1) {
1870
        ptr -= 6;
1871
        v = ptr[0] | ptr[1] | ptr[2] | ptr[3] | ptr[4] | ptr[5];
1872
        if (v != 0)
1873
            break;
1874
    }
1875
    sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
1876

    
1877
    if (g->block_type == 2) {
1878
        /* XXX: check for 8000 Hz */
1879
        if (g->switch_point)
1880
            mdct_long_end = 2;
1881
        else
1882
            mdct_long_end = 0;
1883
    } else {
1884
        mdct_long_end = sblimit;
1885
    }
1886

    
1887
    buf = mdct_buf;
1888
    ptr = g->sb_hybrid;
1889
    for(j=0;j<mdct_long_end;j++) {
1890
        /* apply window & overlap with previous buffer */
1891
        out_ptr = sb_samples + j;
1892
        /* select window */
1893
        if (g->switch_point && j < 2)
1894
            win1 = mdct_win[0];
1895
        else
1896
            win1 = mdct_win[g->block_type];
1897
        /* select frequency inversion */
1898
        win = win1 + ((4 * 36) & -(j & 1));
1899
        imdct36(out_ptr, buf, ptr, win);
1900
        out_ptr += 18*SBLIMIT;
1901
        ptr += 18;
1902
        buf += 18;
1903
    }
1904
    for(j=mdct_long_end;j<sblimit;j++) {
1905
        /* select frequency inversion */
1906
        win = mdct_win[2] + ((4 * 36) & -(j & 1));
1907
        out_ptr = sb_samples + j;
1908

    
1909
        for(i=0; i<6; i++){
1910
            *out_ptr = buf[i];
1911
            out_ptr += SBLIMIT;
1912
        }
1913
        imdct12(out2, ptr + 0);
1914
        for(i=0;i<6;i++) {
1915
            *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*1];
1916
            buf[i + 6*2] = MULH(out2[i + 6], win[i + 6]);
1917
            out_ptr += SBLIMIT;
1918
        }
1919
        imdct12(out2, ptr + 1);
1920
        for(i=0;i<6;i++) {
1921
            *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*2];
1922
            buf[i + 6*0] = MULH(out2[i + 6], win[i + 6]);
1923
            out_ptr += SBLIMIT;
1924
        }
1925
        imdct12(out2, ptr + 2);
1926
        for(i=0;i<6;i++) {
1927
            buf[i + 6*0] = MULH(out2[i], win[i]) + buf[i + 6*0];
1928
            buf[i + 6*1] = MULH(out2[i + 6], win[i + 6]);
1929
            buf[i + 6*2] = 0;
1930
        }
1931
        ptr += 18;
1932
        buf += 18;
1933
    }
1934
    /* zero bands */
1935
    for(j=sblimit;j<SBLIMIT;j++) {
1936
        /* overlap */
1937
        out_ptr = sb_samples + j;
1938
        for(i=0;i<18;i++) {
1939
            *out_ptr = buf[i];
1940
            buf[i] = 0;
1941
            out_ptr += SBLIMIT;
1942
        }
1943
        buf += 18;
1944
    }
1945
}
1946

    
1947
#if defined(DEBUG)
1948
void sample_dump(int fnum, int32_t *tab, int n)
1949
{
1950
    static FILE *files[16], *f;
1951
    char buf[512];
1952
    int i;
1953
    int32_t v;
1954

    
1955
    f = files[fnum];
1956
    if (!f) {
1957
        snprintf(buf, sizeof(buf), "/tmp/out%d.%s.pcm",
1958
                fnum,
1959
#ifdef USE_HIGHPRECISION
1960
                "hp"
1961
#else
1962
                "lp"
1963
#endif
1964
                );
1965
        f = fopen(buf, "w");
1966
        if (!f)
1967
            return;
1968
        files[fnum] = f;
1969
    }
1970

    
1971
    if (fnum == 0) {
1972
        static int pos = 0;
1973
        av_log(NULL, AV_LOG_DEBUG, "pos=%d\n", pos);
1974
        for(i=0;i<n;i++) {
1975
            av_log(NULL, AV_LOG_DEBUG, " %0.4f", (double)tab[i] / FRAC_ONE);
1976
            if ((i % 18) == 17)
1977
                av_log(NULL, AV_LOG_DEBUG, "\n");
1978
        }
1979
        pos += n;
1980
    }
1981
    for(i=0;i<n;i++) {
1982
        /* normalize to 23 frac bits */
1983
        v = tab[i] << (23 - FRAC_BITS);
1984
        fwrite(&v, 1, sizeof(int32_t), f);
1985
    }
1986
}
1987
#endif
1988

    
1989

    
1990
/* main layer3 decoding function */
1991
static int mp_decode_layer3(MPADecodeContext *s)
1992
{
1993
    int nb_granules, main_data_begin, private_bits;
1994
    int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;
1995
    GranuleDef granules[2][2], *g;
1996
    int16_t exponents[576];
1997

    
1998
    /* read side info */
1999
    if (s->lsf) {
2000
        main_data_begin = get_bits(&s->gb, 8);
2001
        private_bits = get_bits(&s->gb, s->nb_channels);
2002
        nb_granules = 1;
2003
    } else {
2004
        main_data_begin = get_bits(&s->gb, 9);
2005
        if (s->nb_channels == 2)
2006
            private_bits = get_bits(&s->gb, 3);
2007
        else
2008
            private_bits = get_bits(&s->gb, 5);
2009
        nb_granules = 2;
2010
        for(ch=0;ch<s->nb_channels;ch++) {
2011
            granules[ch][0].scfsi = 0; /* all scale factors are transmitted */
2012
            granules[ch][1].scfsi = get_bits(&s->gb, 4);
2013
        }
2014
    }
2015

    
2016
    for(gr=0;gr<nb_granules;gr++) {
2017
        for(ch=0;ch<s->nb_channels;ch++) {
2018
            dprintf(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);
2019
            g = &granules[ch][gr];
2020
            g->part2_3_length = get_bits(&s->gb, 12);
2021
            g->big_values = get_bits(&s->gb, 9);
2022
            if(g->big_values > 288){
2023
                av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");
2024
                return -1;
2025
            }
2026

    
2027
            g->global_gain = get_bits(&s->gb, 8);
2028
            /* if MS stereo only is selected, we precompute the
2029
               1/sqrt(2) renormalization factor */
2030
            if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
2031
                MODE_EXT_MS_STEREO)
2032
                g->global_gain -= 2;
2033
            if (s->lsf)
2034
                g->scalefac_compress = get_bits(&s->gb, 9);
2035
            else
2036
                g->scalefac_compress = get_bits(&s->gb, 4);
2037
            blocksplit_flag = get_bits(&s->gb, 1);
2038
            if (blocksplit_flag) {
2039
                g->block_type = get_bits(&s->gb, 2);
2040
                if (g->block_type == 0){
2041
                    av_log(NULL, AV_LOG_ERROR, "invalid block type\n");
2042
                    return -1;
2043
                }
2044
                g->switch_point = get_bits(&s->gb, 1);
2045
                for(i=0;i<2;i++)
2046
                    g->table_select[i] = get_bits(&s->gb, 5);
2047
                for(i=0;i<3;i++)
2048
                    g->subblock_gain[i] = get_bits(&s->gb, 3);
2049
                /* compute huffman coded region sizes */
2050
                if (g->block_type == 2)
2051
                    g->region_size[0] = (36 / 2);
2052
                else {
2053
                    if (s->sample_rate_index <= 2)
2054
                        g->region_size[0] = (36 / 2);
2055
                    else if (s->sample_rate_index != 8)
2056
                        g->region_size[0] = (54 / 2);
2057
                    else
2058
                        g->region_size[0] = (108 / 2);
2059
                }
2060
                g->region_size[1] = (576 / 2);
2061
            } else {
2062
                int region_address1, region_address2, l;
2063
                g->block_type = 0;
2064
                g->switch_point = 0;
2065
                for(i=0;i<3;i++)
2066
                    g->table_select[i] = get_bits(&s->gb, 5);
2067
                /* compute huffman coded region sizes */
2068
                region_address1 = get_bits(&s->gb, 4);
2069
                region_address2 = get_bits(&s->gb, 3);
2070
                dprintf(s->avctx, "region1=%d region2=%d\n",
2071
                        region_address1, region_address2);
2072
                g->region_size[0] =
2073
                    band_index_long[s->sample_rate_index][region_address1 + 1] >> 1;
2074
                l = region_address1 + region_address2 + 2;
2075
                /* should not overflow */
2076
                if (l > 22)
2077
                    l = 22;
2078
                g->region_size[1] =
2079
                    band_index_long[s->sample_rate_index][l] >> 1;
2080
            }
2081
            /* convert region offsets to region sizes and truncate
2082
               size to big_values */
2083
            g->region_size[2] = (576 / 2);
2084
            j = 0;
2085
            for(i=0;i<3;i++) {
2086
                k = FFMIN(g->region_size[i], g->big_values);
2087
                g->region_size[i] = k - j;
2088
                j = k;
2089
            }
2090

    
2091
            /* compute band indexes */
2092
            if (g->block_type == 2) {
2093
                if (g->switch_point) {
2094
                    /* if switched mode, we handle the 36 first samples as
2095
                       long blocks.  For 8000Hz, we handle the 48 first
2096
                       exponents as long blocks (XXX: check this!) */
2097
                    if (s->sample_rate_index <= 2)
2098
                        g->long_end = 8;
2099
                    else if (s->sample_rate_index != 8)
2100
                        g->long_end = 6;
2101
                    else
2102
                        g->long_end = 4; /* 8000 Hz */
2103

    
2104
                    g->short_start = 2 + (s->sample_rate_index != 8);
2105
                } else {
2106
                    g->long_end = 0;
2107
                    g->short_start = 0;
2108
                }
2109
            } else {
2110
                g->short_start = 13;
2111
                g->long_end = 22;
2112
            }
2113

    
2114
            g->preflag = 0;
2115
            if (!s->lsf)
2116
                g->preflag = get_bits(&s->gb, 1);
2117
            g->scalefac_scale = get_bits(&s->gb, 1);
2118
            g->count1table_select = get_bits(&s->gb, 1);
2119
            dprintf(s->avctx, "block_type=%d switch_point=%d\n",
2120
                    g->block_type, g->switch_point);
2121
        }
2122
    }
2123

    
2124
  if (!s->adu_mode) {
2125
    const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);
2126
    assert((get_bits_count(&s->gb) & 7) == 0);
2127
    /* now we get bits from the main_data_begin offset */
2128
    dprintf(s->avctx, "seekback: %d\n", main_data_begin);
2129
//av_log(NULL, AV_LOG_ERROR, "backstep:%d, lastbuf:%d\n", main_data_begin, s->last_buf_size);
2130

    
2131
    memcpy(s->last_buf + s->last_buf_size, ptr, EXTRABYTES);
2132
    s->in_gb= s->gb;
2133
        init_get_bits(&s->gb, s->last_buf, s->last_buf_size*8);
2134
        skip_bits_long(&s->gb, 8*(s->last_buf_size - main_data_begin));
2135
  }
2136

    
2137
    for(gr=0;gr<nb_granules;gr++) {
2138
        for(ch=0;ch<s->nb_channels;ch++) {
2139
            g = &granules[ch][gr];
2140
            if(get_bits_count(&s->gb)<0){
2141
                av_log(NULL, AV_LOG_ERROR, "mdb:%d, lastbuf:%d skipping granule %d\n",
2142
                                            main_data_begin, s->last_buf_size, gr);
2143
                skip_bits_long(&s->gb, g->part2_3_length);
2144
                memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));
2145
                if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->in_gb.buffer){
2146
                    skip_bits_long(&s->in_gb, get_bits_count(&s->gb) - s->gb.size_in_bits);
2147
                    s->gb= s->in_gb;
2148
                    s->in_gb.buffer=NULL;
2149
                }
2150
                continue;
2151
            }
2152

    
2153
            bits_pos = get_bits_count(&s->gb);
2154

    
2155
            if (!s->lsf) {
2156
                uint8_t *sc;
2157
                int slen, slen1, slen2;
2158

    
2159
                /* MPEG1 scale factors */
2160
                slen1 = slen_table[0][g->scalefac_compress];
2161
                slen2 = slen_table[1][g->scalefac_compress];
2162
                dprintf(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);
2163
                if (g->block_type == 2) {
2164
                    n = g->switch_point ? 17 : 18;
2165
                    j = 0;
2166
                    if(slen1){
2167
                        for(i=0;i<n;i++)
2168
                            g->scale_factors[j++] = get_bits(&s->gb, slen1);
2169
                    }else{
2170
                        for(i=0;i<n;i++)
2171
                            g->scale_factors[j++] = 0;
2172
                    }
2173
                    if(slen2){
2174
                        for(i=0;i<18;i++)
2175
                            g->scale_factors[j++] = get_bits(&s->gb, slen2);
2176
                        for(i=0;i<3;i++)
2177
                            g->scale_factors[j++] = 0;
2178
                    }else{
2179
                        for(i=0;i<21;i++)
2180
                            g->scale_factors[j++] = 0;
2181
                    }
2182
                } else {
2183
                    sc = granules[ch][0].scale_factors;
2184
                    j = 0;
2185
                    for(k=0;k<4;k++) {
2186
                        n = (k == 0 ? 6 : 5);
2187
                        if ((g->scfsi & (0x8 >> k)) == 0) {
2188
                            slen = (k < 2) ? slen1 : slen2;
2189
                            if(slen){
2190
                                for(i=0;i<n;i++)
2191
                                    g->scale_factors[j++] = get_bits(&s->gb, slen);
2192
                            }else{
2193
                                for(i=0;i<n;i++)
2194
                                    g->scale_factors[j++] = 0;
2195
                            }
2196
                        } else {
2197
                            /* simply copy from last granule */
2198
                            for(i=0;i<n;i++) {
2199
                                g->scale_factors[j] = sc[j];
2200
                                j++;
2201
                            }
2202
                        }
2203
                    }
2204
                    g->scale_factors[j++] = 0;
2205
                }
2206
#if defined(DEBUG)
2207
                {
2208
                    dprintf(s->avctx, "scfsi=%x gr=%d ch=%d scale_factors:\n",
2209
                           g->scfsi, gr, ch);
2210
                    for(i=0;i<j;i++)
2211
                        dprintf(s->avctx, " %d", g->scale_factors[i]);
2212
                    dprintf(s->avctx, "\n");
2213
                }
2214
#endif
2215
            } else {
2216
                int tindex, tindex2, slen[4], sl, sf;
2217

    
2218
                /* LSF scale factors */
2219
                if (g->block_type == 2) {
2220
                    tindex = g->switch_point ? 2 : 1;
2221
                } else {
2222
                    tindex = 0;
2223
                }
2224
                sf = g->scalefac_compress;
2225
                if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
2226
                    /* intensity stereo case */
2227
                    sf >>= 1;
2228
                    if (sf < 180) {
2229
                        lsf_sf_expand(slen, sf, 6, 6, 0);
2230
                        tindex2 = 3;
2231
                    } else if (sf < 244) {
2232
                        lsf_sf_expand(slen, sf - 180, 4, 4, 0);
2233
                        tindex2 = 4;
2234
                    } else {
2235
                        lsf_sf_expand(slen, sf - 244, 3, 0, 0);
2236
                        tindex2 = 5;
2237
                    }
2238
                } else {
2239
                    /* normal case */
2240
                    if (sf < 400) {
2241
                        lsf_sf_expand(slen, sf, 5, 4, 4);
2242
                        tindex2 = 0;
2243
                    } else if (sf < 500) {
2244
                        lsf_sf_expand(slen, sf - 400, 5, 4, 0);
2245
                        tindex2 = 1;
2246
                    } else {
2247
                        lsf_sf_expand(slen, sf - 500, 3, 0, 0);
2248
                        tindex2 = 2;
2249
                        g->preflag = 1;
2250
                    }
2251
                }
2252

    
2253
                j = 0;
2254
                for(k=0;k<4;k++) {
2255
                    n = lsf_nsf_table[tindex2][tindex][k];
2256
                    sl = slen[k];
2257
                    if(sl){
2258
                        for(i=0;i<n;i++)
2259
                            g->scale_factors[j++] = get_bits(&s->gb, sl);
2260
                    }else{
2261
                        for(i=0;i<n;i++)
2262
                            g->scale_factors[j++] = 0;
2263
                    }
2264
                }
2265
                /* XXX: should compute exact size */
2266
                for(;j<40;j++)
2267
                    g->scale_factors[j] = 0;
2268
#if defined(DEBUG)
2269
                {
2270
                    dprintf(s->avctx, "gr=%d ch=%d scale_factors:\n",
2271
                           gr, ch);
2272
                    for(i=0;i<40;i++)
2273
                        dprintf(s->avctx, " %d", g->scale_factors[i]);
2274
                    dprintf(s->avctx, "\n");
2275
                }
2276
#endif
2277
            }
2278

    
2279
            exponents_from_scale_factors(s, g, exponents);
2280

    
2281
            /* read Huffman coded residue */
2282
            huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);
2283
#if defined(DEBUG)
2284
            sample_dump(0, g->sb_hybrid, 576);
2285
#endif
2286
        } /* ch */
2287

    
2288
        if (s->nb_channels == 2)
2289
            compute_stereo(s, &granules[0][gr], &granules[1][gr]);
2290

    
2291
        for(ch=0;ch<s->nb_channels;ch++) {
2292
            g = &granules[ch][gr];
2293

    
2294
            reorder_block(s, g);
2295
#if defined(DEBUG)
2296
            sample_dump(0, g->sb_hybrid, 576);
2297
#endif
2298
            s->compute_antialias(s, g);
2299
#if defined(DEBUG)
2300
            sample_dump(1, g->sb_hybrid, 576);
2301
#endif
2302
            compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
2303
#if defined(DEBUG)
2304
            sample_dump(2, &s->sb_samples[ch][18 * gr][0], 576);
2305
#endif
2306
        }
2307
    } /* gr */
2308
    if(get_bits_count(&s->gb)<0)
2309
        skip_bits_long(&s->gb, -get_bits_count(&s->gb));
2310
    return nb_granules * 18;
2311
}
2312

    
2313
static int mp_decode_frame(MPADecodeContext *s,
2314
                           OUT_INT *samples, const uint8_t *buf, int buf_size)
2315
{
2316
    int i, nb_frames, ch;
2317
    OUT_INT *samples_ptr;
2318

    
2319
    init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE)*8);
2320

    
2321
    /* skip error protection field */
2322
    if (s->error_protection)
2323
        get_bits(&s->gb, 16);
2324

    
2325
    dprintf(s->avctx, "frame %d:\n", s->frame_count);
2326
    switch(s->layer) {
2327
    case 1:
2328
        nb_frames = mp_decode_layer1(s);
2329
        break;
2330
    case 2:
2331
        nb_frames = mp_decode_layer2(s);
2332
        break;
2333
    case 3:
2334
    default:
2335
        nb_frames = mp_decode_layer3(s);
2336

    
2337
        s->last_buf_size=0;
2338
        if(s->in_gb.buffer){
2339
            align_get_bits(&s->gb);
2340
            i= (s->gb.size_in_bits - get_bits_count(&s->gb))>>3;
2341
            if(i >= 0 && i <= BACKSTEP_SIZE){
2342
                memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);
2343
                s->last_buf_size=i;
2344
            }else
2345
                av_log(NULL, AV_LOG_ERROR, "invalid old backstep %d\n", i);
2346
            s->gb= s->in_gb;
2347
            s->in_gb.buffer= NULL;
2348
        }
2349

    
2350
        align_get_bits(&s->gb);
2351
        assert((get_bits_count(&s->gb) & 7) == 0);
2352
        i= (s->gb.size_in_bits - get_bits_count(&s->gb))>>3;
2353

    
2354
        if(i<0 || i > BACKSTEP_SIZE || nb_frames<0){
2355
            av_log(NULL, AV_LOG_ERROR, "invalid new backstep %d\n", i);
2356
            i= FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);
2357
        }
2358
        assert(i <= buf_size - HEADER_SIZE && i>= 0);
2359
        memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);
2360
        s->last_buf_size += i;
2361

    
2362
        break;
2363
    }
2364
#if defined(DEBUG)
2365
    for(i=0;i<nb_frames;i++) {
2366
        for(ch=0;ch<s->nb_channels;ch++) {
2367
            int j;
2368
            dprintf(s->avctx, "%d-%d:", i, ch);
2369
            for(j=0;j<SBLIMIT;j++)
2370
                dprintf(s->avctx, " %0.6f", (double)s->sb_samples[ch][i][j] / FRAC_ONE);
2371
            dprintf(s->avctx, "\n");
2372
        }
2373
    }
2374
#endif
2375
    /* apply the synthesis filter */
2376
    for(ch=0;ch<s->nb_channels;ch++) {
2377
        samples_ptr = samples + ch;
2378
        for(i=0;i<nb_frames;i++) {
2379
            ff_mpa_synth_filter(s->synth_buf[ch], &(s->synth_buf_offset[ch]),
2380
                         window, &s->dither_state,
2381
                         samples_ptr, s->nb_channels,
2382
                         s->sb_samples[ch][i]);
2383
            samples_ptr += 32 * s->nb_channels;
2384
        }
2385
    }
2386
#ifdef DEBUG
2387
    s->frame_count++;
2388
#endif
2389
    return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
2390
}
2391

    
2392
static int decode_frame(AVCodecContext * avctx,
2393
                        void *data, int *data_size,
2394
                        uint8_t * buf, int buf_size)
2395
{
2396
    MPADecodeContext *s = avctx->priv_data;
2397
    uint32_t header;
2398
    int out_size;
2399
    OUT_INT *out_samples = data;
2400

    
2401
retry:
2402
    if(buf_size < HEADER_SIZE)
2403
        return -1;
2404

    
2405
    header = (buf[0] << 24) | (buf[1] << 16) | (buf[2] << 8) | buf[3];
2406
    if(ff_mpa_check_header(header) < 0){
2407
        buf++;
2408
//        buf_size--;
2409
        av_log(avctx, AV_LOG_ERROR, "Header missing skipping one byte.\n");
2410
        goto retry;
2411
    }
2412

    
2413
    if (decode_header(s, header) == 1) {
2414
        /* free format: prepare to compute frame size */
2415
        s->frame_size = -1;
2416
        return -1;
2417
    }
2418
    /* update codec info */
2419
    avctx->channels = s->nb_channels;
2420
    avctx->bit_rate = s->bit_rate;
2421
    avctx->sub_id = s->layer;
2422
    switch(s->layer) {
2423
    case 1:
2424
        avctx->frame_size = 384;
2425
        break;
2426
    case 2:
2427
        avctx->frame_size = 1152;
2428
        break;
2429
    case 3:
2430
        if (s->lsf)
2431
            avctx->frame_size = 576;
2432
        else
2433
            avctx->frame_size = 1152;
2434
        break;
2435
    }
2436

    
2437
    if(s->frame_size<=0 || s->frame_size > buf_size){
2438
        av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
2439
        return -1;
2440
    }else if(s->frame_size < buf_size){
2441
        av_log(avctx, AV_LOG_ERROR, "incorrect frame size\n");
2442
    }
2443

    
2444
    out_size = mp_decode_frame(s, out_samples, buf, buf_size);
2445
    if(out_size>=0){
2446
        *data_size = out_size;
2447
        avctx->sample_rate = s->sample_rate;
2448
        //FIXME maybe move the other codec info stuff from above here too
2449
    }else
2450
        av_log(avctx, AV_LOG_DEBUG, "Error while decoding MPEG audio frame.\n"); //FIXME return -1 / but also return the number of bytes consumed
2451
    s->frame_size = 0;
2452
    return buf_size;
2453
}
2454

    
2455
static void flush(AVCodecContext *avctx){
2456
    MPADecodeContext *s = avctx->priv_data;
2457
    s->last_buf_size= 0;
2458
}
2459

    
2460
#ifdef CONFIG_MP3ADU_DECODER
2461
static int decode_frame_adu(AVCodecContext * avctx,
2462
                        void *data, int *data_size,
2463
                        uint8_t * buf, int buf_size)
2464
{
2465
    MPADecodeContext *s = avctx->priv_data;
2466
    uint32_t header;
2467
    int len, out_size;
2468
    OUT_INT *out_samples = data;
2469

    
2470
    len = buf_size;
2471

    
2472
    // Discard too short frames
2473
    if (buf_size < HEADER_SIZE) {
2474
        *data_size = 0;
2475
        return buf_size;
2476
    }
2477

    
2478

    
2479
    if (len > MPA_MAX_CODED_FRAME_SIZE)
2480
        len = MPA_MAX_CODED_FRAME_SIZE;
2481

    
2482
    // Get header and restore sync word
2483
    header = (buf[0] << 24) | (buf[1] << 16) | (buf[2] << 8) | buf[3] | 0xffe00000;
2484

    
2485
    if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame
2486
        *data_size = 0;
2487
        return buf_size;
2488
    }
2489

    
2490
    decode_header(s, header);
2491
    /* update codec info */
2492
    avctx->sample_rate = s->sample_rate;
2493
    avctx->channels = s->nb_channels;
2494
    avctx->bit_rate = s->bit_rate;
2495
    avctx->sub_id = s->layer;
2496

    
2497
    avctx->frame_size=s->frame_size = len;
2498

    
2499
    if (avctx->parse_only) {
2500
        out_size = buf_size;
2501
    } else {
2502
        out_size = mp_decode_frame(s, out_samples, buf, buf_size);
2503
    }
2504

    
2505
    *data_size = out_size;
2506
    return buf_size;
2507
}
2508
#endif /* CONFIG_MP3ADU_DECODER */
2509

    
2510
#ifdef CONFIG_MP3ON4_DECODER
2511
/* Next 3 arrays are indexed by channel config number (passed via codecdata) */
2512
static int mp3Frames[16] = {0,1,1,2,3,3,4,5,2};   /* number of mp3 decoder instances */
2513
static int mp3Channels[16] = {0,1,2,3,4,5,6,8,4}; /* total output channels */
2514
/* offsets into output buffer, assume output order is FL FR BL BR C LFE */
2515
static int chan_offset[9][5] = {
2516
    {0},
2517
    {0},            // C
2518
    {0},            // FLR
2519
    {2,0},          // C FLR
2520
    {2,0,3},        // C FLR BS
2521
    {4,0,2},        // C FLR BLRS
2522
    {4,0,2,5},      // C FLR BLRS LFE
2523
    {4,0,2,6,5},    // C FLR BLRS BLR LFE
2524
    {0,2}           // FLR BLRS
2525
};
2526

    
2527

    
2528
static int decode_init_mp3on4(AVCodecContext * avctx)
2529
{
2530
    MP3On4DecodeContext *s = avctx->priv_data;
2531
    int i;
2532

    
2533
    if ((avctx->extradata_size < 2) || (avctx->extradata == NULL)) {
2534
        av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
2535
        return -1;
2536
    }
2537

    
2538
    s->chan_cfg = (((unsigned char *)avctx->extradata)[1] >> 3) & 0x0f;
2539
    s->frames = mp3Frames[s->chan_cfg];
2540
    if(!s->frames) {
2541
        av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
2542
        return -1;
2543
    }
2544
    avctx->channels = mp3Channels[s->chan_cfg];
2545

    
2546
    /* Init the first mp3 decoder in standard way, so that all tables get builded
2547
     * We replace avctx->priv_data with the context of the first decoder so that
2548
     * decode_init() does not have to be changed.
2549
     * Other decoders will be inited here copying data from the first context
2550
     */
2551
    // Allocate zeroed memory for the first decoder context
2552
    s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
2553
    // Put decoder context in place to make init_decode() happy
2554
    avctx->priv_data = s->mp3decctx[0];
2555
    decode_init(avctx);
2556
    // Restore mp3on4 context pointer
2557
    avctx->priv_data = s;
2558
    s->mp3decctx[0]->adu_mode = 1; // Set adu mode
2559

    
2560
    /* Create a separate codec/context for each frame (first is already ok).
2561
     * Each frame is 1 or 2 channels - up to 5 frames allowed
2562
     */
2563
    for (i = 1; i < s->frames; i++) {
2564
        s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
2565
        s->mp3decctx[i]->compute_antialias = s->mp3decctx[0]->compute_antialias;
2566
        s->mp3decctx[i]->adu_mode = 1;
2567
        s->mp3decctx[i]->avctx = avctx;
2568
    }
2569

    
2570
    return 0;
2571
}
2572

    
2573

    
2574
static int decode_close_mp3on4(AVCodecContext * avctx)
2575
{
2576
    MP3On4DecodeContext *s = avctx->priv_data;
2577
    int i;
2578

    
2579
    for (i = 0; i < s->frames; i++)
2580
        if (s->mp3decctx[i])
2581
            av_free(s->mp3decctx[i]);
2582

    
2583
    return 0;
2584
}
2585

    
2586

    
2587
static int decode_frame_mp3on4(AVCodecContext * avctx,
2588
                        void *data, int *data_size,
2589
                        uint8_t * buf, int buf_size)
2590
{
2591
    MP3On4DecodeContext *s = avctx->priv_data;
2592
    MPADecodeContext *m;
2593
    int len, out_size = 0;
2594
    uint32_t header;
2595
    OUT_INT *out_samples = data;
2596
    OUT_INT decoded_buf[MPA_FRAME_SIZE * MPA_MAX_CHANNELS];
2597
    OUT_INT *outptr, *bp;
2598
    int fsize;
2599
    unsigned char *start2 = buf, *start;
2600
    int fr, i, j, n;
2601
    int off = avctx->channels;
2602
    int *coff = chan_offset[s->chan_cfg];
2603

    
2604
    len = buf_size;
2605

    
2606
    // Discard too short frames
2607
    if (buf_size < HEADER_SIZE) {
2608
        *data_size = 0;
2609
        return buf_size;
2610
    }
2611

    
2612
    // If only one decoder interleave is not needed
2613
    outptr = s->frames == 1 ? out_samples : decoded_buf;
2614

    
2615
    for (fr = 0; fr < s->frames; fr++) {
2616
        start = start2;
2617
        fsize = (start[0] << 4) | (start[1] >> 4);
2618
        start2 += fsize;
2619
        if (fsize > len)
2620
            fsize = len;
2621
        len -= fsize;
2622
        if (fsize > MPA_MAX_CODED_FRAME_SIZE)
2623
            fsize = MPA_MAX_CODED_FRAME_SIZE;
2624
        m = s->mp3decctx[fr];
2625
        assert (m != NULL);
2626

    
2627
        // Get header
2628
        header = (start[0] << 24) | (start[1] << 16) | (start[2] << 8) | start[3] | 0xfff00000;
2629

    
2630
        if (ff_mpa_check_header(header) < 0) { // Bad header, discard block
2631
            *data_size = 0;
2632
            return buf_size;
2633
        }
2634

    
2635
        decode_header(m, header);
2636
        mp_decode_frame(m, decoded_buf, start, fsize);
2637

    
2638
        n = MPA_FRAME_SIZE * m->nb_channels;
2639
        out_size += n * sizeof(OUT_INT);
2640
        if(s->frames > 1) {
2641
            /* interleave output data */
2642
            bp = out_samples + coff[fr];
2643
            if(m->nb_channels == 1) {
2644
                for(j = 0; j < n; j++) {
2645
                    *bp = decoded_buf[j];
2646
                    bp += off;
2647
                }
2648
            } else {
2649
                for(j = 0; j < n; j++) {
2650
                    bp[0] = decoded_buf[j++];
2651
                    bp[1] = decoded_buf[j];
2652
                    bp += off;
2653
                }
2654
            }
2655
        }
2656
    }
2657

    
2658
    /* update codec info */
2659
    avctx->sample_rate = s->mp3decctx[0]->sample_rate;
2660
    avctx->frame_size= buf_size;
2661
    avctx->bit_rate = 0;
2662
    for (i = 0; i < s->frames; i++)
2663
        avctx->bit_rate += s->mp3decctx[i]->bit_rate;
2664

    
2665
    *data_size = out_size;
2666
    return buf_size;
2667
}
2668
#endif /* CONFIG_MP3ON4_DECODER */
2669

    
2670
#ifdef CONFIG_MP2_DECODER
2671
AVCodec mp2_decoder =
2672
{
2673
    "mp2",
2674
    CODEC_TYPE_AUDIO,
2675
    CODEC_ID_MP2,
2676
    sizeof(MPADecodeContext),
2677
    decode_init,
2678
    NULL,
2679
    NULL,
2680
    decode_frame,
2681
    CODEC_CAP_PARSE_ONLY,
2682
};
2683
#endif
2684
#ifdef CONFIG_MP3_DECODER
2685
AVCodec mp3_decoder =
2686
{
2687
    "mp3",
2688
    CODEC_TYPE_AUDIO,
2689
    CODEC_ID_MP3,
2690
    sizeof(MPADecodeContext),
2691
    decode_init,
2692
    NULL,
2693
    NULL,
2694
    decode_frame,
2695
    CODEC_CAP_PARSE_ONLY,
2696
    .flush= flush,
2697
};
2698
#endif
2699
#ifdef CONFIG_MP3ADU_DECODER
2700
AVCodec mp3adu_decoder =
2701
{
2702
    "mp3adu",
2703
    CODEC_TYPE_AUDIO,
2704
    CODEC_ID_MP3ADU,
2705
    sizeof(MPADecodeContext),
2706
    decode_init,
2707
    NULL,
2708
    NULL,
2709
    decode_frame_adu,
2710
    CODEC_CAP_PARSE_ONLY,
2711
    .flush= flush,
2712
};
2713
#endif
2714
#ifdef CONFIG_MP3ON4_DECODER
2715
AVCodec mp3on4_decoder =
2716
{
2717
    "mp3on4",
2718
    CODEC_TYPE_AUDIO,
2719
    CODEC_ID_MP3ON4,
2720
    sizeof(MP3On4DecodeContext),
2721
    decode_init_mp3on4,
2722
    NULL,
2723
    decode_close_mp3on4,
2724
    decode_frame_mp3on4,
2725
    .flush= flush,
2726
};
2727
#endif