Statistics
| Branch: | Revision:

ffmpeg / libavcodec / mpegaudiodec.c @ 4a69055b

History | View | Annotate | Download (80.2 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
/* WARNING: only correct for posititive numbers */
50
#define FIXR(a)   ((int)((a) * FRAC_ONE + 0.5))
51
#define FRAC_RND(a) (((a) + (FRAC_ONE/2)) >> FRAC_BITS)
52

    
53
#define FIXHR(a) ((int)((a) * (1LL<<32) + 0.5))
54

    
55
/****************/
56

    
57
#define HEADER_SIZE 4
58

    
59
/**
60
 * Context for MP3On4 decoder
61
 */
62
typedef struct MP3On4DecodeContext {
63
    int frames;   ///< number of mp3 frames per block (number of mp3 decoder instances)
64
    int chan_cfg; ///< channel config number
65
    MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance
66
} MP3On4DecodeContext;
67

    
68
/* layer 3 "granule" */
69
typedef struct GranuleDef {
70
    uint8_t scfsi;
71
    int part2_3_length;
72
    int big_values;
73
    int global_gain;
74
    int scalefac_compress;
75
    uint8_t block_type;
76
    uint8_t switch_point;
77
    int table_select[3];
78
    int subblock_gain[3];
79
    uint8_t scalefac_scale;
80
    uint8_t count1table_select;
81
    int region_size[3]; /* number of huffman codes in each region */
82
    int preflag;
83
    int short_start, long_end; /* long/short band indexes */
84
    uint8_t scale_factors[40];
85
    int32_t sb_hybrid[SBLIMIT * 18]; /* 576 samples */
86
} GranuleDef;
87

    
88
#include "mpegaudiodata.h"
89
#include "mpegaudiodectab.h"
90

    
91
static void compute_antialias_integer(MPADecodeContext *s, GranuleDef *g);
92
static void compute_antialias_float(MPADecodeContext *s, GranuleDef *g);
93

    
94
/* vlc structure for decoding layer 3 huffman tables */
95
static VLC huff_vlc[16];
96
static VLC huff_quad_vlc[2];
97
/* computed from band_size_long */
98
static uint16_t band_index_long[9][23];
99
/* XXX: free when all decoders are closed */
100
#define TABLE_4_3_SIZE (8191 + 16)*4
101
static int8_t  table_4_3_exp[TABLE_4_3_SIZE];
102
static uint32_t table_4_3_value[TABLE_4_3_SIZE];
103
static uint32_t exp_table[512];
104
static uint32_t expval_table[512][16];
105
/* intensity stereo coef table */
106
static int32_t is_table[2][16];
107
static int32_t is_table_lsf[2][2][16];
108
static int32_t csa_table[8][4];
109
static float csa_table_float[8][4];
110
static int32_t mdct_win[8][36];
111

    
112
/* lower 2 bits: modulo 3, higher bits: shift */
113
static uint16_t scale_factor_modshift[64];
114
/* [i][j]:  2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
115
static int32_t scale_factor_mult[15][3];
116
/* mult table for layer 2 group quantization */
117

    
118
#define SCALE_GEN(v) \
119
{ FIXR(1.0 * (v)), FIXR(0.7937005259 * (v)), FIXR(0.6299605249 * (v)) }
120

    
121
static const int32_t scale_factor_mult2[3][3] = {
122
    SCALE_GEN(4.0 / 3.0), /* 3 steps */
123
    SCALE_GEN(4.0 / 5.0), /* 5 steps */
124
    SCALE_GEN(4.0 / 9.0), /* 9 steps */
125
};
126

    
127
static DECLARE_ALIGNED_16(MPA_INT, window[512]);
128

    
129
/* layer 1 unscaling */
130
/* n = number of bits of the mantissa minus 1 */
131
static inline int l1_unscale(int n, int mant, int scale_factor)
132
{
133
    int shift, mod;
134
    int64_t val;
135

    
136
    shift = scale_factor_modshift[scale_factor];
137
    mod = shift & 3;
138
    shift >>= 2;
139
    val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);
140
    shift += n;
141
    /* NOTE: at this point, 1 <= shift >= 21 + 15 */
142
    return (int)((val + (1LL << (shift - 1))) >> shift);
143
}
144

    
145
static inline int l2_unscale_group(int steps, int mant, int scale_factor)
146
{
147
    int shift, mod, val;
148

    
149
    shift = scale_factor_modshift[scale_factor];
150
    mod = shift & 3;
151
    shift >>= 2;
152

    
153
    val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
154
    /* NOTE: at this point, 0 <= shift <= 21 */
155
    if (shift > 0)
156
        val = (val + (1 << (shift - 1))) >> shift;
157
    return val;
158
}
159

    
160
/* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
161
static inline int l3_unscale(int value, int exponent)
162
{
163
    unsigned int m;
164
    int e;
165

    
166
    e = table_4_3_exp  [4*value + (exponent&3)];
167
    m = table_4_3_value[4*value + (exponent&3)];
168
    e -= (exponent >> 2);
169
    assert(e>=1);
170
    if (e > 31)
171
        return 0;
172
    m = (m + (1 << (e-1))) >> e;
173

    
174
    return m;
175
}
176

    
177
/* all integer n^(4/3) computation code */
178
#define DEV_ORDER 13
179

    
180
#define POW_FRAC_BITS 24
181
#define POW_FRAC_ONE    (1 << POW_FRAC_BITS)
182
#define POW_FIX(a)   ((int)((a) * POW_FRAC_ONE))
183
#define POW_MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> POW_FRAC_BITS)
184

    
185
static int dev_4_3_coefs[DEV_ORDER];
186

    
187
#if 0 /* unused */
188
static int pow_mult3[3] = {
189
    POW_FIX(1.0),
190
    POW_FIX(1.25992104989487316476),
191
    POW_FIX(1.58740105196819947474),
192
};
193
#endif
194

    
195
static void int_pow_init(void)
196
{
197
    int i, a;
198

    
199
    a = POW_FIX(1.0);
200
    for(i=0;i<DEV_ORDER;i++) {
201
        a = POW_MULL(a, POW_FIX(4.0 / 3.0) - i * POW_FIX(1.0)) / (i + 1);
202
        dev_4_3_coefs[i] = a;
203
    }
204
}
205

    
206
#if 0 /* unused, remove? */
207
/* return the mantissa and the binary exponent */
208
static int int_pow(int i, int *exp_ptr)
209
{
210
    int e, er, eq, j;
211
    int a, a1;
212

213
    /* renormalize */
214
    a = i;
215
    e = POW_FRAC_BITS;
216
    while (a < (1 << (POW_FRAC_BITS - 1))) {
217
        a = a << 1;
218
        e--;
219
    }
220
    a -= (1 << POW_FRAC_BITS);
221
    a1 = 0;
222
    for(j = DEV_ORDER - 1; j >= 0; j--)
223
        a1 = POW_MULL(a, dev_4_3_coefs[j] + a1);
224
    a = (1 << POW_FRAC_BITS) + a1;
225
    /* exponent compute (exact) */
226
    e = e * 4;
227
    er = e % 3;
228
    eq = e / 3;
229
    a = POW_MULL(a, pow_mult3[er]);
230
    while (a >= 2 * POW_FRAC_ONE) {
231
        a = a >> 1;
232
        eq++;
233
    }
234
    /* convert to float */
235
    while (a < POW_FRAC_ONE) {
236
        a = a << 1;
237
        eq--;
238
    }
239
    /* now POW_FRAC_ONE <= a < 2 * POW_FRAC_ONE */
240
#if POW_FRAC_BITS > FRAC_BITS
241
    a = (a + (1 << (POW_FRAC_BITS - FRAC_BITS - 1))) >> (POW_FRAC_BITS - FRAC_BITS);
242
    /* correct overflow */
243
    if (a >= 2 * (1 << FRAC_BITS)) {
244
        a = a >> 1;
245
        eq++;
246
    }
247
#endif
248
    *exp_ptr = eq;
249
    return a;
250
}
251
#endif
252

    
253
static int decode_init(AVCodecContext * avctx)
254
{
255
    MPADecodeContext *s = avctx->priv_data;
256
    static int init=0;
257
    int i, j, k;
258

    
259
    s->avctx = avctx;
260

    
261
#if defined(USE_HIGHPRECISION) && defined(CONFIG_AUDIO_NONSHORT)
262
    avctx->sample_fmt= SAMPLE_FMT_S32;
263
#else
264
    avctx->sample_fmt= SAMPLE_FMT_S16;
265
#endif
266
    s->error_resilience= avctx->error_resilience;
267

    
268
    if(avctx->antialias_algo != FF_AA_FLOAT)
269
        s->compute_antialias= compute_antialias_integer;
270
    else
271
        s->compute_antialias= compute_antialias_float;
272

    
273
    if (!init && !avctx->parse_only) {
274
        /* scale factors table for layer 1/2 */
275
        for(i=0;i<64;i++) {
276
            int shift, mod;
277
            /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */
278
            shift = (i / 3);
279
            mod = i % 3;
280
            scale_factor_modshift[i] = mod | (shift << 2);
281
        }
282

    
283
        /* scale factor multiply for layer 1 */
284
        for(i=0;i<15;i++) {
285
            int n, norm;
286
            n = i + 2;
287
            norm = ((INT64_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
288
            scale_factor_mult[i][0] = MULL(FIXR(1.0 * 2.0), norm);
289
            scale_factor_mult[i][1] = MULL(FIXR(0.7937005259 * 2.0), norm);
290
            scale_factor_mult[i][2] = MULL(FIXR(0.6299605249 * 2.0), norm);
291
            dprintf(avctx, "%d: norm=%x s=%x %x %x\n",
292
                    i, norm,
293
                    scale_factor_mult[i][0],
294
                    scale_factor_mult[i][1],
295
                    scale_factor_mult[i][2]);
296
        }
297

    
298
        ff_mpa_synth_init(window);
299

    
300
        /* huffman decode tables */
301
        for(i=1;i<16;i++) {
302
            const HuffTable *h = &mpa_huff_tables[i];
303
            int xsize, x, y;
304
            unsigned int n;
305
            uint8_t  tmp_bits [512];
306
            uint16_t tmp_codes[512];
307

    
308
            memset(tmp_bits , 0, sizeof(tmp_bits ));
309
            memset(tmp_codes, 0, sizeof(tmp_codes));
310

    
311
            xsize = h->xsize;
312
            n = xsize * xsize;
313

    
314
            j = 0;
315
            for(x=0;x<xsize;x++) {
316
                for(y=0;y<xsize;y++){
317
                    tmp_bits [(x << 5) | y | ((x&&y)<<4)]= h->bits [j  ];
318
                    tmp_codes[(x << 5) | y | ((x&&y)<<4)]= h->codes[j++];
319
                }
320
            }
321

    
322
            /* XXX: fail test */
323
            init_vlc(&huff_vlc[i], 7, 512,
324
                     tmp_bits, 1, 1, tmp_codes, 2, 2, 1);
325
        }
326
        for(i=0;i<2;i++) {
327
            init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
328
                     mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1, 1);
329
        }
330

    
331
        for(i=0;i<9;i++) {
332
            k = 0;
333
            for(j=0;j<22;j++) {
334
                band_index_long[i][j] = k;
335
                k += band_size_long[i][j];
336
            }
337
            band_index_long[i][22] = k;
338
        }
339

    
340
        /* compute n ^ (4/3) and store it in mantissa/exp format */
341

    
342
        int_pow_init();
343
        for(i=1;i<TABLE_4_3_SIZE;i++) {
344
            double f, fm;
345
            int e, m;
346
            f = pow((double)(i/4), 4.0 / 3.0) * pow(2, (i&3)*0.25);
347
            fm = frexp(f, &e);
348
            m = (uint32_t)(fm*(1LL<<31) + 0.5);
349
            e+= FRAC_BITS - 31 + 5 - 100;
350

    
351
            /* normalized to FRAC_BITS */
352
            table_4_3_value[i] = m;
353
//            av_log(NULL, AV_LOG_DEBUG, "%d %d %f\n", i, m, pow((double)i, 4.0 / 3.0));
354
            table_4_3_exp[i] = -e;
355
        }
356
        for(i=0; i<512*16; i++){
357
            int exponent= (i>>4);
358
            double f= pow(i&15, 4.0 / 3.0) * pow(2, (exponent-400)*0.25 + FRAC_BITS + 5);
359
            expval_table[exponent][i&15]= llrint(f);
360
            if((i&15)==1)
361
                exp_table[exponent]= llrint(f);
362
        }
363

    
364
        for(i=0;i<7;i++) {
365
            float f;
366
            int v;
367
            if (i != 6) {
368
                f = tan((double)i * M_PI / 12.0);
369
                v = FIXR(f / (1.0 + f));
370
            } else {
371
                v = FIXR(1.0);
372
            }
373
            is_table[0][i] = v;
374
            is_table[1][6 - i] = v;
375
        }
376
        /* invalid values */
377
        for(i=7;i<16;i++)
378
            is_table[0][i] = is_table[1][i] = 0.0;
379

    
380
        for(i=0;i<16;i++) {
381
            double f;
382
            int e, k;
383

    
384
            for(j=0;j<2;j++) {
385
                e = -(j + 1) * ((i + 1) >> 1);
386
                f = pow(2.0, e / 4.0);
387
                k = i & 1;
388
                is_table_lsf[j][k ^ 1][i] = FIXR(f);
389
                is_table_lsf[j][k][i] = FIXR(1.0);
390
                dprintf(avctx, "is_table_lsf %d %d: %x %x\n",
391
                        i, j, is_table_lsf[j][0][i], is_table_lsf[j][1][i]);
392
            }
393
        }
394

    
395
        for(i=0;i<8;i++) {
396
            float ci, cs, ca;
397
            ci = ci_table[i];
398
            cs = 1.0 / sqrt(1.0 + ci * ci);
399
            ca = cs * ci;
400
            csa_table[i][0] = FIXHR(cs/4);
401
            csa_table[i][1] = FIXHR(ca/4);
402
            csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);
403
            csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4);
404
            csa_table_float[i][0] = cs;
405
            csa_table_float[i][1] = ca;
406
            csa_table_float[i][2] = ca + cs;
407
            csa_table_float[i][3] = ca - cs;
408
//            printf("%d %d %d %d\n", FIX(cs), FIX(cs-1), FIX(ca), FIX(cs)-FIX(ca));
409
//            av_log(NULL, AV_LOG_DEBUG,"%f %f %f %f\n", cs, ca, ca+cs, ca-cs);
410
        }
411

    
412
        /* compute mdct windows */
413
        for(i=0;i<36;i++) {
414
            for(j=0; j<4; j++){
415
                double d;
416

    
417
                if(j==2 && i%3 != 1)
418
                    continue;
419

    
420
                d= sin(M_PI * (i + 0.5) / 36.0);
421
                if(j==1){
422
                    if     (i>=30) d= 0;
423
                    else if(i>=24) d= sin(M_PI * (i - 18 + 0.5) / 12.0);
424
                    else if(i>=18) d= 1;
425
                }else if(j==3){
426
                    if     (i<  6) d= 0;
427
                    else if(i< 12) d= sin(M_PI * (i -  6 + 0.5) / 12.0);
428
                    else if(i< 18) d= 1;
429
                }
430
                //merge last stage of imdct into the window coefficients
431
                d*= 0.5 / cos(M_PI*(2*i + 19)/72);
432

    
433
                if(j==2)
434
                    mdct_win[j][i/3] = FIXHR((d / (1<<5)));
435
                else
436
                    mdct_win[j][i  ] = FIXHR((d / (1<<5)));
437
//                av_log(NULL, AV_LOG_DEBUG, "%2d %d %f\n", i,j,d / (1<<5));
438
            }
439
        }
440

    
441
        /* NOTE: we do frequency inversion adter the MDCT by changing
442
           the sign of the right window coefs */
443
        for(j=0;j<4;j++) {
444
            for(i=0;i<36;i+=2) {
445
                mdct_win[j + 4][i] = mdct_win[j][i];
446
                mdct_win[j + 4][i + 1] = -mdct_win[j][i + 1];
447
            }
448
        }
449

    
450
#if defined(DEBUG)
451
        for(j=0;j<8;j++) {
452
            av_log(avctx, AV_LOG_DEBUG, "win%d=\n", j);
453
            for(i=0;i<36;i++)
454
                av_log(avctx, AV_LOG_DEBUG, "%f, ", (double)mdct_win[j][i] / FRAC_ONE);
455
            av_log(avctx, AV_LOG_DEBUG, "\n");
456
        }
457
#endif
458
        init = 1;
459
    }
460

    
461
#ifdef DEBUG
462
    s->frame_count = 0;
463
#endif
464
    if (avctx->codec_id == CODEC_ID_MP3ADU)
465
        s->adu_mode = 1;
466
    return 0;
467
}
468

    
469
/* tab[i][j] = 1.0 / (2.0 * cos(pi*(2*k+1) / 2^(6 - j))) */
470

    
471
/* cos(i*pi/64) */
472

    
473
#define COS0_0  FIXHR(0.50060299823519630134/2)
474
#define COS0_1  FIXHR(0.50547095989754365998/2)
475
#define COS0_2  FIXHR(0.51544730992262454697/2)
476
#define COS0_3  FIXHR(0.53104259108978417447/2)
477
#define COS0_4  FIXHR(0.55310389603444452782/2)
478
#define COS0_5  FIXHR(0.58293496820613387367/2)
479
#define COS0_6  FIXHR(0.62250412303566481615/2)
480
#define COS0_7  FIXHR(0.67480834145500574602/2)
481
#define COS0_8  FIXHR(0.74453627100229844977/2)
482
#define COS0_9  FIXHR(0.83934964541552703873/2)
483
#define COS0_10 FIXHR(0.97256823786196069369/2)
484
#define COS0_11 FIXHR(1.16943993343288495515/4)
485
#define COS0_12 FIXHR(1.48416461631416627724/4)
486
#define COS0_13 FIXHR(2.05778100995341155085/8)
487
#define COS0_14 FIXHR(3.40760841846871878570/8)
488
#define COS0_15 FIXHR(10.19000812354805681150/32)
489

    
490
#define COS1_0 FIXHR(0.50241928618815570551/2)
491
#define COS1_1 FIXHR(0.52249861493968888062/2)
492
#define COS1_2 FIXHR(0.56694403481635770368/2)
493
#define COS1_3 FIXHR(0.64682178335999012954/2)
494
#define COS1_4 FIXHR(0.78815462345125022473/2)
495
#define COS1_5 FIXHR(1.06067768599034747134/4)
496
#define COS1_6 FIXHR(1.72244709823833392782/4)
497
#define COS1_7 FIXHR(5.10114861868916385802/16)
498

    
499
#define COS2_0 FIXHR(0.50979557910415916894/2)
500
#define COS2_1 FIXHR(0.60134488693504528054/2)
501
#define COS2_2 FIXHR(0.89997622313641570463/2)
502
#define COS2_3 FIXHR(2.56291544774150617881/8)
503

    
504
#define COS3_0 FIXHR(0.54119610014619698439/2)
505
#define COS3_1 FIXHR(1.30656296487637652785/4)
506

    
507
#define COS4_0 FIXHR(0.70710678118654752439/2)
508

    
509
/* butterfly operator */
510
#define BF(a, b, c, s)\
511
{\
512
    tmp0 = tab[a] + tab[b];\
513
    tmp1 = tab[a] - tab[b];\
514
    tab[a] = tmp0;\
515
    tab[b] = MULH(tmp1<<(s), c);\
516
}
517

    
518
#define BF1(a, b, c, d)\
519
{\
520
    BF(a, b, COS4_0, 1);\
521
    BF(c, d,-COS4_0, 1);\
522
    tab[c] += tab[d];\
523
}
524

    
525
#define BF2(a, b, c, d)\
526
{\
527
    BF(a, b, COS4_0, 1);\
528
    BF(c, d,-COS4_0, 1);\
529
    tab[c] += tab[d];\
530
    tab[a] += tab[c];\
531
    tab[c] += tab[b];\
532
    tab[b] += tab[d];\
533
}
534

    
535
#define ADD(a, b) tab[a] += tab[b]
536

    
537
/* DCT32 without 1/sqrt(2) coef zero scaling. */
538
static void dct32(int32_t *out, int32_t *tab)
539
{
540
    int tmp0, tmp1;
541

    
542
    /* pass 1 */
543
    BF( 0, 31, COS0_0 , 1);
544
    BF(15, 16, COS0_15, 5);
545
    /* pass 2 */
546
    BF( 0, 15, COS1_0 , 1);
547
    BF(16, 31,-COS1_0 , 1);
548
    /* pass 1 */
549
    BF( 7, 24, COS0_7 , 1);
550
    BF( 8, 23, COS0_8 , 1);
551
    /* pass 2 */
552
    BF( 7,  8, COS1_7 , 4);
553
    BF(23, 24,-COS1_7 , 4);
554
    /* pass 3 */
555
    BF( 0,  7, COS2_0 , 1);
556
    BF( 8, 15,-COS2_0 , 1);
557
    BF(16, 23, COS2_0 , 1);
558
    BF(24, 31,-COS2_0 , 1);
559
    /* pass 1 */
560
    BF( 3, 28, COS0_3 , 1);
561
    BF(12, 19, COS0_12, 2);
562
    /* pass 2 */
563
    BF( 3, 12, COS1_3 , 1);
564
    BF(19, 28,-COS1_3 , 1);
565
    /* pass 1 */
566
    BF( 4, 27, COS0_4 , 1);
567
    BF(11, 20, COS0_11, 2);
568
    /* pass 2 */
569
    BF( 4, 11, COS1_4 , 1);
570
    BF(20, 27,-COS1_4 , 1);
571
    /* pass 3 */
572
    BF( 3,  4, COS2_3 , 3);
573
    BF(11, 12,-COS2_3 , 3);
574
    BF(19, 20, COS2_3 , 3);
575
    BF(27, 28,-COS2_3 , 3);
576
    /* pass 4 */
577
    BF( 0,  3, COS3_0 , 1);
578
    BF( 4,  7,-COS3_0 , 1);
579
    BF( 8, 11, COS3_0 , 1);
580
    BF(12, 15,-COS3_0 , 1);
581
    BF(16, 19, COS3_0 , 1);
582
    BF(20, 23,-COS3_0 , 1);
583
    BF(24, 27, COS3_0 , 1);
584
    BF(28, 31,-COS3_0 , 1);
585

    
586

    
587

    
588
    /* pass 1 */
589
    BF( 1, 30, COS0_1 , 1);
590
    BF(14, 17, COS0_14, 3);
591
    /* pass 2 */
592
    BF( 1, 14, COS1_1 , 1);
593
    BF(17, 30,-COS1_1 , 1);
594
    /* pass 1 */
595
    BF( 6, 25, COS0_6 , 1);
596
    BF( 9, 22, COS0_9 , 1);
597
    /* pass 2 */
598
    BF( 6,  9, COS1_6 , 2);
599
    BF(22, 25,-COS1_6 , 2);
600
    /* pass 3 */
601
    BF( 1,  6, COS2_1 , 1);
602
    BF( 9, 14,-COS2_1 , 1);
603
    BF(17, 22, COS2_1 , 1);
604
    BF(25, 30,-COS2_1 , 1);
605

    
606
    /* pass 1 */
607
    BF( 2, 29, COS0_2 , 1);
608
    BF(13, 18, COS0_13, 3);
609
    /* pass 2 */
610
    BF( 2, 13, COS1_2 , 1);
611
    BF(18, 29,-COS1_2 , 1);
612
    /* pass 1 */
613
    BF( 5, 26, COS0_5 , 1);
614
    BF(10, 21, COS0_10, 1);
615
    /* pass 2 */
616
    BF( 5, 10, COS1_5 , 2);
617
    BF(21, 26,-COS1_5 , 2);
618
    /* pass 3 */
619
    BF( 2,  5, COS2_2 , 1);
620
    BF(10, 13,-COS2_2 , 1);
621
    BF(18, 21, COS2_2 , 1);
622
    BF(26, 29,-COS2_2 , 1);
623
    /* pass 4 */
624
    BF( 1,  2, COS3_1 , 2);
625
    BF( 5,  6,-COS3_1 , 2);
626
    BF( 9, 10, COS3_1 , 2);
627
    BF(13, 14,-COS3_1 , 2);
628
    BF(17, 18, COS3_1 , 2);
629
    BF(21, 22,-COS3_1 , 2);
630
    BF(25, 26, COS3_1 , 2);
631
    BF(29, 30,-COS3_1 , 2);
632

    
633
    /* pass 5 */
634
    BF1( 0,  1,  2,  3);
635
    BF2( 4,  5,  6,  7);
636
    BF1( 8,  9, 10, 11);
637
    BF2(12, 13, 14, 15);
638
    BF1(16, 17, 18, 19);
639
    BF2(20, 21, 22, 23);
640
    BF1(24, 25, 26, 27);
641
    BF2(28, 29, 30, 31);
642

    
643
    /* pass 6 */
644

    
645
    ADD( 8, 12);
646
    ADD(12, 10);
647
    ADD(10, 14);
648
    ADD(14,  9);
649
    ADD( 9, 13);
650
    ADD(13, 11);
651
    ADD(11, 15);
652

    
653
    out[ 0] = tab[0];
654
    out[16] = tab[1];
655
    out[ 8] = tab[2];
656
    out[24] = tab[3];
657
    out[ 4] = tab[4];
658
    out[20] = tab[5];
659
    out[12] = tab[6];
660
    out[28] = tab[7];
661
    out[ 2] = tab[8];
662
    out[18] = tab[9];
663
    out[10] = tab[10];
664
    out[26] = tab[11];
665
    out[ 6] = tab[12];
666
    out[22] = tab[13];
667
    out[14] = tab[14];
668
    out[30] = tab[15];
669

    
670
    ADD(24, 28);
671
    ADD(28, 26);
672
    ADD(26, 30);
673
    ADD(30, 25);
674
    ADD(25, 29);
675
    ADD(29, 27);
676
    ADD(27, 31);
677

    
678
    out[ 1] = tab[16] + tab[24];
679
    out[17] = tab[17] + tab[25];
680
    out[ 9] = tab[18] + tab[26];
681
    out[25] = tab[19] + tab[27];
682
    out[ 5] = tab[20] + tab[28];
683
    out[21] = tab[21] + tab[29];
684
    out[13] = tab[22] + tab[30];
685
    out[29] = tab[23] + tab[31];
686
    out[ 3] = tab[24] + tab[20];
687
    out[19] = tab[25] + tab[21];
688
    out[11] = tab[26] + tab[22];
689
    out[27] = tab[27] + tab[23];
690
    out[ 7] = tab[28] + tab[18];
691
    out[23] = tab[29] + tab[19];
692
    out[15] = tab[30] + tab[17];
693
    out[31] = tab[31];
694
}
695

    
696
#if FRAC_BITS <= 15
697

    
698
static inline int round_sample(int *sum)
699
{
700
    int sum1;
701
    sum1 = (*sum) >> OUT_SHIFT;
702
    *sum &= (1<<OUT_SHIFT)-1;
703
    if (sum1 < OUT_MIN)
704
        sum1 = OUT_MIN;
705
    else if (sum1 > OUT_MAX)
706
        sum1 = OUT_MAX;
707
    return sum1;
708
}
709

    
710
/* signed 16x16 -> 32 multiply add accumulate */
711
#define MACS(rt, ra, rb) MAC16(rt, ra, rb)
712

    
713
/* signed 16x16 -> 32 multiply */
714
#define MULS(ra, rb) MUL16(ra, rb)
715

    
716
#else
717

    
718
static inline int round_sample(int64_t *sum)
719
{
720
    int sum1;
721
    sum1 = (int)((*sum) >> OUT_SHIFT);
722
    *sum &= (1<<OUT_SHIFT)-1;
723
    if (sum1 < OUT_MIN)
724
        sum1 = OUT_MIN;
725
    else if (sum1 > OUT_MAX)
726
        sum1 = OUT_MAX;
727
    return sum1;
728
}
729

    
730
#   define MULS(ra, rb) MUL64(ra, rb)
731
#endif
732

    
733
#define SUM8(sum, op, w, p) \
734
{                                               \
735
    sum op MULS((w)[0 * 64], p[0 * 64]);\
736
    sum op MULS((w)[1 * 64], p[1 * 64]);\
737
    sum op MULS((w)[2 * 64], p[2 * 64]);\
738
    sum op MULS((w)[3 * 64], p[3 * 64]);\
739
    sum op MULS((w)[4 * 64], p[4 * 64]);\
740
    sum op MULS((w)[5 * 64], p[5 * 64]);\
741
    sum op MULS((w)[6 * 64], p[6 * 64]);\
742
    sum op MULS((w)[7 * 64], p[7 * 64]);\
743
}
744

    
745
#define SUM8P2(sum1, op1, sum2, op2, w1, w2, p) \
746
{                                               \
747
    int tmp;\
748
    tmp = p[0 * 64];\
749
    sum1 op1 MULS((w1)[0 * 64], tmp);\
750
    sum2 op2 MULS((w2)[0 * 64], tmp);\
751
    tmp = p[1 * 64];\
752
    sum1 op1 MULS((w1)[1 * 64], tmp);\
753
    sum2 op2 MULS((w2)[1 * 64], tmp);\
754
    tmp = p[2 * 64];\
755
    sum1 op1 MULS((w1)[2 * 64], tmp);\
756
    sum2 op2 MULS((w2)[2 * 64], tmp);\
757
    tmp = p[3 * 64];\
758
    sum1 op1 MULS((w1)[3 * 64], tmp);\
759
    sum2 op2 MULS((w2)[3 * 64], tmp);\
760
    tmp = p[4 * 64];\
761
    sum1 op1 MULS((w1)[4 * 64], tmp);\
762
    sum2 op2 MULS((w2)[4 * 64], tmp);\
763
    tmp = p[5 * 64];\
764
    sum1 op1 MULS((w1)[5 * 64], tmp);\
765
    sum2 op2 MULS((w2)[5 * 64], tmp);\
766
    tmp = p[6 * 64];\
767
    sum1 op1 MULS((w1)[6 * 64], tmp);\
768
    sum2 op2 MULS((w2)[6 * 64], tmp);\
769
    tmp = p[7 * 64];\
770
    sum1 op1 MULS((w1)[7 * 64], tmp);\
771
    sum2 op2 MULS((w2)[7 * 64], tmp);\
772
}
773

    
774
void ff_mpa_synth_init(MPA_INT *window)
775
{
776
    int i;
777

    
778
    /* max = 18760, max sum over all 16 coefs : 44736 */
779
    for(i=0;i<257;i++) {
780
        int v;
781
        v = ff_mpa_enwindow[i];
782
#if WFRAC_BITS < 16
783
        v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS);
784
#endif
785
        window[i] = v;
786
        if ((i & 63) != 0)
787
            v = -v;
788
        if (i != 0)
789
            window[512 - i] = v;
790
    }
791
}
792

    
793
/* 32 sub band synthesis filter. Input: 32 sub band samples, Output:
794
   32 samples. */
795
/* XXX: optimize by avoiding ring buffer usage */
796
void ff_mpa_synth_filter(MPA_INT *synth_buf_ptr, int *synth_buf_offset,
797
                         MPA_INT *window, int *dither_state,
798
                         OUT_INT *samples, int incr,
799
                         int32_t sb_samples[SBLIMIT])
800
{
801
    int32_t tmp[32];
802
    register MPA_INT *synth_buf;
803
    register const MPA_INT *w, *w2, *p;
804
    int j, offset, v;
805
    OUT_INT *samples2;
806
#if FRAC_BITS <= 15
807
    int sum, sum2;
808
#else
809
    int64_t sum, sum2;
810
#endif
811

    
812
    dct32(tmp, sb_samples);
813

    
814
    offset = *synth_buf_offset;
815
    synth_buf = synth_buf_ptr + offset;
816

    
817
    for(j=0;j<32;j++) {
818
        v = tmp[j];
819
#if FRAC_BITS <= 15
820
        /* NOTE: can cause a loss in precision if very high amplitude
821
           sound */
822
        v = av_clip_int16(v);
823
#endif
824
        synth_buf[j] = v;
825
    }
826
    /* copy to avoid wrap */
827
    memcpy(synth_buf + 512, synth_buf, 32 * sizeof(MPA_INT));
828

    
829
    samples2 = samples + 31 * incr;
830
    w = window;
831
    w2 = window + 31;
832

    
833
    sum = *dither_state;
834
    p = synth_buf + 16;
835
    SUM8(sum, +=, w, p);
836
    p = synth_buf + 48;
837
    SUM8(sum, -=, w + 32, p);
838
    *samples = round_sample(&sum);
839
    samples += incr;
840
    w++;
841

    
842
    /* we calculate two samples at the same time to avoid one memory
843
       access per two sample */
844
    for(j=1;j<16;j++) {
845
        sum2 = 0;
846
        p = synth_buf + 16 + j;
847
        SUM8P2(sum, +=, sum2, -=, w, w2, p);
848
        p = synth_buf + 48 - j;
849
        SUM8P2(sum, -=, sum2, -=, w + 32, w2 + 32, p);
850

    
851
        *samples = round_sample(&sum);
852
        samples += incr;
853
        sum += sum2;
854
        *samples2 = round_sample(&sum);
855
        samples2 -= incr;
856
        w++;
857
        w2--;
858
    }
859

    
860
    p = synth_buf + 32;
861
    SUM8(sum, -=, w + 32, p);
862
    *samples = round_sample(&sum);
863
    *dither_state= sum;
864

    
865
    offset = (offset - 32) & 511;
866
    *synth_buf_offset = offset;
867
}
868

    
869
#define C3 FIXHR(0.86602540378443864676/2)
870

    
871
/* 0.5 / cos(pi*(2*i+1)/36) */
872
static const int icos36[9] = {
873
    FIXR(0.50190991877167369479),
874
    FIXR(0.51763809020504152469), //0
875
    FIXR(0.55168895948124587824),
876
    FIXR(0.61038729438072803416),
877
    FIXR(0.70710678118654752439), //1
878
    FIXR(0.87172339781054900991),
879
    FIXR(1.18310079157624925896),
880
    FIXR(1.93185165257813657349), //2
881
    FIXR(5.73685662283492756461),
882
};
883

    
884
/* 0.5 / cos(pi*(2*i+1)/36) */
885
static const int icos36h[9] = {
886
    FIXHR(0.50190991877167369479/2),
887
    FIXHR(0.51763809020504152469/2), //0
888
    FIXHR(0.55168895948124587824/2),
889
    FIXHR(0.61038729438072803416/2),
890
    FIXHR(0.70710678118654752439/2), //1
891
    FIXHR(0.87172339781054900991/2),
892
    FIXHR(1.18310079157624925896/4),
893
    FIXHR(1.93185165257813657349/4), //2
894
//    FIXHR(5.73685662283492756461),
895
};
896

    
897
/* 12 points IMDCT. We compute it "by hand" by factorizing obvious
898
   cases. */
899
static void imdct12(int *out, int *in)
900
{
901
    int in0, in1, in2, in3, in4, in5, t1, t2;
902

    
903
    in0= in[0*3];
904
    in1= in[1*3] + in[0*3];
905
    in2= in[2*3] + in[1*3];
906
    in3= in[3*3] + in[2*3];
907
    in4= in[4*3] + in[3*3];
908
    in5= in[5*3] + in[4*3];
909
    in5 += in3;
910
    in3 += in1;
911

    
912
    in2= MULH(2*in2, C3);
913
    in3= MULH(4*in3, C3);
914

    
915
    t1 = in0 - in4;
916
    t2 = MULH(2*(in1 - in5), icos36h[4]);
917

    
918
    out[ 7]=
919
    out[10]= t1 + t2;
920
    out[ 1]=
921
    out[ 4]= t1 - t2;
922

    
923
    in0 += in4>>1;
924
    in4 = in0 + in2;
925
    in5 += 2*in1;
926
    in1 = MULH(in5 + in3, icos36h[1]);
927
    out[ 8]=
928
    out[ 9]= in4 + in1;
929
    out[ 2]=
930
    out[ 3]= in4 - in1;
931

    
932
    in0 -= in2;
933
    in5 = MULH(2*(in5 - in3), icos36h[7]);
934
    out[ 0]=
935
    out[ 5]= in0 - in5;
936
    out[ 6]=
937
    out[11]= in0 + in5;
938
}
939

    
940
/* cos(pi*i/18) */
941
#define C1 FIXHR(0.98480775301220805936/2)
942
#define C2 FIXHR(0.93969262078590838405/2)
943
#define C3 FIXHR(0.86602540378443864676/2)
944
#define C4 FIXHR(0.76604444311897803520/2)
945
#define C5 FIXHR(0.64278760968653932632/2)
946
#define C6 FIXHR(0.5/2)
947
#define C7 FIXHR(0.34202014332566873304/2)
948
#define C8 FIXHR(0.17364817766693034885/2)
949

    
950

    
951
/* using Lee like decomposition followed by hand coded 9 points DCT */
952
static void imdct36(int *out, int *buf, int *in, int *win)
953
{
954
    int i, j, t0, t1, t2, t3, s0, s1, s2, s3;
955
    int tmp[18], *tmp1, *in1;
956

    
957
    for(i=17;i>=1;i--)
958
        in[i] += in[i-1];
959
    for(i=17;i>=3;i-=2)
960
        in[i] += in[i-2];
961

    
962
    for(j=0;j<2;j++) {
963
        tmp1 = tmp + j;
964
        in1 = in + j;
965
#if 0
966
//more accurate but slower
967
        int64_t t0, t1, t2, t3;
968
        t2 = in1[2*4] + in1[2*8] - in1[2*2];
969

970
        t3 = (in1[2*0] + (int64_t)(in1[2*6]>>1))<<32;
971
        t1 = in1[2*0] - in1[2*6];
972
        tmp1[ 6] = t1 - (t2>>1);
973
        tmp1[16] = t1 + t2;
974

975
        t0 = MUL64(2*(in1[2*2] + in1[2*4]),    C2);
976
        t1 = MUL64(   in1[2*4] - in1[2*8] , -2*C8);
977
        t2 = MUL64(2*(in1[2*2] + in1[2*8]),   -C4);
978

979
        tmp1[10] = (t3 - t0 - t2) >> 32;
980
        tmp1[ 2] = (t3 + t0 + t1) >> 32;
981
        tmp1[14] = (t3 + t2 - t1) >> 32;
982

983
        tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
984
        t2 = MUL64(2*(in1[2*1] + in1[2*5]),    C1);
985
        t3 = MUL64(   in1[2*5] - in1[2*7] , -2*C7);
986
        t0 = MUL64(2*in1[2*3], C3);
987

988
        t1 = MUL64(2*(in1[2*1] + in1[2*7]),   -C5);
989

990
        tmp1[ 0] = (t2 + t3 + t0) >> 32;
991
        tmp1[12] = (t2 + t1 - t0) >> 32;
992
        tmp1[ 8] = (t3 - t1 - t0) >> 32;
993
#else
994
        t2 = in1[2*4] + in1[2*8] - in1[2*2];
995

    
996
        t3 = in1[2*0] + (in1[2*6]>>1);
997
        t1 = in1[2*0] - in1[2*6];
998
        tmp1[ 6] = t1 - (t2>>1);
999
        tmp1[16] = t1 + t2;
1000

    
1001
        t0 = MULH(2*(in1[2*2] + in1[2*4]),    C2);
1002
        t1 = MULH(   in1[2*4] - in1[2*8] , -2*C8);
1003
        t2 = MULH(2*(in1[2*2] + in1[2*8]),   -C4);
1004

    
1005
        tmp1[10] = t3 - t0 - t2;
1006
        tmp1[ 2] = t3 + t0 + t1;
1007
        tmp1[14] = t3 + t2 - t1;
1008

    
1009
        tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
1010
        t2 = MULH(2*(in1[2*1] + in1[2*5]),    C1);
1011
        t3 = MULH(   in1[2*5] - in1[2*7] , -2*C7);
1012
        t0 = MULH(2*in1[2*3], C3);
1013

    
1014
        t1 = MULH(2*(in1[2*1] + in1[2*7]),   -C5);
1015

    
1016
        tmp1[ 0] = t2 + t3 + t0;
1017
        tmp1[12] = t2 + t1 - t0;
1018
        tmp1[ 8] = t3 - t1 - t0;
1019
#endif
1020
    }
1021

    
1022
    i = 0;
1023
    for(j=0;j<4;j++) {
1024
        t0 = tmp[i];
1025
        t1 = tmp[i + 2];
1026
        s0 = t1 + t0;
1027
        s2 = t1 - t0;
1028

    
1029
        t2 = tmp[i + 1];
1030
        t3 = tmp[i + 3];
1031
        s1 = MULH(2*(t3 + t2), icos36h[j]);
1032
        s3 = MULL(t3 - t2, icos36[8 - j]);
1033

    
1034
        t0 = s0 + s1;
1035
        t1 = s0 - s1;
1036
        out[(9 + j)*SBLIMIT] =  MULH(t1, win[9 + j]) + buf[9 + j];
1037
        out[(8 - j)*SBLIMIT] =  MULH(t1, win[8 - j]) + buf[8 - j];
1038
        buf[9 + j] = MULH(t0, win[18 + 9 + j]);
1039
        buf[8 - j] = MULH(t0, win[18 + 8 - j]);
1040

    
1041
        t0 = s2 + s3;
1042
        t1 = s2 - s3;
1043
        out[(9 + 8 - j)*SBLIMIT] =  MULH(t1, win[9 + 8 - j]) + buf[9 + 8 - j];
1044
        out[(        j)*SBLIMIT] =  MULH(t1, win[        j]) + buf[        j];
1045
        buf[9 + 8 - j] = MULH(t0, win[18 + 9 + 8 - j]);
1046
        buf[      + j] = MULH(t0, win[18         + j]);
1047
        i += 4;
1048
    }
1049

    
1050
    s0 = tmp[16];
1051
    s1 = MULH(2*tmp[17], icos36h[4]);
1052
    t0 = s0 + s1;
1053
    t1 = s0 - s1;
1054
    out[(9 + 4)*SBLIMIT] =  MULH(t1, win[9 + 4]) + buf[9 + 4];
1055
    out[(8 - 4)*SBLIMIT] =  MULH(t1, win[8 - 4]) + buf[8 - 4];
1056
    buf[9 + 4] = MULH(t0, win[18 + 9 + 4]);
1057
    buf[8 - 4] = MULH(t0, win[18 + 8 - 4]);
1058
}
1059

    
1060
/* return the number of decoded frames */
1061
static int mp_decode_layer1(MPADecodeContext *s)
1062
{
1063
    int bound, i, v, n, ch, j, mant;
1064
    uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
1065
    uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
1066

    
1067
    if (s->mode == MPA_JSTEREO)
1068
        bound = (s->mode_ext + 1) * 4;
1069
    else
1070
        bound = SBLIMIT;
1071

    
1072
    /* allocation bits */
1073
    for(i=0;i<bound;i++) {
1074
        for(ch=0;ch<s->nb_channels;ch++) {
1075
            allocation[ch][i] = get_bits(&s->gb, 4);
1076
        }
1077
    }
1078
    for(i=bound;i<SBLIMIT;i++) {
1079
        allocation[0][i] = get_bits(&s->gb, 4);
1080
    }
1081

    
1082
    /* scale factors */
1083
    for(i=0;i<bound;i++) {
1084
        for(ch=0;ch<s->nb_channels;ch++) {
1085
            if (allocation[ch][i])
1086
                scale_factors[ch][i] = get_bits(&s->gb, 6);
1087
        }
1088
    }
1089
    for(i=bound;i<SBLIMIT;i++) {
1090
        if (allocation[0][i]) {
1091
            scale_factors[0][i] = get_bits(&s->gb, 6);
1092
            scale_factors[1][i] = get_bits(&s->gb, 6);
1093
        }
1094
    }
1095

    
1096
    /* compute samples */
1097
    for(j=0;j<12;j++) {
1098
        for(i=0;i<bound;i++) {
1099
            for(ch=0;ch<s->nb_channels;ch++) {
1100
                n = allocation[ch][i];
1101
                if (n) {
1102
                    mant = get_bits(&s->gb, n + 1);
1103
                    v = l1_unscale(n, mant, scale_factors[ch][i]);
1104
                } else {
1105
                    v = 0;
1106
                }
1107
                s->sb_samples[ch][j][i] = v;
1108
            }
1109
        }
1110
        for(i=bound;i<SBLIMIT;i++) {
1111
            n = allocation[0][i];
1112
            if (n) {
1113
                mant = get_bits(&s->gb, n + 1);
1114
                v = l1_unscale(n, mant, scale_factors[0][i]);
1115
                s->sb_samples[0][j][i] = v;
1116
                v = l1_unscale(n, mant, scale_factors[1][i]);
1117
                s->sb_samples[1][j][i] = v;
1118
            } else {
1119
                s->sb_samples[0][j][i] = 0;
1120
                s->sb_samples[1][j][i] = 0;
1121
            }
1122
        }
1123
    }
1124
    return 12;
1125
}
1126

    
1127
static int mp_decode_layer2(MPADecodeContext *s)
1128
{
1129
    int sblimit; /* number of used subbands */
1130
    const unsigned char *alloc_table;
1131
    int table, bit_alloc_bits, i, j, ch, bound, v;
1132
    unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
1133
    unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
1134
    unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
1135
    int scale, qindex, bits, steps, k, l, m, b;
1136

    
1137
    /* select decoding table */
1138
    table = ff_mpa_l2_select_table(s->bit_rate / 1000, s->nb_channels,
1139
                            s->sample_rate, s->lsf);
1140
    sblimit = ff_mpa_sblimit_table[table];
1141
    alloc_table = ff_mpa_alloc_tables[table];
1142

    
1143
    if (s->mode == MPA_JSTEREO)
1144
        bound = (s->mode_ext + 1) * 4;
1145
    else
1146
        bound = sblimit;
1147

    
1148
    dprintf(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit);
1149

    
1150
    /* sanity check */
1151
    if( bound > sblimit ) bound = sblimit;
1152

    
1153
    /* parse bit allocation */
1154
    j = 0;
1155
    for(i=0;i<bound;i++) {
1156
        bit_alloc_bits = alloc_table[j];
1157
        for(ch=0;ch<s->nb_channels;ch++) {
1158
            bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
1159
        }
1160
        j += 1 << bit_alloc_bits;
1161
    }
1162
    for(i=bound;i<sblimit;i++) {
1163
        bit_alloc_bits = alloc_table[j];
1164
        v = get_bits(&s->gb, bit_alloc_bits);
1165
        bit_alloc[0][i] = v;
1166
        bit_alloc[1][i] = v;
1167
        j += 1 << bit_alloc_bits;
1168
    }
1169

    
1170
#ifdef DEBUG
1171
    {
1172
        for(ch=0;ch<s->nb_channels;ch++) {
1173
            for(i=0;i<sblimit;i++)
1174
                dprintf(s->avctx, " %d", bit_alloc[ch][i]);
1175
            dprintf(s->avctx, "\n");
1176
        }
1177
    }
1178
#endif
1179

    
1180
    /* scale codes */
1181
    for(i=0;i<sblimit;i++) {
1182
        for(ch=0;ch<s->nb_channels;ch++) {
1183
            if (bit_alloc[ch][i])
1184
                scale_code[ch][i] = get_bits(&s->gb, 2);
1185
        }
1186
    }
1187

    
1188
    /* scale factors */
1189
    for(i=0;i<sblimit;i++) {
1190
        for(ch=0;ch<s->nb_channels;ch++) {
1191
            if (bit_alloc[ch][i]) {
1192
                sf = scale_factors[ch][i];
1193
                switch(scale_code[ch][i]) {
1194
                default:
1195
                case 0:
1196
                    sf[0] = get_bits(&s->gb, 6);
1197
                    sf[1] = get_bits(&s->gb, 6);
1198
                    sf[2] = get_bits(&s->gb, 6);
1199
                    break;
1200
                case 2:
1201
                    sf[0] = get_bits(&s->gb, 6);
1202
                    sf[1] = sf[0];
1203
                    sf[2] = sf[0];
1204
                    break;
1205
                case 1:
1206
                    sf[0] = get_bits(&s->gb, 6);
1207
                    sf[2] = get_bits(&s->gb, 6);
1208
                    sf[1] = sf[0];
1209
                    break;
1210
                case 3:
1211
                    sf[0] = get_bits(&s->gb, 6);
1212
                    sf[2] = get_bits(&s->gb, 6);
1213
                    sf[1] = sf[2];
1214
                    break;
1215
                }
1216
            }
1217
        }
1218
    }
1219

    
1220
#ifdef DEBUG
1221
    for(ch=0;ch<s->nb_channels;ch++) {
1222
        for(i=0;i<sblimit;i++) {
1223
            if (bit_alloc[ch][i]) {
1224
                sf = scale_factors[ch][i];
1225
                dprintf(s->avctx, " %d %d %d", sf[0], sf[1], sf[2]);
1226
            } else {
1227
                dprintf(s->avctx, " -");
1228
            }
1229
        }
1230
        dprintf(s->avctx, "\n");
1231
    }
1232
#endif
1233

    
1234
    /* samples */
1235
    for(k=0;k<3;k++) {
1236
        for(l=0;l<12;l+=3) {
1237
            j = 0;
1238
            for(i=0;i<bound;i++) {
1239
                bit_alloc_bits = alloc_table[j];
1240
                for(ch=0;ch<s->nb_channels;ch++) {
1241
                    b = bit_alloc[ch][i];
1242
                    if (b) {
1243
                        scale = scale_factors[ch][i][k];
1244
                        qindex = alloc_table[j+b];
1245
                        bits = ff_mpa_quant_bits[qindex];
1246
                        if (bits < 0) {
1247
                            /* 3 values at the same time */
1248
                            v = get_bits(&s->gb, -bits);
1249
                            steps = ff_mpa_quant_steps[qindex];
1250
                            s->sb_samples[ch][k * 12 + l + 0][i] =
1251
                                l2_unscale_group(steps, v % steps, scale);
1252
                            v = v / steps;
1253
                            s->sb_samples[ch][k * 12 + l + 1][i] =
1254
                                l2_unscale_group(steps, v % steps, scale);
1255
                            v = v / steps;
1256
                            s->sb_samples[ch][k * 12 + l + 2][i] =
1257
                                l2_unscale_group(steps, v, scale);
1258
                        } else {
1259
                            for(m=0;m<3;m++) {
1260
                                v = get_bits(&s->gb, bits);
1261
                                v = l1_unscale(bits - 1, v, scale);
1262
                                s->sb_samples[ch][k * 12 + l + m][i] = v;
1263
                            }
1264
                        }
1265
                    } else {
1266
                        s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1267
                        s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1268
                        s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1269
                    }
1270
                }
1271
                /* next subband in alloc table */
1272
                j += 1 << bit_alloc_bits;
1273
            }
1274
            /* XXX: find a way to avoid this duplication of code */
1275
            for(i=bound;i<sblimit;i++) {
1276
                bit_alloc_bits = alloc_table[j];
1277
                b = bit_alloc[0][i];
1278
                if (b) {
1279
                    int mant, scale0, scale1;
1280
                    scale0 = scale_factors[0][i][k];
1281
                    scale1 = scale_factors[1][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
                        mant = v % steps;
1289
                        v = v / steps;
1290
                        s->sb_samples[0][k * 12 + l + 0][i] =
1291
                            l2_unscale_group(steps, mant, scale0);
1292
                        s->sb_samples[1][k * 12 + l + 0][i] =
1293
                            l2_unscale_group(steps, mant, scale1);
1294
                        mant = v % steps;
1295
                        v = v / steps;
1296
                        s->sb_samples[0][k * 12 + l + 1][i] =
1297
                            l2_unscale_group(steps, mant, scale0);
1298
                        s->sb_samples[1][k * 12 + l + 1][i] =
1299
                            l2_unscale_group(steps, mant, scale1);
1300
                        s->sb_samples[0][k * 12 + l + 2][i] =
1301
                            l2_unscale_group(steps, v, scale0);
1302
                        s->sb_samples[1][k * 12 + l + 2][i] =
1303
                            l2_unscale_group(steps, v, scale1);
1304
                    } else {
1305
                        for(m=0;m<3;m++) {
1306
                            mant = get_bits(&s->gb, bits);
1307
                            s->sb_samples[0][k * 12 + l + m][i] =
1308
                                l1_unscale(bits - 1, mant, scale0);
1309
                            s->sb_samples[1][k * 12 + l + m][i] =
1310
                                l1_unscale(bits - 1, mant, scale1);
1311
                        }
1312
                    }
1313
                } else {
1314
                    s->sb_samples[0][k * 12 + l + 0][i] = 0;
1315
                    s->sb_samples[0][k * 12 + l + 1][i] = 0;
1316
                    s->sb_samples[0][k * 12 + l + 2][i] = 0;
1317
                    s->sb_samples[1][k * 12 + l + 0][i] = 0;
1318
                    s->sb_samples[1][k * 12 + l + 1][i] = 0;
1319
                    s->sb_samples[1][k * 12 + l + 2][i] = 0;
1320
                }
1321
                /* next subband in alloc table */
1322
                j += 1 << bit_alloc_bits;
1323
            }
1324
            /* fill remaining samples to zero */
1325
            for(i=sblimit;i<SBLIMIT;i++) {
1326
                for(ch=0;ch<s->nb_channels;ch++) {
1327
                    s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1328
                    s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1329
                    s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1330
                }
1331
            }
1332
        }
1333
    }
1334
    return 3 * 12;
1335
}
1336

    
1337
static inline void lsf_sf_expand(int *slen,
1338
                                 int sf, int n1, int n2, int n3)
1339
{
1340
    if (n3) {
1341
        slen[3] = sf % n3;
1342
        sf /= n3;
1343
    } else {
1344
        slen[3] = 0;
1345
    }
1346
    if (n2) {
1347
        slen[2] = sf % n2;
1348
        sf /= n2;
1349
    } else {
1350
        slen[2] = 0;
1351
    }
1352
    slen[1] = sf % n1;
1353
    sf /= n1;
1354
    slen[0] = sf;
1355
}
1356

    
1357
static void exponents_from_scale_factors(MPADecodeContext *s,
1358
                                         GranuleDef *g,
1359
                                         int16_t *exponents)
1360
{
1361
    const uint8_t *bstab, *pretab;
1362
    int len, i, j, k, l, v0, shift, gain, gains[3];
1363
    int16_t *exp_ptr;
1364

    
1365
    exp_ptr = exponents;
1366
    gain = g->global_gain - 210;
1367
    shift = g->scalefac_scale + 1;
1368

    
1369
    bstab = band_size_long[s->sample_rate_index];
1370
    pretab = mpa_pretab[g->preflag];
1371
    for(i=0;i<g->long_end;i++) {
1372
        v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400;
1373
        len = bstab[i];
1374
        for(j=len;j>0;j--)
1375
            *exp_ptr++ = v0;
1376
    }
1377

    
1378
    if (g->short_start < 13) {
1379
        bstab = band_size_short[s->sample_rate_index];
1380
        gains[0] = gain - (g->subblock_gain[0] << 3);
1381
        gains[1] = gain - (g->subblock_gain[1] << 3);
1382
        gains[2] = gain - (g->subblock_gain[2] << 3);
1383
        k = g->long_end;
1384
        for(i=g->short_start;i<13;i++) {
1385
            len = bstab[i];
1386
            for(l=0;l<3;l++) {
1387
                v0 = gains[l] - (g->scale_factors[k++] << shift) + 400;
1388
                for(j=len;j>0;j--)
1389
                *exp_ptr++ = v0;
1390
            }
1391
        }
1392
    }
1393
}
1394

    
1395
/* handle n = 0 too */
1396
static inline int get_bitsz(GetBitContext *s, int n)
1397
{
1398
    if (n == 0)
1399
        return 0;
1400
    else
1401
        return get_bits(s, n);
1402
}
1403

    
1404

    
1405
static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos, int *end_pos2){
1406
    if(s->in_gb.buffer && *pos >= s->gb.size_in_bits){
1407
        s->gb= s->in_gb;
1408
        s->in_gb.buffer=NULL;
1409
        assert((get_bits_count(&s->gb) & 7) == 0);
1410
        skip_bits_long(&s->gb, *pos - *end_pos);
1411
        *end_pos2=
1412
        *end_pos= *end_pos2 + get_bits_count(&s->gb) - *pos;
1413
        *pos= get_bits_count(&s->gb);
1414
    }
1415
}
1416

    
1417
static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
1418
                          int16_t *exponents, int end_pos2)
1419
{
1420
    int s_index;
1421
    int i;
1422
    int last_pos, bits_left;
1423
    VLC *vlc;
1424
    int end_pos= FFMIN(end_pos2, s->gb.size_in_bits);
1425

    
1426
    /* low frequencies (called big values) */
1427
    s_index = 0;
1428
    for(i=0;i<3;i++) {
1429
        int j, k, l, linbits;
1430
        j = g->region_size[i];
1431
        if (j == 0)
1432
            continue;
1433
        /* select vlc table */
1434
        k = g->table_select[i];
1435
        l = mpa_huff_data[k][0];
1436
        linbits = mpa_huff_data[k][1];
1437
        vlc = &huff_vlc[l];
1438

    
1439
        if(!l){
1440
            memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*2*j);
1441
            s_index += 2*j;
1442
            continue;
1443
        }
1444

    
1445
        /* read huffcode and compute each couple */
1446
        for(;j>0;j--) {
1447
            int exponent, x, y, v;
1448
            int pos= get_bits_count(&s->gb);
1449

    
1450
            if (pos >= end_pos){
1451
//                av_log(NULL, AV_LOG_ERROR, "pos: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);
1452
                switch_buffer(s, &pos, &end_pos, &end_pos2);
1453
//                av_log(NULL, AV_LOG_ERROR, "new pos: %d %d\n", pos, end_pos);
1454
                if(pos >= end_pos)
1455
                    break;
1456
            }
1457
            y = get_vlc2(&s->gb, vlc->table, 7, 3);
1458

    
1459
            if(!y){
1460
                g->sb_hybrid[s_index  ] =
1461
                g->sb_hybrid[s_index+1] = 0;
1462
                s_index += 2;
1463
                continue;
1464
            }
1465

    
1466
            exponent= exponents[s_index];
1467

    
1468
            dprintf(s->avctx, "region=%d n=%d x=%d y=%d exp=%d\n",
1469
                    i, g->region_size[i] - j, x, y, exponent);
1470
            if(y&16){
1471
                x = y >> 5;
1472
                y = y & 0x0f;
1473
                if (x < 15){
1474
                    v = expval_table[ exponent ][ x ];
1475
//                      v = expval_table[ (exponent&3) ][ x ] >> FFMIN(0 - (exponent>>2), 31);
1476
                }else{
1477
                    x += get_bitsz(&s->gb, linbits);
1478
                    v = l3_unscale(x, exponent);
1479
                }
1480
                if (get_bits1(&s->gb))
1481
                    v = -v;
1482
                g->sb_hybrid[s_index] = v;
1483
                if (y < 15){
1484
                    v = expval_table[ exponent ][ y ];
1485
                }else{
1486
                    y += get_bitsz(&s->gb, linbits);
1487
                    v = l3_unscale(y, exponent);
1488
                }
1489
                if (get_bits1(&s->gb))
1490
                    v = -v;
1491
                g->sb_hybrid[s_index+1] = v;
1492
            }else{
1493
                x = y >> 5;
1494
                y = y & 0x0f;
1495
                x += y;
1496
                if (x < 15){
1497
                    v = expval_table[ exponent ][ x ];
1498
                }else{
1499
                    x += get_bitsz(&s->gb, linbits);
1500
                    v = l3_unscale(x, exponent);
1501
                }
1502
                if (get_bits1(&s->gb))
1503
                    v = -v;
1504
                g->sb_hybrid[s_index+!!y] = v;
1505
                g->sb_hybrid[s_index+ !y] = 0;
1506
            }
1507
            s_index+=2;
1508
        }
1509
    }
1510

    
1511
    /* high frequencies */
1512
    vlc = &huff_quad_vlc[g->count1table_select];
1513
    last_pos=0;
1514
    while (s_index <= 572) {
1515
        int pos, code;
1516
        pos = get_bits_count(&s->gb);
1517
        if (pos >= end_pos) {
1518
            if (pos > end_pos2 && last_pos){
1519
                /* some encoders generate an incorrect size for this
1520
                   part. We must go back into the data */
1521
                s_index -= 4;
1522
                skip_bits_long(&s->gb, last_pos - pos);
1523
                av_log(NULL, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos);
1524
                if(s->error_resilience >= FF_ER_COMPLIANT)
1525
                    s_index=0;
1526
                break;
1527
            }
1528
//                av_log(NULL, AV_LOG_ERROR, "pos2: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);
1529
            switch_buffer(s, &pos, &end_pos, &end_pos2);
1530
//                av_log(NULL, AV_LOG_ERROR, "new pos2: %d %d %d\n", pos, end_pos, s_index);
1531
            if(pos >= end_pos)
1532
                break;
1533
        }
1534
        last_pos= pos;
1535

    
1536
        code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);
1537
        dprintf(s->avctx, "t=%d code=%d\n", g->count1table_select, code);
1538
        g->sb_hybrid[s_index+0]=
1539
        g->sb_hybrid[s_index+1]=
1540
        g->sb_hybrid[s_index+2]=
1541
        g->sb_hybrid[s_index+3]= 0;
1542
        while(code){
1543
            static const int idxtab[16]={3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0};
1544
            int v;
1545
            int pos= s_index+idxtab[code];
1546
            code ^= 8>>idxtab[code];
1547
            v = exp_table[ exponents[pos] ];
1548
//            v = exp_table[ (exponents[pos]&3) ] >> FFMIN(0 - (exponents[pos]>>2), 31);
1549
            if(get_bits1(&s->gb))
1550
                v = -v;
1551
            g->sb_hybrid[pos] = v;
1552
        }
1553
        s_index+=4;
1554
    }
1555
    /* skip extension bits */
1556
    bits_left = end_pos2 - get_bits_count(&s->gb);
1557
//av_log(NULL, AV_LOG_ERROR, "left:%d buf:%p\n", bits_left, s->in_gb.buffer);
1558
    if (bits_left < 0/* || bits_left > 500*/) {
1559
        av_log(NULL, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
1560
        s_index=0;
1561
    }else if(bits_left > 0 && s->error_resilience >= FF_ER_AGGRESSIVE){
1562
        av_log(NULL, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
1563
        s_index=0;
1564
    }
1565
    memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*(576 - s_index));
1566
    skip_bits_long(&s->gb, bits_left);
1567

    
1568
    i= get_bits_count(&s->gb);
1569
    switch_buffer(s, &i, &end_pos, &end_pos2);
1570

    
1571
    return 0;
1572
}
1573

    
1574
/* Reorder short blocks from bitstream order to interleaved order. It
1575
   would be faster to do it in parsing, but the code would be far more
1576
   complicated */
1577
static void reorder_block(MPADecodeContext *s, GranuleDef *g)
1578
{
1579
    int i, j, len;
1580
    int32_t *ptr, *dst, *ptr1;
1581
    int32_t tmp[576];
1582

    
1583
    if (g->block_type != 2)
1584
        return;
1585

    
1586
    if (g->switch_point) {
1587
        if (s->sample_rate_index != 8) {
1588
            ptr = g->sb_hybrid + 36;
1589
        } else {
1590
            ptr = g->sb_hybrid + 48;
1591
        }
1592
    } else {
1593
        ptr = g->sb_hybrid;
1594
    }
1595

    
1596
    for(i=g->short_start;i<13;i++) {
1597
        len = band_size_short[s->sample_rate_index][i];
1598
        ptr1 = ptr;
1599
        dst = tmp;
1600
        for(j=len;j>0;j--) {
1601
            *dst++ = ptr[0*len];
1602
            *dst++ = ptr[1*len];
1603
            *dst++ = ptr[2*len];
1604
            ptr++;
1605
        }
1606
        ptr+=2*len;
1607
        memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));
1608
    }
1609
}
1610

    
1611
#define ISQRT2 FIXR(0.70710678118654752440)
1612

    
1613
static void compute_stereo(MPADecodeContext *s,
1614
                           GranuleDef *g0, GranuleDef *g1)
1615
{
1616
    int i, j, k, l;
1617
    int32_t v1, v2;
1618
    int sf_max, tmp0, tmp1, sf, len, non_zero_found;
1619
    int32_t (*is_tab)[16];
1620
    int32_t *tab0, *tab1;
1621
    int non_zero_found_short[3];
1622

    
1623
    /* intensity stereo */
1624
    if (s->mode_ext & MODE_EXT_I_STEREO) {
1625
        if (!s->lsf) {
1626
            is_tab = is_table;
1627
            sf_max = 7;
1628
        } else {
1629
            is_tab = is_table_lsf[g1->scalefac_compress & 1];
1630
            sf_max = 16;
1631
        }
1632

    
1633
        tab0 = g0->sb_hybrid + 576;
1634
        tab1 = g1->sb_hybrid + 576;
1635

    
1636
        non_zero_found_short[0] = 0;
1637
        non_zero_found_short[1] = 0;
1638
        non_zero_found_short[2] = 0;
1639
        k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1640
        for(i = 12;i >= g1->short_start;i--) {
1641
            /* for last band, use previous scale factor */
1642
            if (i != 11)
1643
                k -= 3;
1644
            len = band_size_short[s->sample_rate_index][i];
1645
            for(l=2;l>=0;l--) {
1646
                tab0 -= len;
1647
                tab1 -= len;
1648
                if (!non_zero_found_short[l]) {
1649
                    /* test if non zero band. if so, stop doing i-stereo */
1650
                    for(j=0;j<len;j++) {
1651
                        if (tab1[j] != 0) {
1652
                            non_zero_found_short[l] = 1;
1653
                            goto found1;
1654
                        }
1655
                    }
1656
                    sf = g1->scale_factors[k + l];
1657
                    if (sf >= sf_max)
1658
                        goto found1;
1659

    
1660
                    v1 = is_tab[0][sf];
1661
                    v2 = is_tab[1][sf];
1662
                    for(j=0;j<len;j++) {
1663
                        tmp0 = tab0[j];
1664
                        tab0[j] = MULL(tmp0, v1);
1665
                        tab1[j] = MULL(tmp0, v2);
1666
                    }
1667
                } else {
1668
                found1:
1669
                    if (s->mode_ext & MODE_EXT_MS_STEREO) {
1670
                        /* lower part of the spectrum : do ms stereo
1671
                           if enabled */
1672
                        for(j=0;j<len;j++) {
1673
                            tmp0 = tab0[j];
1674
                            tmp1 = tab1[j];
1675
                            tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
1676
                            tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
1677
                        }
1678
                    }
1679
                }
1680
            }
1681
        }
1682

    
1683
        non_zero_found = non_zero_found_short[0] |
1684
            non_zero_found_short[1] |
1685
            non_zero_found_short[2];
1686

    
1687
        for(i = g1->long_end - 1;i >= 0;i--) {
1688
            len = band_size_long[s->sample_rate_index][i];
1689
            tab0 -= len;
1690
            tab1 -= len;
1691
            /* test if non zero band. if so, stop doing i-stereo */
1692
            if (!non_zero_found) {
1693
                for(j=0;j<len;j++) {
1694
                    if (tab1[j] != 0) {
1695
                        non_zero_found = 1;
1696
                        goto found2;
1697
                    }
1698
                }
1699
                /* for last band, use previous scale factor */
1700
                k = (i == 21) ? 20 : i;
1701
                sf = g1->scale_factors[k];
1702
                if (sf >= sf_max)
1703
                    goto found2;
1704
                v1 = is_tab[0][sf];
1705
                v2 = is_tab[1][sf];
1706
                for(j=0;j<len;j++) {
1707
                    tmp0 = tab0[j];
1708
                    tab0[j] = MULL(tmp0, v1);
1709
                    tab1[j] = MULL(tmp0, v2);
1710
                }
1711
            } else {
1712
            found2:
1713
                if (s->mode_ext & MODE_EXT_MS_STEREO) {
1714
                    /* lower part of the spectrum : do ms stereo
1715
                       if enabled */
1716
                    for(j=0;j<len;j++) {
1717
                        tmp0 = tab0[j];
1718
                        tmp1 = tab1[j];
1719
                        tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
1720
                        tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
1721
                    }
1722
                }
1723
            }
1724
        }
1725
    } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1726
        /* ms stereo ONLY */
1727
        /* NOTE: the 1/sqrt(2) normalization factor is included in the
1728
           global gain */
1729
        tab0 = g0->sb_hybrid;
1730
        tab1 = g1->sb_hybrid;
1731
        for(i=0;i<576;i++) {
1732
            tmp0 = tab0[i];
1733
            tmp1 = tab1[i];
1734
            tab0[i] = tmp0 + tmp1;
1735
            tab1[i] = tmp0 - tmp1;
1736
        }
1737
    }
1738
}
1739

    
1740
static void compute_antialias_integer(MPADecodeContext *s,
1741
                              GranuleDef *g)
1742
{
1743
    int32_t *ptr, *csa;
1744
    int n, i;
1745

    
1746
    /* we antialias only "long" bands */
1747
    if (g->block_type == 2) {
1748
        if (!g->switch_point)
1749
            return;
1750
        /* XXX: check this for 8000Hz case */
1751
        n = 1;
1752
    } else {
1753
        n = SBLIMIT - 1;
1754
    }
1755

    
1756
    ptr = g->sb_hybrid + 18;
1757
    for(i = n;i > 0;i--) {
1758
        int tmp0, tmp1, tmp2;
1759
        csa = &csa_table[0][0];
1760
#define INT_AA(j) \
1761
            tmp0 = ptr[-1-j];\
1762
            tmp1 = ptr[   j];\
1763
            tmp2= MULH(tmp0 + tmp1, csa[0+4*j]);\
1764
            ptr[-1-j] = 4*(tmp2 - MULH(tmp1, csa[2+4*j]));\
1765
            ptr[   j] = 4*(tmp2 + MULH(tmp0, csa[3+4*j]));
1766

    
1767
        INT_AA(0)
1768
        INT_AA(1)
1769
        INT_AA(2)
1770
        INT_AA(3)
1771
        INT_AA(4)
1772
        INT_AA(5)
1773
        INT_AA(6)
1774
        INT_AA(7)
1775

    
1776
        ptr += 18;
1777
    }
1778
}
1779

    
1780
static void compute_antialias_float(MPADecodeContext *s,
1781
                              GranuleDef *g)
1782
{
1783
    int32_t *ptr;
1784
    int n, i;
1785

    
1786
    /* we antialias only "long" bands */
1787
    if (g->block_type == 2) {
1788
        if (!g->switch_point)
1789
            return;
1790
        /* XXX: check this for 8000Hz case */
1791
        n = 1;
1792
    } else {
1793
        n = SBLIMIT - 1;
1794
    }
1795

    
1796
    ptr = g->sb_hybrid + 18;
1797
    for(i = n;i > 0;i--) {
1798
        float tmp0, tmp1;
1799
        float *csa = &csa_table_float[0][0];
1800
#define FLOAT_AA(j)\
1801
        tmp0= ptr[-1-j];\
1802
        tmp1= ptr[   j];\
1803
        ptr[-1-j] = lrintf(tmp0 * csa[0+4*j] - tmp1 * csa[1+4*j]);\
1804
        ptr[   j] = lrintf(tmp0 * csa[1+4*j] + tmp1 * csa[0+4*j]);
1805

    
1806
        FLOAT_AA(0)
1807
        FLOAT_AA(1)
1808
        FLOAT_AA(2)
1809
        FLOAT_AA(3)
1810
        FLOAT_AA(4)
1811
        FLOAT_AA(5)
1812
        FLOAT_AA(6)
1813
        FLOAT_AA(7)
1814

    
1815
        ptr += 18;
1816
    }
1817
}
1818

    
1819
static void compute_imdct(MPADecodeContext *s,
1820
                          GranuleDef *g,
1821
                          int32_t *sb_samples,
1822
                          int32_t *mdct_buf)
1823
{
1824
    int32_t *ptr, *win, *win1, *buf, *out_ptr, *ptr1;
1825
    int32_t out2[12];
1826
    int i, j, mdct_long_end, v, sblimit;
1827

    
1828
    /* find last non zero block */
1829
    ptr = g->sb_hybrid + 576;
1830
    ptr1 = g->sb_hybrid + 2 * 18;
1831
    while (ptr >= ptr1) {
1832
        ptr -= 6;
1833
        v = ptr[0] | ptr[1] | ptr[2] | ptr[3] | ptr[4] | ptr[5];
1834
        if (v != 0)
1835
            break;
1836
    }
1837
    sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
1838

    
1839
    if (g->block_type == 2) {
1840
        /* XXX: check for 8000 Hz */
1841
        if (g->switch_point)
1842
            mdct_long_end = 2;
1843
        else
1844
            mdct_long_end = 0;
1845
    } else {
1846
        mdct_long_end = sblimit;
1847
    }
1848

    
1849
    buf = mdct_buf;
1850
    ptr = g->sb_hybrid;
1851
    for(j=0;j<mdct_long_end;j++) {
1852
        /* apply window & overlap with previous buffer */
1853
        out_ptr = sb_samples + j;
1854
        /* select window */
1855
        if (g->switch_point && j < 2)
1856
            win1 = mdct_win[0];
1857
        else
1858
            win1 = mdct_win[g->block_type];
1859
        /* select frequency inversion */
1860
        win = win1 + ((4 * 36) & -(j & 1));
1861
        imdct36(out_ptr, buf, ptr, win);
1862
        out_ptr += 18*SBLIMIT;
1863
        ptr += 18;
1864
        buf += 18;
1865
    }
1866
    for(j=mdct_long_end;j<sblimit;j++) {
1867
        /* select frequency inversion */
1868
        win = mdct_win[2] + ((4 * 36) & -(j & 1));
1869
        out_ptr = sb_samples + j;
1870

    
1871
        for(i=0; i<6; i++){
1872
            *out_ptr = buf[i];
1873
            out_ptr += SBLIMIT;
1874
        }
1875
        imdct12(out2, ptr + 0);
1876
        for(i=0;i<6;i++) {
1877
            *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*1];
1878
            buf[i + 6*2] = MULH(out2[i + 6], win[i + 6]);
1879
            out_ptr += SBLIMIT;
1880
        }
1881
        imdct12(out2, ptr + 1);
1882
        for(i=0;i<6;i++) {
1883
            *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*2];
1884
            buf[i + 6*0] = MULH(out2[i + 6], win[i + 6]);
1885
            out_ptr += SBLIMIT;
1886
        }
1887
        imdct12(out2, ptr + 2);
1888
        for(i=0;i<6;i++) {
1889
            buf[i + 6*0] = MULH(out2[i], win[i]) + buf[i + 6*0];
1890
            buf[i + 6*1] = MULH(out2[i + 6], win[i + 6]);
1891
            buf[i + 6*2] = 0;
1892
        }
1893
        ptr += 18;
1894
        buf += 18;
1895
    }
1896
    /* zero bands */
1897
    for(j=sblimit;j<SBLIMIT;j++) {
1898
        /* overlap */
1899
        out_ptr = sb_samples + j;
1900
        for(i=0;i<18;i++) {
1901
            *out_ptr = buf[i];
1902
            buf[i] = 0;
1903
            out_ptr += SBLIMIT;
1904
        }
1905
        buf += 18;
1906
    }
1907
}
1908

    
1909
#if defined(DEBUG)
1910
void sample_dump(int fnum, int32_t *tab, int n)
1911
{
1912
    static FILE *files[16], *f;
1913
    char buf[512];
1914
    int i;
1915
    int32_t v;
1916

    
1917
    f = files[fnum];
1918
    if (!f) {
1919
        snprintf(buf, sizeof(buf), "/tmp/out%d.%s.pcm",
1920
                fnum,
1921
#ifdef USE_HIGHPRECISION
1922
                "hp"
1923
#else
1924
                "lp"
1925
#endif
1926
                );
1927
        f = fopen(buf, "w");
1928
        if (!f)
1929
            return;
1930
        files[fnum] = f;
1931
    }
1932

    
1933
    if (fnum == 0) {
1934
        static int pos = 0;
1935
        av_log(NULL, AV_LOG_DEBUG, "pos=%d\n", pos);
1936
        for(i=0;i<n;i++) {
1937
            av_log(NULL, AV_LOG_DEBUG, " %0.4f", (double)tab[i] / FRAC_ONE);
1938
            if ((i % 18) == 17)
1939
                av_log(NULL, AV_LOG_DEBUG, "\n");
1940
        }
1941
        pos += n;
1942
    }
1943
    for(i=0;i<n;i++) {
1944
        /* normalize to 23 frac bits */
1945
        v = tab[i] << (23 - FRAC_BITS);
1946
        fwrite(&v, 1, sizeof(int32_t), f);
1947
    }
1948
}
1949
#endif
1950

    
1951

    
1952
/* main layer3 decoding function */
1953
static int mp_decode_layer3(MPADecodeContext *s)
1954
{
1955
    int nb_granules, main_data_begin, private_bits;
1956
    int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;
1957
    GranuleDef granules[2][2], *g;
1958
    int16_t exponents[576];
1959

    
1960
    /* read side info */
1961
    if (s->lsf) {
1962
        main_data_begin = get_bits(&s->gb, 8);
1963
        private_bits = get_bits(&s->gb, s->nb_channels);
1964
        nb_granules = 1;
1965
    } else {
1966
        main_data_begin = get_bits(&s->gb, 9);
1967
        if (s->nb_channels == 2)
1968
            private_bits = get_bits(&s->gb, 3);
1969
        else
1970
            private_bits = get_bits(&s->gb, 5);
1971
        nb_granules = 2;
1972
        for(ch=0;ch<s->nb_channels;ch++) {
1973
            granules[ch][0].scfsi = 0; /* all scale factors are transmitted */
1974
            granules[ch][1].scfsi = get_bits(&s->gb, 4);
1975
        }
1976
    }
1977

    
1978
    for(gr=0;gr<nb_granules;gr++) {
1979
        for(ch=0;ch<s->nb_channels;ch++) {
1980
            dprintf(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);
1981
            g = &granules[ch][gr];
1982
            g->part2_3_length = get_bits(&s->gb, 12);
1983
            g->big_values = get_bits(&s->gb, 9);
1984
            if(g->big_values > 288){
1985
                av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");
1986
                return -1;
1987
            }
1988

    
1989
            g->global_gain = get_bits(&s->gb, 8);
1990
            /* if MS stereo only is selected, we precompute the
1991
               1/sqrt(2) renormalization factor */
1992
            if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
1993
                MODE_EXT_MS_STEREO)
1994
                g->global_gain -= 2;
1995
            if (s->lsf)
1996
                g->scalefac_compress = get_bits(&s->gb, 9);
1997
            else
1998
                g->scalefac_compress = get_bits(&s->gb, 4);
1999
            blocksplit_flag = get_bits1(&s->gb);
2000
            if (blocksplit_flag) {
2001
                g->block_type = get_bits(&s->gb, 2);
2002
                if (g->block_type == 0){
2003
                    av_log(NULL, AV_LOG_ERROR, "invalid block type\n");
2004
                    return -1;
2005
                }
2006
                g->switch_point = get_bits1(&s->gb);
2007
                for(i=0;i<2;i++)
2008
                    g->table_select[i] = get_bits(&s->gb, 5);
2009
                for(i=0;i<3;i++)
2010
                    g->subblock_gain[i] = get_bits(&s->gb, 3);
2011
                /* compute huffman coded region sizes */
2012
                if (g->block_type == 2)
2013
                    g->region_size[0] = (36 / 2);
2014
                else {
2015
                    if (s->sample_rate_index <= 2)
2016
                        g->region_size[0] = (36 / 2);
2017
                    else if (s->sample_rate_index != 8)
2018
                        g->region_size[0] = (54 / 2);
2019
                    else
2020
                        g->region_size[0] = (108 / 2);
2021
                }
2022
                g->region_size[1] = (576 / 2);
2023
            } else {
2024
                int region_address1, region_address2, l;
2025
                g->block_type = 0;
2026
                g->switch_point = 0;
2027
                for(i=0;i<3;i++)
2028
                    g->table_select[i] = get_bits(&s->gb, 5);
2029
                /* compute huffman coded region sizes */
2030
                region_address1 = get_bits(&s->gb, 4);
2031
                region_address2 = get_bits(&s->gb, 3);
2032
                dprintf(s->avctx, "region1=%d region2=%d\n",
2033
                        region_address1, region_address2);
2034
                g->region_size[0] =
2035
                    band_index_long[s->sample_rate_index][region_address1 + 1] >> 1;
2036
                l = region_address1 + region_address2 + 2;
2037
                /* should not overflow */
2038
                if (l > 22)
2039
                    l = 22;
2040
                g->region_size[1] =
2041
                    band_index_long[s->sample_rate_index][l] >> 1;
2042
            }
2043
            /* convert region offsets to region sizes and truncate
2044
               size to big_values */
2045
            g->region_size[2] = (576 / 2);
2046
            j = 0;
2047
            for(i=0;i<3;i++) {
2048
                k = FFMIN(g->region_size[i], g->big_values);
2049
                g->region_size[i] = k - j;
2050
                j = k;
2051
            }
2052

    
2053
            /* compute band indexes */
2054
            if (g->block_type == 2) {
2055
                if (g->switch_point) {
2056
                    /* if switched mode, we handle the 36 first samples as
2057
                       long blocks.  For 8000Hz, we handle the 48 first
2058
                       exponents as long blocks (XXX: check this!) */
2059
                    if (s->sample_rate_index <= 2)
2060
                        g->long_end = 8;
2061
                    else if (s->sample_rate_index != 8)
2062
                        g->long_end = 6;
2063
                    else
2064
                        g->long_end = 4; /* 8000 Hz */
2065

    
2066
                    g->short_start = 2 + (s->sample_rate_index != 8);
2067
                } else {
2068
                    g->long_end = 0;
2069
                    g->short_start = 0;
2070
                }
2071
            } else {
2072
                g->short_start = 13;
2073
                g->long_end = 22;
2074
            }
2075

    
2076
            g->preflag = 0;
2077
            if (!s->lsf)
2078
                g->preflag = get_bits1(&s->gb);
2079
            g->scalefac_scale = get_bits1(&s->gb);
2080
            g->count1table_select = get_bits1(&s->gb);
2081
            dprintf(s->avctx, "block_type=%d switch_point=%d\n",
2082
                    g->block_type, g->switch_point);
2083
        }
2084
    }
2085

    
2086
  if (!s->adu_mode) {
2087
    const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);
2088
    assert((get_bits_count(&s->gb) & 7) == 0);
2089
    /* now we get bits from the main_data_begin offset */
2090
    dprintf(s->avctx, "seekback: %d\n", main_data_begin);
2091
//av_log(NULL, AV_LOG_ERROR, "backstep:%d, lastbuf:%d\n", main_data_begin, s->last_buf_size);
2092

    
2093
    memcpy(s->last_buf + s->last_buf_size, ptr, EXTRABYTES);
2094
    s->in_gb= s->gb;
2095
        init_get_bits(&s->gb, s->last_buf, s->last_buf_size*8);
2096
        skip_bits_long(&s->gb, 8*(s->last_buf_size - main_data_begin));
2097
  }
2098

    
2099
    for(gr=0;gr<nb_granules;gr++) {
2100
        for(ch=0;ch<s->nb_channels;ch++) {
2101
            g = &granules[ch][gr];
2102
            if(get_bits_count(&s->gb)<0){
2103
                av_log(NULL, AV_LOG_ERROR, "mdb:%d, lastbuf:%d skipping granule %d\n",
2104
                                            main_data_begin, s->last_buf_size, gr);
2105
                skip_bits_long(&s->gb, g->part2_3_length);
2106
                memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));
2107
                if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->in_gb.buffer){
2108
                    skip_bits_long(&s->in_gb, get_bits_count(&s->gb) - s->gb.size_in_bits);
2109
                    s->gb= s->in_gb;
2110
                    s->in_gb.buffer=NULL;
2111
                }
2112
                continue;
2113
            }
2114

    
2115
            bits_pos = get_bits_count(&s->gb);
2116

    
2117
            if (!s->lsf) {
2118
                uint8_t *sc;
2119
                int slen, slen1, slen2;
2120

    
2121
                /* MPEG1 scale factors */
2122
                slen1 = slen_table[0][g->scalefac_compress];
2123
                slen2 = slen_table[1][g->scalefac_compress];
2124
                dprintf(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);
2125
                if (g->block_type == 2) {
2126
                    n = g->switch_point ? 17 : 18;
2127
                    j = 0;
2128
                    if(slen1){
2129
                        for(i=0;i<n;i++)
2130
                            g->scale_factors[j++] = get_bits(&s->gb, slen1);
2131
                    }else{
2132
                        for(i=0;i<n;i++)
2133
                            g->scale_factors[j++] = 0;
2134
                    }
2135
                    if(slen2){
2136
                        for(i=0;i<18;i++)
2137
                            g->scale_factors[j++] = get_bits(&s->gb, slen2);
2138
                        for(i=0;i<3;i++)
2139
                            g->scale_factors[j++] = 0;
2140
                    }else{
2141
                        for(i=0;i<21;i++)
2142
                            g->scale_factors[j++] = 0;
2143
                    }
2144
                } else {
2145
                    sc = granules[ch][0].scale_factors;
2146
                    j = 0;
2147
                    for(k=0;k<4;k++) {
2148
                        n = (k == 0 ? 6 : 5);
2149
                        if ((g->scfsi & (0x8 >> k)) == 0) {
2150
                            slen = (k < 2) ? slen1 : slen2;
2151
                            if(slen){
2152
                                for(i=0;i<n;i++)
2153
                                    g->scale_factors[j++] = get_bits(&s->gb, slen);
2154
                            }else{
2155
                                for(i=0;i<n;i++)
2156
                                    g->scale_factors[j++] = 0;
2157
                            }
2158
                        } else {
2159
                            /* simply copy from last granule */
2160
                            for(i=0;i<n;i++) {
2161
                                g->scale_factors[j] = sc[j];
2162
                                j++;
2163
                            }
2164
                        }
2165
                    }
2166
                    g->scale_factors[j++] = 0;
2167
                }
2168
#if defined(DEBUG)
2169
                {
2170
                    dprintf(s->avctx, "scfsi=%x gr=%d ch=%d scale_factors:\n",
2171
                           g->scfsi, gr, ch);
2172
                    for(i=0;i<j;i++)
2173
                        dprintf(s->avctx, " %d", g->scale_factors[i]);
2174
                    dprintf(s->avctx, "\n");
2175
                }
2176
#endif
2177
            } else {
2178
                int tindex, tindex2, slen[4], sl, sf;
2179

    
2180
                /* LSF scale factors */
2181
                if (g->block_type == 2) {
2182
                    tindex = g->switch_point ? 2 : 1;
2183
                } else {
2184
                    tindex = 0;
2185
                }
2186
                sf = g->scalefac_compress;
2187
                if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
2188
                    /* intensity stereo case */
2189
                    sf >>= 1;
2190
                    if (sf < 180) {
2191
                        lsf_sf_expand(slen, sf, 6, 6, 0);
2192
                        tindex2 = 3;
2193
                    } else if (sf < 244) {
2194
                        lsf_sf_expand(slen, sf - 180, 4, 4, 0);
2195
                        tindex2 = 4;
2196
                    } else {
2197
                        lsf_sf_expand(slen, sf - 244, 3, 0, 0);
2198
                        tindex2 = 5;
2199
                    }
2200
                } else {
2201
                    /* normal case */
2202
                    if (sf < 400) {
2203
                        lsf_sf_expand(slen, sf, 5, 4, 4);
2204
                        tindex2 = 0;
2205
                    } else if (sf < 500) {
2206
                        lsf_sf_expand(slen, sf - 400, 5, 4, 0);
2207
                        tindex2 = 1;
2208
                    } else {
2209
                        lsf_sf_expand(slen, sf - 500, 3, 0, 0);
2210
                        tindex2 = 2;
2211
                        g->preflag = 1;
2212
                    }
2213
                }
2214

    
2215
                j = 0;
2216
                for(k=0;k<4;k++) {
2217
                    n = lsf_nsf_table[tindex2][tindex][k];
2218
                    sl = slen[k];
2219
                    if(sl){
2220
                        for(i=0;i<n;i++)
2221
                            g->scale_factors[j++] = get_bits(&s->gb, sl);
2222
                    }else{
2223
                        for(i=0;i<n;i++)
2224
                            g->scale_factors[j++] = 0;
2225
                    }
2226
                }
2227
                /* XXX: should compute exact size */
2228
                for(;j<40;j++)
2229
                    g->scale_factors[j] = 0;
2230
#if defined(DEBUG)
2231
                {
2232
                    dprintf(s->avctx, "gr=%d ch=%d scale_factors:\n",
2233
                           gr, ch);
2234
                    for(i=0;i<40;i++)
2235
                        dprintf(s->avctx, " %d", g->scale_factors[i]);
2236
                    dprintf(s->avctx, "\n");
2237
                }
2238
#endif
2239
            }
2240

    
2241
            exponents_from_scale_factors(s, g, exponents);
2242

    
2243
            /* read Huffman coded residue */
2244
            huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);
2245
#if defined(DEBUG)
2246
            sample_dump(0, g->sb_hybrid, 576);
2247
#endif
2248
        } /* ch */
2249

    
2250
        if (s->nb_channels == 2)
2251
            compute_stereo(s, &granules[0][gr], &granules[1][gr]);
2252

    
2253
        for(ch=0;ch<s->nb_channels;ch++) {
2254
            g = &granules[ch][gr];
2255

    
2256
            reorder_block(s, g);
2257
#if defined(DEBUG)
2258
            sample_dump(0, g->sb_hybrid, 576);
2259
#endif
2260
            s->compute_antialias(s, g);
2261
#if defined(DEBUG)
2262
            sample_dump(1, g->sb_hybrid, 576);
2263
#endif
2264
            compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
2265
#if defined(DEBUG)
2266
            sample_dump(2, &s->sb_samples[ch][18 * gr][0], 576);
2267
#endif
2268
        }
2269
    } /* gr */
2270
    if(get_bits_count(&s->gb)<0)
2271
        skip_bits_long(&s->gb, -get_bits_count(&s->gb));
2272
    return nb_granules * 18;
2273
}
2274

    
2275
static int mp_decode_frame(MPADecodeContext *s,
2276
                           OUT_INT *samples, const uint8_t *buf, int buf_size)
2277
{
2278
    int i, nb_frames, ch;
2279
    OUT_INT *samples_ptr;
2280

    
2281
    init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE)*8);
2282

    
2283
    /* skip error protection field */
2284
    if (s->error_protection)
2285
        skip_bits(&s->gb, 16);
2286

    
2287
    dprintf(s->avctx, "frame %d:\n", s->frame_count);
2288
    switch(s->layer) {
2289
    case 1:
2290
        nb_frames = mp_decode_layer1(s);
2291
        break;
2292
    case 2:
2293
        nb_frames = mp_decode_layer2(s);
2294
        break;
2295
    case 3:
2296
    default:
2297
        nb_frames = mp_decode_layer3(s);
2298

    
2299
        s->last_buf_size=0;
2300
        if(s->in_gb.buffer){
2301
            align_get_bits(&s->gb);
2302
            i= (s->gb.size_in_bits - get_bits_count(&s->gb))>>3;
2303
            if(i >= 0 && i <= BACKSTEP_SIZE){
2304
                memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);
2305
                s->last_buf_size=i;
2306
            }else
2307
                av_log(NULL, AV_LOG_ERROR, "invalid old backstep %d\n", i);
2308
            s->gb= s->in_gb;
2309
            s->in_gb.buffer= NULL;
2310
        }
2311

    
2312
        align_get_bits(&s->gb);
2313
        assert((get_bits_count(&s->gb) & 7) == 0);
2314
        i= (s->gb.size_in_bits - get_bits_count(&s->gb))>>3;
2315

    
2316
        if(i<0 || i > BACKSTEP_SIZE || nb_frames<0){
2317
            av_log(NULL, AV_LOG_ERROR, "invalid new backstep %d\n", i);
2318
            i= FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);
2319
        }
2320
        assert(i <= buf_size - HEADER_SIZE && i>= 0);
2321
        memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);
2322
        s->last_buf_size += i;
2323

    
2324
        break;
2325
    }
2326
#if defined(DEBUG)
2327
    for(i=0;i<nb_frames;i++) {
2328
        for(ch=0;ch<s->nb_channels;ch++) {
2329
            int j;
2330
            dprintf(s->avctx, "%d-%d:", i, ch);
2331
            for(j=0;j<SBLIMIT;j++)
2332
                dprintf(s->avctx, " %0.6f", (double)s->sb_samples[ch][i][j] / FRAC_ONE);
2333
            dprintf(s->avctx, "\n");
2334
        }
2335
    }
2336
#endif
2337
    /* apply the synthesis filter */
2338
    for(ch=0;ch<s->nb_channels;ch++) {
2339
        samples_ptr = samples + ch;
2340
        for(i=0;i<nb_frames;i++) {
2341
            ff_mpa_synth_filter(s->synth_buf[ch], &(s->synth_buf_offset[ch]),
2342
                         window, &s->dither_state,
2343
                         samples_ptr, s->nb_channels,
2344
                         s->sb_samples[ch][i]);
2345
            samples_ptr += 32 * s->nb_channels;
2346
        }
2347
    }
2348
#ifdef DEBUG
2349
    s->frame_count++;
2350
#endif
2351
    return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
2352
}
2353

    
2354
static int decode_frame(AVCodecContext * avctx,
2355
                        void *data, int *data_size,
2356
                        uint8_t * buf, int buf_size)
2357
{
2358
    MPADecodeContext *s = avctx->priv_data;
2359
    uint32_t header;
2360
    int out_size;
2361
    OUT_INT *out_samples = data;
2362

    
2363
retry:
2364
    if(buf_size < HEADER_SIZE)
2365
        return -1;
2366

    
2367
    header = AV_RB32(buf);
2368
    if(ff_mpa_check_header(header) < 0){
2369
        buf++;
2370
//        buf_size--;
2371
        av_log(avctx, AV_LOG_ERROR, "Header missing skipping one byte.\n");
2372
        goto retry;
2373
    }
2374

    
2375
    if (ff_mpegaudio_decode_header(s, header) == 1) {
2376
        /* free format: prepare to compute frame size */
2377
        s->frame_size = -1;
2378
        return -1;
2379
    }
2380
    /* update codec info */
2381
    avctx->channels = s->nb_channels;
2382
    avctx->bit_rate = s->bit_rate;
2383
    avctx->sub_id = s->layer;
2384
    switch(s->layer) {
2385
    case 1:
2386
        avctx->frame_size = 384;
2387
        break;
2388
    case 2:
2389
        avctx->frame_size = 1152;
2390
        break;
2391
    case 3:
2392
        if (s->lsf)
2393
            avctx->frame_size = 576;
2394
        else
2395
            avctx->frame_size = 1152;
2396
        break;
2397
    }
2398

    
2399
    if(s->frame_size<=0 || s->frame_size > buf_size){
2400
        av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
2401
        return -1;
2402
    }else if(s->frame_size < buf_size){
2403
        av_log(avctx, AV_LOG_ERROR, "incorrect frame size\n");
2404
        buf_size= s->frame_size;
2405
    }
2406

    
2407
    out_size = mp_decode_frame(s, out_samples, buf, buf_size);
2408
    if(out_size>=0){
2409
        *data_size = out_size;
2410
        avctx->sample_rate = s->sample_rate;
2411
        //FIXME maybe move the other codec info stuff from above here too
2412
    }else
2413
        av_log(avctx, AV_LOG_DEBUG, "Error while decoding MPEG audio frame.\n"); //FIXME return -1 / but also return the number of bytes consumed
2414
    s->frame_size = 0;
2415
    return buf_size;
2416
}
2417

    
2418
static void flush(AVCodecContext *avctx){
2419
    MPADecodeContext *s = avctx->priv_data;
2420
    s->last_buf_size= 0;
2421
}
2422

    
2423
#ifdef CONFIG_MP3ADU_DECODER
2424
static int decode_frame_adu(AVCodecContext * avctx,
2425
                        void *data, int *data_size,
2426
                        uint8_t * buf, int buf_size)
2427
{
2428
    MPADecodeContext *s = avctx->priv_data;
2429
    uint32_t header;
2430
    int len, out_size;
2431
    OUT_INT *out_samples = data;
2432

    
2433
    len = buf_size;
2434

    
2435
    // Discard too short frames
2436
    if (buf_size < HEADER_SIZE) {
2437
        *data_size = 0;
2438
        return buf_size;
2439
    }
2440

    
2441

    
2442
    if (len > MPA_MAX_CODED_FRAME_SIZE)
2443
        len = MPA_MAX_CODED_FRAME_SIZE;
2444

    
2445
    // Get header and restore sync word
2446
    header = AV_RB32(buf) | 0xffe00000;
2447

    
2448
    if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame
2449
        *data_size = 0;
2450
        return buf_size;
2451
    }
2452

    
2453
    ff_mpegaudio_decode_header(s, header);
2454
    /* update codec info */
2455
    avctx->sample_rate = s->sample_rate;
2456
    avctx->channels = s->nb_channels;
2457
    avctx->bit_rate = s->bit_rate;
2458
    avctx->sub_id = s->layer;
2459

    
2460
    avctx->frame_size=s->frame_size = len;
2461

    
2462
    if (avctx->parse_only) {
2463
        out_size = buf_size;
2464
    } else {
2465
        out_size = mp_decode_frame(s, out_samples, buf, buf_size);
2466
    }
2467

    
2468
    *data_size = out_size;
2469
    return buf_size;
2470
}
2471
#endif /* CONFIG_MP3ADU_DECODER */
2472

    
2473
#ifdef CONFIG_MP3ON4_DECODER
2474
/* Next 3 arrays are indexed by channel config number (passed via codecdata) */
2475
static int mp3Frames[16] = {0,1,1,2,3,3,4,5,2};   /* number of mp3 decoder instances */
2476
static int mp3Channels[16] = {0,1,2,3,4,5,6,8,4}; /* total output channels */
2477
/* offsets into output buffer, assume output order is FL FR BL BR C LFE */
2478
static int chan_offset[9][5] = {
2479
    {0},
2480
    {0},            // C
2481
    {0},            // FLR
2482
    {2,0},          // C FLR
2483
    {2,0,3},        // C FLR BS
2484
    {4,0,2},        // C FLR BLRS
2485
    {4,0,2,5},      // C FLR BLRS LFE
2486
    {4,0,2,6,5},    // C FLR BLRS BLR LFE
2487
    {0,2}           // FLR BLRS
2488
};
2489

    
2490

    
2491
static int decode_init_mp3on4(AVCodecContext * avctx)
2492
{
2493
    MP3On4DecodeContext *s = avctx->priv_data;
2494
    int i;
2495

    
2496
    if ((avctx->extradata_size < 2) || (avctx->extradata == NULL)) {
2497
        av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
2498
        return -1;
2499
    }
2500

    
2501
    s->chan_cfg = (((unsigned char *)avctx->extradata)[1] >> 3) & 0x0f;
2502
    s->frames = mp3Frames[s->chan_cfg];
2503
    if(!s->frames) {
2504
        av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
2505
        return -1;
2506
    }
2507
    avctx->channels = mp3Channels[s->chan_cfg];
2508

    
2509
    /* Init the first mp3 decoder in standard way, so that all tables get builded
2510
     * We replace avctx->priv_data with the context of the first decoder so that
2511
     * decode_init() does not have to be changed.
2512
     * Other decoders will be inited here copying data from the first context
2513
     */
2514
    // Allocate zeroed memory for the first decoder context
2515
    s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
2516
    // Put decoder context in place to make init_decode() happy
2517
    avctx->priv_data = s->mp3decctx[0];
2518
    decode_init(avctx);
2519
    // Restore mp3on4 context pointer
2520
    avctx->priv_data = s;
2521
    s->mp3decctx[0]->adu_mode = 1; // Set adu mode
2522

    
2523
    /* Create a separate codec/context for each frame (first is already ok).
2524
     * Each frame is 1 or 2 channels - up to 5 frames allowed
2525
     */
2526
    for (i = 1; i < s->frames; i++) {
2527
        s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
2528
        s->mp3decctx[i]->compute_antialias = s->mp3decctx[0]->compute_antialias;
2529
        s->mp3decctx[i]->adu_mode = 1;
2530
        s->mp3decctx[i]->avctx = avctx;
2531
    }
2532

    
2533
    return 0;
2534
}
2535

    
2536

    
2537
static int decode_close_mp3on4(AVCodecContext * avctx)
2538
{
2539
    MP3On4DecodeContext *s = avctx->priv_data;
2540
    int i;
2541

    
2542
    for (i = 0; i < s->frames; i++)
2543
        if (s->mp3decctx[i])
2544
            av_free(s->mp3decctx[i]);
2545

    
2546
    return 0;
2547
}
2548

    
2549

    
2550
static int decode_frame_mp3on4(AVCodecContext * avctx,
2551
                        void *data, int *data_size,
2552
                        uint8_t * buf, int buf_size)
2553
{
2554
    MP3On4DecodeContext *s = avctx->priv_data;
2555
    MPADecodeContext *m;
2556
    int len, out_size = 0;
2557
    uint32_t header;
2558
    OUT_INT *out_samples = data;
2559
    OUT_INT decoded_buf[MPA_FRAME_SIZE * MPA_MAX_CHANNELS];
2560
    OUT_INT *outptr, *bp;
2561
    int fsize;
2562
    unsigned char *start2 = buf, *start;
2563
    int fr, i, j, n;
2564
    int off = avctx->channels;
2565
    int *coff = chan_offset[s->chan_cfg];
2566

    
2567
    len = buf_size;
2568

    
2569
    // Discard too short frames
2570
    if (buf_size < HEADER_SIZE) {
2571
        *data_size = 0;
2572
        return buf_size;
2573
    }
2574

    
2575
    // If only one decoder interleave is not needed
2576
    outptr = s->frames == 1 ? out_samples : decoded_buf;
2577

    
2578
    for (fr = 0; fr < s->frames; fr++) {
2579
        start = start2;
2580
        fsize = (start[0] << 4) | (start[1] >> 4);
2581
        start2 += fsize;
2582
        if (fsize > len)
2583
            fsize = len;
2584
        len -= fsize;
2585
        if (fsize > MPA_MAX_CODED_FRAME_SIZE)
2586
            fsize = MPA_MAX_CODED_FRAME_SIZE;
2587
        m = s->mp3decctx[fr];
2588
        assert (m != NULL);
2589

    
2590
        // Get header
2591
        header = AV_RB32(start) | 0xfff00000;
2592

    
2593
        if (ff_mpa_check_header(header) < 0) { // Bad header, discard block
2594
            *data_size = 0;
2595
            return buf_size;
2596
        }
2597

    
2598
        ff_mpegaudio_decode_header(m, header);
2599
        mp_decode_frame(m, decoded_buf, start, fsize);
2600

    
2601
        n = MPA_FRAME_SIZE * m->nb_channels;
2602
        out_size += n * sizeof(OUT_INT);
2603
        if(s->frames > 1) {
2604
            /* interleave output data */
2605
            bp = out_samples + coff[fr];
2606
            if(m->nb_channels == 1) {
2607
                for(j = 0; j < n; j++) {
2608
                    *bp = decoded_buf[j];
2609
                    bp += off;
2610
                }
2611
            } else {
2612
                for(j = 0; j < n; j++) {
2613
                    bp[0] = decoded_buf[j++];
2614
                    bp[1] = decoded_buf[j];
2615
                    bp += off;
2616
                }
2617
            }
2618
        }
2619
    }
2620

    
2621
    /* update codec info */
2622
    avctx->sample_rate = s->mp3decctx[0]->sample_rate;
2623
    avctx->frame_size= buf_size;
2624
    avctx->bit_rate = 0;
2625
    for (i = 0; i < s->frames; i++)
2626
        avctx->bit_rate += s->mp3decctx[i]->bit_rate;
2627

    
2628
    *data_size = out_size;
2629
    return buf_size;
2630
}
2631
#endif /* CONFIG_MP3ON4_DECODER */
2632

    
2633
#ifdef CONFIG_MP2_DECODER
2634
AVCodec mp2_decoder =
2635
{
2636
    "mp2",
2637
    CODEC_TYPE_AUDIO,
2638
    CODEC_ID_MP2,
2639
    sizeof(MPADecodeContext),
2640
    decode_init,
2641
    NULL,
2642
    NULL,
2643
    decode_frame,
2644
    CODEC_CAP_PARSE_ONLY,
2645
    .flush= flush,
2646
};
2647
#endif
2648
#ifdef CONFIG_MP3_DECODER
2649
AVCodec mp3_decoder =
2650
{
2651
    "mp3",
2652
    CODEC_TYPE_AUDIO,
2653
    CODEC_ID_MP3,
2654
    sizeof(MPADecodeContext),
2655
    decode_init,
2656
    NULL,
2657
    NULL,
2658
    decode_frame,
2659
    CODEC_CAP_PARSE_ONLY,
2660
    .flush= flush,
2661
};
2662
#endif
2663
#ifdef CONFIG_MP3ADU_DECODER
2664
AVCodec mp3adu_decoder =
2665
{
2666
    "mp3adu",
2667
    CODEC_TYPE_AUDIO,
2668
    CODEC_ID_MP3ADU,
2669
    sizeof(MPADecodeContext),
2670
    decode_init,
2671
    NULL,
2672
    NULL,
2673
    decode_frame_adu,
2674
    CODEC_CAP_PARSE_ONLY,
2675
    .flush= flush,
2676
};
2677
#endif
2678
#ifdef CONFIG_MP3ON4_DECODER
2679
AVCodec mp3on4_decoder =
2680
{
2681
    "mp3on4",
2682
    CODEC_TYPE_AUDIO,
2683
    CODEC_ID_MP3ON4,
2684
    sizeof(MP3On4DecodeContext),
2685
    decode_init_mp3on4,
2686
    NULL,
2687
    decode_close_mp3on4,
2688
    decode_frame_mp3on4,
2689
    .flush= flush,
2690
};
2691
#endif