Statistics
| Branch: | Revision:

ffmpeg / libavcodec / mpegaudiodec.c @ 6122b733

History | View | Annotate | Download (79.9 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
/**
130
 * Convert region offsets to region sizes and truncate
131
 * size to big_values.
132
 */
133
void ff_region_offset2size(GranuleDef *g){
134
    int i, k, j=0;
135
    g->region_size[2] = (576 / 2);
136
    for(i=0;i<3;i++) {
137
        k = FFMIN(g->region_size[i], g->big_values);
138
        g->region_size[i] = k - j;
139
        j = k;
140
    }
141
}
142

    
143
void ff_init_short_region(MPADecodeContext *s, GranuleDef *g){
144
    if (g->block_type == 2)
145
        g->region_size[0] = (36 / 2);
146
    else {
147
        if (s->sample_rate_index <= 2)
148
            g->region_size[0] = (36 / 2);
149
        else if (s->sample_rate_index != 8)
150
            g->region_size[0] = (54 / 2);
151
        else
152
            g->region_size[0] = (108 / 2);
153
    }
154
    g->region_size[1] = (576 / 2);
155
}
156

    
157
void ff_init_long_region(MPADecodeContext *s, GranuleDef *g, int ra1, int ra2){
158
    int l;
159
    g->region_size[0] =
160
        band_index_long[s->sample_rate_index][ra1 + 1] >> 1;
161
    /* should not overflow */
162
    l = FFMIN(ra1 + ra2 + 2, 22);
163
    g->region_size[1] =
164
        band_index_long[s->sample_rate_index][l] >> 1;
165
}
166

    
167
void ff_compute_band_indexes(MPADecodeContext *s, GranuleDef *g){
168
    if (g->block_type == 2) {
169
        if (g->switch_point) {
170
            /* if switched mode, we handle the 36 first samples as
171
                long blocks.  For 8000Hz, we handle the 48 first
172
                exponents as long blocks (XXX: check this!) */
173
            if (s->sample_rate_index <= 2)
174
                g->long_end = 8;
175
            else if (s->sample_rate_index != 8)
176
                g->long_end = 6;
177
            else
178
                g->long_end = 4; /* 8000 Hz */
179

    
180
            g->short_start = 2 + (s->sample_rate_index != 8);
181
        } else {
182
            g->long_end = 0;
183
            g->short_start = 0;
184
        }
185
    } else {
186
        g->short_start = 13;
187
        g->long_end = 22;
188
    }
189
}
190

    
191
/* layer 1 unscaling */
192
/* n = number of bits of the mantissa minus 1 */
193
static inline int l1_unscale(int n, int mant, int scale_factor)
194
{
195
    int shift, mod;
196
    int64_t val;
197

    
198
    shift = scale_factor_modshift[scale_factor];
199
    mod = shift & 3;
200
    shift >>= 2;
201
    val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);
202
    shift += n;
203
    /* NOTE: at this point, 1 <= shift >= 21 + 15 */
204
    return (int)((val + (1LL << (shift - 1))) >> shift);
205
}
206

    
207
static inline int l2_unscale_group(int steps, int mant, int scale_factor)
208
{
209
    int shift, mod, val;
210

    
211
    shift = scale_factor_modshift[scale_factor];
212
    mod = shift & 3;
213
    shift >>= 2;
214

    
215
    val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
216
    /* NOTE: at this point, 0 <= shift <= 21 */
217
    if (shift > 0)
218
        val = (val + (1 << (shift - 1))) >> shift;
219
    return val;
220
}
221

    
222
/* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
223
static inline int l3_unscale(int value, int exponent)
224
{
225
    unsigned int m;
226
    int e;
227

    
228
    e = table_4_3_exp  [4*value + (exponent&3)];
229
    m = table_4_3_value[4*value + (exponent&3)];
230
    e -= (exponent >> 2);
231
    assert(e>=1);
232
    if (e > 31)
233
        return 0;
234
    m = (m + (1 << (e-1))) >> e;
235

    
236
    return m;
237
}
238

    
239
/* all integer n^(4/3) computation code */
240
#define DEV_ORDER 13
241

    
242
#define POW_FRAC_BITS 24
243
#define POW_FRAC_ONE    (1 << POW_FRAC_BITS)
244
#define POW_FIX(a)   ((int)((a) * POW_FRAC_ONE))
245
#define POW_MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> POW_FRAC_BITS)
246

    
247
static int dev_4_3_coefs[DEV_ORDER];
248

    
249
#if 0 /* unused */
250
static int pow_mult3[3] = {
251
    POW_FIX(1.0),
252
    POW_FIX(1.25992104989487316476),
253
    POW_FIX(1.58740105196819947474),
254
};
255
#endif
256

    
257
static void int_pow_init(void)
258
{
259
    int i, a;
260

    
261
    a = POW_FIX(1.0);
262
    for(i=0;i<DEV_ORDER;i++) {
263
        a = POW_MULL(a, POW_FIX(4.0 / 3.0) - i * POW_FIX(1.0)) / (i + 1);
264
        dev_4_3_coefs[i] = a;
265
    }
266
}
267

    
268
#if 0 /* unused, remove? */
269
/* return the mantissa and the binary exponent */
270
static int int_pow(int i, int *exp_ptr)
271
{
272
    int e, er, eq, j;
273
    int a, a1;
274

275
    /* renormalize */
276
    a = i;
277
    e = POW_FRAC_BITS;
278
    while (a < (1 << (POW_FRAC_BITS - 1))) {
279
        a = a << 1;
280
        e--;
281
    }
282
    a -= (1 << POW_FRAC_BITS);
283
    a1 = 0;
284
    for(j = DEV_ORDER - 1; j >= 0; j--)
285
        a1 = POW_MULL(a, dev_4_3_coefs[j] + a1);
286
    a = (1 << POW_FRAC_BITS) + a1;
287
    /* exponent compute (exact) */
288
    e = e * 4;
289
    er = e % 3;
290
    eq = e / 3;
291
    a = POW_MULL(a, pow_mult3[er]);
292
    while (a >= 2 * POW_FRAC_ONE) {
293
        a = a >> 1;
294
        eq++;
295
    }
296
    /* convert to float */
297
    while (a < POW_FRAC_ONE) {
298
        a = a << 1;
299
        eq--;
300
    }
301
    /* now POW_FRAC_ONE <= a < 2 * POW_FRAC_ONE */
302
#if POW_FRAC_BITS > FRAC_BITS
303
    a = (a + (1 << (POW_FRAC_BITS - FRAC_BITS - 1))) >> (POW_FRAC_BITS - FRAC_BITS);
304
    /* correct overflow */
305
    if (a >= 2 * (1 << FRAC_BITS)) {
306
        a = a >> 1;
307
        eq++;
308
    }
309
#endif
310
    *exp_ptr = eq;
311
    return a;
312
}
313
#endif
314

    
315
static int decode_init(AVCodecContext * avctx)
316
{
317
    MPADecodeContext *s = avctx->priv_data;
318
    static int init=0;
319
    int i, j, k;
320

    
321
    s->avctx = avctx;
322

    
323
#if defined(USE_HIGHPRECISION) && defined(CONFIG_AUDIO_NONSHORT)
324
    avctx->sample_fmt= SAMPLE_FMT_S32;
325
#else
326
    avctx->sample_fmt= SAMPLE_FMT_S16;
327
#endif
328
    s->error_resilience= avctx->error_resilience;
329

    
330
    if(avctx->antialias_algo != FF_AA_FLOAT)
331
        s->compute_antialias= compute_antialias_integer;
332
    else
333
        s->compute_antialias= compute_antialias_float;
334

    
335
    if (!init && !avctx->parse_only) {
336
        /* scale factors table for layer 1/2 */
337
        for(i=0;i<64;i++) {
338
            int shift, mod;
339
            /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */
340
            shift = (i / 3);
341
            mod = i % 3;
342
            scale_factor_modshift[i] = mod | (shift << 2);
343
        }
344

    
345
        /* scale factor multiply for layer 1 */
346
        for(i=0;i<15;i++) {
347
            int n, norm;
348
            n = i + 2;
349
            norm = ((INT64_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
350
            scale_factor_mult[i][0] = MULL(FIXR(1.0 * 2.0), norm);
351
            scale_factor_mult[i][1] = MULL(FIXR(0.7937005259 * 2.0), norm);
352
            scale_factor_mult[i][2] = MULL(FIXR(0.6299605249 * 2.0), norm);
353
            dprintf(avctx, "%d: norm=%x s=%x %x %x\n",
354
                    i, norm,
355
                    scale_factor_mult[i][0],
356
                    scale_factor_mult[i][1],
357
                    scale_factor_mult[i][2]);
358
        }
359

    
360
        ff_mpa_synth_init(window);
361

    
362
        /* huffman decode tables */
363
        for(i=1;i<16;i++) {
364
            const HuffTable *h = &mpa_huff_tables[i];
365
            int xsize, x, y;
366
            unsigned int n;
367
            uint8_t  tmp_bits [512];
368
            uint16_t tmp_codes[512];
369

    
370
            memset(tmp_bits , 0, sizeof(tmp_bits ));
371
            memset(tmp_codes, 0, sizeof(tmp_codes));
372

    
373
            xsize = h->xsize;
374
            n = xsize * xsize;
375

    
376
            j = 0;
377
            for(x=0;x<xsize;x++) {
378
                for(y=0;y<xsize;y++){
379
                    tmp_bits [(x << 5) | y | ((x&&y)<<4)]= h->bits [j  ];
380
                    tmp_codes[(x << 5) | y | ((x&&y)<<4)]= h->codes[j++];
381
                }
382
            }
383

    
384
            /* XXX: fail test */
385
            init_vlc(&huff_vlc[i], 7, 512,
386
                     tmp_bits, 1, 1, tmp_codes, 2, 2, 1);
387
        }
388
        for(i=0;i<2;i++) {
389
            init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
390
                     mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1, 1);
391
        }
392

    
393
        for(i=0;i<9;i++) {
394
            k = 0;
395
            for(j=0;j<22;j++) {
396
                band_index_long[i][j] = k;
397
                k += band_size_long[i][j];
398
            }
399
            band_index_long[i][22] = k;
400
        }
401

    
402
        /* compute n ^ (4/3) and store it in mantissa/exp format */
403

    
404
        int_pow_init();
405
        for(i=1;i<TABLE_4_3_SIZE;i++) {
406
            double f, fm;
407
            int e, m;
408
            f = pow((double)(i/4), 4.0 / 3.0) * pow(2, (i&3)*0.25);
409
            fm = frexp(f, &e);
410
            m = (uint32_t)(fm*(1LL<<31) + 0.5);
411
            e+= FRAC_BITS - 31 + 5 - 100;
412

    
413
            /* normalized to FRAC_BITS */
414
            table_4_3_value[i] = m;
415
//            av_log(NULL, AV_LOG_DEBUG, "%d %d %f\n", i, m, pow((double)i, 4.0 / 3.0));
416
            table_4_3_exp[i] = -e;
417
        }
418
        for(i=0; i<512*16; i++){
419
            int exponent= (i>>4);
420
            double f= pow(i&15, 4.0 / 3.0) * pow(2, (exponent-400)*0.25 + FRAC_BITS + 5);
421
            expval_table[exponent][i&15]= llrint(f);
422
            if((i&15)==1)
423
                exp_table[exponent]= llrint(f);
424
        }
425

    
426
        for(i=0;i<7;i++) {
427
            float f;
428
            int v;
429
            if (i != 6) {
430
                f = tan((double)i * M_PI / 12.0);
431
                v = FIXR(f / (1.0 + f));
432
            } else {
433
                v = FIXR(1.0);
434
            }
435
            is_table[0][i] = v;
436
            is_table[1][6 - i] = v;
437
        }
438
        /* invalid values */
439
        for(i=7;i<16;i++)
440
            is_table[0][i] = is_table[1][i] = 0.0;
441

    
442
        for(i=0;i<16;i++) {
443
            double f;
444
            int e, k;
445

    
446
            for(j=0;j<2;j++) {
447
                e = -(j + 1) * ((i + 1) >> 1);
448
                f = pow(2.0, e / 4.0);
449
                k = i & 1;
450
                is_table_lsf[j][k ^ 1][i] = FIXR(f);
451
                is_table_lsf[j][k][i] = FIXR(1.0);
452
                dprintf(avctx, "is_table_lsf %d %d: %x %x\n",
453
                        i, j, is_table_lsf[j][0][i], is_table_lsf[j][1][i]);
454
            }
455
        }
456

    
457
        for(i=0;i<8;i++) {
458
            float ci, cs, ca;
459
            ci = ci_table[i];
460
            cs = 1.0 / sqrt(1.0 + ci * ci);
461
            ca = cs * ci;
462
            csa_table[i][0] = FIXHR(cs/4);
463
            csa_table[i][1] = FIXHR(ca/4);
464
            csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);
465
            csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4);
466
            csa_table_float[i][0] = cs;
467
            csa_table_float[i][1] = ca;
468
            csa_table_float[i][2] = ca + cs;
469
            csa_table_float[i][3] = ca - cs;
470
//            printf("%d %d %d %d\n", FIX(cs), FIX(cs-1), FIX(ca), FIX(cs)-FIX(ca));
471
//            av_log(NULL, AV_LOG_DEBUG,"%f %f %f %f\n", cs, ca, ca+cs, ca-cs);
472
        }
473

    
474
        /* compute mdct windows */
475
        for(i=0;i<36;i++) {
476
            for(j=0; j<4; j++){
477
                double d;
478

    
479
                if(j==2 && i%3 != 1)
480
                    continue;
481

    
482
                d= sin(M_PI * (i + 0.5) / 36.0);
483
                if(j==1){
484
                    if     (i>=30) d= 0;
485
                    else if(i>=24) d= sin(M_PI * (i - 18 + 0.5) / 12.0);
486
                    else if(i>=18) d= 1;
487
                }else if(j==3){
488
                    if     (i<  6) d= 0;
489
                    else if(i< 12) d= sin(M_PI * (i -  6 + 0.5) / 12.0);
490
                    else if(i< 18) d= 1;
491
                }
492
                //merge last stage of imdct into the window coefficients
493
                d*= 0.5 / cos(M_PI*(2*i + 19)/72);
494

    
495
                if(j==2)
496
                    mdct_win[j][i/3] = FIXHR((d / (1<<5)));
497
                else
498
                    mdct_win[j][i  ] = FIXHR((d / (1<<5)));
499
//                av_log(NULL, AV_LOG_DEBUG, "%2d %d %f\n", i,j,d / (1<<5));
500
            }
501
        }
502

    
503
        /* NOTE: we do frequency inversion adter the MDCT by changing
504
           the sign of the right window coefs */
505
        for(j=0;j<4;j++) {
506
            for(i=0;i<36;i+=2) {
507
                mdct_win[j + 4][i] = mdct_win[j][i];
508
                mdct_win[j + 4][i + 1] = -mdct_win[j][i + 1];
509
            }
510
        }
511

    
512
#if defined(DEBUG)
513
        for(j=0;j<8;j++) {
514
            av_log(avctx, AV_LOG_DEBUG, "win%d=\n", j);
515
            for(i=0;i<36;i++)
516
                av_log(avctx, AV_LOG_DEBUG, "%f, ", (double)mdct_win[j][i] / FRAC_ONE);
517
            av_log(avctx, AV_LOG_DEBUG, "\n");
518
        }
519
#endif
520
        init = 1;
521
    }
522

    
523
#ifdef DEBUG
524
    s->frame_count = 0;
525
#endif
526
    if (avctx->codec_id == CODEC_ID_MP3ADU)
527
        s->adu_mode = 1;
528
    return 0;
529
}
530

    
531
/* tab[i][j] = 1.0 / (2.0 * cos(pi*(2*k+1) / 2^(6 - j))) */
532

    
533
/* cos(i*pi/64) */
534

    
535
#define COS0_0  FIXHR(0.50060299823519630134/2)
536
#define COS0_1  FIXHR(0.50547095989754365998/2)
537
#define COS0_2  FIXHR(0.51544730992262454697/2)
538
#define COS0_3  FIXHR(0.53104259108978417447/2)
539
#define COS0_4  FIXHR(0.55310389603444452782/2)
540
#define COS0_5  FIXHR(0.58293496820613387367/2)
541
#define COS0_6  FIXHR(0.62250412303566481615/2)
542
#define COS0_7  FIXHR(0.67480834145500574602/2)
543
#define COS0_8  FIXHR(0.74453627100229844977/2)
544
#define COS0_9  FIXHR(0.83934964541552703873/2)
545
#define COS0_10 FIXHR(0.97256823786196069369/2)
546
#define COS0_11 FIXHR(1.16943993343288495515/4)
547
#define COS0_12 FIXHR(1.48416461631416627724/4)
548
#define COS0_13 FIXHR(2.05778100995341155085/8)
549
#define COS0_14 FIXHR(3.40760841846871878570/8)
550
#define COS0_15 FIXHR(10.19000812354805681150/32)
551

    
552
#define COS1_0 FIXHR(0.50241928618815570551/2)
553
#define COS1_1 FIXHR(0.52249861493968888062/2)
554
#define COS1_2 FIXHR(0.56694403481635770368/2)
555
#define COS1_3 FIXHR(0.64682178335999012954/2)
556
#define COS1_4 FIXHR(0.78815462345125022473/2)
557
#define COS1_5 FIXHR(1.06067768599034747134/4)
558
#define COS1_6 FIXHR(1.72244709823833392782/4)
559
#define COS1_7 FIXHR(5.10114861868916385802/16)
560

    
561
#define COS2_0 FIXHR(0.50979557910415916894/2)
562
#define COS2_1 FIXHR(0.60134488693504528054/2)
563
#define COS2_2 FIXHR(0.89997622313641570463/2)
564
#define COS2_3 FIXHR(2.56291544774150617881/8)
565

    
566
#define COS3_0 FIXHR(0.54119610014619698439/2)
567
#define COS3_1 FIXHR(1.30656296487637652785/4)
568

    
569
#define COS4_0 FIXHR(0.70710678118654752439/2)
570

    
571
/* butterfly operator */
572
#define BF(a, b, c, s)\
573
{\
574
    tmp0 = tab[a] + tab[b];\
575
    tmp1 = tab[a] - tab[b];\
576
    tab[a] = tmp0;\
577
    tab[b] = MULH(tmp1<<(s), c);\
578
}
579

    
580
#define BF1(a, b, c, d)\
581
{\
582
    BF(a, b, COS4_0, 1);\
583
    BF(c, d,-COS4_0, 1);\
584
    tab[c] += tab[d];\
585
}
586

    
587
#define BF2(a, b, c, d)\
588
{\
589
    BF(a, b, COS4_0, 1);\
590
    BF(c, d,-COS4_0, 1);\
591
    tab[c] += tab[d];\
592
    tab[a] += tab[c];\
593
    tab[c] += tab[b];\
594
    tab[b] += tab[d];\
595
}
596

    
597
#define ADD(a, b) tab[a] += tab[b]
598

    
599
/* DCT32 without 1/sqrt(2) coef zero scaling. */
600
static void dct32(int32_t *out, int32_t *tab)
601
{
602
    int tmp0, tmp1;
603

    
604
    /* pass 1 */
605
    BF( 0, 31, COS0_0 , 1);
606
    BF(15, 16, COS0_15, 5);
607
    /* pass 2 */
608
    BF( 0, 15, COS1_0 , 1);
609
    BF(16, 31,-COS1_0 , 1);
610
    /* pass 1 */
611
    BF( 7, 24, COS0_7 , 1);
612
    BF( 8, 23, COS0_8 , 1);
613
    /* pass 2 */
614
    BF( 7,  8, COS1_7 , 4);
615
    BF(23, 24,-COS1_7 , 4);
616
    /* pass 3 */
617
    BF( 0,  7, COS2_0 , 1);
618
    BF( 8, 15,-COS2_0 , 1);
619
    BF(16, 23, COS2_0 , 1);
620
    BF(24, 31,-COS2_0 , 1);
621
    /* pass 1 */
622
    BF( 3, 28, COS0_3 , 1);
623
    BF(12, 19, COS0_12, 2);
624
    /* pass 2 */
625
    BF( 3, 12, COS1_3 , 1);
626
    BF(19, 28,-COS1_3 , 1);
627
    /* pass 1 */
628
    BF( 4, 27, COS0_4 , 1);
629
    BF(11, 20, COS0_11, 2);
630
    /* pass 2 */
631
    BF( 4, 11, COS1_4 , 1);
632
    BF(20, 27,-COS1_4 , 1);
633
    /* pass 3 */
634
    BF( 3,  4, COS2_3 , 3);
635
    BF(11, 12,-COS2_3 , 3);
636
    BF(19, 20, COS2_3 , 3);
637
    BF(27, 28,-COS2_3 , 3);
638
    /* pass 4 */
639
    BF( 0,  3, COS3_0 , 1);
640
    BF( 4,  7,-COS3_0 , 1);
641
    BF( 8, 11, COS3_0 , 1);
642
    BF(12, 15,-COS3_0 , 1);
643
    BF(16, 19, COS3_0 , 1);
644
    BF(20, 23,-COS3_0 , 1);
645
    BF(24, 27, COS3_0 , 1);
646
    BF(28, 31,-COS3_0 , 1);
647

    
648

    
649

    
650
    /* pass 1 */
651
    BF( 1, 30, COS0_1 , 1);
652
    BF(14, 17, COS0_14, 3);
653
    /* pass 2 */
654
    BF( 1, 14, COS1_1 , 1);
655
    BF(17, 30,-COS1_1 , 1);
656
    /* pass 1 */
657
    BF( 6, 25, COS0_6 , 1);
658
    BF( 9, 22, COS0_9 , 1);
659
    /* pass 2 */
660
    BF( 6,  9, COS1_6 , 2);
661
    BF(22, 25,-COS1_6 , 2);
662
    /* pass 3 */
663
    BF( 1,  6, COS2_1 , 1);
664
    BF( 9, 14,-COS2_1 , 1);
665
    BF(17, 22, COS2_1 , 1);
666
    BF(25, 30,-COS2_1 , 1);
667

    
668
    /* pass 1 */
669
    BF( 2, 29, COS0_2 , 1);
670
    BF(13, 18, COS0_13, 3);
671
    /* pass 2 */
672
    BF( 2, 13, COS1_2 , 1);
673
    BF(18, 29,-COS1_2 , 1);
674
    /* pass 1 */
675
    BF( 5, 26, COS0_5 , 1);
676
    BF(10, 21, COS0_10, 1);
677
    /* pass 2 */
678
    BF( 5, 10, COS1_5 , 2);
679
    BF(21, 26,-COS1_5 , 2);
680
    /* pass 3 */
681
    BF( 2,  5, COS2_2 , 1);
682
    BF(10, 13,-COS2_2 , 1);
683
    BF(18, 21, COS2_2 , 1);
684
    BF(26, 29,-COS2_2 , 1);
685
    /* pass 4 */
686
    BF( 1,  2, COS3_1 , 2);
687
    BF( 5,  6,-COS3_1 , 2);
688
    BF( 9, 10, COS3_1 , 2);
689
    BF(13, 14,-COS3_1 , 2);
690
    BF(17, 18, COS3_1 , 2);
691
    BF(21, 22,-COS3_1 , 2);
692
    BF(25, 26, COS3_1 , 2);
693
    BF(29, 30,-COS3_1 , 2);
694

    
695
    /* pass 5 */
696
    BF1( 0,  1,  2,  3);
697
    BF2( 4,  5,  6,  7);
698
    BF1( 8,  9, 10, 11);
699
    BF2(12, 13, 14, 15);
700
    BF1(16, 17, 18, 19);
701
    BF2(20, 21, 22, 23);
702
    BF1(24, 25, 26, 27);
703
    BF2(28, 29, 30, 31);
704

    
705
    /* pass 6 */
706

    
707
    ADD( 8, 12);
708
    ADD(12, 10);
709
    ADD(10, 14);
710
    ADD(14,  9);
711
    ADD( 9, 13);
712
    ADD(13, 11);
713
    ADD(11, 15);
714

    
715
    out[ 0] = tab[0];
716
    out[16] = tab[1];
717
    out[ 8] = tab[2];
718
    out[24] = tab[3];
719
    out[ 4] = tab[4];
720
    out[20] = tab[5];
721
    out[12] = tab[6];
722
    out[28] = tab[7];
723
    out[ 2] = tab[8];
724
    out[18] = tab[9];
725
    out[10] = tab[10];
726
    out[26] = tab[11];
727
    out[ 6] = tab[12];
728
    out[22] = tab[13];
729
    out[14] = tab[14];
730
    out[30] = tab[15];
731

    
732
    ADD(24, 28);
733
    ADD(28, 26);
734
    ADD(26, 30);
735
    ADD(30, 25);
736
    ADD(25, 29);
737
    ADD(29, 27);
738
    ADD(27, 31);
739

    
740
    out[ 1] = tab[16] + tab[24];
741
    out[17] = tab[17] + tab[25];
742
    out[ 9] = tab[18] + tab[26];
743
    out[25] = tab[19] + tab[27];
744
    out[ 5] = tab[20] + tab[28];
745
    out[21] = tab[21] + tab[29];
746
    out[13] = tab[22] + tab[30];
747
    out[29] = tab[23] + tab[31];
748
    out[ 3] = tab[24] + tab[20];
749
    out[19] = tab[25] + tab[21];
750
    out[11] = tab[26] + tab[22];
751
    out[27] = tab[27] + tab[23];
752
    out[ 7] = tab[28] + tab[18];
753
    out[23] = tab[29] + tab[19];
754
    out[15] = tab[30] + tab[17];
755
    out[31] = tab[31];
756
}
757

    
758
#if FRAC_BITS <= 15
759

    
760
static inline int round_sample(int *sum)
761
{
762
    int sum1;
763
    sum1 = (*sum) >> OUT_SHIFT;
764
    *sum &= (1<<OUT_SHIFT)-1;
765
    if (sum1 < OUT_MIN)
766
        sum1 = OUT_MIN;
767
    else if (sum1 > OUT_MAX)
768
        sum1 = OUT_MAX;
769
    return sum1;
770
}
771

    
772
/* signed 16x16 -> 32 multiply add accumulate */
773
#define MACS(rt, ra, rb) MAC16(rt, ra, rb)
774

    
775
/* signed 16x16 -> 32 multiply */
776
#define MULS(ra, rb) MUL16(ra, rb)
777

    
778
#else
779

    
780
static inline int round_sample(int64_t *sum)
781
{
782
    int sum1;
783
    sum1 = (int)((*sum) >> OUT_SHIFT);
784
    *sum &= (1<<OUT_SHIFT)-1;
785
    if (sum1 < OUT_MIN)
786
        sum1 = OUT_MIN;
787
    else if (sum1 > OUT_MAX)
788
        sum1 = OUT_MAX;
789
    return sum1;
790
}
791

    
792
#   define MULS(ra, rb) MUL64(ra, rb)
793
#endif
794

    
795
#define SUM8(sum, op, w, p) \
796
{                                               \
797
    sum op MULS((w)[0 * 64], p[0 * 64]);\
798
    sum op MULS((w)[1 * 64], p[1 * 64]);\
799
    sum op MULS((w)[2 * 64], p[2 * 64]);\
800
    sum op MULS((w)[3 * 64], p[3 * 64]);\
801
    sum op MULS((w)[4 * 64], p[4 * 64]);\
802
    sum op MULS((w)[5 * 64], p[5 * 64]);\
803
    sum op MULS((w)[6 * 64], p[6 * 64]);\
804
    sum op MULS((w)[7 * 64], p[7 * 64]);\
805
}
806

    
807
#define SUM8P2(sum1, op1, sum2, op2, w1, w2, p) \
808
{                                               \
809
    int tmp;\
810
    tmp = p[0 * 64];\
811
    sum1 op1 MULS((w1)[0 * 64], tmp);\
812
    sum2 op2 MULS((w2)[0 * 64], tmp);\
813
    tmp = p[1 * 64];\
814
    sum1 op1 MULS((w1)[1 * 64], tmp);\
815
    sum2 op2 MULS((w2)[1 * 64], tmp);\
816
    tmp = p[2 * 64];\
817
    sum1 op1 MULS((w1)[2 * 64], tmp);\
818
    sum2 op2 MULS((w2)[2 * 64], tmp);\
819
    tmp = p[3 * 64];\
820
    sum1 op1 MULS((w1)[3 * 64], tmp);\
821
    sum2 op2 MULS((w2)[3 * 64], tmp);\
822
    tmp = p[4 * 64];\
823
    sum1 op1 MULS((w1)[4 * 64], tmp);\
824
    sum2 op2 MULS((w2)[4 * 64], tmp);\
825
    tmp = p[5 * 64];\
826
    sum1 op1 MULS((w1)[5 * 64], tmp);\
827
    sum2 op2 MULS((w2)[5 * 64], tmp);\
828
    tmp = p[6 * 64];\
829
    sum1 op1 MULS((w1)[6 * 64], tmp);\
830
    sum2 op2 MULS((w2)[6 * 64], tmp);\
831
    tmp = p[7 * 64];\
832
    sum1 op1 MULS((w1)[7 * 64], tmp);\
833
    sum2 op2 MULS((w2)[7 * 64], tmp);\
834
}
835

    
836
void ff_mpa_synth_init(MPA_INT *window)
837
{
838
    int i;
839

    
840
    /* max = 18760, max sum over all 16 coefs : 44736 */
841
    for(i=0;i<257;i++) {
842
        int v;
843
        v = ff_mpa_enwindow[i];
844
#if WFRAC_BITS < 16
845
        v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS);
846
#endif
847
        window[i] = v;
848
        if ((i & 63) != 0)
849
            v = -v;
850
        if (i != 0)
851
            window[512 - i] = v;
852
    }
853
}
854

    
855
/* 32 sub band synthesis filter. Input: 32 sub band samples, Output:
856
   32 samples. */
857
/* XXX: optimize by avoiding ring buffer usage */
858
void ff_mpa_synth_filter(MPA_INT *synth_buf_ptr, int *synth_buf_offset,
859
                         MPA_INT *window, int *dither_state,
860
                         OUT_INT *samples, int incr,
861
                         int32_t sb_samples[SBLIMIT])
862
{
863
    int32_t tmp[32];
864
    register MPA_INT *synth_buf;
865
    register const MPA_INT *w, *w2, *p;
866
    int j, offset, v;
867
    OUT_INT *samples2;
868
#if FRAC_BITS <= 15
869
    int sum, sum2;
870
#else
871
    int64_t sum, sum2;
872
#endif
873

    
874
    dct32(tmp, sb_samples);
875

    
876
    offset = *synth_buf_offset;
877
    synth_buf = synth_buf_ptr + offset;
878

    
879
    for(j=0;j<32;j++) {
880
        v = tmp[j];
881
#if FRAC_BITS <= 15
882
        /* NOTE: can cause a loss in precision if very high amplitude
883
           sound */
884
        v = av_clip_int16(v);
885
#endif
886
        synth_buf[j] = v;
887
    }
888
    /* copy to avoid wrap */
889
    memcpy(synth_buf + 512, synth_buf, 32 * sizeof(MPA_INT));
890

    
891
    samples2 = samples + 31 * incr;
892
    w = window;
893
    w2 = window + 31;
894

    
895
    sum = *dither_state;
896
    p = synth_buf + 16;
897
    SUM8(sum, +=, w, p);
898
    p = synth_buf + 48;
899
    SUM8(sum, -=, w + 32, p);
900
    *samples = round_sample(&sum);
901
    samples += incr;
902
    w++;
903

    
904
    /* we calculate two samples at the same time to avoid one memory
905
       access per two sample */
906
    for(j=1;j<16;j++) {
907
        sum2 = 0;
908
        p = synth_buf + 16 + j;
909
        SUM8P2(sum, +=, sum2, -=, w, w2, p);
910
        p = synth_buf + 48 - j;
911
        SUM8P2(sum, -=, sum2, -=, w + 32, w2 + 32, p);
912

    
913
        *samples = round_sample(&sum);
914
        samples += incr;
915
        sum += sum2;
916
        *samples2 = round_sample(&sum);
917
        samples2 -= incr;
918
        w++;
919
        w2--;
920
    }
921

    
922
    p = synth_buf + 32;
923
    SUM8(sum, -=, w + 32, p);
924
    *samples = round_sample(&sum);
925
    *dither_state= sum;
926

    
927
    offset = (offset - 32) & 511;
928
    *synth_buf_offset = offset;
929
}
930

    
931
#define C3 FIXHR(0.86602540378443864676/2)
932

    
933
/* 0.5 / cos(pi*(2*i+1)/36) */
934
static const int icos36[9] = {
935
    FIXR(0.50190991877167369479),
936
    FIXR(0.51763809020504152469), //0
937
    FIXR(0.55168895948124587824),
938
    FIXR(0.61038729438072803416),
939
    FIXR(0.70710678118654752439), //1
940
    FIXR(0.87172339781054900991),
941
    FIXR(1.18310079157624925896),
942
    FIXR(1.93185165257813657349), //2
943
    FIXR(5.73685662283492756461),
944
};
945

    
946
/* 0.5 / cos(pi*(2*i+1)/36) */
947
static const int icos36h[9] = {
948
    FIXHR(0.50190991877167369479/2),
949
    FIXHR(0.51763809020504152469/2), //0
950
    FIXHR(0.55168895948124587824/2),
951
    FIXHR(0.61038729438072803416/2),
952
    FIXHR(0.70710678118654752439/2), //1
953
    FIXHR(0.87172339781054900991/2),
954
    FIXHR(1.18310079157624925896/4),
955
    FIXHR(1.93185165257813657349/4), //2
956
//    FIXHR(5.73685662283492756461),
957
};
958

    
959
/* 12 points IMDCT. We compute it "by hand" by factorizing obvious
960
   cases. */
961
static void imdct12(int *out, int *in)
962
{
963
    int in0, in1, in2, in3, in4, in5, t1, t2;
964

    
965
    in0= in[0*3];
966
    in1= in[1*3] + in[0*3];
967
    in2= in[2*3] + in[1*3];
968
    in3= in[3*3] + in[2*3];
969
    in4= in[4*3] + in[3*3];
970
    in5= in[5*3] + in[4*3];
971
    in5 += in3;
972
    in3 += in1;
973

    
974
    in2= MULH(2*in2, C3);
975
    in3= MULH(4*in3, C3);
976

    
977
    t1 = in0 - in4;
978
    t2 = MULH(2*(in1 - in5), icos36h[4]);
979

    
980
    out[ 7]=
981
    out[10]= t1 + t2;
982
    out[ 1]=
983
    out[ 4]= t1 - t2;
984

    
985
    in0 += in4>>1;
986
    in4 = in0 + in2;
987
    in5 += 2*in1;
988
    in1 = MULH(in5 + in3, icos36h[1]);
989
    out[ 8]=
990
    out[ 9]= in4 + in1;
991
    out[ 2]=
992
    out[ 3]= in4 - in1;
993

    
994
    in0 -= in2;
995
    in5 = MULH(2*(in5 - in3), icos36h[7]);
996
    out[ 0]=
997
    out[ 5]= in0 - in5;
998
    out[ 6]=
999
    out[11]= in0 + in5;
1000
}
1001

    
1002
/* cos(pi*i/18) */
1003
#define C1 FIXHR(0.98480775301220805936/2)
1004
#define C2 FIXHR(0.93969262078590838405/2)
1005
#define C3 FIXHR(0.86602540378443864676/2)
1006
#define C4 FIXHR(0.76604444311897803520/2)
1007
#define C5 FIXHR(0.64278760968653932632/2)
1008
#define C6 FIXHR(0.5/2)
1009
#define C7 FIXHR(0.34202014332566873304/2)
1010
#define C8 FIXHR(0.17364817766693034885/2)
1011

    
1012

    
1013
/* using Lee like decomposition followed by hand coded 9 points DCT */
1014
static void imdct36(int *out, int *buf, int *in, int *win)
1015
{
1016
    int i, j, t0, t1, t2, t3, s0, s1, s2, s3;
1017
    int tmp[18], *tmp1, *in1;
1018

    
1019
    for(i=17;i>=1;i--)
1020
        in[i] += in[i-1];
1021
    for(i=17;i>=3;i-=2)
1022
        in[i] += in[i-2];
1023

    
1024
    for(j=0;j<2;j++) {
1025
        tmp1 = tmp + j;
1026
        in1 = in + j;
1027
#if 0
1028
//more accurate but slower
1029
        int64_t t0, t1, t2, t3;
1030
        t2 = in1[2*4] + in1[2*8] - in1[2*2];
1031

1032
        t3 = (in1[2*0] + (int64_t)(in1[2*6]>>1))<<32;
1033
        t1 = in1[2*0] - in1[2*6];
1034
        tmp1[ 6] = t1 - (t2>>1);
1035
        tmp1[16] = t1 + t2;
1036

1037
        t0 = MUL64(2*(in1[2*2] + in1[2*4]),    C2);
1038
        t1 = MUL64(   in1[2*4] - in1[2*8] , -2*C8);
1039
        t2 = MUL64(2*(in1[2*2] + in1[2*8]),   -C4);
1040

1041
        tmp1[10] = (t3 - t0 - t2) >> 32;
1042
        tmp1[ 2] = (t3 + t0 + t1) >> 32;
1043
        tmp1[14] = (t3 + t2 - t1) >> 32;
1044

1045
        tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
1046
        t2 = MUL64(2*(in1[2*1] + in1[2*5]),    C1);
1047
        t3 = MUL64(   in1[2*5] - in1[2*7] , -2*C7);
1048
        t0 = MUL64(2*in1[2*3], C3);
1049

1050
        t1 = MUL64(2*(in1[2*1] + in1[2*7]),   -C5);
1051

1052
        tmp1[ 0] = (t2 + t3 + t0) >> 32;
1053
        tmp1[12] = (t2 + t1 - t0) >> 32;
1054
        tmp1[ 8] = (t3 - t1 - t0) >> 32;
1055
#else
1056
        t2 = in1[2*4] + in1[2*8] - in1[2*2];
1057

    
1058
        t3 = in1[2*0] + (in1[2*6]>>1);
1059
        t1 = in1[2*0] - in1[2*6];
1060
        tmp1[ 6] = t1 - (t2>>1);
1061
        tmp1[16] = t1 + t2;
1062

    
1063
        t0 = MULH(2*(in1[2*2] + in1[2*4]),    C2);
1064
        t1 = MULH(   in1[2*4] - in1[2*8] , -2*C8);
1065
        t2 = MULH(2*(in1[2*2] + in1[2*8]),   -C4);
1066

    
1067
        tmp1[10] = t3 - t0 - t2;
1068
        tmp1[ 2] = t3 + t0 + t1;
1069
        tmp1[14] = t3 + t2 - t1;
1070

    
1071
        tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
1072
        t2 = MULH(2*(in1[2*1] + in1[2*5]),    C1);
1073
        t3 = MULH(   in1[2*5] - in1[2*7] , -2*C7);
1074
        t0 = MULH(2*in1[2*3], C3);
1075

    
1076
        t1 = MULH(2*(in1[2*1] + in1[2*7]),   -C5);
1077

    
1078
        tmp1[ 0] = t2 + t3 + t0;
1079
        tmp1[12] = t2 + t1 - t0;
1080
        tmp1[ 8] = t3 - t1 - t0;
1081
#endif
1082
    }
1083

    
1084
    i = 0;
1085
    for(j=0;j<4;j++) {
1086
        t0 = tmp[i];
1087
        t1 = tmp[i + 2];
1088
        s0 = t1 + t0;
1089
        s2 = t1 - t0;
1090

    
1091
        t2 = tmp[i + 1];
1092
        t3 = tmp[i + 3];
1093
        s1 = MULH(2*(t3 + t2), icos36h[j]);
1094
        s3 = MULL(t3 - t2, icos36[8 - j]);
1095

    
1096
        t0 = s0 + s1;
1097
        t1 = s0 - s1;
1098
        out[(9 + j)*SBLIMIT] =  MULH(t1, win[9 + j]) + buf[9 + j];
1099
        out[(8 - j)*SBLIMIT] =  MULH(t1, win[8 - j]) + buf[8 - j];
1100
        buf[9 + j] = MULH(t0, win[18 + 9 + j]);
1101
        buf[8 - j] = MULH(t0, win[18 + 8 - j]);
1102

    
1103
        t0 = s2 + s3;
1104
        t1 = s2 - s3;
1105
        out[(9 + 8 - j)*SBLIMIT] =  MULH(t1, win[9 + 8 - j]) + buf[9 + 8 - j];
1106
        out[(        j)*SBLIMIT] =  MULH(t1, win[        j]) + buf[        j];
1107
        buf[9 + 8 - j] = MULH(t0, win[18 + 9 + 8 - j]);
1108
        buf[      + j] = MULH(t0, win[18         + j]);
1109
        i += 4;
1110
    }
1111

    
1112
    s0 = tmp[16];
1113
    s1 = MULH(2*tmp[17], icos36h[4]);
1114
    t0 = s0 + s1;
1115
    t1 = s0 - s1;
1116
    out[(9 + 4)*SBLIMIT] =  MULH(t1, win[9 + 4]) + buf[9 + 4];
1117
    out[(8 - 4)*SBLIMIT] =  MULH(t1, win[8 - 4]) + buf[8 - 4];
1118
    buf[9 + 4] = MULH(t0, win[18 + 9 + 4]);
1119
    buf[8 - 4] = MULH(t0, win[18 + 8 - 4]);
1120
}
1121

    
1122
/* return the number of decoded frames */
1123
static int mp_decode_layer1(MPADecodeContext *s)
1124
{
1125
    int bound, i, v, n, ch, j, mant;
1126
    uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
1127
    uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
1128

    
1129
    if (s->mode == MPA_JSTEREO)
1130
        bound = (s->mode_ext + 1) * 4;
1131
    else
1132
        bound = SBLIMIT;
1133

    
1134
    /* allocation bits */
1135
    for(i=0;i<bound;i++) {
1136
        for(ch=0;ch<s->nb_channels;ch++) {
1137
            allocation[ch][i] = get_bits(&s->gb, 4);
1138
        }
1139
    }
1140
    for(i=bound;i<SBLIMIT;i++) {
1141
        allocation[0][i] = get_bits(&s->gb, 4);
1142
    }
1143

    
1144
    /* scale factors */
1145
    for(i=0;i<bound;i++) {
1146
        for(ch=0;ch<s->nb_channels;ch++) {
1147
            if (allocation[ch][i])
1148
                scale_factors[ch][i] = get_bits(&s->gb, 6);
1149
        }
1150
    }
1151
    for(i=bound;i<SBLIMIT;i++) {
1152
        if (allocation[0][i]) {
1153
            scale_factors[0][i] = get_bits(&s->gb, 6);
1154
            scale_factors[1][i] = get_bits(&s->gb, 6);
1155
        }
1156
    }
1157

    
1158
    /* compute samples */
1159
    for(j=0;j<12;j++) {
1160
        for(i=0;i<bound;i++) {
1161
            for(ch=0;ch<s->nb_channels;ch++) {
1162
                n = allocation[ch][i];
1163
                if (n) {
1164
                    mant = get_bits(&s->gb, n + 1);
1165
                    v = l1_unscale(n, mant, scale_factors[ch][i]);
1166
                } else {
1167
                    v = 0;
1168
                }
1169
                s->sb_samples[ch][j][i] = v;
1170
            }
1171
        }
1172
        for(i=bound;i<SBLIMIT;i++) {
1173
            n = allocation[0][i];
1174
            if (n) {
1175
                mant = get_bits(&s->gb, n + 1);
1176
                v = l1_unscale(n, mant, scale_factors[0][i]);
1177
                s->sb_samples[0][j][i] = v;
1178
                v = l1_unscale(n, mant, scale_factors[1][i]);
1179
                s->sb_samples[1][j][i] = v;
1180
            } else {
1181
                s->sb_samples[0][j][i] = 0;
1182
                s->sb_samples[1][j][i] = 0;
1183
            }
1184
        }
1185
    }
1186
    return 12;
1187
}
1188

    
1189
static int mp_decode_layer2(MPADecodeContext *s)
1190
{
1191
    int sblimit; /* number of used subbands */
1192
    const unsigned char *alloc_table;
1193
    int table, bit_alloc_bits, i, j, ch, bound, v;
1194
    unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
1195
    unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
1196
    unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
1197
    int scale, qindex, bits, steps, k, l, m, b;
1198

    
1199
    /* select decoding table */
1200
    table = ff_mpa_l2_select_table(s->bit_rate / 1000, s->nb_channels,
1201
                            s->sample_rate, s->lsf);
1202
    sblimit = ff_mpa_sblimit_table[table];
1203
    alloc_table = ff_mpa_alloc_tables[table];
1204

    
1205
    if (s->mode == MPA_JSTEREO)
1206
        bound = (s->mode_ext + 1) * 4;
1207
    else
1208
        bound = sblimit;
1209

    
1210
    dprintf(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit);
1211

    
1212
    /* sanity check */
1213
    if( bound > sblimit ) bound = sblimit;
1214

    
1215
    /* parse bit allocation */
1216
    j = 0;
1217
    for(i=0;i<bound;i++) {
1218
        bit_alloc_bits = alloc_table[j];
1219
        for(ch=0;ch<s->nb_channels;ch++) {
1220
            bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
1221
        }
1222
        j += 1 << bit_alloc_bits;
1223
    }
1224
    for(i=bound;i<sblimit;i++) {
1225
        bit_alloc_bits = alloc_table[j];
1226
        v = get_bits(&s->gb, bit_alloc_bits);
1227
        bit_alloc[0][i] = v;
1228
        bit_alloc[1][i] = v;
1229
        j += 1 << bit_alloc_bits;
1230
    }
1231

    
1232
#ifdef DEBUG
1233
    {
1234
        for(ch=0;ch<s->nb_channels;ch++) {
1235
            for(i=0;i<sblimit;i++)
1236
                dprintf(s->avctx, " %d", bit_alloc[ch][i]);
1237
            dprintf(s->avctx, "\n");
1238
        }
1239
    }
1240
#endif
1241

    
1242
    /* scale codes */
1243
    for(i=0;i<sblimit;i++) {
1244
        for(ch=0;ch<s->nb_channels;ch++) {
1245
            if (bit_alloc[ch][i])
1246
                scale_code[ch][i] = get_bits(&s->gb, 2);
1247
        }
1248
    }
1249

    
1250
    /* scale factors */
1251
    for(i=0;i<sblimit;i++) {
1252
        for(ch=0;ch<s->nb_channels;ch++) {
1253
            if (bit_alloc[ch][i]) {
1254
                sf = scale_factors[ch][i];
1255
                switch(scale_code[ch][i]) {
1256
                default:
1257
                case 0:
1258
                    sf[0] = get_bits(&s->gb, 6);
1259
                    sf[1] = get_bits(&s->gb, 6);
1260
                    sf[2] = get_bits(&s->gb, 6);
1261
                    break;
1262
                case 2:
1263
                    sf[0] = get_bits(&s->gb, 6);
1264
                    sf[1] = sf[0];
1265
                    sf[2] = sf[0];
1266
                    break;
1267
                case 1:
1268
                    sf[0] = get_bits(&s->gb, 6);
1269
                    sf[2] = get_bits(&s->gb, 6);
1270
                    sf[1] = sf[0];
1271
                    break;
1272
                case 3:
1273
                    sf[0] = get_bits(&s->gb, 6);
1274
                    sf[2] = get_bits(&s->gb, 6);
1275
                    sf[1] = sf[2];
1276
                    break;
1277
                }
1278
            }
1279
        }
1280
    }
1281

    
1282
#ifdef DEBUG
1283
    for(ch=0;ch<s->nb_channels;ch++) {
1284
        for(i=0;i<sblimit;i++) {
1285
            if (bit_alloc[ch][i]) {
1286
                sf = scale_factors[ch][i];
1287
                dprintf(s->avctx, " %d %d %d", sf[0], sf[1], sf[2]);
1288
            } else {
1289
                dprintf(s->avctx, " -");
1290
            }
1291
        }
1292
        dprintf(s->avctx, "\n");
1293
    }
1294
#endif
1295

    
1296
    /* samples */
1297
    for(k=0;k<3;k++) {
1298
        for(l=0;l<12;l+=3) {
1299
            j = 0;
1300
            for(i=0;i<bound;i++) {
1301
                bit_alloc_bits = alloc_table[j];
1302
                for(ch=0;ch<s->nb_channels;ch++) {
1303
                    b = bit_alloc[ch][i];
1304
                    if (b) {
1305
                        scale = scale_factors[ch][i][k];
1306
                        qindex = alloc_table[j+b];
1307
                        bits = ff_mpa_quant_bits[qindex];
1308
                        if (bits < 0) {
1309
                            /* 3 values at the same time */
1310
                            v = get_bits(&s->gb, -bits);
1311
                            steps = ff_mpa_quant_steps[qindex];
1312
                            s->sb_samples[ch][k * 12 + l + 0][i] =
1313
                                l2_unscale_group(steps, v % steps, scale);
1314
                            v = v / steps;
1315
                            s->sb_samples[ch][k * 12 + l + 1][i] =
1316
                                l2_unscale_group(steps, v % steps, scale);
1317
                            v = v / steps;
1318
                            s->sb_samples[ch][k * 12 + l + 2][i] =
1319
                                l2_unscale_group(steps, v, scale);
1320
                        } else {
1321
                            for(m=0;m<3;m++) {
1322
                                v = get_bits(&s->gb, bits);
1323
                                v = l1_unscale(bits - 1, v, scale);
1324
                                s->sb_samples[ch][k * 12 + l + m][i] = v;
1325
                            }
1326
                        }
1327
                    } else {
1328
                        s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1329
                        s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1330
                        s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1331
                    }
1332
                }
1333
                /* next subband in alloc table */
1334
                j += 1 << bit_alloc_bits;
1335
            }
1336
            /* XXX: find a way to avoid this duplication of code */
1337
            for(i=bound;i<sblimit;i++) {
1338
                bit_alloc_bits = alloc_table[j];
1339
                b = bit_alloc[0][i];
1340
                if (b) {
1341
                    int mant, scale0, scale1;
1342
                    scale0 = scale_factors[0][i][k];
1343
                    scale1 = scale_factors[1][i][k];
1344
                    qindex = alloc_table[j+b];
1345
                    bits = ff_mpa_quant_bits[qindex];
1346
                    if (bits < 0) {
1347
                        /* 3 values at the same time */
1348
                        v = get_bits(&s->gb, -bits);
1349
                        steps = ff_mpa_quant_steps[qindex];
1350
                        mant = v % steps;
1351
                        v = v / steps;
1352
                        s->sb_samples[0][k * 12 + l + 0][i] =
1353
                            l2_unscale_group(steps, mant, scale0);
1354
                        s->sb_samples[1][k * 12 + l + 0][i] =
1355
                            l2_unscale_group(steps, mant, scale1);
1356
                        mant = v % steps;
1357
                        v = v / steps;
1358
                        s->sb_samples[0][k * 12 + l + 1][i] =
1359
                            l2_unscale_group(steps, mant, scale0);
1360
                        s->sb_samples[1][k * 12 + l + 1][i] =
1361
                            l2_unscale_group(steps, mant, scale1);
1362
                        s->sb_samples[0][k * 12 + l + 2][i] =
1363
                            l2_unscale_group(steps, v, scale0);
1364
                        s->sb_samples[1][k * 12 + l + 2][i] =
1365
                            l2_unscale_group(steps, v, scale1);
1366
                    } else {
1367
                        for(m=0;m<3;m++) {
1368
                            mant = get_bits(&s->gb, bits);
1369
                            s->sb_samples[0][k * 12 + l + m][i] =
1370
                                l1_unscale(bits - 1, mant, scale0);
1371
                            s->sb_samples[1][k * 12 + l + m][i] =
1372
                                l1_unscale(bits - 1, mant, scale1);
1373
                        }
1374
                    }
1375
                } else {
1376
                    s->sb_samples[0][k * 12 + l + 0][i] = 0;
1377
                    s->sb_samples[0][k * 12 + l + 1][i] = 0;
1378
                    s->sb_samples[0][k * 12 + l + 2][i] = 0;
1379
                    s->sb_samples[1][k * 12 + l + 0][i] = 0;
1380
                    s->sb_samples[1][k * 12 + l + 1][i] = 0;
1381
                    s->sb_samples[1][k * 12 + l + 2][i] = 0;
1382
                }
1383
                /* next subband in alloc table */
1384
                j += 1 << bit_alloc_bits;
1385
            }
1386
            /* fill remaining samples to zero */
1387
            for(i=sblimit;i<SBLIMIT;i++) {
1388
                for(ch=0;ch<s->nb_channels;ch++) {
1389
                    s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1390
                    s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1391
                    s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1392
                }
1393
            }
1394
        }
1395
    }
1396
    return 3 * 12;
1397
}
1398

    
1399
static inline void lsf_sf_expand(int *slen,
1400
                                 int sf, int n1, int n2, int n3)
1401
{
1402
    if (n3) {
1403
        slen[3] = sf % n3;
1404
        sf /= n3;
1405
    } else {
1406
        slen[3] = 0;
1407
    }
1408
    if (n2) {
1409
        slen[2] = sf % n2;
1410
        sf /= n2;
1411
    } else {
1412
        slen[2] = 0;
1413
    }
1414
    slen[1] = sf % n1;
1415
    sf /= n1;
1416
    slen[0] = sf;
1417
}
1418

    
1419
static void exponents_from_scale_factors(MPADecodeContext *s,
1420
                                         GranuleDef *g,
1421
                                         int16_t *exponents)
1422
{
1423
    const uint8_t *bstab, *pretab;
1424
    int len, i, j, k, l, v0, shift, gain, gains[3];
1425
    int16_t *exp_ptr;
1426

    
1427
    exp_ptr = exponents;
1428
    gain = g->global_gain - 210;
1429
    shift = g->scalefac_scale + 1;
1430

    
1431
    bstab = band_size_long[s->sample_rate_index];
1432
    pretab = mpa_pretab[g->preflag];
1433
    for(i=0;i<g->long_end;i++) {
1434
        v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400;
1435
        len = bstab[i];
1436
        for(j=len;j>0;j--)
1437
            *exp_ptr++ = v0;
1438
    }
1439

    
1440
    if (g->short_start < 13) {
1441
        bstab = band_size_short[s->sample_rate_index];
1442
        gains[0] = gain - (g->subblock_gain[0] << 3);
1443
        gains[1] = gain - (g->subblock_gain[1] << 3);
1444
        gains[2] = gain - (g->subblock_gain[2] << 3);
1445
        k = g->long_end;
1446
        for(i=g->short_start;i<13;i++) {
1447
            len = bstab[i];
1448
            for(l=0;l<3;l++) {
1449
                v0 = gains[l] - (g->scale_factors[k++] << shift) + 400;
1450
                for(j=len;j>0;j--)
1451
                *exp_ptr++ = v0;
1452
            }
1453
        }
1454
    }
1455
}
1456

    
1457
/* handle n = 0 too */
1458
static inline int get_bitsz(GetBitContext *s, int n)
1459
{
1460
    if (n == 0)
1461
        return 0;
1462
    else
1463
        return get_bits(s, n);
1464
}
1465

    
1466

    
1467
static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos, int *end_pos2){
1468
    if(s->in_gb.buffer && *pos >= s->gb.size_in_bits){
1469
        s->gb= s->in_gb;
1470
        s->in_gb.buffer=NULL;
1471
        assert((get_bits_count(&s->gb) & 7) == 0);
1472
        skip_bits_long(&s->gb, *pos - *end_pos);
1473
        *end_pos2=
1474
        *end_pos= *end_pos2 + get_bits_count(&s->gb) - *pos;
1475
        *pos= get_bits_count(&s->gb);
1476
    }
1477
}
1478

    
1479
static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
1480
                          int16_t *exponents, int end_pos2)
1481
{
1482
    int s_index;
1483
    int i;
1484
    int last_pos, bits_left;
1485
    VLC *vlc;
1486
    int end_pos= FFMIN(end_pos2, s->gb.size_in_bits);
1487

    
1488
    /* low frequencies (called big values) */
1489
    s_index = 0;
1490
    for(i=0;i<3;i++) {
1491
        int j, k, l, linbits;
1492
        j = g->region_size[i];
1493
        if (j == 0)
1494
            continue;
1495
        /* select vlc table */
1496
        k = g->table_select[i];
1497
        l = mpa_huff_data[k][0];
1498
        linbits = mpa_huff_data[k][1];
1499
        vlc = &huff_vlc[l];
1500

    
1501
        if(!l){
1502
            memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*2*j);
1503
            s_index += 2*j;
1504
            continue;
1505
        }
1506

    
1507
        /* read huffcode and compute each couple */
1508
        for(;j>0;j--) {
1509
            int exponent, x, y, v;
1510
            int pos= get_bits_count(&s->gb);
1511

    
1512
            if (pos >= end_pos){
1513
//                av_log(NULL, AV_LOG_ERROR, "pos: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);
1514
                switch_buffer(s, &pos, &end_pos, &end_pos2);
1515
//                av_log(NULL, AV_LOG_ERROR, "new pos: %d %d\n", pos, end_pos);
1516
                if(pos >= end_pos)
1517
                    break;
1518
            }
1519
            y = get_vlc2(&s->gb, vlc->table, 7, 3);
1520

    
1521
            if(!y){
1522
                g->sb_hybrid[s_index  ] =
1523
                g->sb_hybrid[s_index+1] = 0;
1524
                s_index += 2;
1525
                continue;
1526
            }
1527

    
1528
            exponent= exponents[s_index];
1529

    
1530
            dprintf(s->avctx, "region=%d n=%d x=%d y=%d exp=%d\n",
1531
                    i, g->region_size[i] - j, x, y, exponent);
1532
            if(y&16){
1533
                x = y >> 5;
1534
                y = y & 0x0f;
1535
                if (x < 15){
1536
                    v = expval_table[ exponent ][ x ];
1537
//                      v = expval_table[ (exponent&3) ][ x ] >> FFMIN(0 - (exponent>>2), 31);
1538
                }else{
1539
                    x += get_bitsz(&s->gb, linbits);
1540
                    v = l3_unscale(x, exponent);
1541
                }
1542
                if (get_bits1(&s->gb))
1543
                    v = -v;
1544
                g->sb_hybrid[s_index] = v;
1545
                if (y < 15){
1546
                    v = expval_table[ exponent ][ y ];
1547
                }else{
1548
                    y += get_bitsz(&s->gb, linbits);
1549
                    v = l3_unscale(y, exponent);
1550
                }
1551
                if (get_bits1(&s->gb))
1552
                    v = -v;
1553
                g->sb_hybrid[s_index+1] = v;
1554
            }else{
1555
                x = y >> 5;
1556
                y = y & 0x0f;
1557
                x += y;
1558
                if (x < 15){
1559
                    v = expval_table[ exponent ][ x ];
1560
                }else{
1561
                    x += get_bitsz(&s->gb, linbits);
1562
                    v = l3_unscale(x, exponent);
1563
                }
1564
                if (get_bits1(&s->gb))
1565
                    v = -v;
1566
                g->sb_hybrid[s_index+!!y] = v;
1567
                g->sb_hybrid[s_index+ !y] = 0;
1568
            }
1569
            s_index+=2;
1570
        }
1571
    }
1572

    
1573
    /* high frequencies */
1574
    vlc = &huff_quad_vlc[g->count1table_select];
1575
    last_pos=0;
1576
    while (s_index <= 572) {
1577
        int pos, code;
1578
        pos = get_bits_count(&s->gb);
1579
        if (pos >= end_pos) {
1580
            if (pos > end_pos2 && last_pos){
1581
                /* some encoders generate an incorrect size for this
1582
                   part. We must go back into the data */
1583
                s_index -= 4;
1584
                skip_bits_long(&s->gb, last_pos - pos);
1585
                av_log(NULL, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos);
1586
                if(s->error_resilience >= FF_ER_COMPLIANT)
1587
                    s_index=0;
1588
                break;
1589
            }
1590
//                av_log(NULL, AV_LOG_ERROR, "pos2: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);
1591
            switch_buffer(s, &pos, &end_pos, &end_pos2);
1592
//                av_log(NULL, AV_LOG_ERROR, "new pos2: %d %d %d\n", pos, end_pos, s_index);
1593
            if(pos >= end_pos)
1594
                break;
1595
        }
1596
        last_pos= pos;
1597

    
1598
        code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);
1599
        dprintf(s->avctx, "t=%d code=%d\n", g->count1table_select, code);
1600
        g->sb_hybrid[s_index+0]=
1601
        g->sb_hybrid[s_index+1]=
1602
        g->sb_hybrid[s_index+2]=
1603
        g->sb_hybrid[s_index+3]= 0;
1604
        while(code){
1605
            static const int idxtab[16]={3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0};
1606
            int v;
1607
            int pos= s_index+idxtab[code];
1608
            code ^= 8>>idxtab[code];
1609
            v = exp_table[ exponents[pos] ];
1610
//            v = exp_table[ (exponents[pos]&3) ] >> FFMIN(0 - (exponents[pos]>>2), 31);
1611
            if(get_bits1(&s->gb))
1612
                v = -v;
1613
            g->sb_hybrid[pos] = v;
1614
        }
1615
        s_index+=4;
1616
    }
1617
    /* skip extension bits */
1618
    bits_left = end_pos2 - get_bits_count(&s->gb);
1619
//av_log(NULL, AV_LOG_ERROR, "left:%d buf:%p\n", bits_left, s->in_gb.buffer);
1620
    if (bits_left < 0/* || bits_left > 500*/) {
1621
        av_log(NULL, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
1622
        s_index=0;
1623
    }else if(bits_left > 0 && s->error_resilience >= FF_ER_AGGRESSIVE){
1624
        av_log(NULL, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
1625
        s_index=0;
1626
    }
1627
    memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*(576 - s_index));
1628
    skip_bits_long(&s->gb, bits_left);
1629

    
1630
    i= get_bits_count(&s->gb);
1631
    switch_buffer(s, &i, &end_pos, &end_pos2);
1632

    
1633
    return 0;
1634
}
1635

    
1636
/* Reorder short blocks from bitstream order to interleaved order. It
1637
   would be faster to do it in parsing, but the code would be far more
1638
   complicated */
1639
static void reorder_block(MPADecodeContext *s, GranuleDef *g)
1640
{
1641
    int i, j, len;
1642
    int32_t *ptr, *dst, *ptr1;
1643
    int32_t tmp[576];
1644

    
1645
    if (g->block_type != 2)
1646
        return;
1647

    
1648
    if (g->switch_point) {
1649
        if (s->sample_rate_index != 8) {
1650
            ptr = g->sb_hybrid + 36;
1651
        } else {
1652
            ptr = g->sb_hybrid + 48;
1653
        }
1654
    } else {
1655
        ptr = g->sb_hybrid;
1656
    }
1657

    
1658
    for(i=g->short_start;i<13;i++) {
1659
        len = band_size_short[s->sample_rate_index][i];
1660
        ptr1 = ptr;
1661
        dst = tmp;
1662
        for(j=len;j>0;j--) {
1663
            *dst++ = ptr[0*len];
1664
            *dst++ = ptr[1*len];
1665
            *dst++ = ptr[2*len];
1666
            ptr++;
1667
        }
1668
        ptr+=2*len;
1669
        memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));
1670
    }
1671
}
1672

    
1673
#define ISQRT2 FIXR(0.70710678118654752440)
1674

    
1675
static void compute_stereo(MPADecodeContext *s,
1676
                           GranuleDef *g0, GranuleDef *g1)
1677
{
1678
    int i, j, k, l;
1679
    int32_t v1, v2;
1680
    int sf_max, tmp0, tmp1, sf, len, non_zero_found;
1681
    int32_t (*is_tab)[16];
1682
    int32_t *tab0, *tab1;
1683
    int non_zero_found_short[3];
1684

    
1685
    /* intensity stereo */
1686
    if (s->mode_ext & MODE_EXT_I_STEREO) {
1687
        if (!s->lsf) {
1688
            is_tab = is_table;
1689
            sf_max = 7;
1690
        } else {
1691
            is_tab = is_table_lsf[g1->scalefac_compress & 1];
1692
            sf_max = 16;
1693
        }
1694

    
1695
        tab0 = g0->sb_hybrid + 576;
1696
        tab1 = g1->sb_hybrid + 576;
1697

    
1698
        non_zero_found_short[0] = 0;
1699
        non_zero_found_short[1] = 0;
1700
        non_zero_found_short[2] = 0;
1701
        k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1702
        for(i = 12;i >= g1->short_start;i--) {
1703
            /* for last band, use previous scale factor */
1704
            if (i != 11)
1705
                k -= 3;
1706
            len = band_size_short[s->sample_rate_index][i];
1707
            for(l=2;l>=0;l--) {
1708
                tab0 -= len;
1709
                tab1 -= len;
1710
                if (!non_zero_found_short[l]) {
1711
                    /* test if non zero band. if so, stop doing i-stereo */
1712
                    for(j=0;j<len;j++) {
1713
                        if (tab1[j] != 0) {
1714
                            non_zero_found_short[l] = 1;
1715
                            goto found1;
1716
                        }
1717
                    }
1718
                    sf = g1->scale_factors[k + l];
1719
                    if (sf >= sf_max)
1720
                        goto found1;
1721

    
1722
                    v1 = is_tab[0][sf];
1723
                    v2 = is_tab[1][sf];
1724
                    for(j=0;j<len;j++) {
1725
                        tmp0 = tab0[j];
1726
                        tab0[j] = MULL(tmp0, v1);
1727
                        tab1[j] = MULL(tmp0, v2);
1728
                    }
1729
                } else {
1730
                found1:
1731
                    if (s->mode_ext & MODE_EXT_MS_STEREO) {
1732
                        /* lower part of the spectrum : do ms stereo
1733
                           if enabled */
1734
                        for(j=0;j<len;j++) {
1735
                            tmp0 = tab0[j];
1736
                            tmp1 = tab1[j];
1737
                            tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
1738
                            tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
1739
                        }
1740
                    }
1741
                }
1742
            }
1743
        }
1744

    
1745
        non_zero_found = non_zero_found_short[0] |
1746
            non_zero_found_short[1] |
1747
            non_zero_found_short[2];
1748

    
1749
        for(i = g1->long_end - 1;i >= 0;i--) {
1750
            len = band_size_long[s->sample_rate_index][i];
1751
            tab0 -= len;
1752
            tab1 -= len;
1753
            /* test if non zero band. if so, stop doing i-stereo */
1754
            if (!non_zero_found) {
1755
                for(j=0;j<len;j++) {
1756
                    if (tab1[j] != 0) {
1757
                        non_zero_found = 1;
1758
                        goto found2;
1759
                    }
1760
                }
1761
                /* for last band, use previous scale factor */
1762
                k = (i == 21) ? 20 : i;
1763
                sf = g1->scale_factors[k];
1764
                if (sf >= sf_max)
1765
                    goto found2;
1766
                v1 = is_tab[0][sf];
1767
                v2 = is_tab[1][sf];
1768
                for(j=0;j<len;j++) {
1769
                    tmp0 = tab0[j];
1770
                    tab0[j] = MULL(tmp0, v1);
1771
                    tab1[j] = MULL(tmp0, v2);
1772
                }
1773
            } else {
1774
            found2:
1775
                if (s->mode_ext & MODE_EXT_MS_STEREO) {
1776
                    /* lower part of the spectrum : do ms stereo
1777
                       if enabled */
1778
                    for(j=0;j<len;j++) {
1779
                        tmp0 = tab0[j];
1780
                        tmp1 = tab1[j];
1781
                        tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
1782
                        tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
1783
                    }
1784
                }
1785
            }
1786
        }
1787
    } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1788
        /* ms stereo ONLY */
1789
        /* NOTE: the 1/sqrt(2) normalization factor is included in the
1790
           global gain */
1791
        tab0 = g0->sb_hybrid;
1792
        tab1 = g1->sb_hybrid;
1793
        for(i=0;i<576;i++) {
1794
            tmp0 = tab0[i];
1795
            tmp1 = tab1[i];
1796
            tab0[i] = tmp0 + tmp1;
1797
            tab1[i] = tmp0 - tmp1;
1798
        }
1799
    }
1800
}
1801

    
1802
static void compute_antialias_integer(MPADecodeContext *s,
1803
                              GranuleDef *g)
1804
{
1805
    int32_t *ptr, *csa;
1806
    int n, i;
1807

    
1808
    /* we antialias only "long" bands */
1809
    if (g->block_type == 2) {
1810
        if (!g->switch_point)
1811
            return;
1812
        /* XXX: check this for 8000Hz case */
1813
        n = 1;
1814
    } else {
1815
        n = SBLIMIT - 1;
1816
    }
1817

    
1818
    ptr = g->sb_hybrid + 18;
1819
    for(i = n;i > 0;i--) {
1820
        int tmp0, tmp1, tmp2;
1821
        csa = &csa_table[0][0];
1822
#define INT_AA(j) \
1823
            tmp0 = ptr[-1-j];\
1824
            tmp1 = ptr[   j];\
1825
            tmp2= MULH(tmp0 + tmp1, csa[0+4*j]);\
1826
            ptr[-1-j] = 4*(tmp2 - MULH(tmp1, csa[2+4*j]));\
1827
            ptr[   j] = 4*(tmp2 + MULH(tmp0, csa[3+4*j]));
1828

    
1829
        INT_AA(0)
1830
        INT_AA(1)
1831
        INT_AA(2)
1832
        INT_AA(3)
1833
        INT_AA(4)
1834
        INT_AA(5)
1835
        INT_AA(6)
1836
        INT_AA(7)
1837

    
1838
        ptr += 18;
1839
    }
1840
}
1841

    
1842
static void compute_antialias_float(MPADecodeContext *s,
1843
                              GranuleDef *g)
1844
{
1845
    int32_t *ptr;
1846
    int n, i;
1847

    
1848
    /* we antialias only "long" bands */
1849
    if (g->block_type == 2) {
1850
        if (!g->switch_point)
1851
            return;
1852
        /* XXX: check this for 8000Hz case */
1853
        n = 1;
1854
    } else {
1855
        n = SBLIMIT - 1;
1856
    }
1857

    
1858
    ptr = g->sb_hybrid + 18;
1859
    for(i = n;i > 0;i--) {
1860
        float tmp0, tmp1;
1861
        float *csa = &csa_table_float[0][0];
1862
#define FLOAT_AA(j)\
1863
        tmp0= ptr[-1-j];\
1864
        tmp1= ptr[   j];\
1865
        ptr[-1-j] = lrintf(tmp0 * csa[0+4*j] - tmp1 * csa[1+4*j]);\
1866
        ptr[   j] = lrintf(tmp0 * csa[1+4*j] + tmp1 * csa[0+4*j]);
1867

    
1868
        FLOAT_AA(0)
1869
        FLOAT_AA(1)
1870
        FLOAT_AA(2)
1871
        FLOAT_AA(3)
1872
        FLOAT_AA(4)
1873
        FLOAT_AA(5)
1874
        FLOAT_AA(6)
1875
        FLOAT_AA(7)
1876

    
1877
        ptr += 18;
1878
    }
1879
}
1880

    
1881
static void compute_imdct(MPADecodeContext *s,
1882
                          GranuleDef *g,
1883
                          int32_t *sb_samples,
1884
                          int32_t *mdct_buf)
1885
{
1886
    int32_t *ptr, *win, *win1, *buf, *out_ptr, *ptr1;
1887
    int32_t out2[12];
1888
    int i, j, mdct_long_end, v, sblimit;
1889

    
1890
    /* find last non zero block */
1891
    ptr = g->sb_hybrid + 576;
1892
    ptr1 = g->sb_hybrid + 2 * 18;
1893
    while (ptr >= ptr1) {
1894
        ptr -= 6;
1895
        v = ptr[0] | ptr[1] | ptr[2] | ptr[3] | ptr[4] | ptr[5];
1896
        if (v != 0)
1897
            break;
1898
    }
1899
    sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
1900

    
1901
    if (g->block_type == 2) {
1902
        /* XXX: check for 8000 Hz */
1903
        if (g->switch_point)
1904
            mdct_long_end = 2;
1905
        else
1906
            mdct_long_end = 0;
1907
    } else {
1908
        mdct_long_end = sblimit;
1909
    }
1910

    
1911
    buf = mdct_buf;
1912
    ptr = g->sb_hybrid;
1913
    for(j=0;j<mdct_long_end;j++) {
1914
        /* apply window & overlap with previous buffer */
1915
        out_ptr = sb_samples + j;
1916
        /* select window */
1917
        if (g->switch_point && j < 2)
1918
            win1 = mdct_win[0];
1919
        else
1920
            win1 = mdct_win[g->block_type];
1921
        /* select frequency inversion */
1922
        win = win1 + ((4 * 36) & -(j & 1));
1923
        imdct36(out_ptr, buf, ptr, win);
1924
        out_ptr += 18*SBLIMIT;
1925
        ptr += 18;
1926
        buf += 18;
1927
    }
1928
    for(j=mdct_long_end;j<sblimit;j++) {
1929
        /* select frequency inversion */
1930
        win = mdct_win[2] + ((4 * 36) & -(j & 1));
1931
        out_ptr = sb_samples + j;
1932

    
1933
        for(i=0; i<6; i++){
1934
            *out_ptr = buf[i];
1935
            out_ptr += SBLIMIT;
1936
        }
1937
        imdct12(out2, ptr + 0);
1938
        for(i=0;i<6;i++) {
1939
            *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*1];
1940
            buf[i + 6*2] = MULH(out2[i + 6], win[i + 6]);
1941
            out_ptr += SBLIMIT;
1942
        }
1943
        imdct12(out2, ptr + 1);
1944
        for(i=0;i<6;i++) {
1945
            *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*2];
1946
            buf[i + 6*0] = MULH(out2[i + 6], win[i + 6]);
1947
            out_ptr += SBLIMIT;
1948
        }
1949
        imdct12(out2, ptr + 2);
1950
        for(i=0;i<6;i++) {
1951
            buf[i + 6*0] = MULH(out2[i], win[i]) + buf[i + 6*0];
1952
            buf[i + 6*1] = MULH(out2[i + 6], win[i + 6]);
1953
            buf[i + 6*2] = 0;
1954
        }
1955
        ptr += 18;
1956
        buf += 18;
1957
    }
1958
    /* zero bands */
1959
    for(j=sblimit;j<SBLIMIT;j++) {
1960
        /* overlap */
1961
        out_ptr = sb_samples + j;
1962
        for(i=0;i<18;i++) {
1963
            *out_ptr = buf[i];
1964
            buf[i] = 0;
1965
            out_ptr += SBLIMIT;
1966
        }
1967
        buf += 18;
1968
    }
1969
}
1970

    
1971
#if defined(DEBUG)
1972
void sample_dump(int fnum, int32_t *tab, int n)
1973
{
1974
    static FILE *files[16], *f;
1975
    char buf[512];
1976
    int i;
1977
    int32_t v;
1978

    
1979
    f = files[fnum];
1980
    if (!f) {
1981
        snprintf(buf, sizeof(buf), "/tmp/out%d.%s.pcm",
1982
                fnum,
1983
#ifdef USE_HIGHPRECISION
1984
                "hp"
1985
#else
1986
                "lp"
1987
#endif
1988
                );
1989
        f = fopen(buf, "w");
1990
        if (!f)
1991
            return;
1992
        files[fnum] = f;
1993
    }
1994

    
1995
    if (fnum == 0) {
1996
        static int pos = 0;
1997
        av_log(NULL, AV_LOG_DEBUG, "pos=%d\n", pos);
1998
        for(i=0;i<n;i++) {
1999
            av_log(NULL, AV_LOG_DEBUG, " %0.4f", (double)tab[i] / FRAC_ONE);
2000
            if ((i % 18) == 17)
2001
                av_log(NULL, AV_LOG_DEBUG, "\n");
2002
        }
2003
        pos += n;
2004
    }
2005
    for(i=0;i<n;i++) {
2006
        /* normalize to 23 frac bits */
2007
        v = tab[i] << (23 - FRAC_BITS);
2008
        fwrite(&v, 1, sizeof(int32_t), f);
2009
    }
2010
}
2011
#endif
2012

    
2013

    
2014
/* main layer3 decoding function */
2015
static int mp_decode_layer3(MPADecodeContext *s)
2016
{
2017
    int nb_granules, main_data_begin, private_bits;
2018
    int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;
2019
    GranuleDef granules[2][2], *g;
2020
    int16_t exponents[576];
2021

    
2022
    /* read side info */
2023
    if (s->lsf) {
2024
        main_data_begin = get_bits(&s->gb, 8);
2025
        private_bits = get_bits(&s->gb, s->nb_channels);
2026
        nb_granules = 1;
2027
    } else {
2028
        main_data_begin = get_bits(&s->gb, 9);
2029
        if (s->nb_channels == 2)
2030
            private_bits = get_bits(&s->gb, 3);
2031
        else
2032
            private_bits = get_bits(&s->gb, 5);
2033
        nb_granules = 2;
2034
        for(ch=0;ch<s->nb_channels;ch++) {
2035
            granules[ch][0].scfsi = 0; /* all scale factors are transmitted */
2036
            granules[ch][1].scfsi = get_bits(&s->gb, 4);
2037
        }
2038
    }
2039

    
2040
    for(gr=0;gr<nb_granules;gr++) {
2041
        for(ch=0;ch<s->nb_channels;ch++) {
2042
            dprintf(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);
2043
            g = &granules[ch][gr];
2044
            g->part2_3_length = get_bits(&s->gb, 12);
2045
            g->big_values = get_bits(&s->gb, 9);
2046
            if(g->big_values > 288){
2047
                av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");
2048
                return -1;
2049
            }
2050

    
2051
            g->global_gain = get_bits(&s->gb, 8);
2052
            /* if MS stereo only is selected, we precompute the
2053
               1/sqrt(2) renormalization factor */
2054
            if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
2055
                MODE_EXT_MS_STEREO)
2056
                g->global_gain -= 2;
2057
            if (s->lsf)
2058
                g->scalefac_compress = get_bits(&s->gb, 9);
2059
            else
2060
                g->scalefac_compress = get_bits(&s->gb, 4);
2061
            blocksplit_flag = get_bits1(&s->gb);
2062
            if (blocksplit_flag) {
2063
                g->block_type = get_bits(&s->gb, 2);
2064
                if (g->block_type == 0){
2065
                    av_log(NULL, AV_LOG_ERROR, "invalid block type\n");
2066
                    return -1;
2067
                }
2068
                g->switch_point = get_bits1(&s->gb);
2069
                for(i=0;i<2;i++)
2070
                    g->table_select[i] = get_bits(&s->gb, 5);
2071
                for(i=0;i<3;i++)
2072
                    g->subblock_gain[i] = get_bits(&s->gb, 3);
2073
                ff_init_short_region(s, g);
2074
            } else {
2075
                int region_address1, region_address2;
2076
                g->block_type = 0;
2077
                g->switch_point = 0;
2078
                for(i=0;i<3;i++)
2079
                    g->table_select[i] = get_bits(&s->gb, 5);
2080
                /* compute huffman coded region sizes */
2081
                region_address1 = get_bits(&s->gb, 4);
2082
                region_address2 = get_bits(&s->gb, 3);
2083
                dprintf(s->avctx, "region1=%d region2=%d\n",
2084
                        region_address1, region_address2);
2085
                ff_init_long_region(s, g, region_address1, region_address2);
2086
            }
2087
            ff_region_offset2size(g);
2088
            ff_compute_band_indexes(s, g);
2089

    
2090
            g->preflag = 0;
2091
            if (!s->lsf)
2092
                g->preflag = get_bits1(&s->gb);
2093
            g->scalefac_scale = get_bits1(&s->gb);
2094
            g->count1table_select = get_bits1(&s->gb);
2095
            dprintf(s->avctx, "block_type=%d switch_point=%d\n",
2096
                    g->block_type, g->switch_point);
2097
        }
2098
    }
2099

    
2100
  if (!s->adu_mode) {
2101
    const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);
2102
    assert((get_bits_count(&s->gb) & 7) == 0);
2103
    /* now we get bits from the main_data_begin offset */
2104
    dprintf(s->avctx, "seekback: %d\n", main_data_begin);
2105
//av_log(NULL, AV_LOG_ERROR, "backstep:%d, lastbuf:%d\n", main_data_begin, s->last_buf_size);
2106

    
2107
    memcpy(s->last_buf + s->last_buf_size, ptr, EXTRABYTES);
2108
    s->in_gb= s->gb;
2109
        init_get_bits(&s->gb, s->last_buf, s->last_buf_size*8);
2110
        skip_bits_long(&s->gb, 8*(s->last_buf_size - main_data_begin));
2111
  }
2112

    
2113
    for(gr=0;gr<nb_granules;gr++) {
2114
        for(ch=0;ch<s->nb_channels;ch++) {
2115
            g = &granules[ch][gr];
2116
            if(get_bits_count(&s->gb)<0){
2117
                av_log(NULL, AV_LOG_ERROR, "mdb:%d, lastbuf:%d skipping granule %d\n",
2118
                                            main_data_begin, s->last_buf_size, gr);
2119
                skip_bits_long(&s->gb, g->part2_3_length);
2120
                memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));
2121
                if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->in_gb.buffer){
2122
                    skip_bits_long(&s->in_gb, get_bits_count(&s->gb) - s->gb.size_in_bits);
2123
                    s->gb= s->in_gb;
2124
                    s->in_gb.buffer=NULL;
2125
                }
2126
                continue;
2127
            }
2128

    
2129
            bits_pos = get_bits_count(&s->gb);
2130

    
2131
            if (!s->lsf) {
2132
                uint8_t *sc;
2133
                int slen, slen1, slen2;
2134

    
2135
                /* MPEG1 scale factors */
2136
                slen1 = slen_table[0][g->scalefac_compress];
2137
                slen2 = slen_table[1][g->scalefac_compress];
2138
                dprintf(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);
2139
                if (g->block_type == 2) {
2140
                    n = g->switch_point ? 17 : 18;
2141
                    j = 0;
2142
                    if(slen1){
2143
                        for(i=0;i<n;i++)
2144
                            g->scale_factors[j++] = get_bits(&s->gb, slen1);
2145
                    }else{
2146
                        for(i=0;i<n;i++)
2147
                            g->scale_factors[j++] = 0;
2148
                    }
2149
                    if(slen2){
2150
                        for(i=0;i<18;i++)
2151
                            g->scale_factors[j++] = get_bits(&s->gb, slen2);
2152
                        for(i=0;i<3;i++)
2153
                            g->scale_factors[j++] = 0;
2154
                    }else{
2155
                        for(i=0;i<21;i++)
2156
                            g->scale_factors[j++] = 0;
2157
                    }
2158
                } else {
2159
                    sc = granules[ch][0].scale_factors;
2160
                    j = 0;
2161
                    for(k=0;k<4;k++) {
2162
                        n = (k == 0 ? 6 : 5);
2163
                        if ((g->scfsi & (0x8 >> k)) == 0) {
2164
                            slen = (k < 2) ? slen1 : slen2;
2165
                            if(slen){
2166
                                for(i=0;i<n;i++)
2167
                                    g->scale_factors[j++] = get_bits(&s->gb, slen);
2168
                            }else{
2169
                                for(i=0;i<n;i++)
2170
                                    g->scale_factors[j++] = 0;
2171
                            }
2172
                        } else {
2173
                            /* simply copy from last granule */
2174
                            for(i=0;i<n;i++) {
2175
                                g->scale_factors[j] = sc[j];
2176
                                j++;
2177
                            }
2178
                        }
2179
                    }
2180
                    g->scale_factors[j++] = 0;
2181
                }
2182
#if defined(DEBUG)
2183
                {
2184
                    dprintf(s->avctx, "scfsi=%x gr=%d ch=%d scale_factors:\n",
2185
                           g->scfsi, gr, ch);
2186
                    for(i=0;i<j;i++)
2187
                        dprintf(s->avctx, " %d", g->scale_factors[i]);
2188
                    dprintf(s->avctx, "\n");
2189
                }
2190
#endif
2191
            } else {
2192
                int tindex, tindex2, slen[4], sl, sf;
2193

    
2194
                /* LSF scale factors */
2195
                if (g->block_type == 2) {
2196
                    tindex = g->switch_point ? 2 : 1;
2197
                } else {
2198
                    tindex = 0;
2199
                }
2200
                sf = g->scalefac_compress;
2201
                if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
2202
                    /* intensity stereo case */
2203
                    sf >>= 1;
2204
                    if (sf < 180) {
2205
                        lsf_sf_expand(slen, sf, 6, 6, 0);
2206
                        tindex2 = 3;
2207
                    } else if (sf < 244) {
2208
                        lsf_sf_expand(slen, sf - 180, 4, 4, 0);
2209
                        tindex2 = 4;
2210
                    } else {
2211
                        lsf_sf_expand(slen, sf - 244, 3, 0, 0);
2212
                        tindex2 = 5;
2213
                    }
2214
                } else {
2215
                    /* normal case */
2216
                    if (sf < 400) {
2217
                        lsf_sf_expand(slen, sf, 5, 4, 4);
2218
                        tindex2 = 0;
2219
                    } else if (sf < 500) {
2220
                        lsf_sf_expand(slen, sf - 400, 5, 4, 0);
2221
                        tindex2 = 1;
2222
                    } else {
2223
                        lsf_sf_expand(slen, sf - 500, 3, 0, 0);
2224
                        tindex2 = 2;
2225
                        g->preflag = 1;
2226
                    }
2227
                }
2228

    
2229
                j = 0;
2230
                for(k=0;k<4;k++) {
2231
                    n = lsf_nsf_table[tindex2][tindex][k];
2232
                    sl = slen[k];
2233
                    if(sl){
2234
                        for(i=0;i<n;i++)
2235
                            g->scale_factors[j++] = get_bits(&s->gb, sl);
2236
                    }else{
2237
                        for(i=0;i<n;i++)
2238
                            g->scale_factors[j++] = 0;
2239
                    }
2240
                }
2241
                /* XXX: should compute exact size */
2242
                for(;j<40;j++)
2243
                    g->scale_factors[j] = 0;
2244
#if defined(DEBUG)
2245
                {
2246
                    dprintf(s->avctx, "gr=%d ch=%d scale_factors:\n",
2247
                           gr, ch);
2248
                    for(i=0;i<40;i++)
2249
                        dprintf(s->avctx, " %d", g->scale_factors[i]);
2250
                    dprintf(s->avctx, "\n");
2251
                }
2252
#endif
2253
            }
2254

    
2255
            exponents_from_scale_factors(s, g, exponents);
2256

    
2257
            /* read Huffman coded residue */
2258
            huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);
2259
#if defined(DEBUG)
2260
            sample_dump(0, g->sb_hybrid, 576);
2261
#endif
2262
        } /* ch */
2263

    
2264
        if (s->nb_channels == 2)
2265
            compute_stereo(s, &granules[0][gr], &granules[1][gr]);
2266

    
2267
        for(ch=0;ch<s->nb_channels;ch++) {
2268
            g = &granules[ch][gr];
2269

    
2270
            reorder_block(s, g);
2271
#if defined(DEBUG)
2272
            sample_dump(0, g->sb_hybrid, 576);
2273
#endif
2274
            s->compute_antialias(s, g);
2275
#if defined(DEBUG)
2276
            sample_dump(1, g->sb_hybrid, 576);
2277
#endif
2278
            compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
2279
#if defined(DEBUG)
2280
            sample_dump(2, &s->sb_samples[ch][18 * gr][0], 576);
2281
#endif
2282
        }
2283
    } /* gr */
2284
    if(get_bits_count(&s->gb)<0)
2285
        skip_bits_long(&s->gb, -get_bits_count(&s->gb));
2286
    return nb_granules * 18;
2287
}
2288

    
2289
static int mp_decode_frame(MPADecodeContext *s,
2290
                           OUT_INT *samples, const uint8_t *buf, int buf_size)
2291
{
2292
    int i, nb_frames, ch;
2293
    OUT_INT *samples_ptr;
2294

    
2295
    init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE)*8);
2296

    
2297
    /* skip error protection field */
2298
    if (s->error_protection)
2299
        skip_bits(&s->gb, 16);
2300

    
2301
    dprintf(s->avctx, "frame %d:\n", s->frame_count);
2302
    switch(s->layer) {
2303
    case 1:
2304
        s->avctx->frame_size = 384;
2305
        nb_frames = mp_decode_layer1(s);
2306
        break;
2307
    case 2:
2308
        s->avctx->frame_size = 1152;
2309
        nb_frames = mp_decode_layer2(s);
2310
        break;
2311
    case 3:
2312
        s->avctx->frame_size = s->lsf ? 576 : 1152;
2313
    default:
2314
        nb_frames = mp_decode_layer3(s);
2315

    
2316
        s->last_buf_size=0;
2317
        if(s->in_gb.buffer){
2318
            align_get_bits(&s->gb);
2319
            i= (s->gb.size_in_bits - get_bits_count(&s->gb))>>3;
2320
            if(i >= 0 && i <= BACKSTEP_SIZE){
2321
                memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);
2322
                s->last_buf_size=i;
2323
            }else
2324
                av_log(NULL, AV_LOG_ERROR, "invalid old backstep %d\n", i);
2325
            s->gb= s->in_gb;
2326
            s->in_gb.buffer= NULL;
2327
        }
2328

    
2329
        align_get_bits(&s->gb);
2330
        assert((get_bits_count(&s->gb) & 7) == 0);
2331
        i= (s->gb.size_in_bits - get_bits_count(&s->gb))>>3;
2332

    
2333
        if(i<0 || i > BACKSTEP_SIZE || nb_frames<0){
2334
            av_log(NULL, AV_LOG_ERROR, "invalid new backstep %d\n", i);
2335
            i= FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);
2336
        }
2337
        assert(i <= buf_size - HEADER_SIZE && i>= 0);
2338
        memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);
2339
        s->last_buf_size += i;
2340

    
2341
        break;
2342
    }
2343
#if defined(DEBUG)
2344
    for(i=0;i<nb_frames;i++) {
2345
        for(ch=0;ch<s->nb_channels;ch++) {
2346
            int j;
2347
            dprintf(s->avctx, "%d-%d:", i, ch);
2348
            for(j=0;j<SBLIMIT;j++)
2349
                dprintf(s->avctx, " %0.6f", (double)s->sb_samples[ch][i][j] / FRAC_ONE);
2350
            dprintf(s->avctx, "\n");
2351
        }
2352
    }
2353
#endif
2354
    /* apply the synthesis filter */
2355
    for(ch=0;ch<s->nb_channels;ch++) {
2356
        samples_ptr = samples + ch;
2357
        for(i=0;i<nb_frames;i++) {
2358
            ff_mpa_synth_filter(s->synth_buf[ch], &(s->synth_buf_offset[ch]),
2359
                         window, &s->dither_state,
2360
                         samples_ptr, s->nb_channels,
2361
                         s->sb_samples[ch][i]);
2362
            samples_ptr += 32 * s->nb_channels;
2363
        }
2364
    }
2365
#ifdef DEBUG
2366
    s->frame_count++;
2367
#endif
2368
    return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
2369
}
2370

    
2371
static int decode_frame(AVCodecContext * avctx,
2372
                        void *data, int *data_size,
2373
                        const uint8_t * buf, int buf_size)
2374
{
2375
    MPADecodeContext *s = avctx->priv_data;
2376
    uint32_t header;
2377
    int out_size;
2378
    OUT_INT *out_samples = data;
2379

    
2380
retry:
2381
    if(buf_size < HEADER_SIZE)
2382
        return -1;
2383

    
2384
    header = AV_RB32(buf);
2385
    if(ff_mpa_check_header(header) < 0){
2386
        buf++;
2387
//        buf_size--;
2388
        av_log(avctx, AV_LOG_ERROR, "Header missing skipping one byte.\n");
2389
        goto retry;
2390
    }
2391

    
2392
    if (ff_mpegaudio_decode_header(s, header) == 1) {
2393
        /* free format: prepare to compute frame size */
2394
        s->frame_size = -1;
2395
        return -1;
2396
    }
2397
    /* update codec info */
2398
    avctx->channels = s->nb_channels;
2399
    avctx->bit_rate = s->bit_rate;
2400
    avctx->sub_id = s->layer;
2401

    
2402
    if(s->frame_size<=0 || s->frame_size > buf_size){
2403
        av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
2404
        return -1;
2405
    }else if(s->frame_size < buf_size){
2406
        av_log(avctx, AV_LOG_ERROR, "incorrect frame size\n");
2407
        buf_size= s->frame_size;
2408
    }
2409

    
2410
    out_size = mp_decode_frame(s, out_samples, buf, buf_size);
2411
    if(out_size>=0){
2412
        *data_size = out_size;
2413
        avctx->sample_rate = s->sample_rate;
2414
        //FIXME maybe move the other codec info stuff from above here too
2415
    }else
2416
        av_log(avctx, AV_LOG_DEBUG, "Error while decoding MPEG audio frame.\n"); //FIXME return -1 / but also return the number of bytes consumed
2417
    s->frame_size = 0;
2418
    return buf_size;
2419
}
2420

    
2421
static void flush(AVCodecContext *avctx){
2422
    MPADecodeContext *s = avctx->priv_data;
2423
    memset(s->synth_buf, 0, sizeof(s->synth_buf));
2424
    s->last_buf_size= 0;
2425
}
2426

    
2427
#ifdef CONFIG_MP3ADU_DECODER
2428
static int decode_frame_adu(AVCodecContext * avctx,
2429
                        void *data, int *data_size,
2430
                        const uint8_t * buf, int buf_size)
2431
{
2432
    MPADecodeContext *s = avctx->priv_data;
2433
    uint32_t header;
2434
    int len, out_size;
2435
    OUT_INT *out_samples = data;
2436

    
2437
    len = buf_size;
2438

    
2439
    // Discard too short frames
2440
    if (buf_size < HEADER_SIZE) {
2441
        *data_size = 0;
2442
        return buf_size;
2443
    }
2444

    
2445

    
2446
    if (len > MPA_MAX_CODED_FRAME_SIZE)
2447
        len = MPA_MAX_CODED_FRAME_SIZE;
2448

    
2449
    // Get header and restore sync word
2450
    header = AV_RB32(buf) | 0xffe00000;
2451

    
2452
    if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame
2453
        *data_size = 0;
2454
        return buf_size;
2455
    }
2456

    
2457
    ff_mpegaudio_decode_header(s, header);
2458
    /* update codec info */
2459
    avctx->sample_rate = s->sample_rate;
2460
    avctx->channels = s->nb_channels;
2461
    avctx->bit_rate = s->bit_rate;
2462
    avctx->sub_id = s->layer;
2463

    
2464
    s->frame_size = len;
2465

    
2466
    if (avctx->parse_only) {
2467
        out_size = buf_size;
2468
    } else {
2469
        out_size = mp_decode_frame(s, out_samples, buf, buf_size);
2470
    }
2471

    
2472
    *data_size = out_size;
2473
    return buf_size;
2474
}
2475
#endif /* CONFIG_MP3ADU_DECODER */
2476

    
2477
#ifdef CONFIG_MP3ON4_DECODER
2478
/* Next 3 arrays are indexed by channel config number (passed via codecdata) */
2479
static int mp3Frames[16] = {0,1,1,2,3,3,4,5,2};   /* number of mp3 decoder instances */
2480
static int mp3Channels[16] = {0,1,2,3,4,5,6,8,4}; /* total output channels */
2481
/* offsets into output buffer, assume output order is FL FR BL BR C LFE */
2482
static int chan_offset[9][5] = {
2483
    {0},
2484
    {0},            // C
2485
    {0},            // FLR
2486
    {2,0},          // C FLR
2487
    {2,0,3},        // C FLR BS
2488
    {4,0,2},        // C FLR BLRS
2489
    {4,0,2,5},      // C FLR BLRS LFE
2490
    {4,0,2,6,5},    // C FLR BLRS BLR LFE
2491
    {0,2}           // FLR BLRS
2492
};
2493

    
2494

    
2495
static int decode_init_mp3on4(AVCodecContext * avctx)
2496
{
2497
    MP3On4DecodeContext *s = avctx->priv_data;
2498
    int i;
2499

    
2500
    if ((avctx->extradata_size < 2) || (avctx->extradata == NULL)) {
2501
        av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
2502
        return -1;
2503
    }
2504

    
2505
    s->chan_cfg = (((unsigned char *)avctx->extradata)[1] >> 3) & 0x0f;
2506
    s->frames = mp3Frames[s->chan_cfg];
2507
    if(!s->frames) {
2508
        av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
2509
        return -1;
2510
    }
2511
    avctx->channels = mp3Channels[s->chan_cfg];
2512

    
2513
    /* Init the first mp3 decoder in standard way, so that all tables get builded
2514
     * We replace avctx->priv_data with the context of the first decoder so that
2515
     * decode_init() does not have to be changed.
2516
     * Other decoders will be initialized here copying data from the first context
2517
     */
2518
    // Allocate zeroed memory for the first decoder context
2519
    s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
2520
    // Put decoder context in place to make init_decode() happy
2521
    avctx->priv_data = s->mp3decctx[0];
2522
    decode_init(avctx);
2523
    // Restore mp3on4 context pointer
2524
    avctx->priv_data = s;
2525
    s->mp3decctx[0]->adu_mode = 1; // Set adu mode
2526

    
2527
    /* Create a separate codec/context for each frame (first is already ok).
2528
     * Each frame is 1 or 2 channels - up to 5 frames allowed
2529
     */
2530
    for (i = 1; i < s->frames; i++) {
2531
        s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
2532
        s->mp3decctx[i]->compute_antialias = s->mp3decctx[0]->compute_antialias;
2533
        s->mp3decctx[i]->adu_mode = 1;
2534
        s->mp3decctx[i]->avctx = avctx;
2535
    }
2536

    
2537
    return 0;
2538
}
2539

    
2540

    
2541
static int decode_close_mp3on4(AVCodecContext * avctx)
2542
{
2543
    MP3On4DecodeContext *s = avctx->priv_data;
2544
    int i;
2545

    
2546
    for (i = 0; i < s->frames; i++)
2547
        if (s->mp3decctx[i])
2548
            av_free(s->mp3decctx[i]);
2549

    
2550
    return 0;
2551
}
2552

    
2553

    
2554
static int decode_frame_mp3on4(AVCodecContext * avctx,
2555
                        void *data, int *data_size,
2556
                        const uint8_t * buf, int buf_size)
2557
{
2558
    MP3On4DecodeContext *s = avctx->priv_data;
2559
    MPADecodeContext *m;
2560
    int len, out_size = 0;
2561
    uint32_t header;
2562
    OUT_INT *out_samples = data;
2563
    OUT_INT decoded_buf[MPA_FRAME_SIZE * MPA_MAX_CHANNELS];
2564
    OUT_INT *outptr, *bp;
2565
    int fsize;
2566
    const unsigned char *start2 = buf, *start;
2567
    int fr, i, j, n;
2568
    int off = avctx->channels;
2569
    int *coff = chan_offset[s->chan_cfg];
2570

    
2571
    len = buf_size;
2572

    
2573
    // Discard too short frames
2574
    if (buf_size < HEADER_SIZE) {
2575
        *data_size = 0;
2576
        return buf_size;
2577
    }
2578

    
2579
    // If only one decoder interleave is not needed
2580
    outptr = s->frames == 1 ? out_samples : decoded_buf;
2581

    
2582
    for (fr = 0; fr < s->frames; fr++) {
2583
        start = start2;
2584
        fsize = (start[0] << 4) | (start[1] >> 4);
2585
        start2 += fsize;
2586
        if (fsize > len)
2587
            fsize = len;
2588
        len -= fsize;
2589
        if (fsize > MPA_MAX_CODED_FRAME_SIZE)
2590
            fsize = MPA_MAX_CODED_FRAME_SIZE;
2591
        m = s->mp3decctx[fr];
2592
        assert (m != NULL);
2593

    
2594
        // Get header
2595
        header = AV_RB32(start) | 0xfff00000;
2596

    
2597
        if (ff_mpa_check_header(header) < 0) { // Bad header, discard block
2598
            *data_size = 0;
2599
            return buf_size;
2600
        }
2601

    
2602
        ff_mpegaudio_decode_header(m, header);
2603
        mp_decode_frame(m, decoded_buf, start, fsize);
2604

    
2605
        n = MPA_FRAME_SIZE * m->nb_channels;
2606
        out_size += n * sizeof(OUT_INT);
2607
        if(s->frames > 1) {
2608
            /* interleave output data */
2609
            bp = out_samples + coff[fr];
2610
            if(m->nb_channels == 1) {
2611
                for(j = 0; j < n; j++) {
2612
                    *bp = decoded_buf[j];
2613
                    bp += off;
2614
                }
2615
            } else {
2616
                for(j = 0; j < n; j++) {
2617
                    bp[0] = decoded_buf[j++];
2618
                    bp[1] = decoded_buf[j];
2619
                    bp += off;
2620
                }
2621
            }
2622
        }
2623
    }
2624

    
2625
    /* update codec info */
2626
    avctx->sample_rate = s->mp3decctx[0]->sample_rate;
2627
    avctx->bit_rate = 0;
2628
    for (i = 0; i < s->frames; i++)
2629
        avctx->bit_rate += s->mp3decctx[i]->bit_rate;
2630

    
2631
    *data_size = out_size;
2632
    return buf_size;
2633
}
2634
#endif /* CONFIG_MP3ON4_DECODER */
2635

    
2636
#ifdef CONFIG_MP2_DECODER
2637
AVCodec mp2_decoder =
2638
{
2639
    "mp2",
2640
    CODEC_TYPE_AUDIO,
2641
    CODEC_ID_MP2,
2642
    sizeof(MPADecodeContext),
2643
    decode_init,
2644
    NULL,
2645
    NULL,
2646
    decode_frame,
2647
    CODEC_CAP_PARSE_ONLY,
2648
    .flush= flush,
2649
};
2650
#endif
2651
#ifdef CONFIG_MP3_DECODER
2652
AVCodec mp3_decoder =
2653
{
2654
    "mp3",
2655
    CODEC_TYPE_AUDIO,
2656
    CODEC_ID_MP3,
2657
    sizeof(MPADecodeContext),
2658
    decode_init,
2659
    NULL,
2660
    NULL,
2661
    decode_frame,
2662
    CODEC_CAP_PARSE_ONLY,
2663
    .flush= flush,
2664
};
2665
#endif
2666
#ifdef CONFIG_MP3ADU_DECODER
2667
AVCodec mp3adu_decoder =
2668
{
2669
    "mp3adu",
2670
    CODEC_TYPE_AUDIO,
2671
    CODEC_ID_MP3ADU,
2672
    sizeof(MPADecodeContext),
2673
    decode_init,
2674
    NULL,
2675
    NULL,
2676
    decode_frame_adu,
2677
    CODEC_CAP_PARSE_ONLY,
2678
    .flush= flush,
2679
};
2680
#endif
2681
#ifdef CONFIG_MP3ON4_DECODER
2682
AVCodec mp3on4_decoder =
2683
{
2684
    "mp3on4",
2685
    CODEC_TYPE_AUDIO,
2686
    CODEC_ID_MP3ON4,
2687
    sizeof(MP3On4DecodeContext),
2688
    decode_init_mp3on4,
2689
    NULL,
2690
    decode_close_mp3on4,
2691
    decode_frame_mp3on4,
2692
    .flush= flush,
2693
};
2694
#endif