Statistics
| Branch: | Revision:

ffmpeg / libavcodec / mpegaudiodec.c @ 0d5b2eb4

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
/* layer 3 "granule" */
60
typedef struct GranuleDef {
61
    uint8_t scfsi;
62
    int part2_3_length;
63
    int big_values;
64
    int global_gain;
65
    int scalefac_compress;
66
    uint8_t block_type;
67
    uint8_t switch_point;
68
    int table_select[3];
69
    int subblock_gain[3];
70
    uint8_t scalefac_scale;
71
    uint8_t count1table_select;
72
    int region_size[3]; /* number of huffman codes in each region */
73
    int preflag;
74
    int short_start, long_end; /* long/short band indexes */
75
    uint8_t scale_factors[40];
76
    int32_t sb_hybrid[SBLIMIT * 18]; /* 576 samples */
77
} GranuleDef;
78

    
79
#include "mpegaudiodata.h"
80
#include "mpegaudiodectab.h"
81

    
82
static void compute_antialias_integer(MPADecodeContext *s, GranuleDef *g);
83
static void compute_antialias_float(MPADecodeContext *s, GranuleDef *g);
84

    
85
/* vlc structure for decoding layer 3 huffman tables */
86
static VLC huff_vlc[16];
87
static VLC huff_quad_vlc[2];
88
/* computed from band_size_long */
89
static uint16_t band_index_long[9][23];
90
/* XXX: free when all decoders are closed */
91
#define TABLE_4_3_SIZE (8191 + 16)*4
92
static int8_t  table_4_3_exp[TABLE_4_3_SIZE];
93
static uint32_t table_4_3_value[TABLE_4_3_SIZE];
94
static uint32_t exp_table[512];
95
static uint32_t expval_table[512][16];
96
/* intensity stereo coef table */
97
static int32_t is_table[2][16];
98
static int32_t is_table_lsf[2][2][16];
99
static int32_t csa_table[8][4];
100
static float csa_table_float[8][4];
101
static int32_t mdct_win[8][36];
102

    
103
/* lower 2 bits: modulo 3, higher bits: shift */
104
static uint16_t scale_factor_modshift[64];
105
/* [i][j]:  2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
106
static int32_t scale_factor_mult[15][3];
107
/* mult table for layer 2 group quantization */
108

    
109
#define SCALE_GEN(v) \
110
{ FIXR(1.0 * (v)), FIXR(0.7937005259 * (v)), FIXR(0.6299605249 * (v)) }
111

    
112
static const int32_t scale_factor_mult2[3][3] = {
113
    SCALE_GEN(4.0 / 3.0), /* 3 steps */
114
    SCALE_GEN(4.0 / 5.0), /* 5 steps */
115
    SCALE_GEN(4.0 / 9.0), /* 9 steps */
116
};
117

    
118
static DECLARE_ALIGNED_16(MPA_INT, window[512]);
119

    
120
/**
121
 * Convert region offsets to region sizes and truncate
122
 * size to big_values.
123
 */
124
void ff_region_offset2size(GranuleDef *g){
125
    int i, k, j=0;
126
    g->region_size[2] = (576 / 2);
127
    for(i=0;i<3;i++) {
128
        k = FFMIN(g->region_size[i], g->big_values);
129
        g->region_size[i] = k - j;
130
        j = k;
131
    }
132
}
133

    
134
void ff_init_short_region(MPADecodeContext *s, GranuleDef *g){
135
    if (g->block_type == 2)
136
        g->region_size[0] = (36 / 2);
137
    else {
138
        if (s->sample_rate_index <= 2)
139
            g->region_size[0] = (36 / 2);
140
        else if (s->sample_rate_index != 8)
141
            g->region_size[0] = (54 / 2);
142
        else
143
            g->region_size[0] = (108 / 2);
144
    }
145
    g->region_size[1] = (576 / 2);
146
}
147

    
148
void ff_init_long_region(MPADecodeContext *s, GranuleDef *g, int ra1, int ra2){
149
    int l;
150
    g->region_size[0] =
151
        band_index_long[s->sample_rate_index][ra1 + 1] >> 1;
152
    /* should not overflow */
153
    l = FFMIN(ra1 + ra2 + 2, 22);
154
    g->region_size[1] =
155
        band_index_long[s->sample_rate_index][l] >> 1;
156
}
157

    
158
void ff_compute_band_indexes(MPADecodeContext *s, GranuleDef *g){
159
    if (g->block_type == 2) {
160
        if (g->switch_point) {
161
            /* if switched mode, we handle the 36 first samples as
162
                long blocks.  For 8000Hz, we handle the 48 first
163
                exponents as long blocks (XXX: check this!) */
164
            if (s->sample_rate_index <= 2)
165
                g->long_end = 8;
166
            else if (s->sample_rate_index != 8)
167
                g->long_end = 6;
168
            else
169
                g->long_end = 4; /* 8000 Hz */
170

    
171
            g->short_start = 2 + (s->sample_rate_index != 8);
172
        } else {
173
            g->long_end = 0;
174
            g->short_start = 0;
175
        }
176
    } else {
177
        g->short_start = 13;
178
        g->long_end = 22;
179
    }
180
}
181

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

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

    
198
static inline int l2_unscale_group(int steps, int mant, int scale_factor)
199
{
200
    int shift, mod, val;
201

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

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

    
213
/* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
214
static inline int l3_unscale(int value, int exponent)
215
{
216
    unsigned int m;
217
    int e;
218

    
219
    e = table_4_3_exp  [4*value + (exponent&3)];
220
    m = table_4_3_value[4*value + (exponent&3)];
221
    e -= (exponent >> 2);
222
    assert(e>=1);
223
    if (e > 31)
224
        return 0;
225
    m = (m + (1 << (e-1))) >> e;
226

    
227
    return m;
228
}
229

    
230
/* all integer n^(4/3) computation code */
231
#define DEV_ORDER 13
232

    
233
#define POW_FRAC_BITS 24
234
#define POW_FRAC_ONE    (1 << POW_FRAC_BITS)
235
#define POW_FIX(a)   ((int)((a) * POW_FRAC_ONE))
236
#define POW_MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> POW_FRAC_BITS)
237

    
238
static int dev_4_3_coefs[DEV_ORDER];
239

    
240
#if 0 /* unused */
241
static int pow_mult3[3] = {
242
    POW_FIX(1.0),
243
    POW_FIX(1.25992104989487316476),
244
    POW_FIX(1.58740105196819947474),
245
};
246
#endif
247

    
248
static void int_pow_init(void)
249
{
250
    int i, a;
251

    
252
    a = POW_FIX(1.0);
253
    for(i=0;i<DEV_ORDER;i++) {
254
        a = POW_MULL(a, POW_FIX(4.0 / 3.0) - i * POW_FIX(1.0)) / (i + 1);
255
        dev_4_3_coefs[i] = a;
256
    }
257
}
258

    
259
#if 0 /* unused, remove? */
260
/* return the mantissa and the binary exponent */
261
static int int_pow(int i, int *exp_ptr)
262
{
263
    int e, er, eq, j;
264
    int a, a1;
265

266
    /* renormalize */
267
    a = i;
268
    e = POW_FRAC_BITS;
269
    while (a < (1 << (POW_FRAC_BITS - 1))) {
270
        a = a << 1;
271
        e--;
272
    }
273
    a -= (1 << POW_FRAC_BITS);
274
    a1 = 0;
275
    for(j = DEV_ORDER - 1; j >= 0; j--)
276
        a1 = POW_MULL(a, dev_4_3_coefs[j] + a1);
277
    a = (1 << POW_FRAC_BITS) + a1;
278
    /* exponent compute (exact) */
279
    e = e * 4;
280
    er = e % 3;
281
    eq = e / 3;
282
    a = POW_MULL(a, pow_mult3[er]);
283
    while (a >= 2 * POW_FRAC_ONE) {
284
        a = a >> 1;
285
        eq++;
286
    }
287
    /* convert to float */
288
    while (a < POW_FRAC_ONE) {
289
        a = a << 1;
290
        eq--;
291
    }
292
    /* now POW_FRAC_ONE <= a < 2 * POW_FRAC_ONE */
293
#if POW_FRAC_BITS > FRAC_BITS
294
    a = (a + (1 << (POW_FRAC_BITS - FRAC_BITS - 1))) >> (POW_FRAC_BITS - FRAC_BITS);
295
    /* correct overflow */
296
    if (a >= 2 * (1 << FRAC_BITS)) {
297
        a = a >> 1;
298
        eq++;
299
    }
300
#endif
301
    *exp_ptr = eq;
302
    return a;
303
}
304
#endif
305

    
306
static int decode_init(AVCodecContext * avctx)
307
{
308
    MPADecodeContext *s = avctx->priv_data;
309
    static int init=0;
310
    int i, j, k;
311

    
312
    s->avctx = avctx;
313

    
314
#if defined(USE_HIGHPRECISION) && defined(CONFIG_AUDIO_NONSHORT)
315
    avctx->sample_fmt= SAMPLE_FMT_S32;
316
#else
317
    avctx->sample_fmt= SAMPLE_FMT_S16;
318
#endif
319
    s->error_resilience= avctx->error_resilience;
320

    
321
    if(avctx->antialias_algo != FF_AA_FLOAT)
322
        s->compute_antialias= compute_antialias_integer;
323
    else
324
        s->compute_antialias= compute_antialias_float;
325

    
326
    if (!init && !avctx->parse_only) {
327
        /* scale factors table for layer 1/2 */
328
        for(i=0;i<64;i++) {
329
            int shift, mod;
330
            /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */
331
            shift = (i / 3);
332
            mod = i % 3;
333
            scale_factor_modshift[i] = mod | (shift << 2);
334
        }
335

    
336
        /* scale factor multiply for layer 1 */
337
        for(i=0;i<15;i++) {
338
            int n, norm;
339
            n = i + 2;
340
            norm = ((INT64_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
341
            scale_factor_mult[i][0] = MULL(FIXR(1.0 * 2.0), norm);
342
            scale_factor_mult[i][1] = MULL(FIXR(0.7937005259 * 2.0), norm);
343
            scale_factor_mult[i][2] = MULL(FIXR(0.6299605249 * 2.0), norm);
344
            dprintf(avctx, "%d: norm=%x s=%x %x %x\n",
345
                    i, norm,
346
                    scale_factor_mult[i][0],
347
                    scale_factor_mult[i][1],
348
                    scale_factor_mult[i][2]);
349
        }
350

    
351
        ff_mpa_synth_init(window);
352

    
353
        /* huffman decode tables */
354
        for(i=1;i<16;i++) {
355
            const HuffTable *h = &mpa_huff_tables[i];
356
            int xsize, x, y;
357
            unsigned int n;
358
            uint8_t  tmp_bits [512];
359
            uint16_t tmp_codes[512];
360

    
361
            memset(tmp_bits , 0, sizeof(tmp_bits ));
362
            memset(tmp_codes, 0, sizeof(tmp_codes));
363

    
364
            xsize = h->xsize;
365
            n = xsize * xsize;
366

    
367
            j = 0;
368
            for(x=0;x<xsize;x++) {
369
                for(y=0;y<xsize;y++){
370
                    tmp_bits [(x << 5) | y | ((x&&y)<<4)]= h->bits [j  ];
371
                    tmp_codes[(x << 5) | y | ((x&&y)<<4)]= h->codes[j++];
372
                }
373
            }
374

    
375
            /* XXX: fail test */
376
            init_vlc(&huff_vlc[i], 7, 512,
377
                     tmp_bits, 1, 1, tmp_codes, 2, 2, 1);
378
        }
379
        for(i=0;i<2;i++) {
380
            init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
381
                     mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1, 1);
382
        }
383

    
384
        for(i=0;i<9;i++) {
385
            k = 0;
386
            for(j=0;j<22;j++) {
387
                band_index_long[i][j] = k;
388
                k += band_size_long[i][j];
389
            }
390
            band_index_long[i][22] = k;
391
        }
392

    
393
        /* compute n ^ (4/3) and store it in mantissa/exp format */
394

    
395
        int_pow_init();
396
        for(i=1;i<TABLE_4_3_SIZE;i++) {
397
            double f, fm;
398
            int e, m;
399
            f = pow((double)(i/4), 4.0 / 3.0) * pow(2, (i&3)*0.25);
400
            fm = frexp(f, &e);
401
            m = (uint32_t)(fm*(1LL<<31) + 0.5);
402
            e+= FRAC_BITS - 31 + 5 - 100;
403

    
404
            /* normalized to FRAC_BITS */
405
            table_4_3_value[i] = m;
406
//            av_log(NULL, AV_LOG_DEBUG, "%d %d %f\n", i, m, pow((double)i, 4.0 / 3.0));
407
            table_4_3_exp[i] = -e;
408
        }
409
        for(i=0; i<512*16; i++){
410
            int exponent= (i>>4);
411
            double f= pow(i&15, 4.0 / 3.0) * pow(2, (exponent-400)*0.25 + FRAC_BITS + 5);
412
            expval_table[exponent][i&15]= llrint(f);
413
            if((i&15)==1)
414
                exp_table[exponent]= llrint(f);
415
        }
416

    
417
        for(i=0;i<7;i++) {
418
            float f;
419
            int v;
420
            if (i != 6) {
421
                f = tan((double)i * M_PI / 12.0);
422
                v = FIXR(f / (1.0 + f));
423
            } else {
424
                v = FIXR(1.0);
425
            }
426
            is_table[0][i] = v;
427
            is_table[1][6 - i] = v;
428
        }
429
        /* invalid values */
430
        for(i=7;i<16;i++)
431
            is_table[0][i] = is_table[1][i] = 0.0;
432

    
433
        for(i=0;i<16;i++) {
434
            double f;
435
            int e, k;
436

    
437
            for(j=0;j<2;j++) {
438
                e = -(j + 1) * ((i + 1) >> 1);
439
                f = pow(2.0, e / 4.0);
440
                k = i & 1;
441
                is_table_lsf[j][k ^ 1][i] = FIXR(f);
442
                is_table_lsf[j][k][i] = FIXR(1.0);
443
                dprintf(avctx, "is_table_lsf %d %d: %x %x\n",
444
                        i, j, is_table_lsf[j][0][i], is_table_lsf[j][1][i]);
445
            }
446
        }
447

    
448
        for(i=0;i<8;i++) {
449
            float ci, cs, ca;
450
            ci = ci_table[i];
451
            cs = 1.0 / sqrt(1.0 + ci * ci);
452
            ca = cs * ci;
453
            csa_table[i][0] = FIXHR(cs/4);
454
            csa_table[i][1] = FIXHR(ca/4);
455
            csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);
456
            csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4);
457
            csa_table_float[i][0] = cs;
458
            csa_table_float[i][1] = ca;
459
            csa_table_float[i][2] = ca + cs;
460
            csa_table_float[i][3] = ca - cs;
461
//            printf("%d %d %d %d\n", FIX(cs), FIX(cs-1), FIX(ca), FIX(cs)-FIX(ca));
462
//            av_log(NULL, AV_LOG_DEBUG,"%f %f %f %f\n", cs, ca, ca+cs, ca-cs);
463
        }
464

    
465
        /* compute mdct windows */
466
        for(i=0;i<36;i++) {
467
            for(j=0; j<4; j++){
468
                double d;
469

    
470
                if(j==2 && i%3 != 1)
471
                    continue;
472

    
473
                d= sin(M_PI * (i + 0.5) / 36.0);
474
                if(j==1){
475
                    if     (i>=30) d= 0;
476
                    else if(i>=24) d= sin(M_PI * (i - 18 + 0.5) / 12.0);
477
                    else if(i>=18) d= 1;
478
                }else if(j==3){
479
                    if     (i<  6) d= 0;
480
                    else if(i< 12) d= sin(M_PI * (i -  6 + 0.5) / 12.0);
481
                    else if(i< 18) d= 1;
482
                }
483
                //merge last stage of imdct into the window coefficients
484
                d*= 0.5 / cos(M_PI*(2*i + 19)/72);
485

    
486
                if(j==2)
487
                    mdct_win[j][i/3] = FIXHR((d / (1<<5)));
488
                else
489
                    mdct_win[j][i  ] = FIXHR((d / (1<<5)));
490
//                av_log(NULL, AV_LOG_DEBUG, "%2d %d %f\n", i,j,d / (1<<5));
491
            }
492
        }
493

    
494
        /* NOTE: we do frequency inversion adter the MDCT by changing
495
           the sign of the right window coefs */
496
        for(j=0;j<4;j++) {
497
            for(i=0;i<36;i+=2) {
498
                mdct_win[j + 4][i] = mdct_win[j][i];
499
                mdct_win[j + 4][i + 1] = -mdct_win[j][i + 1];
500
            }
501
        }
502

    
503
#if defined(DEBUG)
504
        for(j=0;j<8;j++) {
505
            av_log(avctx, AV_LOG_DEBUG, "win%d=\n", j);
506
            for(i=0;i<36;i++)
507
                av_log(avctx, AV_LOG_DEBUG, "%f, ", (double)mdct_win[j][i] / FRAC_ONE);
508
            av_log(avctx, AV_LOG_DEBUG, "\n");
509
        }
510
#endif
511
        init = 1;
512
    }
513

    
514
#ifdef DEBUG
515
    s->frame_count = 0;
516
#endif
517
    if (avctx->codec_id == CODEC_ID_MP3ADU)
518
        s->adu_mode = 1;
519
    return 0;
520
}
521

    
522
/* tab[i][j] = 1.0 / (2.0 * cos(pi*(2*k+1) / 2^(6 - j))) */
523

    
524
/* cos(i*pi/64) */
525

    
526
#define COS0_0  FIXHR(0.50060299823519630134/2)
527
#define COS0_1  FIXHR(0.50547095989754365998/2)
528
#define COS0_2  FIXHR(0.51544730992262454697/2)
529
#define COS0_3  FIXHR(0.53104259108978417447/2)
530
#define COS0_4  FIXHR(0.55310389603444452782/2)
531
#define COS0_5  FIXHR(0.58293496820613387367/2)
532
#define COS0_6  FIXHR(0.62250412303566481615/2)
533
#define COS0_7  FIXHR(0.67480834145500574602/2)
534
#define COS0_8  FIXHR(0.74453627100229844977/2)
535
#define COS0_9  FIXHR(0.83934964541552703873/2)
536
#define COS0_10 FIXHR(0.97256823786196069369/2)
537
#define COS0_11 FIXHR(1.16943993343288495515/4)
538
#define COS0_12 FIXHR(1.48416461631416627724/4)
539
#define COS0_13 FIXHR(2.05778100995341155085/8)
540
#define COS0_14 FIXHR(3.40760841846871878570/8)
541
#define COS0_15 FIXHR(10.19000812354805681150/32)
542

    
543
#define COS1_0 FIXHR(0.50241928618815570551/2)
544
#define COS1_1 FIXHR(0.52249861493968888062/2)
545
#define COS1_2 FIXHR(0.56694403481635770368/2)
546
#define COS1_3 FIXHR(0.64682178335999012954/2)
547
#define COS1_4 FIXHR(0.78815462345125022473/2)
548
#define COS1_5 FIXHR(1.06067768599034747134/4)
549
#define COS1_6 FIXHR(1.72244709823833392782/4)
550
#define COS1_7 FIXHR(5.10114861868916385802/16)
551

    
552
#define COS2_0 FIXHR(0.50979557910415916894/2)
553
#define COS2_1 FIXHR(0.60134488693504528054/2)
554
#define COS2_2 FIXHR(0.89997622313641570463/2)
555
#define COS2_3 FIXHR(2.56291544774150617881/8)
556

    
557
#define COS3_0 FIXHR(0.54119610014619698439/2)
558
#define COS3_1 FIXHR(1.30656296487637652785/4)
559

    
560
#define COS4_0 FIXHR(0.70710678118654752439/2)
561

    
562
/* butterfly operator */
563
#define BF(a, b, c, s)\
564
{\
565
    tmp0 = tab[a] + tab[b];\
566
    tmp1 = tab[a] - tab[b];\
567
    tab[a] = tmp0;\
568
    tab[b] = MULH(tmp1<<(s), c);\
569
}
570

    
571
#define BF1(a, b, c, d)\
572
{\
573
    BF(a, b, COS4_0, 1);\
574
    BF(c, d,-COS4_0, 1);\
575
    tab[c] += tab[d];\
576
}
577

    
578
#define BF2(a, b, c, d)\
579
{\
580
    BF(a, b, COS4_0, 1);\
581
    BF(c, d,-COS4_0, 1);\
582
    tab[c] += tab[d];\
583
    tab[a] += tab[c];\
584
    tab[c] += tab[b];\
585
    tab[b] += tab[d];\
586
}
587

    
588
#define ADD(a, b) tab[a] += tab[b]
589

    
590
/* DCT32 without 1/sqrt(2) coef zero scaling. */
591
static void dct32(int32_t *out, int32_t *tab)
592
{
593
    int tmp0, tmp1;
594

    
595
    /* pass 1 */
596
    BF( 0, 31, COS0_0 , 1);
597
    BF(15, 16, COS0_15, 5);
598
    /* pass 2 */
599
    BF( 0, 15, COS1_0 , 1);
600
    BF(16, 31,-COS1_0 , 1);
601
    /* pass 1 */
602
    BF( 7, 24, COS0_7 , 1);
603
    BF( 8, 23, COS0_8 , 1);
604
    /* pass 2 */
605
    BF( 7,  8, COS1_7 , 4);
606
    BF(23, 24,-COS1_7 , 4);
607
    /* pass 3 */
608
    BF( 0,  7, COS2_0 , 1);
609
    BF( 8, 15,-COS2_0 , 1);
610
    BF(16, 23, COS2_0 , 1);
611
    BF(24, 31,-COS2_0 , 1);
612
    /* pass 1 */
613
    BF( 3, 28, COS0_3 , 1);
614
    BF(12, 19, COS0_12, 2);
615
    /* pass 2 */
616
    BF( 3, 12, COS1_3 , 1);
617
    BF(19, 28,-COS1_3 , 1);
618
    /* pass 1 */
619
    BF( 4, 27, COS0_4 , 1);
620
    BF(11, 20, COS0_11, 2);
621
    /* pass 2 */
622
    BF( 4, 11, COS1_4 , 1);
623
    BF(20, 27,-COS1_4 , 1);
624
    /* pass 3 */
625
    BF( 3,  4, COS2_3 , 3);
626
    BF(11, 12,-COS2_3 , 3);
627
    BF(19, 20, COS2_3 , 3);
628
    BF(27, 28,-COS2_3 , 3);
629
    /* pass 4 */
630
    BF( 0,  3, COS3_0 , 1);
631
    BF( 4,  7,-COS3_0 , 1);
632
    BF( 8, 11, COS3_0 , 1);
633
    BF(12, 15,-COS3_0 , 1);
634
    BF(16, 19, COS3_0 , 1);
635
    BF(20, 23,-COS3_0 , 1);
636
    BF(24, 27, COS3_0 , 1);
637
    BF(28, 31,-COS3_0 , 1);
638

    
639

    
640

    
641
    /* pass 1 */
642
    BF( 1, 30, COS0_1 , 1);
643
    BF(14, 17, COS0_14, 3);
644
    /* pass 2 */
645
    BF( 1, 14, COS1_1 , 1);
646
    BF(17, 30,-COS1_1 , 1);
647
    /* pass 1 */
648
    BF( 6, 25, COS0_6 , 1);
649
    BF( 9, 22, COS0_9 , 1);
650
    /* pass 2 */
651
    BF( 6,  9, COS1_6 , 2);
652
    BF(22, 25,-COS1_6 , 2);
653
    /* pass 3 */
654
    BF( 1,  6, COS2_1 , 1);
655
    BF( 9, 14,-COS2_1 , 1);
656
    BF(17, 22, COS2_1 , 1);
657
    BF(25, 30,-COS2_1 , 1);
658

    
659
    /* pass 1 */
660
    BF( 2, 29, COS0_2 , 1);
661
    BF(13, 18, COS0_13, 3);
662
    /* pass 2 */
663
    BF( 2, 13, COS1_2 , 1);
664
    BF(18, 29,-COS1_2 , 1);
665
    /* pass 1 */
666
    BF( 5, 26, COS0_5 , 1);
667
    BF(10, 21, COS0_10, 1);
668
    /* pass 2 */
669
    BF( 5, 10, COS1_5 , 2);
670
    BF(21, 26,-COS1_5 , 2);
671
    /* pass 3 */
672
    BF( 2,  5, COS2_2 , 1);
673
    BF(10, 13,-COS2_2 , 1);
674
    BF(18, 21, COS2_2 , 1);
675
    BF(26, 29,-COS2_2 , 1);
676
    /* pass 4 */
677
    BF( 1,  2, COS3_1 , 2);
678
    BF( 5,  6,-COS3_1 , 2);
679
    BF( 9, 10, COS3_1 , 2);
680
    BF(13, 14,-COS3_1 , 2);
681
    BF(17, 18, COS3_1 , 2);
682
    BF(21, 22,-COS3_1 , 2);
683
    BF(25, 26, COS3_1 , 2);
684
    BF(29, 30,-COS3_1 , 2);
685

    
686
    /* pass 5 */
687
    BF1( 0,  1,  2,  3);
688
    BF2( 4,  5,  6,  7);
689
    BF1( 8,  9, 10, 11);
690
    BF2(12, 13, 14, 15);
691
    BF1(16, 17, 18, 19);
692
    BF2(20, 21, 22, 23);
693
    BF1(24, 25, 26, 27);
694
    BF2(28, 29, 30, 31);
695

    
696
    /* pass 6 */
697

    
698
    ADD( 8, 12);
699
    ADD(12, 10);
700
    ADD(10, 14);
701
    ADD(14,  9);
702
    ADD( 9, 13);
703
    ADD(13, 11);
704
    ADD(11, 15);
705

    
706
    out[ 0] = tab[0];
707
    out[16] = tab[1];
708
    out[ 8] = tab[2];
709
    out[24] = tab[3];
710
    out[ 4] = tab[4];
711
    out[20] = tab[5];
712
    out[12] = tab[6];
713
    out[28] = tab[7];
714
    out[ 2] = tab[8];
715
    out[18] = tab[9];
716
    out[10] = tab[10];
717
    out[26] = tab[11];
718
    out[ 6] = tab[12];
719
    out[22] = tab[13];
720
    out[14] = tab[14];
721
    out[30] = tab[15];
722

    
723
    ADD(24, 28);
724
    ADD(28, 26);
725
    ADD(26, 30);
726
    ADD(30, 25);
727
    ADD(25, 29);
728
    ADD(29, 27);
729
    ADD(27, 31);
730

    
731
    out[ 1] = tab[16] + tab[24];
732
    out[17] = tab[17] + tab[25];
733
    out[ 9] = tab[18] + tab[26];
734
    out[25] = tab[19] + tab[27];
735
    out[ 5] = tab[20] + tab[28];
736
    out[21] = tab[21] + tab[29];
737
    out[13] = tab[22] + tab[30];
738
    out[29] = tab[23] + tab[31];
739
    out[ 3] = tab[24] + tab[20];
740
    out[19] = tab[25] + tab[21];
741
    out[11] = tab[26] + tab[22];
742
    out[27] = tab[27] + tab[23];
743
    out[ 7] = tab[28] + tab[18];
744
    out[23] = tab[29] + tab[19];
745
    out[15] = tab[30] + tab[17];
746
    out[31] = tab[31];
747
}
748

    
749
#if FRAC_BITS <= 15
750

    
751
static inline int round_sample(int *sum)
752
{
753
    int sum1;
754
    sum1 = (*sum) >> OUT_SHIFT;
755
    *sum &= (1<<OUT_SHIFT)-1;
756
    if (sum1 < OUT_MIN)
757
        sum1 = OUT_MIN;
758
    else if (sum1 > OUT_MAX)
759
        sum1 = OUT_MAX;
760
    return sum1;
761
}
762

    
763
/* signed 16x16 -> 32 multiply add accumulate */
764
#define MACS(rt, ra, rb) MAC16(rt, ra, rb)
765

    
766
/* signed 16x16 -> 32 multiply */
767
#define MULS(ra, rb) MUL16(ra, rb)
768

    
769
#else
770

    
771
static inline int round_sample(int64_t *sum)
772
{
773
    int sum1;
774
    sum1 = (int)((*sum) >> OUT_SHIFT);
775
    *sum &= (1<<OUT_SHIFT)-1;
776
    if (sum1 < OUT_MIN)
777
        sum1 = OUT_MIN;
778
    else if (sum1 > OUT_MAX)
779
        sum1 = OUT_MAX;
780
    return sum1;
781
}
782

    
783
#   define MULS(ra, rb) MUL64(ra, rb)
784
#endif
785

    
786
#define SUM8(sum, op, w, p) \
787
{                                               \
788
    sum op MULS((w)[0 * 64], p[0 * 64]);\
789
    sum op MULS((w)[1 * 64], p[1 * 64]);\
790
    sum op MULS((w)[2 * 64], p[2 * 64]);\
791
    sum op MULS((w)[3 * 64], p[3 * 64]);\
792
    sum op MULS((w)[4 * 64], p[4 * 64]);\
793
    sum op MULS((w)[5 * 64], p[5 * 64]);\
794
    sum op MULS((w)[6 * 64], p[6 * 64]);\
795
    sum op MULS((w)[7 * 64], p[7 * 64]);\
796
}
797

    
798
#define SUM8P2(sum1, op1, sum2, op2, w1, w2, p) \
799
{                                               \
800
    int tmp;\
801
    tmp = p[0 * 64];\
802
    sum1 op1 MULS((w1)[0 * 64], tmp);\
803
    sum2 op2 MULS((w2)[0 * 64], tmp);\
804
    tmp = p[1 * 64];\
805
    sum1 op1 MULS((w1)[1 * 64], tmp);\
806
    sum2 op2 MULS((w2)[1 * 64], tmp);\
807
    tmp = p[2 * 64];\
808
    sum1 op1 MULS((w1)[2 * 64], tmp);\
809
    sum2 op2 MULS((w2)[2 * 64], tmp);\
810
    tmp = p[3 * 64];\
811
    sum1 op1 MULS((w1)[3 * 64], tmp);\
812
    sum2 op2 MULS((w2)[3 * 64], tmp);\
813
    tmp = p[4 * 64];\
814
    sum1 op1 MULS((w1)[4 * 64], tmp);\
815
    sum2 op2 MULS((w2)[4 * 64], tmp);\
816
    tmp = p[5 * 64];\
817
    sum1 op1 MULS((w1)[5 * 64], tmp);\
818
    sum2 op2 MULS((w2)[5 * 64], tmp);\
819
    tmp = p[6 * 64];\
820
    sum1 op1 MULS((w1)[6 * 64], tmp);\
821
    sum2 op2 MULS((w2)[6 * 64], tmp);\
822
    tmp = p[7 * 64];\
823
    sum1 op1 MULS((w1)[7 * 64], tmp);\
824
    sum2 op2 MULS((w2)[7 * 64], tmp);\
825
}
826

    
827
void ff_mpa_synth_init(MPA_INT *window)
828
{
829
    int i;
830

    
831
    /* max = 18760, max sum over all 16 coefs : 44736 */
832
    for(i=0;i<257;i++) {
833
        int v;
834
        v = ff_mpa_enwindow[i];
835
#if WFRAC_BITS < 16
836
        v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS);
837
#endif
838
        window[i] = v;
839
        if ((i & 63) != 0)
840
            v = -v;
841
        if (i != 0)
842
            window[512 - i] = v;
843
    }
844
}
845

    
846
/* 32 sub band synthesis filter. Input: 32 sub band samples, Output:
847
   32 samples. */
848
/* XXX: optimize by avoiding ring buffer usage */
849
void ff_mpa_synth_filter(MPA_INT *synth_buf_ptr, int *synth_buf_offset,
850
                         MPA_INT *window, int *dither_state,
851
                         OUT_INT *samples, int incr,
852
                         int32_t sb_samples[SBLIMIT])
853
{
854
    int32_t tmp[32];
855
    register MPA_INT *synth_buf;
856
    register const MPA_INT *w, *w2, *p;
857
    int j, offset, v;
858
    OUT_INT *samples2;
859
#if FRAC_BITS <= 15
860
    int sum, sum2;
861
#else
862
    int64_t sum, sum2;
863
#endif
864

    
865
    dct32(tmp, sb_samples);
866

    
867
    offset = *synth_buf_offset;
868
    synth_buf = synth_buf_ptr + offset;
869

    
870
    for(j=0;j<32;j++) {
871
        v = tmp[j];
872
#if FRAC_BITS <= 15
873
        /* NOTE: can cause a loss in precision if very high amplitude
874
           sound */
875
        v = av_clip_int16(v);
876
#endif
877
        synth_buf[j] = v;
878
    }
879
    /* copy to avoid wrap */
880
    memcpy(synth_buf + 512, synth_buf, 32 * sizeof(MPA_INT));
881

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

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

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

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

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

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

    
922
#define C3 FIXHR(0.86602540378443864676/2)
923

    
924
/* 0.5 / cos(pi*(2*i+1)/36) */
925
static const int icos36[9] = {
926
    FIXR(0.50190991877167369479),
927
    FIXR(0.51763809020504152469), //0
928
    FIXR(0.55168895948124587824),
929
    FIXR(0.61038729438072803416),
930
    FIXR(0.70710678118654752439), //1
931
    FIXR(0.87172339781054900991),
932
    FIXR(1.18310079157624925896),
933
    FIXR(1.93185165257813657349), //2
934
    FIXR(5.73685662283492756461),
935
};
936

    
937
/* 0.5 / cos(pi*(2*i+1)/36) */
938
static const int icos36h[9] = {
939
    FIXHR(0.50190991877167369479/2),
940
    FIXHR(0.51763809020504152469/2), //0
941
    FIXHR(0.55168895948124587824/2),
942
    FIXHR(0.61038729438072803416/2),
943
    FIXHR(0.70710678118654752439/2), //1
944
    FIXHR(0.87172339781054900991/2),
945
    FIXHR(1.18310079157624925896/4),
946
    FIXHR(1.93185165257813657349/4), //2
947
//    FIXHR(5.73685662283492756461),
948
};
949

    
950
/* 12 points IMDCT. We compute it "by hand" by factorizing obvious
951
   cases. */
952
static void imdct12(int *out, int *in)
953
{
954
    int in0, in1, in2, in3, in4, in5, t1, t2;
955

    
956
    in0= in[0*3];
957
    in1= in[1*3] + in[0*3];
958
    in2= in[2*3] + in[1*3];
959
    in3= in[3*3] + in[2*3];
960
    in4= in[4*3] + in[3*3];
961
    in5= in[5*3] + in[4*3];
962
    in5 += in3;
963
    in3 += in1;
964

    
965
    in2= MULH(2*in2, C3);
966
    in3= MULH(4*in3, C3);
967

    
968
    t1 = in0 - in4;
969
    t2 = MULH(2*(in1 - in5), icos36h[4]);
970

    
971
    out[ 7]=
972
    out[10]= t1 + t2;
973
    out[ 1]=
974
    out[ 4]= t1 - t2;
975

    
976
    in0 += in4>>1;
977
    in4 = in0 + in2;
978
    in5 += 2*in1;
979
    in1 = MULH(in5 + in3, icos36h[1]);
980
    out[ 8]=
981
    out[ 9]= in4 + in1;
982
    out[ 2]=
983
    out[ 3]= in4 - in1;
984

    
985
    in0 -= in2;
986
    in5 = MULH(2*(in5 - in3), icos36h[7]);
987
    out[ 0]=
988
    out[ 5]= in0 - in5;
989
    out[ 6]=
990
    out[11]= in0 + in5;
991
}
992

    
993
/* cos(pi*i/18) */
994
#define C1 FIXHR(0.98480775301220805936/2)
995
#define C2 FIXHR(0.93969262078590838405/2)
996
#define C3 FIXHR(0.86602540378443864676/2)
997
#define C4 FIXHR(0.76604444311897803520/2)
998
#define C5 FIXHR(0.64278760968653932632/2)
999
#define C6 FIXHR(0.5/2)
1000
#define C7 FIXHR(0.34202014332566873304/2)
1001
#define C8 FIXHR(0.17364817766693034885/2)
1002

    
1003

    
1004
/* using Lee like decomposition followed by hand coded 9 points DCT */
1005
static void imdct36(int *out, int *buf, int *in, int *win)
1006
{
1007
    int i, j, t0, t1, t2, t3, s0, s1, s2, s3;
1008
    int tmp[18], *tmp1, *in1;
1009

    
1010
    for(i=17;i>=1;i--)
1011
        in[i] += in[i-1];
1012
    for(i=17;i>=3;i-=2)
1013
        in[i] += in[i-2];
1014

    
1015
    for(j=0;j<2;j++) {
1016
        tmp1 = tmp + j;
1017
        in1 = in + j;
1018
#if 0
1019
//more accurate but slower
1020
        int64_t t0, t1, t2, t3;
1021
        t2 = in1[2*4] + in1[2*8] - in1[2*2];
1022

1023
        t3 = (in1[2*0] + (int64_t)(in1[2*6]>>1))<<32;
1024
        t1 = in1[2*0] - in1[2*6];
1025
        tmp1[ 6] = t1 - (t2>>1);
1026
        tmp1[16] = t1 + t2;
1027

1028
        t0 = MUL64(2*(in1[2*2] + in1[2*4]),    C2);
1029
        t1 = MUL64(   in1[2*4] - in1[2*8] , -2*C8);
1030
        t2 = MUL64(2*(in1[2*2] + in1[2*8]),   -C4);
1031

1032
        tmp1[10] = (t3 - t0 - t2) >> 32;
1033
        tmp1[ 2] = (t3 + t0 + t1) >> 32;
1034
        tmp1[14] = (t3 + t2 - t1) >> 32;
1035

1036
        tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
1037
        t2 = MUL64(2*(in1[2*1] + in1[2*5]),    C1);
1038
        t3 = MUL64(   in1[2*5] - in1[2*7] , -2*C7);
1039
        t0 = MUL64(2*in1[2*3], C3);
1040

1041
        t1 = MUL64(2*(in1[2*1] + in1[2*7]),   -C5);
1042

1043
        tmp1[ 0] = (t2 + t3 + t0) >> 32;
1044
        tmp1[12] = (t2 + t1 - t0) >> 32;
1045
        tmp1[ 8] = (t3 - t1 - t0) >> 32;
1046
#else
1047
        t2 = in1[2*4] + in1[2*8] - in1[2*2];
1048

    
1049
        t3 = in1[2*0] + (in1[2*6]>>1);
1050
        t1 = in1[2*0] - in1[2*6];
1051
        tmp1[ 6] = t1 - (t2>>1);
1052
        tmp1[16] = t1 + t2;
1053

    
1054
        t0 = MULH(2*(in1[2*2] + in1[2*4]),    C2);
1055
        t1 = MULH(   in1[2*4] - in1[2*8] , -2*C8);
1056
        t2 = MULH(2*(in1[2*2] + in1[2*8]),   -C4);
1057

    
1058
        tmp1[10] = t3 - t0 - t2;
1059
        tmp1[ 2] = t3 + t0 + t1;
1060
        tmp1[14] = t3 + t2 - t1;
1061

    
1062
        tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
1063
        t2 = MULH(2*(in1[2*1] + in1[2*5]),    C1);
1064
        t3 = MULH(   in1[2*5] - in1[2*7] , -2*C7);
1065
        t0 = MULH(2*in1[2*3], C3);
1066

    
1067
        t1 = MULH(2*(in1[2*1] + in1[2*7]),   -C5);
1068

    
1069
        tmp1[ 0] = t2 + t3 + t0;
1070
        tmp1[12] = t2 + t1 - t0;
1071
        tmp1[ 8] = t3 - t1 - t0;
1072
#endif
1073
    }
1074

    
1075
    i = 0;
1076
    for(j=0;j<4;j++) {
1077
        t0 = tmp[i];
1078
        t1 = tmp[i + 2];
1079
        s0 = t1 + t0;
1080
        s2 = t1 - t0;
1081

    
1082
        t2 = tmp[i + 1];
1083
        t3 = tmp[i + 3];
1084
        s1 = MULH(2*(t3 + t2), icos36h[j]);
1085
        s3 = MULL(t3 - t2, icos36[8 - j]);
1086

    
1087
        t0 = s0 + s1;
1088
        t1 = s0 - s1;
1089
        out[(9 + j)*SBLIMIT] =  MULH(t1, win[9 + j]) + buf[9 + j];
1090
        out[(8 - j)*SBLIMIT] =  MULH(t1, win[8 - j]) + buf[8 - j];
1091
        buf[9 + j] = MULH(t0, win[18 + 9 + j]);
1092
        buf[8 - j] = MULH(t0, win[18 + 8 - j]);
1093

    
1094
        t0 = s2 + s3;
1095
        t1 = s2 - s3;
1096
        out[(9 + 8 - j)*SBLIMIT] =  MULH(t1, win[9 + 8 - j]) + buf[9 + 8 - j];
1097
        out[(        j)*SBLIMIT] =  MULH(t1, win[        j]) + buf[        j];
1098
        buf[9 + 8 - j] = MULH(t0, win[18 + 9 + 8 - j]);
1099
        buf[      + j] = MULH(t0, win[18         + j]);
1100
        i += 4;
1101
    }
1102

    
1103
    s0 = tmp[16];
1104
    s1 = MULH(2*tmp[17], icos36h[4]);
1105
    t0 = s0 + s1;
1106
    t1 = s0 - s1;
1107
    out[(9 + 4)*SBLIMIT] =  MULH(t1, win[9 + 4]) + buf[9 + 4];
1108
    out[(8 - 4)*SBLIMIT] =  MULH(t1, win[8 - 4]) + buf[8 - 4];
1109
    buf[9 + 4] = MULH(t0, win[18 + 9 + 4]);
1110
    buf[8 - 4] = MULH(t0, win[18 + 8 - 4]);
1111
}
1112

    
1113
/* return the number of decoded frames */
1114
static int mp_decode_layer1(MPADecodeContext *s)
1115
{
1116
    int bound, i, v, n, ch, j, mant;
1117
    uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
1118
    uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
1119

    
1120
    if (s->mode == MPA_JSTEREO)
1121
        bound = (s->mode_ext + 1) * 4;
1122
    else
1123
        bound = SBLIMIT;
1124

    
1125
    /* allocation bits */
1126
    for(i=0;i<bound;i++) {
1127
        for(ch=0;ch<s->nb_channels;ch++) {
1128
            allocation[ch][i] = get_bits(&s->gb, 4);
1129
        }
1130
    }
1131
    for(i=bound;i<SBLIMIT;i++) {
1132
        allocation[0][i] = get_bits(&s->gb, 4);
1133
    }
1134

    
1135
    /* scale factors */
1136
    for(i=0;i<bound;i++) {
1137
        for(ch=0;ch<s->nb_channels;ch++) {
1138
            if (allocation[ch][i])
1139
                scale_factors[ch][i] = get_bits(&s->gb, 6);
1140
        }
1141
    }
1142
    for(i=bound;i<SBLIMIT;i++) {
1143
        if (allocation[0][i]) {
1144
            scale_factors[0][i] = get_bits(&s->gb, 6);
1145
            scale_factors[1][i] = get_bits(&s->gb, 6);
1146
        }
1147
    }
1148

    
1149
    /* compute samples */
1150
    for(j=0;j<12;j++) {
1151
        for(i=0;i<bound;i++) {
1152
            for(ch=0;ch<s->nb_channels;ch++) {
1153
                n = allocation[ch][i];
1154
                if (n) {
1155
                    mant = get_bits(&s->gb, n + 1);
1156
                    v = l1_unscale(n, mant, scale_factors[ch][i]);
1157
                } else {
1158
                    v = 0;
1159
                }
1160
                s->sb_samples[ch][j][i] = v;
1161
            }
1162
        }
1163
        for(i=bound;i<SBLIMIT;i++) {
1164
            n = allocation[0][i];
1165
            if (n) {
1166
                mant = get_bits(&s->gb, n + 1);
1167
                v = l1_unscale(n, mant, scale_factors[0][i]);
1168
                s->sb_samples[0][j][i] = v;
1169
                v = l1_unscale(n, mant, scale_factors[1][i]);
1170
                s->sb_samples[1][j][i] = v;
1171
            } else {
1172
                s->sb_samples[0][j][i] = 0;
1173
                s->sb_samples[1][j][i] = 0;
1174
            }
1175
        }
1176
    }
1177
    return 12;
1178
}
1179

    
1180
static int mp_decode_layer2(MPADecodeContext *s)
1181
{
1182
    int sblimit; /* number of used subbands */
1183
    const unsigned char *alloc_table;
1184
    int table, bit_alloc_bits, i, j, ch, bound, v;
1185
    unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
1186
    unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
1187
    unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
1188
    int scale, qindex, bits, steps, k, l, m, b;
1189

    
1190
    /* select decoding table */
1191
    table = ff_mpa_l2_select_table(s->bit_rate / 1000, s->nb_channels,
1192
                            s->sample_rate, s->lsf);
1193
    sblimit = ff_mpa_sblimit_table[table];
1194
    alloc_table = ff_mpa_alloc_tables[table];
1195

    
1196
    if (s->mode == MPA_JSTEREO)
1197
        bound = (s->mode_ext + 1) * 4;
1198
    else
1199
        bound = sblimit;
1200

    
1201
    dprintf(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit);
1202

    
1203
    /* sanity check */
1204
    if( bound > sblimit ) bound = sblimit;
1205

    
1206
    /* parse bit allocation */
1207
    j = 0;
1208
    for(i=0;i<bound;i++) {
1209
        bit_alloc_bits = alloc_table[j];
1210
        for(ch=0;ch<s->nb_channels;ch++) {
1211
            bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
1212
        }
1213
        j += 1 << bit_alloc_bits;
1214
    }
1215
    for(i=bound;i<sblimit;i++) {
1216
        bit_alloc_bits = alloc_table[j];
1217
        v = get_bits(&s->gb, bit_alloc_bits);
1218
        bit_alloc[0][i] = v;
1219
        bit_alloc[1][i] = v;
1220
        j += 1 << bit_alloc_bits;
1221
    }
1222

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

    
1233
    /* scale codes */
1234
    for(i=0;i<sblimit;i++) {
1235
        for(ch=0;ch<s->nb_channels;ch++) {
1236
            if (bit_alloc[ch][i])
1237
                scale_code[ch][i] = get_bits(&s->gb, 2);
1238
        }
1239
    }
1240

    
1241
    /* scale factors */
1242
    for(i=0;i<sblimit;i++) {
1243
        for(ch=0;ch<s->nb_channels;ch++) {
1244
            if (bit_alloc[ch][i]) {
1245
                sf = scale_factors[ch][i];
1246
                switch(scale_code[ch][i]) {
1247
                default:
1248
                case 0:
1249
                    sf[0] = get_bits(&s->gb, 6);
1250
                    sf[1] = get_bits(&s->gb, 6);
1251
                    sf[2] = get_bits(&s->gb, 6);
1252
                    break;
1253
                case 2:
1254
                    sf[0] = get_bits(&s->gb, 6);
1255
                    sf[1] = sf[0];
1256
                    sf[2] = sf[0];
1257
                    break;
1258
                case 1:
1259
                    sf[0] = get_bits(&s->gb, 6);
1260
                    sf[2] = get_bits(&s->gb, 6);
1261
                    sf[1] = sf[0];
1262
                    break;
1263
                case 3:
1264
                    sf[0] = get_bits(&s->gb, 6);
1265
                    sf[2] = get_bits(&s->gb, 6);
1266
                    sf[1] = sf[2];
1267
                    break;
1268
                }
1269
            }
1270
        }
1271
    }
1272

    
1273
#ifdef DEBUG
1274
    for(ch=0;ch<s->nb_channels;ch++) {
1275
        for(i=0;i<sblimit;i++) {
1276
            if (bit_alloc[ch][i]) {
1277
                sf = scale_factors[ch][i];
1278
                dprintf(s->avctx, " %d %d %d", sf[0], sf[1], sf[2]);
1279
            } else {
1280
                dprintf(s->avctx, " -");
1281
            }
1282
        }
1283
        dprintf(s->avctx, "\n");
1284
    }
1285
#endif
1286

    
1287
    /* samples */
1288
    for(k=0;k<3;k++) {
1289
        for(l=0;l<12;l+=3) {
1290
            j = 0;
1291
            for(i=0;i<bound;i++) {
1292
                bit_alloc_bits = alloc_table[j];
1293
                for(ch=0;ch<s->nb_channels;ch++) {
1294
                    b = bit_alloc[ch][i];
1295
                    if (b) {
1296
                        scale = scale_factors[ch][i][k];
1297
                        qindex = alloc_table[j+b];
1298
                        bits = ff_mpa_quant_bits[qindex];
1299
                        if (bits < 0) {
1300
                            /* 3 values at the same time */
1301
                            v = get_bits(&s->gb, -bits);
1302
                            steps = ff_mpa_quant_steps[qindex];
1303
                            s->sb_samples[ch][k * 12 + l + 0][i] =
1304
                                l2_unscale_group(steps, v % steps, scale);
1305
                            v = v / steps;
1306
                            s->sb_samples[ch][k * 12 + l + 1][i] =
1307
                                l2_unscale_group(steps, v % steps, scale);
1308
                            v = v / steps;
1309
                            s->sb_samples[ch][k * 12 + l + 2][i] =
1310
                                l2_unscale_group(steps, v, scale);
1311
                        } else {
1312
                            for(m=0;m<3;m++) {
1313
                                v = get_bits(&s->gb, bits);
1314
                                v = l1_unscale(bits - 1, v, scale);
1315
                                s->sb_samples[ch][k * 12 + l + m][i] = v;
1316
                            }
1317
                        }
1318
                    } else {
1319
                        s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1320
                        s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1321
                        s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1322
                    }
1323
                }
1324
                /* next subband in alloc table */
1325
                j += 1 << bit_alloc_bits;
1326
            }
1327
            /* XXX: find a way to avoid this duplication of code */
1328
            for(i=bound;i<sblimit;i++) {
1329
                bit_alloc_bits = alloc_table[j];
1330
                b = bit_alloc[0][i];
1331
                if (b) {
1332
                    int mant, scale0, scale1;
1333
                    scale0 = scale_factors[0][i][k];
1334
                    scale1 = scale_factors[1][i][k];
1335
                    qindex = alloc_table[j+b];
1336
                    bits = ff_mpa_quant_bits[qindex];
1337
                    if (bits < 0) {
1338
                        /* 3 values at the same time */
1339
                        v = get_bits(&s->gb, -bits);
1340
                        steps = ff_mpa_quant_steps[qindex];
1341
                        mant = v % steps;
1342
                        v = v / steps;
1343
                        s->sb_samples[0][k * 12 + l + 0][i] =
1344
                            l2_unscale_group(steps, mant, scale0);
1345
                        s->sb_samples[1][k * 12 + l + 0][i] =
1346
                            l2_unscale_group(steps, mant, scale1);
1347
                        mant = v % steps;
1348
                        v = v / steps;
1349
                        s->sb_samples[0][k * 12 + l + 1][i] =
1350
                            l2_unscale_group(steps, mant, scale0);
1351
                        s->sb_samples[1][k * 12 + l + 1][i] =
1352
                            l2_unscale_group(steps, mant, scale1);
1353
                        s->sb_samples[0][k * 12 + l + 2][i] =
1354
                            l2_unscale_group(steps, v, scale0);
1355
                        s->sb_samples[1][k * 12 + l + 2][i] =
1356
                            l2_unscale_group(steps, v, scale1);
1357
                    } else {
1358
                        for(m=0;m<3;m++) {
1359
                            mant = get_bits(&s->gb, bits);
1360
                            s->sb_samples[0][k * 12 + l + m][i] =
1361
                                l1_unscale(bits - 1, mant, scale0);
1362
                            s->sb_samples[1][k * 12 + l + m][i] =
1363
                                l1_unscale(bits - 1, mant, scale1);
1364
                        }
1365
                    }
1366
                } else {
1367
                    s->sb_samples[0][k * 12 + l + 0][i] = 0;
1368
                    s->sb_samples[0][k * 12 + l + 1][i] = 0;
1369
                    s->sb_samples[0][k * 12 + l + 2][i] = 0;
1370
                    s->sb_samples[1][k * 12 + l + 0][i] = 0;
1371
                    s->sb_samples[1][k * 12 + l + 1][i] = 0;
1372
                    s->sb_samples[1][k * 12 + l + 2][i] = 0;
1373
                }
1374
                /* next subband in alloc table */
1375
                j += 1 << bit_alloc_bits;
1376
            }
1377
            /* fill remaining samples to zero */
1378
            for(i=sblimit;i<SBLIMIT;i++) {
1379
                for(ch=0;ch<s->nb_channels;ch++) {
1380
                    s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1381
                    s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1382
                    s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1383
                }
1384
            }
1385
        }
1386
    }
1387
    return 3 * 12;
1388
}
1389

    
1390
static inline void lsf_sf_expand(int *slen,
1391
                                 int sf, int n1, int n2, int n3)
1392
{
1393
    if (n3) {
1394
        slen[3] = sf % n3;
1395
        sf /= n3;
1396
    } else {
1397
        slen[3] = 0;
1398
    }
1399
    if (n2) {
1400
        slen[2] = sf % n2;
1401
        sf /= n2;
1402
    } else {
1403
        slen[2] = 0;
1404
    }
1405
    slen[1] = sf % n1;
1406
    sf /= n1;
1407
    slen[0] = sf;
1408
}
1409

    
1410
static void exponents_from_scale_factors(MPADecodeContext *s,
1411
                                         GranuleDef *g,
1412
                                         int16_t *exponents)
1413
{
1414
    const uint8_t *bstab, *pretab;
1415
    int len, i, j, k, l, v0, shift, gain, gains[3];
1416
    int16_t *exp_ptr;
1417

    
1418
    exp_ptr = exponents;
1419
    gain = g->global_gain - 210;
1420
    shift = g->scalefac_scale + 1;
1421

    
1422
    bstab = band_size_long[s->sample_rate_index];
1423
    pretab = mpa_pretab[g->preflag];
1424
    for(i=0;i<g->long_end;i++) {
1425
        v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400;
1426
        len = bstab[i];
1427
        for(j=len;j>0;j--)
1428
            *exp_ptr++ = v0;
1429
    }
1430

    
1431
    if (g->short_start < 13) {
1432
        bstab = band_size_short[s->sample_rate_index];
1433
        gains[0] = gain - (g->subblock_gain[0] << 3);
1434
        gains[1] = gain - (g->subblock_gain[1] << 3);
1435
        gains[2] = gain - (g->subblock_gain[2] << 3);
1436
        k = g->long_end;
1437
        for(i=g->short_start;i<13;i++) {
1438
            len = bstab[i];
1439
            for(l=0;l<3;l++) {
1440
                v0 = gains[l] - (g->scale_factors[k++] << shift) + 400;
1441
                for(j=len;j>0;j--)
1442
                *exp_ptr++ = v0;
1443
            }
1444
        }
1445
    }
1446
}
1447

    
1448
/* handle n = 0 too */
1449
static inline int get_bitsz(GetBitContext *s, int n)
1450
{
1451
    if (n == 0)
1452
        return 0;
1453
    else
1454
        return get_bits(s, n);
1455
}
1456

    
1457

    
1458
static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos, int *end_pos2){
1459
    if(s->in_gb.buffer && *pos >= s->gb.size_in_bits){
1460
        s->gb= s->in_gb;
1461
        s->in_gb.buffer=NULL;
1462
        assert((get_bits_count(&s->gb) & 7) == 0);
1463
        skip_bits_long(&s->gb, *pos - *end_pos);
1464
        *end_pos2=
1465
        *end_pos= *end_pos2 + get_bits_count(&s->gb) - *pos;
1466
        *pos= get_bits_count(&s->gb);
1467
    }
1468
}
1469

    
1470
static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
1471
                          int16_t *exponents, int end_pos2)
1472
{
1473
    int s_index;
1474
    int i;
1475
    int last_pos, bits_left;
1476
    VLC *vlc;
1477
    int end_pos= FFMIN(end_pos2, s->gb.size_in_bits);
1478

    
1479
    /* low frequencies (called big values) */
1480
    s_index = 0;
1481
    for(i=0;i<3;i++) {
1482
        int j, k, l, linbits;
1483
        j = g->region_size[i];
1484
        if (j == 0)
1485
            continue;
1486
        /* select vlc table */
1487
        k = g->table_select[i];
1488
        l = mpa_huff_data[k][0];
1489
        linbits = mpa_huff_data[k][1];
1490
        vlc = &huff_vlc[l];
1491

    
1492
        if(!l){
1493
            memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*2*j);
1494
            s_index += 2*j;
1495
            continue;
1496
        }
1497

    
1498
        /* read huffcode and compute each couple */
1499
        for(;j>0;j--) {
1500
            int exponent, x, y, v;
1501
            int pos= get_bits_count(&s->gb);
1502

    
1503
            if (pos >= end_pos){
1504
//                av_log(NULL, AV_LOG_ERROR, "pos: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);
1505
                switch_buffer(s, &pos, &end_pos, &end_pos2);
1506
//                av_log(NULL, AV_LOG_ERROR, "new pos: %d %d\n", pos, end_pos);
1507
                if(pos >= end_pos)
1508
                    break;
1509
            }
1510
            y = get_vlc2(&s->gb, vlc->table, 7, 3);
1511

    
1512
            if(!y){
1513
                g->sb_hybrid[s_index  ] =
1514
                g->sb_hybrid[s_index+1] = 0;
1515
                s_index += 2;
1516
                continue;
1517
            }
1518

    
1519
            exponent= exponents[s_index];
1520

    
1521
            dprintf(s->avctx, "region=%d n=%d x=%d y=%d exp=%d\n",
1522
                    i, g->region_size[i] - j, x, y, exponent);
1523
            if(y&16){
1524
                x = y >> 5;
1525
                y = y & 0x0f;
1526
                if (x < 15){
1527
                    v = expval_table[ exponent ][ x ];
1528
//                      v = expval_table[ (exponent&3) ][ x ] >> FFMIN(0 - (exponent>>2), 31);
1529
                }else{
1530
                    x += get_bitsz(&s->gb, linbits);
1531
                    v = l3_unscale(x, exponent);
1532
                }
1533
                if (get_bits1(&s->gb))
1534
                    v = -v;
1535
                g->sb_hybrid[s_index] = v;
1536
                if (y < 15){
1537
                    v = expval_table[ exponent ][ y ];
1538
                }else{
1539
                    y += get_bitsz(&s->gb, linbits);
1540
                    v = l3_unscale(y, exponent);
1541
                }
1542
                if (get_bits1(&s->gb))
1543
                    v = -v;
1544
                g->sb_hybrid[s_index+1] = v;
1545
            }else{
1546
                x = y >> 5;
1547
                y = y & 0x0f;
1548
                x += y;
1549
                if (x < 15){
1550
                    v = expval_table[ exponent ][ x ];
1551
                }else{
1552
                    x += get_bitsz(&s->gb, linbits);
1553
                    v = l3_unscale(x, exponent);
1554
                }
1555
                if (get_bits1(&s->gb))
1556
                    v = -v;
1557
                g->sb_hybrid[s_index+!!y] = v;
1558
                g->sb_hybrid[s_index+ !y] = 0;
1559
            }
1560
            s_index+=2;
1561
        }
1562
    }
1563

    
1564
    /* high frequencies */
1565
    vlc = &huff_quad_vlc[g->count1table_select];
1566
    last_pos=0;
1567
    while (s_index <= 572) {
1568
        int pos, code;
1569
        pos = get_bits_count(&s->gb);
1570
        if (pos >= end_pos) {
1571
            if (pos > end_pos2 && last_pos){
1572
                /* some encoders generate an incorrect size for this
1573
                   part. We must go back into the data */
1574
                s_index -= 4;
1575
                skip_bits_long(&s->gb, last_pos - pos);
1576
                av_log(NULL, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos);
1577
                if(s->error_resilience >= FF_ER_COMPLIANT)
1578
                    s_index=0;
1579
                break;
1580
            }
1581
//                av_log(NULL, AV_LOG_ERROR, "pos2: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);
1582
            switch_buffer(s, &pos, &end_pos, &end_pos2);
1583
//                av_log(NULL, AV_LOG_ERROR, "new pos2: %d %d %d\n", pos, end_pos, s_index);
1584
            if(pos >= end_pos)
1585
                break;
1586
        }
1587
        last_pos= pos;
1588

    
1589
        code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);
1590
        dprintf(s->avctx, "t=%d code=%d\n", g->count1table_select, code);
1591
        g->sb_hybrid[s_index+0]=
1592
        g->sb_hybrid[s_index+1]=
1593
        g->sb_hybrid[s_index+2]=
1594
        g->sb_hybrid[s_index+3]= 0;
1595
        while(code){
1596
            static const int idxtab[16]={3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0};
1597
            int v;
1598
            int pos= s_index+idxtab[code];
1599
            code ^= 8>>idxtab[code];
1600
            v = exp_table[ exponents[pos] ];
1601
//            v = exp_table[ (exponents[pos]&3) ] >> FFMIN(0 - (exponents[pos]>>2), 31);
1602
            if(get_bits1(&s->gb))
1603
                v = -v;
1604
            g->sb_hybrid[pos] = v;
1605
        }
1606
        s_index+=4;
1607
    }
1608
    /* skip extension bits */
1609
    bits_left = end_pos2 - get_bits_count(&s->gb);
1610
//av_log(NULL, AV_LOG_ERROR, "left:%d buf:%p\n", bits_left, s->in_gb.buffer);
1611
    if (bits_left < 0/* || bits_left > 500*/) {
1612
        av_log(NULL, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
1613
        s_index=0;
1614
    }else if(bits_left > 0 && s->error_resilience >= FF_ER_AGGRESSIVE){
1615
        av_log(NULL, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
1616
        s_index=0;
1617
    }
1618
    memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*(576 - s_index));
1619
    skip_bits_long(&s->gb, bits_left);
1620

    
1621
    i= get_bits_count(&s->gb);
1622
    switch_buffer(s, &i, &end_pos, &end_pos2);
1623

    
1624
    return 0;
1625
}
1626

    
1627
/* Reorder short blocks from bitstream order to interleaved order. It
1628
   would be faster to do it in parsing, but the code would be far more
1629
   complicated */
1630
static void reorder_block(MPADecodeContext *s, GranuleDef *g)
1631
{
1632
    int i, j, len;
1633
    int32_t *ptr, *dst, *ptr1;
1634
    int32_t tmp[576];
1635

    
1636
    if (g->block_type != 2)
1637
        return;
1638

    
1639
    if (g->switch_point) {
1640
        if (s->sample_rate_index != 8) {
1641
            ptr = g->sb_hybrid + 36;
1642
        } else {
1643
            ptr = g->sb_hybrid + 48;
1644
        }
1645
    } else {
1646
        ptr = g->sb_hybrid;
1647
    }
1648

    
1649
    for(i=g->short_start;i<13;i++) {
1650
        len = band_size_short[s->sample_rate_index][i];
1651
        ptr1 = ptr;
1652
        dst = tmp;
1653
        for(j=len;j>0;j--) {
1654
            *dst++ = ptr[0*len];
1655
            *dst++ = ptr[1*len];
1656
            *dst++ = ptr[2*len];
1657
            ptr++;
1658
        }
1659
        ptr+=2*len;
1660
        memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));
1661
    }
1662
}
1663

    
1664
#define ISQRT2 FIXR(0.70710678118654752440)
1665

    
1666
static void compute_stereo(MPADecodeContext *s,
1667
                           GranuleDef *g0, GranuleDef *g1)
1668
{
1669
    int i, j, k, l;
1670
    int32_t v1, v2;
1671
    int sf_max, tmp0, tmp1, sf, len, non_zero_found;
1672
    int32_t (*is_tab)[16];
1673
    int32_t *tab0, *tab1;
1674
    int non_zero_found_short[3];
1675

    
1676
    /* intensity stereo */
1677
    if (s->mode_ext & MODE_EXT_I_STEREO) {
1678
        if (!s->lsf) {
1679
            is_tab = is_table;
1680
            sf_max = 7;
1681
        } else {
1682
            is_tab = is_table_lsf[g1->scalefac_compress & 1];
1683
            sf_max = 16;
1684
        }
1685

    
1686
        tab0 = g0->sb_hybrid + 576;
1687
        tab1 = g1->sb_hybrid + 576;
1688

    
1689
        non_zero_found_short[0] = 0;
1690
        non_zero_found_short[1] = 0;
1691
        non_zero_found_short[2] = 0;
1692
        k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1693
        for(i = 12;i >= g1->short_start;i--) {
1694
            /* for last band, use previous scale factor */
1695
            if (i != 11)
1696
                k -= 3;
1697
            len = band_size_short[s->sample_rate_index][i];
1698
            for(l=2;l>=0;l--) {
1699
                tab0 -= len;
1700
                tab1 -= len;
1701
                if (!non_zero_found_short[l]) {
1702
                    /* test if non zero band. if so, stop doing i-stereo */
1703
                    for(j=0;j<len;j++) {
1704
                        if (tab1[j] != 0) {
1705
                            non_zero_found_short[l] = 1;
1706
                            goto found1;
1707
                        }
1708
                    }
1709
                    sf = g1->scale_factors[k + l];
1710
                    if (sf >= sf_max)
1711
                        goto found1;
1712

    
1713
                    v1 = is_tab[0][sf];
1714
                    v2 = is_tab[1][sf];
1715
                    for(j=0;j<len;j++) {
1716
                        tmp0 = tab0[j];
1717
                        tab0[j] = MULL(tmp0, v1);
1718
                        tab1[j] = MULL(tmp0, v2);
1719
                    }
1720
                } else {
1721
                found1:
1722
                    if (s->mode_ext & MODE_EXT_MS_STEREO) {
1723
                        /* lower part of the spectrum : do ms stereo
1724
                           if enabled */
1725
                        for(j=0;j<len;j++) {
1726
                            tmp0 = tab0[j];
1727
                            tmp1 = tab1[j];
1728
                            tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
1729
                            tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
1730
                        }
1731
                    }
1732
                }
1733
            }
1734
        }
1735

    
1736
        non_zero_found = non_zero_found_short[0] |
1737
            non_zero_found_short[1] |
1738
            non_zero_found_short[2];
1739

    
1740
        for(i = g1->long_end - 1;i >= 0;i--) {
1741
            len = band_size_long[s->sample_rate_index][i];
1742
            tab0 -= len;
1743
            tab1 -= len;
1744
            /* test if non zero band. if so, stop doing i-stereo */
1745
            if (!non_zero_found) {
1746
                for(j=0;j<len;j++) {
1747
                    if (tab1[j] != 0) {
1748
                        non_zero_found = 1;
1749
                        goto found2;
1750
                    }
1751
                }
1752
                /* for last band, use previous scale factor */
1753
                k = (i == 21) ? 20 : i;
1754
                sf = g1->scale_factors[k];
1755
                if (sf >= sf_max)
1756
                    goto found2;
1757
                v1 = is_tab[0][sf];
1758
                v2 = is_tab[1][sf];
1759
                for(j=0;j<len;j++) {
1760
                    tmp0 = tab0[j];
1761
                    tab0[j] = MULL(tmp0, v1);
1762
                    tab1[j] = MULL(tmp0, v2);
1763
                }
1764
            } else {
1765
            found2:
1766
                if (s->mode_ext & MODE_EXT_MS_STEREO) {
1767
                    /* lower part of the spectrum : do ms stereo
1768
                       if enabled */
1769
                    for(j=0;j<len;j++) {
1770
                        tmp0 = tab0[j];
1771
                        tmp1 = tab1[j];
1772
                        tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
1773
                        tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
1774
                    }
1775
                }
1776
            }
1777
        }
1778
    } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1779
        /* ms stereo ONLY */
1780
        /* NOTE: the 1/sqrt(2) normalization factor is included in the
1781
           global gain */
1782
        tab0 = g0->sb_hybrid;
1783
        tab1 = g1->sb_hybrid;
1784
        for(i=0;i<576;i++) {
1785
            tmp0 = tab0[i];
1786
            tmp1 = tab1[i];
1787
            tab0[i] = tmp0 + tmp1;
1788
            tab1[i] = tmp0 - tmp1;
1789
        }
1790
    }
1791
}
1792

    
1793
static void compute_antialias_integer(MPADecodeContext *s,
1794
                              GranuleDef *g)
1795
{
1796
    int32_t *ptr, *csa;
1797
    int n, i;
1798

    
1799
    /* we antialias only "long" bands */
1800
    if (g->block_type == 2) {
1801
        if (!g->switch_point)
1802
            return;
1803
        /* XXX: check this for 8000Hz case */
1804
        n = 1;
1805
    } else {
1806
        n = SBLIMIT - 1;
1807
    }
1808

    
1809
    ptr = g->sb_hybrid + 18;
1810
    for(i = n;i > 0;i--) {
1811
        int tmp0, tmp1, tmp2;
1812
        csa = &csa_table[0][0];
1813
#define INT_AA(j) \
1814
            tmp0 = ptr[-1-j];\
1815
            tmp1 = ptr[   j];\
1816
            tmp2= MULH(tmp0 + tmp1, csa[0+4*j]);\
1817
            ptr[-1-j] = 4*(tmp2 - MULH(tmp1, csa[2+4*j]));\
1818
            ptr[   j] = 4*(tmp2 + MULH(tmp0, csa[3+4*j]));
1819

    
1820
        INT_AA(0)
1821
        INT_AA(1)
1822
        INT_AA(2)
1823
        INT_AA(3)
1824
        INT_AA(4)
1825
        INT_AA(5)
1826
        INT_AA(6)
1827
        INT_AA(7)
1828

    
1829
        ptr += 18;
1830
    }
1831
}
1832

    
1833
static void compute_antialias_float(MPADecodeContext *s,
1834
                              GranuleDef *g)
1835
{
1836
    int32_t *ptr;
1837
    int n, i;
1838

    
1839
    /* we antialias only "long" bands */
1840
    if (g->block_type == 2) {
1841
        if (!g->switch_point)
1842
            return;
1843
        /* XXX: check this for 8000Hz case */
1844
        n = 1;
1845
    } else {
1846
        n = SBLIMIT - 1;
1847
    }
1848

    
1849
    ptr = g->sb_hybrid + 18;
1850
    for(i = n;i > 0;i--) {
1851
        float tmp0, tmp1;
1852
        float *csa = &csa_table_float[0][0];
1853
#define FLOAT_AA(j)\
1854
        tmp0= ptr[-1-j];\
1855
        tmp1= ptr[   j];\
1856
        ptr[-1-j] = lrintf(tmp0 * csa[0+4*j] - tmp1 * csa[1+4*j]);\
1857
        ptr[   j] = lrintf(tmp0 * csa[1+4*j] + tmp1 * csa[0+4*j]);
1858

    
1859
        FLOAT_AA(0)
1860
        FLOAT_AA(1)
1861
        FLOAT_AA(2)
1862
        FLOAT_AA(3)
1863
        FLOAT_AA(4)
1864
        FLOAT_AA(5)
1865
        FLOAT_AA(6)
1866
        FLOAT_AA(7)
1867

    
1868
        ptr += 18;
1869
    }
1870
}
1871

    
1872
static void compute_imdct(MPADecodeContext *s,
1873
                          GranuleDef *g,
1874
                          int32_t *sb_samples,
1875
                          int32_t *mdct_buf)
1876
{
1877
    int32_t *ptr, *win, *win1, *buf, *out_ptr, *ptr1;
1878
    int32_t out2[12];
1879
    int i, j, mdct_long_end, v, sblimit;
1880

    
1881
    /* find last non zero block */
1882
    ptr = g->sb_hybrid + 576;
1883
    ptr1 = g->sb_hybrid + 2 * 18;
1884
    while (ptr >= ptr1) {
1885
        ptr -= 6;
1886
        v = ptr[0] | ptr[1] | ptr[2] | ptr[3] | ptr[4] | ptr[5];
1887
        if (v != 0)
1888
            break;
1889
    }
1890
    sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
1891

    
1892
    if (g->block_type == 2) {
1893
        /* XXX: check for 8000 Hz */
1894
        if (g->switch_point)
1895
            mdct_long_end = 2;
1896
        else
1897
            mdct_long_end = 0;
1898
    } else {
1899
        mdct_long_end = sblimit;
1900
    }
1901

    
1902
    buf = mdct_buf;
1903
    ptr = g->sb_hybrid;
1904
    for(j=0;j<mdct_long_end;j++) {
1905
        /* apply window & overlap with previous buffer */
1906
        out_ptr = sb_samples + j;
1907
        /* select window */
1908
        if (g->switch_point && j < 2)
1909
            win1 = mdct_win[0];
1910
        else
1911
            win1 = mdct_win[g->block_type];
1912
        /* select frequency inversion */
1913
        win = win1 + ((4 * 36) & -(j & 1));
1914
        imdct36(out_ptr, buf, ptr, win);
1915
        out_ptr += 18*SBLIMIT;
1916
        ptr += 18;
1917
        buf += 18;
1918
    }
1919
    for(j=mdct_long_end;j<sblimit;j++) {
1920
        /* select frequency inversion */
1921
        win = mdct_win[2] + ((4 * 36) & -(j & 1));
1922
        out_ptr = sb_samples + j;
1923

    
1924
        for(i=0; i<6; i++){
1925
            *out_ptr = buf[i];
1926
            out_ptr += SBLIMIT;
1927
        }
1928
        imdct12(out2, ptr + 0);
1929
        for(i=0;i<6;i++) {
1930
            *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*1];
1931
            buf[i + 6*2] = MULH(out2[i + 6], win[i + 6]);
1932
            out_ptr += SBLIMIT;
1933
        }
1934
        imdct12(out2, ptr + 1);
1935
        for(i=0;i<6;i++) {
1936
            *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*2];
1937
            buf[i + 6*0] = MULH(out2[i + 6], win[i + 6]);
1938
            out_ptr += SBLIMIT;
1939
        }
1940
        imdct12(out2, ptr + 2);
1941
        for(i=0;i<6;i++) {
1942
            buf[i + 6*0] = MULH(out2[i], win[i]) + buf[i + 6*0];
1943
            buf[i + 6*1] = MULH(out2[i + 6], win[i + 6]);
1944
            buf[i + 6*2] = 0;
1945
        }
1946
        ptr += 18;
1947
        buf += 18;
1948
    }
1949
    /* zero bands */
1950
    for(j=sblimit;j<SBLIMIT;j++) {
1951
        /* overlap */
1952
        out_ptr = sb_samples + j;
1953
        for(i=0;i<18;i++) {
1954
            *out_ptr = buf[i];
1955
            buf[i] = 0;
1956
            out_ptr += SBLIMIT;
1957
        }
1958
        buf += 18;
1959
    }
1960
}
1961

    
1962
#if defined(DEBUG)
1963
void sample_dump(int fnum, int32_t *tab, int n)
1964
{
1965
    static FILE *files[16], *f;
1966
    char buf[512];
1967
    int i;
1968
    int32_t v;
1969

    
1970
    f = files[fnum];
1971
    if (!f) {
1972
        snprintf(buf, sizeof(buf), "/tmp/out%d.%s.pcm",
1973
                fnum,
1974
#ifdef USE_HIGHPRECISION
1975
                "hp"
1976
#else
1977
                "lp"
1978
#endif
1979
                );
1980
        f = fopen(buf, "w");
1981
        if (!f)
1982
            return;
1983
        files[fnum] = f;
1984
    }
1985

    
1986
    if (fnum == 0) {
1987
        static int pos = 0;
1988
        av_log(NULL, AV_LOG_DEBUG, "pos=%d\n", pos);
1989
        for(i=0;i<n;i++) {
1990
            av_log(NULL, AV_LOG_DEBUG, " %0.4f", (double)tab[i] / FRAC_ONE);
1991
            if ((i % 18) == 17)
1992
                av_log(NULL, AV_LOG_DEBUG, "\n");
1993
        }
1994
        pos += n;
1995
    }
1996
    for(i=0;i<n;i++) {
1997
        /* normalize to 23 frac bits */
1998
        v = tab[i] << (23 - FRAC_BITS);
1999
        fwrite(&v, 1, sizeof(int32_t), f);
2000
    }
2001
}
2002
#endif
2003

    
2004

    
2005
/* main layer3 decoding function */
2006
static int mp_decode_layer3(MPADecodeContext *s)
2007
{
2008
    int nb_granules, main_data_begin, private_bits;
2009
    int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;
2010
    GranuleDef granules[2][2], *g;
2011
    int16_t exponents[576];
2012

    
2013
    /* read side info */
2014
    if (s->lsf) {
2015
        main_data_begin = get_bits(&s->gb, 8);
2016
        private_bits = get_bits(&s->gb, s->nb_channels);
2017
        nb_granules = 1;
2018
    } else {
2019
        main_data_begin = get_bits(&s->gb, 9);
2020
        if (s->nb_channels == 2)
2021
            private_bits = get_bits(&s->gb, 3);
2022
        else
2023
            private_bits = get_bits(&s->gb, 5);
2024
        nb_granules = 2;
2025
        for(ch=0;ch<s->nb_channels;ch++) {
2026
            granules[ch][0].scfsi = 0; /* all scale factors are transmitted */
2027
            granules[ch][1].scfsi = get_bits(&s->gb, 4);
2028
        }
2029
    }
2030

    
2031
    for(gr=0;gr<nb_granules;gr++) {
2032
        for(ch=0;ch<s->nb_channels;ch++) {
2033
            dprintf(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);
2034
            g = &granules[ch][gr];
2035
            g->part2_3_length = get_bits(&s->gb, 12);
2036
            g->big_values = get_bits(&s->gb, 9);
2037
            if(g->big_values > 288){
2038
                av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");
2039
                return -1;
2040
            }
2041

    
2042
            g->global_gain = get_bits(&s->gb, 8);
2043
            /* if MS stereo only is selected, we precompute the
2044
               1/sqrt(2) renormalization factor */
2045
            if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
2046
                MODE_EXT_MS_STEREO)
2047
                g->global_gain -= 2;
2048
            if (s->lsf)
2049
                g->scalefac_compress = get_bits(&s->gb, 9);
2050
            else
2051
                g->scalefac_compress = get_bits(&s->gb, 4);
2052
            blocksplit_flag = get_bits1(&s->gb);
2053
            if (blocksplit_flag) {
2054
                g->block_type = get_bits(&s->gb, 2);
2055
                if (g->block_type == 0){
2056
                    av_log(NULL, AV_LOG_ERROR, "invalid block type\n");
2057
                    return -1;
2058
                }
2059
                g->switch_point = get_bits1(&s->gb);
2060
                for(i=0;i<2;i++)
2061
                    g->table_select[i] = get_bits(&s->gb, 5);
2062
                for(i=0;i<3;i++)
2063
                    g->subblock_gain[i] = get_bits(&s->gb, 3);
2064
                ff_init_short_region(s, g);
2065
            } else {
2066
                int region_address1, region_address2;
2067
                g->block_type = 0;
2068
                g->switch_point = 0;
2069
                for(i=0;i<3;i++)
2070
                    g->table_select[i] = get_bits(&s->gb, 5);
2071
                /* compute huffman coded region sizes */
2072
                region_address1 = get_bits(&s->gb, 4);
2073
                region_address2 = get_bits(&s->gb, 3);
2074
                dprintf(s->avctx, "region1=%d region2=%d\n",
2075
                        region_address1, region_address2);
2076
                ff_init_long_region(s, g, region_address1, region_address2);
2077
            }
2078
            ff_region_offset2size(g);
2079
            ff_compute_band_indexes(s, g);
2080

    
2081
            g->preflag = 0;
2082
            if (!s->lsf)
2083
                g->preflag = get_bits1(&s->gb);
2084
            g->scalefac_scale = get_bits1(&s->gb);
2085
            g->count1table_select = get_bits1(&s->gb);
2086
            dprintf(s->avctx, "block_type=%d switch_point=%d\n",
2087
                    g->block_type, g->switch_point);
2088
        }
2089
    }
2090

    
2091
  if (!s->adu_mode) {
2092
    const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);
2093
    assert((get_bits_count(&s->gb) & 7) == 0);
2094
    /* now we get bits from the main_data_begin offset */
2095
    dprintf(s->avctx, "seekback: %d\n", main_data_begin);
2096
//av_log(NULL, AV_LOG_ERROR, "backstep:%d, lastbuf:%d\n", main_data_begin, s->last_buf_size);
2097

    
2098
    memcpy(s->last_buf + s->last_buf_size, ptr, EXTRABYTES);
2099
    s->in_gb= s->gb;
2100
        init_get_bits(&s->gb, s->last_buf, s->last_buf_size*8);
2101
        skip_bits_long(&s->gb, 8*(s->last_buf_size - main_data_begin));
2102
  }
2103

    
2104
    for(gr=0;gr<nb_granules;gr++) {
2105
        for(ch=0;ch<s->nb_channels;ch++) {
2106
            g = &granules[ch][gr];
2107
            if(get_bits_count(&s->gb)<0){
2108
                av_log(NULL, AV_LOG_ERROR, "mdb:%d, lastbuf:%d skipping granule %d\n",
2109
                                            main_data_begin, s->last_buf_size, gr);
2110
                skip_bits_long(&s->gb, g->part2_3_length);
2111
                memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));
2112
                if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->in_gb.buffer){
2113
                    skip_bits_long(&s->in_gb, get_bits_count(&s->gb) - s->gb.size_in_bits);
2114
                    s->gb= s->in_gb;
2115
                    s->in_gb.buffer=NULL;
2116
                }
2117
                continue;
2118
            }
2119

    
2120
            bits_pos = get_bits_count(&s->gb);
2121

    
2122
            if (!s->lsf) {
2123
                uint8_t *sc;
2124
                int slen, slen1, slen2;
2125

    
2126
                /* MPEG1 scale factors */
2127
                slen1 = slen_table[0][g->scalefac_compress];
2128
                slen2 = slen_table[1][g->scalefac_compress];
2129
                dprintf(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);
2130
                if (g->block_type == 2) {
2131
                    n = g->switch_point ? 17 : 18;
2132
                    j = 0;
2133
                    if(slen1){
2134
                        for(i=0;i<n;i++)
2135
                            g->scale_factors[j++] = get_bits(&s->gb, slen1);
2136
                    }else{
2137
                        for(i=0;i<n;i++)
2138
                            g->scale_factors[j++] = 0;
2139
                    }
2140
                    if(slen2){
2141
                        for(i=0;i<18;i++)
2142
                            g->scale_factors[j++] = get_bits(&s->gb, slen2);
2143
                        for(i=0;i<3;i++)
2144
                            g->scale_factors[j++] = 0;
2145
                    }else{
2146
                        for(i=0;i<21;i++)
2147
                            g->scale_factors[j++] = 0;
2148
                    }
2149
                } else {
2150
                    sc = granules[ch][0].scale_factors;
2151
                    j = 0;
2152
                    for(k=0;k<4;k++) {
2153
                        n = (k == 0 ? 6 : 5);
2154
                        if ((g->scfsi & (0x8 >> k)) == 0) {
2155
                            slen = (k < 2) ? slen1 : slen2;
2156
                            if(slen){
2157
                                for(i=0;i<n;i++)
2158
                                    g->scale_factors[j++] = get_bits(&s->gb, slen);
2159
                            }else{
2160
                                for(i=0;i<n;i++)
2161
                                    g->scale_factors[j++] = 0;
2162
                            }
2163
                        } else {
2164
                            /* simply copy from last granule */
2165
                            for(i=0;i<n;i++) {
2166
                                g->scale_factors[j] = sc[j];
2167
                                j++;
2168
                            }
2169
                        }
2170
                    }
2171
                    g->scale_factors[j++] = 0;
2172
                }
2173
#if defined(DEBUG)
2174
                {
2175
                    dprintf(s->avctx, "scfsi=%x gr=%d ch=%d scale_factors:\n",
2176
                           g->scfsi, gr, ch);
2177
                    for(i=0;i<j;i++)
2178
                        dprintf(s->avctx, " %d", g->scale_factors[i]);
2179
                    dprintf(s->avctx, "\n");
2180
                }
2181
#endif
2182
            } else {
2183
                int tindex, tindex2, slen[4], sl, sf;
2184

    
2185
                /* LSF scale factors */
2186
                if (g->block_type == 2) {
2187
                    tindex = g->switch_point ? 2 : 1;
2188
                } else {
2189
                    tindex = 0;
2190
                }
2191
                sf = g->scalefac_compress;
2192
                if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
2193
                    /* intensity stereo case */
2194
                    sf >>= 1;
2195
                    if (sf < 180) {
2196
                        lsf_sf_expand(slen, sf, 6, 6, 0);
2197
                        tindex2 = 3;
2198
                    } else if (sf < 244) {
2199
                        lsf_sf_expand(slen, sf - 180, 4, 4, 0);
2200
                        tindex2 = 4;
2201
                    } else {
2202
                        lsf_sf_expand(slen, sf - 244, 3, 0, 0);
2203
                        tindex2 = 5;
2204
                    }
2205
                } else {
2206
                    /* normal case */
2207
                    if (sf < 400) {
2208
                        lsf_sf_expand(slen, sf, 5, 4, 4);
2209
                        tindex2 = 0;
2210
                    } else if (sf < 500) {
2211
                        lsf_sf_expand(slen, sf - 400, 5, 4, 0);
2212
                        tindex2 = 1;
2213
                    } else {
2214
                        lsf_sf_expand(slen, sf - 500, 3, 0, 0);
2215
                        tindex2 = 2;
2216
                        g->preflag = 1;
2217
                    }
2218
                }
2219

    
2220
                j = 0;
2221
                for(k=0;k<4;k++) {
2222
                    n = lsf_nsf_table[tindex2][tindex][k];
2223
                    sl = slen[k];
2224
                    if(sl){
2225
                        for(i=0;i<n;i++)
2226
                            g->scale_factors[j++] = get_bits(&s->gb, sl);
2227
                    }else{
2228
                        for(i=0;i<n;i++)
2229
                            g->scale_factors[j++] = 0;
2230
                    }
2231
                }
2232
                /* XXX: should compute exact size */
2233
                for(;j<40;j++)
2234
                    g->scale_factors[j] = 0;
2235
#if defined(DEBUG)
2236
                {
2237
                    dprintf(s->avctx, "gr=%d ch=%d scale_factors:\n",
2238
                           gr, ch);
2239
                    for(i=0;i<40;i++)
2240
                        dprintf(s->avctx, " %d", g->scale_factors[i]);
2241
                    dprintf(s->avctx, "\n");
2242
                }
2243
#endif
2244
            }
2245

    
2246
            exponents_from_scale_factors(s, g, exponents);
2247

    
2248
            /* read Huffman coded residue */
2249
            huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);
2250
#if defined(DEBUG)
2251
            sample_dump(0, g->sb_hybrid, 576);
2252
#endif
2253
        } /* ch */
2254

    
2255
        if (s->nb_channels == 2)
2256
            compute_stereo(s, &granules[0][gr], &granules[1][gr]);
2257

    
2258
        for(ch=0;ch<s->nb_channels;ch++) {
2259
            g = &granules[ch][gr];
2260

    
2261
            reorder_block(s, g);
2262
#if defined(DEBUG)
2263
            sample_dump(0, g->sb_hybrid, 576);
2264
#endif
2265
            s->compute_antialias(s, g);
2266
#if defined(DEBUG)
2267
            sample_dump(1, g->sb_hybrid, 576);
2268
#endif
2269
            compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
2270
#if defined(DEBUG)
2271
            sample_dump(2, &s->sb_samples[ch][18 * gr][0], 576);
2272
#endif
2273
        }
2274
    } /* gr */
2275
    if(get_bits_count(&s->gb)<0)
2276
        skip_bits_long(&s->gb, -get_bits_count(&s->gb));
2277
    return nb_granules * 18;
2278
}
2279

    
2280
static int mp_decode_frame(MPADecodeContext *s,
2281
                           OUT_INT *samples, const uint8_t *buf, int buf_size)
2282
{
2283
    int i, nb_frames, ch;
2284
    OUT_INT *samples_ptr;
2285

    
2286
    init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE)*8);
2287

    
2288
    /* skip error protection field */
2289
    if (s->error_protection)
2290
        skip_bits(&s->gb, 16);
2291

    
2292
    dprintf(s->avctx, "frame %d:\n", s->frame_count);
2293
    switch(s->layer) {
2294
    case 1:
2295
        s->avctx->frame_size = 384;
2296
        nb_frames = mp_decode_layer1(s);
2297
        break;
2298
    case 2:
2299
        s->avctx->frame_size = 1152;
2300
        nb_frames = mp_decode_layer2(s);
2301
        break;
2302
    case 3:
2303
        s->avctx->frame_size = s->lsf ? 576 : 1152;
2304
    default:
2305
        nb_frames = mp_decode_layer3(s);
2306

    
2307
        s->last_buf_size=0;
2308
        if(s->in_gb.buffer){
2309
            align_get_bits(&s->gb);
2310
            i= (s->gb.size_in_bits - get_bits_count(&s->gb))>>3;
2311
            if(i >= 0 && i <= BACKSTEP_SIZE){
2312
                memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);
2313
                s->last_buf_size=i;
2314
            }else
2315
                av_log(NULL, AV_LOG_ERROR, "invalid old backstep %d\n", i);
2316
            s->gb= s->in_gb;
2317
            s->in_gb.buffer= NULL;
2318
        }
2319

    
2320
        align_get_bits(&s->gb);
2321
        assert((get_bits_count(&s->gb) & 7) == 0);
2322
        i= (s->gb.size_in_bits - get_bits_count(&s->gb))>>3;
2323

    
2324
        if(i<0 || i > BACKSTEP_SIZE || nb_frames<0){
2325
            av_log(NULL, AV_LOG_ERROR, "invalid new backstep %d\n", i);
2326
            i= FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);
2327
        }
2328
        assert(i <= buf_size - HEADER_SIZE && i>= 0);
2329
        memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);
2330
        s->last_buf_size += i;
2331

    
2332
        break;
2333
    }
2334
#if defined(DEBUG)
2335
    for(i=0;i<nb_frames;i++) {
2336
        for(ch=0;ch<s->nb_channels;ch++) {
2337
            int j;
2338
            dprintf(s->avctx, "%d-%d:", i, ch);
2339
            for(j=0;j<SBLIMIT;j++)
2340
                dprintf(s->avctx, " %0.6f", (double)s->sb_samples[ch][i][j] / FRAC_ONE);
2341
            dprintf(s->avctx, "\n");
2342
        }
2343
    }
2344
#endif
2345
    /* apply the synthesis filter */
2346
    for(ch=0;ch<s->nb_channels;ch++) {
2347
        samples_ptr = samples + ch;
2348
        for(i=0;i<nb_frames;i++) {
2349
            ff_mpa_synth_filter(s->synth_buf[ch], &(s->synth_buf_offset[ch]),
2350
                         window, &s->dither_state,
2351
                         samples_ptr, s->nb_channels,
2352
                         s->sb_samples[ch][i]);
2353
            samples_ptr += 32 * s->nb_channels;
2354
        }
2355
    }
2356
#ifdef DEBUG
2357
    s->frame_count++;
2358
#endif
2359
    return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
2360
}
2361

    
2362
static int decode_frame(AVCodecContext * avctx,
2363
                        void *data, int *data_size,
2364
                        const uint8_t * buf, int buf_size)
2365
{
2366
    MPADecodeContext *s = avctx->priv_data;
2367
    uint32_t header;
2368
    int out_size;
2369
    OUT_INT *out_samples = data;
2370

    
2371
retry:
2372
    if(buf_size < HEADER_SIZE)
2373
        return -1;
2374

    
2375
    header = AV_RB32(buf);
2376
    if(ff_mpa_check_header(header) < 0){
2377
        buf++;
2378
//        buf_size--;
2379
        av_log(avctx, AV_LOG_ERROR, "Header missing skipping one byte.\n");
2380
        goto retry;
2381
    }
2382

    
2383
    if (ff_mpegaudio_decode_header(s, header) == 1) {
2384
        /* free format: prepare to compute frame size */
2385
        s->frame_size = -1;
2386
        return -1;
2387
    }
2388
    /* update codec info */
2389
    avctx->channels = s->nb_channels;
2390
    avctx->bit_rate = s->bit_rate;
2391
    avctx->sub_id = s->layer;
2392

    
2393
    if(s->frame_size<=0 || s->frame_size > buf_size){
2394
        av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
2395
        return -1;
2396
    }else if(s->frame_size < buf_size){
2397
        av_log(avctx, AV_LOG_ERROR, "incorrect frame size\n");
2398
        buf_size= s->frame_size;
2399
    }
2400

    
2401
    out_size = mp_decode_frame(s, out_samples, buf, buf_size);
2402
    if(out_size>=0){
2403
        *data_size = out_size;
2404
        avctx->sample_rate = s->sample_rate;
2405
        //FIXME maybe move the other codec info stuff from above here too
2406
    }else
2407
        av_log(avctx, AV_LOG_DEBUG, "Error while decoding MPEG audio frame.\n"); //FIXME return -1 / but also return the number of bytes consumed
2408
    s->frame_size = 0;
2409
    return buf_size;
2410
}
2411

    
2412
static void flush(AVCodecContext *avctx){
2413
    MPADecodeContext *s = avctx->priv_data;
2414
    memset(s->synth_buf, 0, sizeof(s->synth_buf));
2415
    s->last_buf_size= 0;
2416
}
2417

    
2418
#ifdef CONFIG_MP3ADU_DECODER
2419
static int decode_frame_adu(AVCodecContext * avctx,
2420
                        void *data, int *data_size,
2421
                        const uint8_t * buf, int buf_size)
2422
{
2423
    MPADecodeContext *s = avctx->priv_data;
2424
    uint32_t header;
2425
    int len, out_size;
2426
    OUT_INT *out_samples = data;
2427

    
2428
    len = buf_size;
2429

    
2430
    // Discard too short frames
2431
    if (buf_size < HEADER_SIZE) {
2432
        *data_size = 0;
2433
        return buf_size;
2434
    }
2435

    
2436

    
2437
    if (len > MPA_MAX_CODED_FRAME_SIZE)
2438
        len = MPA_MAX_CODED_FRAME_SIZE;
2439

    
2440
    // Get header and restore sync word
2441
    header = AV_RB32(buf) | 0xffe00000;
2442

    
2443
    if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame
2444
        *data_size = 0;
2445
        return buf_size;
2446
    }
2447

    
2448
    ff_mpegaudio_decode_header(s, header);
2449
    /* update codec info */
2450
    avctx->sample_rate = s->sample_rate;
2451
    avctx->channels = s->nb_channels;
2452
    avctx->bit_rate = s->bit_rate;
2453
    avctx->sub_id = s->layer;
2454

    
2455
    s->frame_size = len;
2456

    
2457
    if (avctx->parse_only) {
2458
        out_size = buf_size;
2459
    } else {
2460
        out_size = mp_decode_frame(s, out_samples, buf, buf_size);
2461
    }
2462

    
2463
    *data_size = out_size;
2464
    return buf_size;
2465
}
2466
#endif /* CONFIG_MP3ADU_DECODER */
2467

    
2468
#ifdef CONFIG_MP3ON4_DECODER
2469

    
2470
/**
2471
 * Context for MP3On4 decoder
2472
 */
2473
typedef struct MP3On4DecodeContext {
2474
    int frames;   ///< number of mp3 frames per block (number of mp3 decoder instances)
2475
    int syncword; ///< syncword patch
2476
    const uint8_t *coff; ///< channels offsets in output buffer
2477
    MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance
2478
} MP3On4DecodeContext;
2479

    
2480
#include "mpeg4audio.h"
2481

    
2482
/* Next 3 arrays are indexed by channel config number (passed via codecdata) */
2483
static const uint8_t mp3Frames[8] = {0,1,1,2,3,3,4,5};   /* number of mp3 decoder instances */
2484
/* offsets into output buffer, assume output order is FL FR BL BR C LFE */
2485
static const uint8_t chan_offset[8][5] = {
2486
    {0},
2487
    {0},            // C
2488
    {0},            // FLR
2489
    {2,0},          // C FLR
2490
    {2,0,3},        // C FLR BS
2491
    {4,0,2},        // C FLR BLRS
2492
    {4,0,2,5},      // C FLR BLRS LFE
2493
    {4,0,2,6,5},    // C FLR BLRS BLR LFE
2494
};
2495

    
2496

    
2497
static int decode_init_mp3on4(AVCodecContext * avctx)
2498
{
2499
    MP3On4DecodeContext *s = avctx->priv_data;
2500
    MPEG4AudioConfig cfg;
2501
    int i;
2502

    
2503
    if ((avctx->extradata_size < 2) || (avctx->extradata == NULL)) {
2504
        av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
2505
        return -1;
2506
    }
2507

    
2508
    ff_mpeg4audio_get_config(&cfg, avctx->extradata, avctx->extradata_size);
2509
    if (!cfg.chan_config || cfg.chan_config > 7) {
2510
        av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
2511
        return -1;
2512
    }
2513
    s->frames = mp3Frames[cfg.chan_config];
2514
    s->coff = chan_offset[cfg.chan_config];
2515
    avctx->channels = ff_mpeg4audio_channels[cfg.chan_config];
2516

    
2517
    if (cfg.sample_rate < 16000)
2518
        s->syncword = 0xffe00000;
2519
    else
2520
        s->syncword = 0xfff00000;
2521

    
2522
    /* Init the first mp3 decoder in standard way, so that all tables get builded
2523
     * We replace avctx->priv_data with the context of the first decoder so that
2524
     * decode_init() does not have to be changed.
2525
     * Other decoders will be initialized here copying data from the first context
2526
     */
2527
    // Allocate zeroed memory for the first decoder context
2528
    s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
2529
    // Put decoder context in place to make init_decode() happy
2530
    avctx->priv_data = s->mp3decctx[0];
2531
    decode_init(avctx);
2532
    // Restore mp3on4 context pointer
2533
    avctx->priv_data = s;
2534
    s->mp3decctx[0]->adu_mode = 1; // Set adu mode
2535

    
2536
    /* Create a separate codec/context for each frame (first is already ok).
2537
     * Each frame is 1 or 2 channels - up to 5 frames allowed
2538
     */
2539
    for (i = 1; i < s->frames; i++) {
2540
        s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
2541
        s->mp3decctx[i]->compute_antialias = s->mp3decctx[0]->compute_antialias;
2542
        s->mp3decctx[i]->adu_mode = 1;
2543
        s->mp3decctx[i]->avctx = avctx;
2544
    }
2545

    
2546
    return 0;
2547
}
2548

    
2549

    
2550
static int decode_close_mp3on4(AVCodecContext * avctx)
2551
{
2552
    MP3On4DecodeContext *s = avctx->priv_data;
2553
    int i;
2554

    
2555
    for (i = 0; i < s->frames; i++)
2556
        if (s->mp3decctx[i])
2557
            av_free(s->mp3decctx[i]);
2558

    
2559
    return 0;
2560
}
2561

    
2562

    
2563
static int decode_frame_mp3on4(AVCodecContext * avctx,
2564
                        void *data, int *data_size,
2565
                        const uint8_t * buf, int buf_size)
2566
{
2567
    MP3On4DecodeContext *s = avctx->priv_data;
2568
    MPADecodeContext *m;
2569
    int len, out_size = 0;
2570
    uint32_t header;
2571
    OUT_INT *out_samples = data;
2572
    OUT_INT decoded_buf[MPA_FRAME_SIZE * MPA_MAX_CHANNELS];
2573
    OUT_INT *outptr, *bp;
2574
    int fsize;
2575
    int fr, i, j, n;
2576
    int off = avctx->channels;
2577

    
2578
    len = buf_size;
2579

    
2580
    *data_size = 0;
2581
    // Discard too short frames
2582
    if (buf_size < HEADER_SIZE)
2583
        return -1;
2584

    
2585
    // If only one decoder interleave is not needed
2586
    outptr = s->frames == 1 ? out_samples : decoded_buf;
2587

    
2588
    for (fr = 0; fr < s->frames; fr++) {
2589
        fsize = AV_RB16(buf) >> 4;
2590
        fsize = FFMIN3(fsize, len, MPA_MAX_CODED_FRAME_SIZE);
2591
        m = s->mp3decctx[fr];
2592
        assert (m != NULL);
2593

    
2594
        header = (AV_RB32(buf) & 0x000fffff) | s->syncword; // patch header
2595

    
2596
        if (ff_mpa_check_header(header) < 0) { // Bad header, discard block
2597
            *data_size = 0;
2598
            return buf_size;
2599
        }
2600

    
2601
        ff_mpegaudio_decode_header(m, header);
2602
        out_size += mp_decode_frame(m, outptr, buf, fsize);
2603
        buf += fsize;
2604
        len -= fsize;
2605

    
2606
        if(s->frames > 1) {
2607
            n = m->avctx->frame_size*m->nb_channels;
2608
            /* interleave output data */
2609
            bp = out_samples + s->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