Statistics
| Branch: | Revision:

ffmpeg / libavcodec / mpegaudiodec.c @ 84dc2d8a

History | View | Annotate | Download (76.6 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 libavcodec/mpegaudiodec.c
24
 * MPEG Audio decoder.
25
 */
26

    
27
#include "avcodec.h"
28
#include "get_bits.h"
29
#include "dsputil.h"
30

    
31
/*
32
 * TODO:
33
 *  - in low precision mode, use more 16 bit multiplies in synth filter
34
 *  - test lsf / mpeg25 extensively.
35
 */
36

    
37
#include "mpegaudio.h"
38
#include "mpegaudiodecheader.h"
39

    
40
#include "mathops.h"
41

    
42
/* WARNING: only correct for posititive numbers */
43
#define FIXR(a)   ((int)((a) * FRAC_ONE + 0.5))
44
#define FRAC_RND(a) (((a) + (FRAC_ONE/2)) >> FRAC_BITS)
45

    
46
#define FIXHR(a) ((int)((a) * (1LL<<32) + 0.5))
47

    
48
/****************/
49

    
50
#define HEADER_SIZE 4
51

    
52
#include "mpegaudiodata.h"
53
#include "mpegaudiodectab.h"
54

    
55
static void compute_antialias_integer(MPADecodeContext *s, GranuleDef *g);
56
static void compute_antialias_float(MPADecodeContext *s, GranuleDef *g);
57

    
58
/* vlc structure for decoding layer 3 huffman tables */
59
static VLC huff_vlc[16];
60
static VLC_TYPE huff_vlc_tables[
61
  0+128+128+128+130+128+154+166+
62
  142+204+190+170+542+460+662+414
63
  ][2];
64
static const int huff_vlc_tables_sizes[16] = {
65
  0, 128, 128, 128, 130, 128, 154, 166,
66
  142, 204, 190, 170, 542, 460, 662, 414
67
};
68
static VLC huff_quad_vlc[2];
69
static VLC_TYPE huff_quad_vlc_tables[128+16][2];
70
static const int huff_quad_vlc_tables_sizes[2] = {
71
  128, 16
72
};
73
/* computed from band_size_long */
74
static uint16_t band_index_long[9][23];
75
#include "mpegaudio_tablegen.h"
76
/* intensity stereo coef table */
77
static int32_t is_table[2][16];
78
static int32_t is_table_lsf[2][2][16];
79
static int32_t csa_table[8][4];
80
static float csa_table_float[8][4];
81
static int32_t mdct_win[8][36];
82

    
83
/* lower 2 bits: modulo 3, higher bits: shift */
84
static uint16_t scale_factor_modshift[64];
85
/* [i][j]:  2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
86
static int32_t scale_factor_mult[15][3];
87
/* mult table for layer 2 group quantization */
88

    
89
#define SCALE_GEN(v) \
90
{ FIXR(1.0 * (v)), FIXR(0.7937005259 * (v)), FIXR(0.6299605249 * (v)) }
91

    
92
static const int32_t scale_factor_mult2[3][3] = {
93
    SCALE_GEN(4.0 / 3.0), /* 3 steps */
94
    SCALE_GEN(4.0 / 5.0), /* 5 steps */
95
    SCALE_GEN(4.0 / 9.0), /* 9 steps */
96
};
97

    
98
DECLARE_ALIGNED(16, MPA_INT, ff_mpa_synth_window)[512];
99

    
100
/**
101
 * Convert region offsets to region sizes and truncate
102
 * size to big_values.
103
 */
104
void ff_region_offset2size(GranuleDef *g){
105
    int i, k, j=0;
106
    g->region_size[2] = (576 / 2);
107
    for(i=0;i<3;i++) {
108
        k = FFMIN(g->region_size[i], g->big_values);
109
        g->region_size[i] = k - j;
110
        j = k;
111
    }
112
}
113

    
114
void ff_init_short_region(MPADecodeContext *s, GranuleDef *g){
115
    if (g->block_type == 2)
116
        g->region_size[0] = (36 / 2);
117
    else {
118
        if (s->sample_rate_index <= 2)
119
            g->region_size[0] = (36 / 2);
120
        else if (s->sample_rate_index != 8)
121
            g->region_size[0] = (54 / 2);
122
        else
123
            g->region_size[0] = (108 / 2);
124
    }
125
    g->region_size[1] = (576 / 2);
126
}
127

    
128
void ff_init_long_region(MPADecodeContext *s, GranuleDef *g, int ra1, int ra2){
129
    int l;
130
    g->region_size[0] =
131
        band_index_long[s->sample_rate_index][ra1 + 1] >> 1;
132
    /* should not overflow */
133
    l = FFMIN(ra1 + ra2 + 2, 22);
134
    g->region_size[1] =
135
        band_index_long[s->sample_rate_index][l] >> 1;
136
}
137

    
138
void ff_compute_band_indexes(MPADecodeContext *s, GranuleDef *g){
139
    if (g->block_type == 2) {
140
        if (g->switch_point) {
141
            /* if switched mode, we handle the 36 first samples as
142
                long blocks.  For 8000Hz, we handle the 48 first
143
                exponents as long blocks (XXX: check this!) */
144
            if (s->sample_rate_index <= 2)
145
                g->long_end = 8;
146
            else if (s->sample_rate_index != 8)
147
                g->long_end = 6;
148
            else
149
                g->long_end = 4; /* 8000 Hz */
150

    
151
            g->short_start = 2 + (s->sample_rate_index != 8);
152
        } else {
153
            g->long_end = 0;
154
            g->short_start = 0;
155
        }
156
    } else {
157
        g->short_start = 13;
158
        g->long_end = 22;
159
    }
160
}
161

    
162
/* layer 1 unscaling */
163
/* n = number of bits of the mantissa minus 1 */
164
static inline int l1_unscale(int n, int mant, int scale_factor)
165
{
166
    int shift, mod;
167
    int64_t val;
168

    
169
    shift = scale_factor_modshift[scale_factor];
170
    mod = shift & 3;
171
    shift >>= 2;
172
    val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);
173
    shift += n;
174
    /* NOTE: at this point, 1 <= shift >= 21 + 15 */
175
    return (int)((val + (1LL << (shift - 1))) >> shift);
176
}
177

    
178
static inline int l2_unscale_group(int steps, int mant, int scale_factor)
179
{
180
    int shift, mod, val;
181

    
182
    shift = scale_factor_modshift[scale_factor];
183
    mod = shift & 3;
184
    shift >>= 2;
185

    
186
    val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
187
    /* NOTE: at this point, 0 <= shift <= 21 */
188
    if (shift > 0)
189
        val = (val + (1 << (shift - 1))) >> shift;
190
    return val;
191
}
192

    
193
/* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
194
static inline int l3_unscale(int value, int exponent)
195
{
196
    unsigned int m;
197
    int e;
198

    
199
    e = table_4_3_exp  [4*value + (exponent&3)];
200
    m = table_4_3_value[4*value + (exponent&3)];
201
    e -= (exponent >> 2);
202
    assert(e>=1);
203
    if (e > 31)
204
        return 0;
205
    m = (m + (1 << (e-1))) >> e;
206

    
207
    return m;
208
}
209

    
210
/* all integer n^(4/3) computation code */
211
#define DEV_ORDER 13
212

    
213
#define POW_FRAC_BITS 24
214
#define POW_FRAC_ONE    (1 << POW_FRAC_BITS)
215
#define POW_FIX(a)   ((int)((a) * POW_FRAC_ONE))
216
#define POW_MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> POW_FRAC_BITS)
217

    
218
static int dev_4_3_coefs[DEV_ORDER];
219

    
220
#if 0 /* unused */
221
static int pow_mult3[3] = {
222
    POW_FIX(1.0),
223
    POW_FIX(1.25992104989487316476),
224
    POW_FIX(1.58740105196819947474),
225
};
226
#endif
227

    
228
static av_cold void int_pow_init(void)
229
{
230
    int i, a;
231

    
232
    a = POW_FIX(1.0);
233
    for(i=0;i<DEV_ORDER;i++) {
234
        a = POW_MULL(a, POW_FIX(4.0 / 3.0) - i * POW_FIX(1.0)) / (i + 1);
235
        dev_4_3_coefs[i] = a;
236
    }
237
}
238

    
239
#if 0 /* unused, remove? */
240
/* return the mantissa and the binary exponent */
241
static int int_pow(int i, int *exp_ptr)
242
{
243
    int e, er, eq, j;
244
    int a, a1;
245

246
    /* renormalize */
247
    a = i;
248
    e = POW_FRAC_BITS;
249
    while (a < (1 << (POW_FRAC_BITS - 1))) {
250
        a = a << 1;
251
        e--;
252
    }
253
    a -= (1 << POW_FRAC_BITS);
254
    a1 = 0;
255
    for(j = DEV_ORDER - 1; j >= 0; j--)
256
        a1 = POW_MULL(a, dev_4_3_coefs[j] + a1);
257
    a = (1 << POW_FRAC_BITS) + a1;
258
    /* exponent compute (exact) */
259
    e = e * 4;
260
    er = e % 3;
261
    eq = e / 3;
262
    a = POW_MULL(a, pow_mult3[er]);
263
    while (a >= 2 * POW_FRAC_ONE) {
264
        a = a >> 1;
265
        eq++;
266
    }
267
    /* convert to float */
268
    while (a < POW_FRAC_ONE) {
269
        a = a << 1;
270
        eq--;
271
    }
272
    /* now POW_FRAC_ONE <= a < 2 * POW_FRAC_ONE */
273
#if POW_FRAC_BITS > FRAC_BITS
274
    a = (a + (1 << (POW_FRAC_BITS - FRAC_BITS - 1))) >> (POW_FRAC_BITS - FRAC_BITS);
275
    /* correct overflow */
276
    if (a >= 2 * (1 << FRAC_BITS)) {
277
        a = a >> 1;
278
        eq++;
279
    }
280
#endif
281
    *exp_ptr = eq;
282
    return a;
283
}
284
#endif
285

    
286
static av_cold int decode_init(AVCodecContext * avctx)
287
{
288
    MPADecodeContext *s = avctx->priv_data;
289
    static int init=0;
290
    int i, j, k;
291

    
292
    s->avctx = avctx;
293

    
294
    avctx->sample_fmt= OUT_FMT;
295
    s->error_recognition= avctx->error_recognition;
296

    
297
    if(avctx->antialias_algo != FF_AA_FLOAT)
298
        s->compute_antialias= compute_antialias_integer;
299
    else
300
        s->compute_antialias= compute_antialias_float;
301

    
302
    if (!init && !avctx->parse_only) {
303
        int offset;
304

    
305
        /* scale factors table for layer 1/2 */
306
        for(i=0;i<64;i++) {
307
            int shift, mod;
308
            /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */
309
            shift = (i / 3);
310
            mod = i % 3;
311
            scale_factor_modshift[i] = mod | (shift << 2);
312
        }
313

    
314
        /* scale factor multiply for layer 1 */
315
        for(i=0;i<15;i++) {
316
            int n, norm;
317
            n = i + 2;
318
            norm = ((INT64_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
319
            scale_factor_mult[i][0] = MULL(FIXR(1.0 * 2.0), norm, FRAC_BITS);
320
            scale_factor_mult[i][1] = MULL(FIXR(0.7937005259 * 2.0), norm, FRAC_BITS);
321
            scale_factor_mult[i][2] = MULL(FIXR(0.6299605249 * 2.0), norm, FRAC_BITS);
322
            dprintf(avctx, "%d: norm=%x s=%x %x %x\n",
323
                    i, norm,
324
                    scale_factor_mult[i][0],
325
                    scale_factor_mult[i][1],
326
                    scale_factor_mult[i][2]);
327
        }
328

    
329
        ff_mpa_synth_init(ff_mpa_synth_window);
330

    
331
        /* huffman decode tables */
332
        offset = 0;
333
        for(i=1;i<16;i++) {
334
            const HuffTable *h = &mpa_huff_tables[i];
335
            int xsize, x, y;
336
            uint8_t  tmp_bits [512];
337
            uint16_t tmp_codes[512];
338

    
339
            memset(tmp_bits , 0, sizeof(tmp_bits ));
340
            memset(tmp_codes, 0, sizeof(tmp_codes));
341

    
342
            xsize = h->xsize;
343

    
344
            j = 0;
345
            for(x=0;x<xsize;x++) {
346
                for(y=0;y<xsize;y++){
347
                    tmp_bits [(x << 5) | y | ((x&&y)<<4)]= h->bits [j  ];
348
                    tmp_codes[(x << 5) | y | ((x&&y)<<4)]= h->codes[j++];
349
                }
350
            }
351

    
352
            /* XXX: fail test */
353
            huff_vlc[i].table = huff_vlc_tables+offset;
354
            huff_vlc[i].table_allocated = huff_vlc_tables_sizes[i];
355
            init_vlc(&huff_vlc[i], 7, 512,
356
                     tmp_bits, 1, 1, tmp_codes, 2, 2,
357
                     INIT_VLC_USE_NEW_STATIC);
358
            offset += huff_vlc_tables_sizes[i];
359
        }
360
        assert(offset == FF_ARRAY_ELEMS(huff_vlc_tables));
361

    
362
        offset = 0;
363
        for(i=0;i<2;i++) {
364
            huff_quad_vlc[i].table = huff_quad_vlc_tables+offset;
365
            huff_quad_vlc[i].table_allocated = huff_quad_vlc_tables_sizes[i];
366
            init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
367
                     mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1,
368
                     INIT_VLC_USE_NEW_STATIC);
369
            offset += huff_quad_vlc_tables_sizes[i];
370
        }
371
        assert(offset == FF_ARRAY_ELEMS(huff_quad_vlc_tables));
372

    
373
        for(i=0;i<9;i++) {
374
            k = 0;
375
            for(j=0;j<22;j++) {
376
                band_index_long[i][j] = k;
377
                k += band_size_long[i][j];
378
            }
379
            band_index_long[i][22] = k;
380
        }
381

    
382
        /* compute n ^ (4/3) and store it in mantissa/exp format */
383

    
384
        int_pow_init();
385
        mpegaudio_tableinit();
386

    
387
        for(i=0;i<7;i++) {
388
            float f;
389
            int v;
390
            if (i != 6) {
391
                f = tan((double)i * M_PI / 12.0);
392
                v = FIXR(f / (1.0 + f));
393
            } else {
394
                v = FIXR(1.0);
395
            }
396
            is_table[0][i] = v;
397
            is_table[1][6 - i] = v;
398
        }
399
        /* invalid values */
400
        for(i=7;i<16;i++)
401
            is_table[0][i] = is_table[1][i] = 0.0;
402

    
403
        for(i=0;i<16;i++) {
404
            double f;
405
            int e, k;
406

    
407
            for(j=0;j<2;j++) {
408
                e = -(j + 1) * ((i + 1) >> 1);
409
                f = pow(2.0, e / 4.0);
410
                k = i & 1;
411
                is_table_lsf[j][k ^ 1][i] = FIXR(f);
412
                is_table_lsf[j][k][i] = FIXR(1.0);
413
                dprintf(avctx, "is_table_lsf %d %d: %x %x\n",
414
                        i, j, is_table_lsf[j][0][i], is_table_lsf[j][1][i]);
415
            }
416
        }
417

    
418
        for(i=0;i<8;i++) {
419
            float ci, cs, ca;
420
            ci = ci_table[i];
421
            cs = 1.0 / sqrt(1.0 + ci * ci);
422
            ca = cs * ci;
423
            csa_table[i][0] = FIXHR(cs/4);
424
            csa_table[i][1] = FIXHR(ca/4);
425
            csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);
426
            csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4);
427
            csa_table_float[i][0] = cs;
428
            csa_table_float[i][1] = ca;
429
            csa_table_float[i][2] = ca + cs;
430
            csa_table_float[i][3] = ca - cs;
431
        }
432

    
433
        /* compute mdct windows */
434
        for(i=0;i<36;i++) {
435
            for(j=0; j<4; j++){
436
                double d;
437

    
438
                if(j==2 && i%3 != 1)
439
                    continue;
440

    
441
                d= sin(M_PI * (i + 0.5) / 36.0);
442
                if(j==1){
443
                    if     (i>=30) d= 0;
444
                    else if(i>=24) d= sin(M_PI * (i - 18 + 0.5) / 12.0);
445
                    else if(i>=18) d= 1;
446
                }else if(j==3){
447
                    if     (i<  6) d= 0;
448
                    else if(i< 12) d= sin(M_PI * (i -  6 + 0.5) / 12.0);
449
                    else if(i< 18) d= 1;
450
                }
451
                //merge last stage of imdct into the window coefficients
452
                d*= 0.5 / cos(M_PI*(2*i + 19)/72);
453

    
454
                if(j==2)
455
                    mdct_win[j][i/3] = FIXHR((d / (1<<5)));
456
                else
457
                    mdct_win[j][i  ] = FIXHR((d / (1<<5)));
458
            }
459
        }
460

    
461
        /* NOTE: we do frequency inversion adter the MDCT by changing
462
           the sign of the right window coefs */
463
        for(j=0;j<4;j++) {
464
            for(i=0;i<36;i+=2) {
465
                mdct_win[j + 4][i] = mdct_win[j][i];
466
                mdct_win[j + 4][i + 1] = -mdct_win[j][i + 1];
467
            }
468
        }
469

    
470
        init = 1;
471
    }
472

    
473
    if (avctx->codec_id == CODEC_ID_MP3ADU)
474
        s->adu_mode = 1;
475
    return 0;
476
}
477

    
478
/* tab[i][j] = 1.0 / (2.0 * cos(pi*(2*k+1) / 2^(6 - j))) */
479

    
480
/* cos(i*pi/64) */
481

    
482
#define COS0_0  FIXHR(0.50060299823519630134/2)
483
#define COS0_1  FIXHR(0.50547095989754365998/2)
484
#define COS0_2  FIXHR(0.51544730992262454697/2)
485
#define COS0_3  FIXHR(0.53104259108978417447/2)
486
#define COS0_4  FIXHR(0.55310389603444452782/2)
487
#define COS0_5  FIXHR(0.58293496820613387367/2)
488
#define COS0_6  FIXHR(0.62250412303566481615/2)
489
#define COS0_7  FIXHR(0.67480834145500574602/2)
490
#define COS0_8  FIXHR(0.74453627100229844977/2)
491
#define COS0_9  FIXHR(0.83934964541552703873/2)
492
#define COS0_10 FIXHR(0.97256823786196069369/2)
493
#define COS0_11 FIXHR(1.16943993343288495515/4)
494
#define COS0_12 FIXHR(1.48416461631416627724/4)
495
#define COS0_13 FIXHR(2.05778100995341155085/8)
496
#define COS0_14 FIXHR(3.40760841846871878570/8)
497
#define COS0_15 FIXHR(10.19000812354805681150/32)
498

    
499
#define COS1_0 FIXHR(0.50241928618815570551/2)
500
#define COS1_1 FIXHR(0.52249861493968888062/2)
501
#define COS1_2 FIXHR(0.56694403481635770368/2)
502
#define COS1_3 FIXHR(0.64682178335999012954/2)
503
#define COS1_4 FIXHR(0.78815462345125022473/2)
504
#define COS1_5 FIXHR(1.06067768599034747134/4)
505
#define COS1_6 FIXHR(1.72244709823833392782/4)
506
#define COS1_7 FIXHR(5.10114861868916385802/16)
507

    
508
#define COS2_0 FIXHR(0.50979557910415916894/2)
509
#define COS2_1 FIXHR(0.60134488693504528054/2)
510
#define COS2_2 FIXHR(0.89997622313641570463/2)
511
#define COS2_3 FIXHR(2.56291544774150617881/8)
512

    
513
#define COS3_0 FIXHR(0.54119610014619698439/2)
514
#define COS3_1 FIXHR(1.30656296487637652785/4)
515

    
516
#define COS4_0 FIXHR(0.70710678118654752439/2)
517

    
518
/* butterfly operator */
519
#define BF(a, b, c, s)\
520
{\
521
    tmp0 = tab[a] + tab[b];\
522
    tmp1 = tab[a] - tab[b];\
523
    tab[a] = tmp0;\
524
    tab[b] = MULH(tmp1<<(s), c);\
525
}
526

    
527
#define BF1(a, b, c, d)\
528
{\
529
    BF(a, b, COS4_0, 1);\
530
    BF(c, d,-COS4_0, 1);\
531
    tab[c] += tab[d];\
532
}
533

    
534
#define BF2(a, b, c, d)\
535
{\
536
    BF(a, b, COS4_0, 1);\
537
    BF(c, d,-COS4_0, 1);\
538
    tab[c] += tab[d];\
539
    tab[a] += tab[c];\
540
    tab[c] += tab[b];\
541
    tab[b] += tab[d];\
542
}
543

    
544
#define ADD(a, b) tab[a] += tab[b]
545

    
546
/* DCT32 without 1/sqrt(2) coef zero scaling. */
547
static void dct32(int32_t *out, int32_t *tab)
548
{
549
    int tmp0, tmp1;
550

    
551
    /* pass 1 */
552
    BF( 0, 31, COS0_0 , 1);
553
    BF(15, 16, COS0_15, 5);
554
    /* pass 2 */
555
    BF( 0, 15, COS1_0 , 1);
556
    BF(16, 31,-COS1_0 , 1);
557
    /* pass 1 */
558
    BF( 7, 24, COS0_7 , 1);
559
    BF( 8, 23, COS0_8 , 1);
560
    /* pass 2 */
561
    BF( 7,  8, COS1_7 , 4);
562
    BF(23, 24,-COS1_7 , 4);
563
    /* pass 3 */
564
    BF( 0,  7, COS2_0 , 1);
565
    BF( 8, 15,-COS2_0 , 1);
566
    BF(16, 23, COS2_0 , 1);
567
    BF(24, 31,-COS2_0 , 1);
568
    /* pass 1 */
569
    BF( 3, 28, COS0_3 , 1);
570
    BF(12, 19, COS0_12, 2);
571
    /* pass 2 */
572
    BF( 3, 12, COS1_3 , 1);
573
    BF(19, 28,-COS1_3 , 1);
574
    /* pass 1 */
575
    BF( 4, 27, COS0_4 , 1);
576
    BF(11, 20, COS0_11, 2);
577
    /* pass 2 */
578
    BF( 4, 11, COS1_4 , 1);
579
    BF(20, 27,-COS1_4 , 1);
580
    /* pass 3 */
581
    BF( 3,  4, COS2_3 , 3);
582
    BF(11, 12,-COS2_3 , 3);
583
    BF(19, 20, COS2_3 , 3);
584
    BF(27, 28,-COS2_3 , 3);
585
    /* pass 4 */
586
    BF( 0,  3, COS3_0 , 1);
587
    BF( 4,  7,-COS3_0 , 1);
588
    BF( 8, 11, COS3_0 , 1);
589
    BF(12, 15,-COS3_0 , 1);
590
    BF(16, 19, COS3_0 , 1);
591
    BF(20, 23,-COS3_0 , 1);
592
    BF(24, 27, COS3_0 , 1);
593
    BF(28, 31,-COS3_0 , 1);
594

    
595

    
596

    
597
    /* pass 1 */
598
    BF( 1, 30, COS0_1 , 1);
599
    BF(14, 17, COS0_14, 3);
600
    /* pass 2 */
601
    BF( 1, 14, COS1_1 , 1);
602
    BF(17, 30,-COS1_1 , 1);
603
    /* pass 1 */
604
    BF( 6, 25, COS0_6 , 1);
605
    BF( 9, 22, COS0_9 , 1);
606
    /* pass 2 */
607
    BF( 6,  9, COS1_6 , 2);
608
    BF(22, 25,-COS1_6 , 2);
609
    /* pass 3 */
610
    BF( 1,  6, COS2_1 , 1);
611
    BF( 9, 14,-COS2_1 , 1);
612
    BF(17, 22, COS2_1 , 1);
613
    BF(25, 30,-COS2_1 , 1);
614

    
615
    /* pass 1 */
616
    BF( 2, 29, COS0_2 , 1);
617
    BF(13, 18, COS0_13, 3);
618
    /* pass 2 */
619
    BF( 2, 13, COS1_2 , 1);
620
    BF(18, 29,-COS1_2 , 1);
621
    /* pass 1 */
622
    BF( 5, 26, COS0_5 , 1);
623
    BF(10, 21, COS0_10, 1);
624
    /* pass 2 */
625
    BF( 5, 10, COS1_5 , 2);
626
    BF(21, 26,-COS1_5 , 2);
627
    /* pass 3 */
628
    BF( 2,  5, COS2_2 , 1);
629
    BF(10, 13,-COS2_2 , 1);
630
    BF(18, 21, COS2_2 , 1);
631
    BF(26, 29,-COS2_2 , 1);
632
    /* pass 4 */
633
    BF( 1,  2, COS3_1 , 2);
634
    BF( 5,  6,-COS3_1 , 2);
635
    BF( 9, 10, COS3_1 , 2);
636
    BF(13, 14,-COS3_1 , 2);
637
    BF(17, 18, COS3_1 , 2);
638
    BF(21, 22,-COS3_1 , 2);
639
    BF(25, 26, COS3_1 , 2);
640
    BF(29, 30,-COS3_1 , 2);
641

    
642
    /* pass 5 */
643
    BF1( 0,  1,  2,  3);
644
    BF2( 4,  5,  6,  7);
645
    BF1( 8,  9, 10, 11);
646
    BF2(12, 13, 14, 15);
647
    BF1(16, 17, 18, 19);
648
    BF2(20, 21, 22, 23);
649
    BF1(24, 25, 26, 27);
650
    BF2(28, 29, 30, 31);
651

    
652
    /* pass 6 */
653

    
654
    ADD( 8, 12);
655
    ADD(12, 10);
656
    ADD(10, 14);
657
    ADD(14,  9);
658
    ADD( 9, 13);
659
    ADD(13, 11);
660
    ADD(11, 15);
661

    
662
    out[ 0] = tab[0];
663
    out[16] = tab[1];
664
    out[ 8] = tab[2];
665
    out[24] = tab[3];
666
    out[ 4] = tab[4];
667
    out[20] = tab[5];
668
    out[12] = tab[6];
669
    out[28] = tab[7];
670
    out[ 2] = tab[8];
671
    out[18] = tab[9];
672
    out[10] = tab[10];
673
    out[26] = tab[11];
674
    out[ 6] = tab[12];
675
    out[22] = tab[13];
676
    out[14] = tab[14];
677
    out[30] = tab[15];
678

    
679
    ADD(24, 28);
680
    ADD(28, 26);
681
    ADD(26, 30);
682
    ADD(30, 25);
683
    ADD(25, 29);
684
    ADD(29, 27);
685
    ADD(27, 31);
686

    
687
    out[ 1] = tab[16] + tab[24];
688
    out[17] = tab[17] + tab[25];
689
    out[ 9] = tab[18] + tab[26];
690
    out[25] = tab[19] + tab[27];
691
    out[ 5] = tab[20] + tab[28];
692
    out[21] = tab[21] + tab[29];
693
    out[13] = tab[22] + tab[30];
694
    out[29] = tab[23] + tab[31];
695
    out[ 3] = tab[24] + tab[20];
696
    out[19] = tab[25] + tab[21];
697
    out[11] = tab[26] + tab[22];
698
    out[27] = tab[27] + tab[23];
699
    out[ 7] = tab[28] + tab[18];
700
    out[23] = tab[29] + tab[19];
701
    out[15] = tab[30] + tab[17];
702
    out[31] = tab[31];
703
}
704

    
705
#if FRAC_BITS <= 15
706

    
707
static inline int round_sample(int *sum)
708
{
709
    int sum1;
710
    sum1 = (*sum) >> OUT_SHIFT;
711
    *sum &= (1<<OUT_SHIFT)-1;
712
    return av_clip(sum1, OUT_MIN, OUT_MAX);
713
}
714

    
715
/* signed 16x16 -> 32 multiply add accumulate */
716
#define MACS(rt, ra, rb) MAC16(rt, ra, rb)
717

    
718
/* signed 16x16 -> 32 multiply */
719
#define MULS(ra, rb) MUL16(ra, rb)
720

    
721
#define MLSS(rt, ra, rb) MLS16(rt, ra, rb)
722

    
723
#else
724

    
725
static inline int round_sample(int64_t *sum)
726
{
727
    int sum1;
728
    sum1 = (int)((*sum) >> OUT_SHIFT);
729
    *sum &= (1<<OUT_SHIFT)-1;
730
    return av_clip(sum1, OUT_MIN, OUT_MAX);
731
}
732

    
733
#   define MULS(ra, rb) MUL64(ra, rb)
734
#   define MACS(rt, ra, rb) MAC64(rt, ra, rb)
735
#   define MLSS(rt, ra, rb) MLS64(rt, ra, rb)
736
#endif
737

    
738
#define SUM8(op, sum, w, p)               \
739
{                                         \
740
    op(sum, (w)[0 * 64], (p)[0 * 64]);    \
741
    op(sum, (w)[1 * 64], (p)[1 * 64]);    \
742
    op(sum, (w)[2 * 64], (p)[2 * 64]);    \
743
    op(sum, (w)[3 * 64], (p)[3 * 64]);    \
744
    op(sum, (w)[4 * 64], (p)[4 * 64]);    \
745
    op(sum, (w)[5 * 64], (p)[5 * 64]);    \
746
    op(sum, (w)[6 * 64], (p)[6 * 64]);    \
747
    op(sum, (w)[7 * 64], (p)[7 * 64]);    \
748
}
749

    
750
#define SUM8P2(sum1, op1, sum2, op2, w1, w2, p) \
751
{                                               \
752
    int tmp;\
753
    tmp = p[0 * 64];\
754
    op1(sum1, (w1)[0 * 64], tmp);\
755
    op2(sum2, (w2)[0 * 64], tmp);\
756
    tmp = p[1 * 64];\
757
    op1(sum1, (w1)[1 * 64], tmp);\
758
    op2(sum2, (w2)[1 * 64], tmp);\
759
    tmp = p[2 * 64];\
760
    op1(sum1, (w1)[2 * 64], tmp);\
761
    op2(sum2, (w2)[2 * 64], tmp);\
762
    tmp = p[3 * 64];\
763
    op1(sum1, (w1)[3 * 64], tmp);\
764
    op2(sum2, (w2)[3 * 64], tmp);\
765
    tmp = p[4 * 64];\
766
    op1(sum1, (w1)[4 * 64], tmp);\
767
    op2(sum2, (w2)[4 * 64], tmp);\
768
    tmp = p[5 * 64];\
769
    op1(sum1, (w1)[5 * 64], tmp);\
770
    op2(sum2, (w2)[5 * 64], tmp);\
771
    tmp = p[6 * 64];\
772
    op1(sum1, (w1)[6 * 64], tmp);\
773
    op2(sum2, (w2)[6 * 64], tmp);\
774
    tmp = p[7 * 64];\
775
    op1(sum1, (w1)[7 * 64], tmp);\
776
    op2(sum2, (w2)[7 * 64], tmp);\
777
}
778

    
779
void av_cold ff_mpa_synth_init(MPA_INT *window)
780
{
781
    int i;
782

    
783
    /* max = 18760, max sum over all 16 coefs : 44736 */
784
    for(i=0;i<257;i++) {
785
        int v;
786
        v = ff_mpa_enwindow[i];
787
#if WFRAC_BITS < 16
788
        v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS);
789
#endif
790
        window[i] = v;
791
        if ((i & 63) != 0)
792
            v = -v;
793
        if (i != 0)
794
            window[512 - i] = v;
795
    }
796
}
797

    
798
/* 32 sub band synthesis filter. Input: 32 sub band samples, Output:
799
   32 samples. */
800
/* XXX: optimize by avoiding ring buffer usage */
801
void ff_mpa_synth_filter(MPA_INT *synth_buf_ptr, int *synth_buf_offset,
802
                         MPA_INT *window, int *dither_state,
803
                         OUT_INT *samples, int incr,
804
                         int32_t sb_samples[SBLIMIT])
805
{
806
    register MPA_INT *synth_buf;
807
    register const MPA_INT *w, *w2, *p;
808
    int j, offset;
809
    OUT_INT *samples2;
810
#if FRAC_BITS <= 15
811
    int32_t tmp[32];
812
    int sum, sum2;
813
#else
814
    int64_t sum, sum2;
815
#endif
816

    
817
    offset = *synth_buf_offset;
818
    synth_buf = synth_buf_ptr + offset;
819

    
820
#if FRAC_BITS <= 15
821
    dct32(tmp, sb_samples);
822
    for(j=0;j<32;j++) {
823
        /* NOTE: can cause a loss in precision if very high amplitude
824
           sound */
825
        synth_buf[j] = av_clip_int16(tmp[j]);
826
    }
827
#else
828
    dct32(synth_buf, sb_samples);
829
#endif
830

    
831
    /* copy to avoid wrap */
832
    memcpy(synth_buf + 512, synth_buf, 32 * sizeof(MPA_INT));
833

    
834
    samples2 = samples + 31 * incr;
835
    w = window;
836
    w2 = window + 31;
837

    
838
    sum = *dither_state;
839
    p = synth_buf + 16;
840
    SUM8(MACS, sum, w, p);
841
    p = synth_buf + 48;
842
    SUM8(MLSS, sum, w + 32, p);
843
    *samples = round_sample(&sum);
844
    samples += incr;
845
    w++;
846

    
847
    /* we calculate two samples at the same time to avoid one memory
848
       access per two sample */
849
    for(j=1;j<16;j++) {
850
        sum2 = 0;
851
        p = synth_buf + 16 + j;
852
        SUM8P2(sum, MACS, sum2, MLSS, w, w2, p);
853
        p = synth_buf + 48 - j;
854
        SUM8P2(sum, MLSS, sum2, MLSS, w + 32, w2 + 32, p);
855

    
856
        *samples = round_sample(&sum);
857
        samples += incr;
858
        sum += sum2;
859
        *samples2 = round_sample(&sum);
860
        samples2 -= incr;
861
        w++;
862
        w2--;
863
    }
864

    
865
    p = synth_buf + 32;
866
    SUM8(MLSS, sum, w + 32, p);
867
    *samples = round_sample(&sum);
868
    *dither_state= sum;
869

    
870
    offset = (offset - 32) & 511;
871
    *synth_buf_offset = offset;
872
}
873

    
874
#define C3 FIXHR(0.86602540378443864676/2)
875

    
876
/* 0.5 / cos(pi*(2*i+1)/36) */
877
static const int icos36[9] = {
878
    FIXR(0.50190991877167369479),
879
    FIXR(0.51763809020504152469), //0
880
    FIXR(0.55168895948124587824),
881
    FIXR(0.61038729438072803416),
882
    FIXR(0.70710678118654752439), //1
883
    FIXR(0.87172339781054900991),
884
    FIXR(1.18310079157624925896),
885
    FIXR(1.93185165257813657349), //2
886
    FIXR(5.73685662283492756461),
887
};
888

    
889
/* 0.5 / cos(pi*(2*i+1)/36) */
890
static const int icos36h[9] = {
891
    FIXHR(0.50190991877167369479/2),
892
    FIXHR(0.51763809020504152469/2), //0
893
    FIXHR(0.55168895948124587824/2),
894
    FIXHR(0.61038729438072803416/2),
895
    FIXHR(0.70710678118654752439/2), //1
896
    FIXHR(0.87172339781054900991/2),
897
    FIXHR(1.18310079157624925896/4),
898
    FIXHR(1.93185165257813657349/4), //2
899
//    FIXHR(5.73685662283492756461),
900
};
901

    
902
/* 12 points IMDCT. We compute it "by hand" by factorizing obvious
903
   cases. */
904
static void imdct12(int *out, int *in)
905
{
906
    int in0, in1, in2, in3, in4, in5, t1, t2;
907

    
908
    in0= in[0*3];
909
    in1= in[1*3] + in[0*3];
910
    in2= in[2*3] + in[1*3];
911
    in3= in[3*3] + in[2*3];
912
    in4= in[4*3] + in[3*3];
913
    in5= in[5*3] + in[4*3];
914
    in5 += in3;
915
    in3 += in1;
916

    
917
    in2= MULH(2*in2, C3);
918
    in3= MULH(4*in3, C3);
919

    
920
    t1 = in0 - in4;
921
    t2 = MULH(2*(in1 - in5), icos36h[4]);
922

    
923
    out[ 7]=
924
    out[10]= t1 + t2;
925
    out[ 1]=
926
    out[ 4]= t1 - t2;
927

    
928
    in0 += in4>>1;
929
    in4 = in0 + in2;
930
    in5 += 2*in1;
931
    in1 = MULH(in5 + in3, icos36h[1]);
932
    out[ 8]=
933
    out[ 9]= in4 + in1;
934
    out[ 2]=
935
    out[ 3]= in4 - in1;
936

    
937
    in0 -= in2;
938
    in5 = MULH(2*(in5 - in3), icos36h[7]);
939
    out[ 0]=
940
    out[ 5]= in0 - in5;
941
    out[ 6]=
942
    out[11]= in0 + in5;
943
}
944

    
945
/* cos(pi*i/18) */
946
#define C1 FIXHR(0.98480775301220805936/2)
947
#define C2 FIXHR(0.93969262078590838405/2)
948
#define C3 FIXHR(0.86602540378443864676/2)
949
#define C4 FIXHR(0.76604444311897803520/2)
950
#define C5 FIXHR(0.64278760968653932632/2)
951
#define C6 FIXHR(0.5/2)
952
#define C7 FIXHR(0.34202014332566873304/2)
953
#define C8 FIXHR(0.17364817766693034885/2)
954

    
955

    
956
/* using Lee like decomposition followed by hand coded 9 points DCT */
957
static void imdct36(int *out, int *buf, int *in, int *win)
958
{
959
    int i, j, t0, t1, t2, t3, s0, s1, s2, s3;
960
    int tmp[18], *tmp1, *in1;
961

    
962
    for(i=17;i>=1;i--)
963
        in[i] += in[i-1];
964
    for(i=17;i>=3;i-=2)
965
        in[i] += in[i-2];
966

    
967
    for(j=0;j<2;j++) {
968
        tmp1 = tmp + j;
969
        in1 = in + j;
970
#if 0
971
//more accurate but slower
972
        int64_t t0, t1, t2, t3;
973
        t2 = in1[2*4] + in1[2*8] - in1[2*2];
974

975
        t3 = (in1[2*0] + (int64_t)(in1[2*6]>>1))<<32;
976
        t1 = in1[2*0] - in1[2*6];
977
        tmp1[ 6] = t1 - (t2>>1);
978
        tmp1[16] = t1 + t2;
979

980
        t0 = MUL64(2*(in1[2*2] + in1[2*4]),    C2);
981
        t1 = MUL64(   in1[2*4] - in1[2*8] , -2*C8);
982
        t2 = MUL64(2*(in1[2*2] + in1[2*8]),   -C4);
983

984
        tmp1[10] = (t3 - t0 - t2) >> 32;
985
        tmp1[ 2] = (t3 + t0 + t1) >> 32;
986
        tmp1[14] = (t3 + t2 - t1) >> 32;
987

988
        tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
989
        t2 = MUL64(2*(in1[2*1] + in1[2*5]),    C1);
990
        t3 = MUL64(   in1[2*5] - in1[2*7] , -2*C7);
991
        t0 = MUL64(2*in1[2*3], C3);
992

993
        t1 = MUL64(2*(in1[2*1] + in1[2*7]),   -C5);
994

995
        tmp1[ 0] = (t2 + t3 + t0) >> 32;
996
        tmp1[12] = (t2 + t1 - t0) >> 32;
997
        tmp1[ 8] = (t3 - t1 - t0) >> 32;
998
#else
999
        t2 = in1[2*4] + in1[2*8] - in1[2*2];
1000

    
1001
        t3 = in1[2*0] + (in1[2*6]>>1);
1002
        t1 = in1[2*0] - in1[2*6];
1003
        tmp1[ 6] = t1 - (t2>>1);
1004
        tmp1[16] = t1 + t2;
1005

    
1006
        t0 = MULH(2*(in1[2*2] + in1[2*4]),    C2);
1007
        t1 = MULH(   in1[2*4] - in1[2*8] , -2*C8);
1008
        t2 = MULH(2*(in1[2*2] + in1[2*8]),   -C4);
1009

    
1010
        tmp1[10] = t3 - t0 - t2;
1011
        tmp1[ 2] = t3 + t0 + t1;
1012
        tmp1[14] = t3 + t2 - t1;
1013

    
1014
        tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
1015
        t2 = MULH(2*(in1[2*1] + in1[2*5]),    C1);
1016
        t3 = MULH(   in1[2*5] - in1[2*7] , -2*C7);
1017
        t0 = MULH(2*in1[2*3], C3);
1018

    
1019
        t1 = MULH(2*(in1[2*1] + in1[2*7]),   -C5);
1020

    
1021
        tmp1[ 0] = t2 + t3 + t0;
1022
        tmp1[12] = t2 + t1 - t0;
1023
        tmp1[ 8] = t3 - t1 - t0;
1024
#endif
1025
    }
1026

    
1027
    i = 0;
1028
    for(j=0;j<4;j++) {
1029
        t0 = tmp[i];
1030
        t1 = tmp[i + 2];
1031
        s0 = t1 + t0;
1032
        s2 = t1 - t0;
1033

    
1034
        t2 = tmp[i + 1];
1035
        t3 = tmp[i + 3];
1036
        s1 = MULH(2*(t3 + t2), icos36h[j]);
1037
        s3 = MULL(t3 - t2, icos36[8 - j], FRAC_BITS);
1038

    
1039
        t0 = s0 + s1;
1040
        t1 = s0 - s1;
1041
        out[(9 + j)*SBLIMIT] =  MULH(t1, win[9 + j]) + buf[9 + j];
1042
        out[(8 - j)*SBLIMIT] =  MULH(t1, win[8 - j]) + buf[8 - j];
1043
        buf[9 + j] = MULH(t0, win[18 + 9 + j]);
1044
        buf[8 - j] = MULH(t0, win[18 + 8 - j]);
1045

    
1046
        t0 = s2 + s3;
1047
        t1 = s2 - s3;
1048
        out[(9 + 8 - j)*SBLIMIT] =  MULH(t1, win[9 + 8 - j]) + buf[9 + 8 - j];
1049
        out[(        j)*SBLIMIT] =  MULH(t1, win[        j]) + buf[        j];
1050
        buf[9 + 8 - j] = MULH(t0, win[18 + 9 + 8 - j]);
1051
        buf[      + j] = MULH(t0, win[18         + j]);
1052
        i += 4;
1053
    }
1054

    
1055
    s0 = tmp[16];
1056
    s1 = MULH(2*tmp[17], icos36h[4]);
1057
    t0 = s0 + s1;
1058
    t1 = s0 - s1;
1059
    out[(9 + 4)*SBLIMIT] =  MULH(t1, win[9 + 4]) + buf[9 + 4];
1060
    out[(8 - 4)*SBLIMIT] =  MULH(t1, win[8 - 4]) + buf[8 - 4];
1061
    buf[9 + 4] = MULH(t0, win[18 + 9 + 4]);
1062
    buf[8 - 4] = MULH(t0, win[18 + 8 - 4]);
1063
}
1064

    
1065
/* return the number of decoded frames */
1066
static int mp_decode_layer1(MPADecodeContext *s)
1067
{
1068
    int bound, i, v, n, ch, j, mant;
1069
    uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
1070
    uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
1071

    
1072
    if (s->mode == MPA_JSTEREO)
1073
        bound = (s->mode_ext + 1) * 4;
1074
    else
1075
        bound = SBLIMIT;
1076

    
1077
    /* allocation bits */
1078
    for(i=0;i<bound;i++) {
1079
        for(ch=0;ch<s->nb_channels;ch++) {
1080
            allocation[ch][i] = get_bits(&s->gb, 4);
1081
        }
1082
    }
1083
    for(i=bound;i<SBLIMIT;i++) {
1084
        allocation[0][i] = get_bits(&s->gb, 4);
1085
    }
1086

    
1087
    /* scale factors */
1088
    for(i=0;i<bound;i++) {
1089
        for(ch=0;ch<s->nb_channels;ch++) {
1090
            if (allocation[ch][i])
1091
                scale_factors[ch][i] = get_bits(&s->gb, 6);
1092
        }
1093
    }
1094
    for(i=bound;i<SBLIMIT;i++) {
1095
        if (allocation[0][i]) {
1096
            scale_factors[0][i] = get_bits(&s->gb, 6);
1097
            scale_factors[1][i] = get_bits(&s->gb, 6);
1098
        }
1099
    }
1100

    
1101
    /* compute samples */
1102
    for(j=0;j<12;j++) {
1103
        for(i=0;i<bound;i++) {
1104
            for(ch=0;ch<s->nb_channels;ch++) {
1105
                n = allocation[ch][i];
1106
                if (n) {
1107
                    mant = get_bits(&s->gb, n + 1);
1108
                    v = l1_unscale(n, mant, scale_factors[ch][i]);
1109
                } else {
1110
                    v = 0;
1111
                }
1112
                s->sb_samples[ch][j][i] = v;
1113
            }
1114
        }
1115
        for(i=bound;i<SBLIMIT;i++) {
1116
            n = allocation[0][i];
1117
            if (n) {
1118
                mant = get_bits(&s->gb, n + 1);
1119
                v = l1_unscale(n, mant, scale_factors[0][i]);
1120
                s->sb_samples[0][j][i] = v;
1121
                v = l1_unscale(n, mant, scale_factors[1][i]);
1122
                s->sb_samples[1][j][i] = v;
1123
            } else {
1124
                s->sb_samples[0][j][i] = 0;
1125
                s->sb_samples[1][j][i] = 0;
1126
            }
1127
        }
1128
    }
1129
    return 12;
1130
}
1131

    
1132
static int mp_decode_layer2(MPADecodeContext *s)
1133
{
1134
    int sblimit; /* number of used subbands */
1135
    const unsigned char *alloc_table;
1136
    int table, bit_alloc_bits, i, j, ch, bound, v;
1137
    unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
1138
    unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
1139
    unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
1140
    int scale, qindex, bits, steps, k, l, m, b;
1141

    
1142
    /* select decoding table */
1143
    table = ff_mpa_l2_select_table(s->bit_rate / 1000, s->nb_channels,
1144
                            s->sample_rate, s->lsf);
1145
    sblimit = ff_mpa_sblimit_table[table];
1146
    alloc_table = ff_mpa_alloc_tables[table];
1147

    
1148
    if (s->mode == MPA_JSTEREO)
1149
        bound = (s->mode_ext + 1) * 4;
1150
    else
1151
        bound = sblimit;
1152

    
1153
    dprintf(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit);
1154

    
1155
    /* sanity check */
1156
    if( bound > sblimit ) bound = sblimit;
1157

    
1158
    /* parse bit allocation */
1159
    j = 0;
1160
    for(i=0;i<bound;i++) {
1161
        bit_alloc_bits = alloc_table[j];
1162
        for(ch=0;ch<s->nb_channels;ch++) {
1163
            bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
1164
        }
1165
        j += 1 << bit_alloc_bits;
1166
    }
1167
    for(i=bound;i<sblimit;i++) {
1168
        bit_alloc_bits = alloc_table[j];
1169
        v = get_bits(&s->gb, bit_alloc_bits);
1170
        bit_alloc[0][i] = v;
1171
        bit_alloc[1][i] = v;
1172
        j += 1 << bit_alloc_bits;
1173
    }
1174

    
1175
    /* scale codes */
1176
    for(i=0;i<sblimit;i++) {
1177
        for(ch=0;ch<s->nb_channels;ch++) {
1178
            if (bit_alloc[ch][i])
1179
                scale_code[ch][i] = get_bits(&s->gb, 2);
1180
        }
1181
    }
1182

    
1183
    /* scale factors */
1184
    for(i=0;i<sblimit;i++) {
1185
        for(ch=0;ch<s->nb_channels;ch++) {
1186
            if (bit_alloc[ch][i]) {
1187
                sf = scale_factors[ch][i];
1188
                switch(scale_code[ch][i]) {
1189
                default:
1190
                case 0:
1191
                    sf[0] = get_bits(&s->gb, 6);
1192
                    sf[1] = get_bits(&s->gb, 6);
1193
                    sf[2] = get_bits(&s->gb, 6);
1194
                    break;
1195
                case 2:
1196
                    sf[0] = get_bits(&s->gb, 6);
1197
                    sf[1] = sf[0];
1198
                    sf[2] = sf[0];
1199
                    break;
1200
                case 1:
1201
                    sf[0] = get_bits(&s->gb, 6);
1202
                    sf[2] = get_bits(&s->gb, 6);
1203
                    sf[1] = sf[0];
1204
                    break;
1205
                case 3:
1206
                    sf[0] = get_bits(&s->gb, 6);
1207
                    sf[2] = get_bits(&s->gb, 6);
1208
                    sf[1] = sf[2];
1209
                    break;
1210
                }
1211
            }
1212
        }
1213
    }
1214

    
1215
    /* samples */
1216
    for(k=0;k<3;k++) {
1217
        for(l=0;l<12;l+=3) {
1218
            j = 0;
1219
            for(i=0;i<bound;i++) {
1220
                bit_alloc_bits = alloc_table[j];
1221
                for(ch=0;ch<s->nb_channels;ch++) {
1222
                    b = bit_alloc[ch][i];
1223
                    if (b) {
1224
                        scale = scale_factors[ch][i][k];
1225
                        qindex = alloc_table[j+b];
1226
                        bits = ff_mpa_quant_bits[qindex];
1227
                        if (bits < 0) {
1228
                            /* 3 values at the same time */
1229
                            v = get_bits(&s->gb, -bits);
1230
                            steps = ff_mpa_quant_steps[qindex];
1231
                            s->sb_samples[ch][k * 12 + l + 0][i] =
1232
                                l2_unscale_group(steps, v % steps, scale);
1233
                            v = v / steps;
1234
                            s->sb_samples[ch][k * 12 + l + 1][i] =
1235
                                l2_unscale_group(steps, v % steps, scale);
1236
                            v = v / steps;
1237
                            s->sb_samples[ch][k * 12 + l + 2][i] =
1238
                                l2_unscale_group(steps, v, scale);
1239
                        } else {
1240
                            for(m=0;m<3;m++) {
1241
                                v = get_bits(&s->gb, bits);
1242
                                v = l1_unscale(bits - 1, v, scale);
1243
                                s->sb_samples[ch][k * 12 + l + m][i] = v;
1244
                            }
1245
                        }
1246
                    } else {
1247
                        s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1248
                        s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1249
                        s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1250
                    }
1251
                }
1252
                /* next subband in alloc table */
1253
                j += 1 << bit_alloc_bits;
1254
            }
1255
            /* XXX: find a way to avoid this duplication of code */
1256
            for(i=bound;i<sblimit;i++) {
1257
                bit_alloc_bits = alloc_table[j];
1258
                b = bit_alloc[0][i];
1259
                if (b) {
1260
                    int mant, scale0, scale1;
1261
                    scale0 = scale_factors[0][i][k];
1262
                    scale1 = scale_factors[1][i][k];
1263
                    qindex = alloc_table[j+b];
1264
                    bits = ff_mpa_quant_bits[qindex];
1265
                    if (bits < 0) {
1266
                        /* 3 values at the same time */
1267
                        v = get_bits(&s->gb, -bits);
1268
                        steps = ff_mpa_quant_steps[qindex];
1269
                        mant = v % steps;
1270
                        v = v / steps;
1271
                        s->sb_samples[0][k * 12 + l + 0][i] =
1272
                            l2_unscale_group(steps, mant, scale0);
1273
                        s->sb_samples[1][k * 12 + l + 0][i] =
1274
                            l2_unscale_group(steps, mant, scale1);
1275
                        mant = v % steps;
1276
                        v = v / steps;
1277
                        s->sb_samples[0][k * 12 + l + 1][i] =
1278
                            l2_unscale_group(steps, mant, scale0);
1279
                        s->sb_samples[1][k * 12 + l + 1][i] =
1280
                            l2_unscale_group(steps, mant, scale1);
1281
                        s->sb_samples[0][k * 12 + l + 2][i] =
1282
                            l2_unscale_group(steps, v, scale0);
1283
                        s->sb_samples[1][k * 12 + l + 2][i] =
1284
                            l2_unscale_group(steps, v, scale1);
1285
                    } else {
1286
                        for(m=0;m<3;m++) {
1287
                            mant = get_bits(&s->gb, bits);
1288
                            s->sb_samples[0][k * 12 + l + m][i] =
1289
                                l1_unscale(bits - 1, mant, scale0);
1290
                            s->sb_samples[1][k * 12 + l + m][i] =
1291
                                l1_unscale(bits - 1, mant, scale1);
1292
                        }
1293
                    }
1294
                } else {
1295
                    s->sb_samples[0][k * 12 + l + 0][i] = 0;
1296
                    s->sb_samples[0][k * 12 + l + 1][i] = 0;
1297
                    s->sb_samples[0][k * 12 + l + 2][i] = 0;
1298
                    s->sb_samples[1][k * 12 + l + 0][i] = 0;
1299
                    s->sb_samples[1][k * 12 + l + 1][i] = 0;
1300
                    s->sb_samples[1][k * 12 + l + 2][i] = 0;
1301
                }
1302
                /* next subband in alloc table */
1303
                j += 1 << bit_alloc_bits;
1304
            }
1305
            /* fill remaining samples to zero */
1306
            for(i=sblimit;i<SBLIMIT;i++) {
1307
                for(ch=0;ch<s->nb_channels;ch++) {
1308
                    s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1309
                    s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1310
                    s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1311
                }
1312
            }
1313
        }
1314
    }
1315
    return 3 * 12;
1316
}
1317

    
1318
static inline void lsf_sf_expand(int *slen,
1319
                                 int sf, int n1, int n2, int n3)
1320
{
1321
    if (n3) {
1322
        slen[3] = sf % n3;
1323
        sf /= n3;
1324
    } else {
1325
        slen[3] = 0;
1326
    }
1327
    if (n2) {
1328
        slen[2] = sf % n2;
1329
        sf /= n2;
1330
    } else {
1331
        slen[2] = 0;
1332
    }
1333
    slen[1] = sf % n1;
1334
    sf /= n1;
1335
    slen[0] = sf;
1336
}
1337

    
1338
static void exponents_from_scale_factors(MPADecodeContext *s,
1339
                                         GranuleDef *g,
1340
                                         int16_t *exponents)
1341
{
1342
    const uint8_t *bstab, *pretab;
1343
    int len, i, j, k, l, v0, shift, gain, gains[3];
1344
    int16_t *exp_ptr;
1345

    
1346
    exp_ptr = exponents;
1347
    gain = g->global_gain - 210;
1348
    shift = g->scalefac_scale + 1;
1349

    
1350
    bstab = band_size_long[s->sample_rate_index];
1351
    pretab = mpa_pretab[g->preflag];
1352
    for(i=0;i<g->long_end;i++) {
1353
        v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400;
1354
        len = bstab[i];
1355
        for(j=len;j>0;j--)
1356
            *exp_ptr++ = v0;
1357
    }
1358

    
1359
    if (g->short_start < 13) {
1360
        bstab = band_size_short[s->sample_rate_index];
1361
        gains[0] = gain - (g->subblock_gain[0] << 3);
1362
        gains[1] = gain - (g->subblock_gain[1] << 3);
1363
        gains[2] = gain - (g->subblock_gain[2] << 3);
1364
        k = g->long_end;
1365
        for(i=g->short_start;i<13;i++) {
1366
            len = bstab[i];
1367
            for(l=0;l<3;l++) {
1368
                v0 = gains[l] - (g->scale_factors[k++] << shift) + 400;
1369
                for(j=len;j>0;j--)
1370
                *exp_ptr++ = v0;
1371
            }
1372
        }
1373
    }
1374
}
1375

    
1376
/* handle n = 0 too */
1377
static inline int get_bitsz(GetBitContext *s, int n)
1378
{
1379
    if (n == 0)
1380
        return 0;
1381
    else
1382
        return get_bits(s, n);
1383
}
1384

    
1385

    
1386
static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos, int *end_pos2){
1387
    if(s->in_gb.buffer && *pos >= s->gb.size_in_bits){
1388
        s->gb= s->in_gb;
1389
        s->in_gb.buffer=NULL;
1390
        assert((get_bits_count(&s->gb) & 7) == 0);
1391
        skip_bits_long(&s->gb, *pos - *end_pos);
1392
        *end_pos2=
1393
        *end_pos= *end_pos2 + get_bits_count(&s->gb) - *pos;
1394
        *pos= get_bits_count(&s->gb);
1395
    }
1396
}
1397

    
1398
static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
1399
                          int16_t *exponents, int end_pos2)
1400
{
1401
    int s_index;
1402
    int i;
1403
    int last_pos, bits_left;
1404
    VLC *vlc;
1405
    int end_pos= FFMIN(end_pos2, s->gb.size_in_bits);
1406

    
1407
    /* low frequencies (called big values) */
1408
    s_index = 0;
1409
    for(i=0;i<3;i++) {
1410
        int j, k, l, linbits;
1411
        j = g->region_size[i];
1412
        if (j == 0)
1413
            continue;
1414
        /* select vlc table */
1415
        k = g->table_select[i];
1416
        l = mpa_huff_data[k][0];
1417
        linbits = mpa_huff_data[k][1];
1418
        vlc = &huff_vlc[l];
1419

    
1420
        if(!l){
1421
            memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*2*j);
1422
            s_index += 2*j;
1423
            continue;
1424
        }
1425

    
1426
        /* read huffcode and compute each couple */
1427
        for(;j>0;j--) {
1428
            int exponent, x, y, v;
1429
            int pos= get_bits_count(&s->gb);
1430

    
1431
            if (pos >= end_pos){
1432
//                av_log(NULL, AV_LOG_ERROR, "pos: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);
1433
                switch_buffer(s, &pos, &end_pos, &end_pos2);
1434
//                av_log(NULL, AV_LOG_ERROR, "new pos: %d %d\n", pos, end_pos);
1435
                if(pos >= end_pos)
1436
                    break;
1437
            }
1438
            y = get_vlc2(&s->gb, vlc->table, 7, 3);
1439

    
1440
            if(!y){
1441
                g->sb_hybrid[s_index  ] =
1442
                g->sb_hybrid[s_index+1] = 0;
1443
                s_index += 2;
1444
                continue;
1445
            }
1446

    
1447
            exponent= exponents[s_index];
1448

    
1449
            dprintf(s->avctx, "region=%d n=%d x=%d y=%d exp=%d\n",
1450
                    i, g->region_size[i] - j, x, y, exponent);
1451
            if(y&16){
1452
                x = y >> 5;
1453
                y = y & 0x0f;
1454
                if (x < 15){
1455
                    v = expval_table[ exponent ][ x ];
1456
//                      v = expval_table[ (exponent&3) ][ x ] >> FFMIN(0 - (exponent>>2), 31);
1457
                }else{
1458
                    x += get_bitsz(&s->gb, linbits);
1459
                    v = l3_unscale(x, exponent);
1460
                }
1461
                if (get_bits1(&s->gb))
1462
                    v = -v;
1463
                g->sb_hybrid[s_index] = v;
1464
                if (y < 15){
1465
                    v = expval_table[ exponent ][ y ];
1466
                }else{
1467
                    y += get_bitsz(&s->gb, linbits);
1468
                    v = l3_unscale(y, exponent);
1469
                }
1470
                if (get_bits1(&s->gb))
1471
                    v = -v;
1472
                g->sb_hybrid[s_index+1] = v;
1473
            }else{
1474
                x = y >> 5;
1475
                y = y & 0x0f;
1476
                x += y;
1477
                if (x < 15){
1478
                    v = expval_table[ exponent ][ x ];
1479
                }else{
1480
                    x += get_bitsz(&s->gb, linbits);
1481
                    v = l3_unscale(x, exponent);
1482
                }
1483
                if (get_bits1(&s->gb))
1484
                    v = -v;
1485
                g->sb_hybrid[s_index+!!y] = v;
1486
                g->sb_hybrid[s_index+ !y] = 0;
1487
            }
1488
            s_index+=2;
1489
        }
1490
    }
1491

    
1492
    /* high frequencies */
1493
    vlc = &huff_quad_vlc[g->count1table_select];
1494
    last_pos=0;
1495
    while (s_index <= 572) {
1496
        int pos, code;
1497
        pos = get_bits_count(&s->gb);
1498
        if (pos >= end_pos) {
1499
            if (pos > end_pos2 && last_pos){
1500
                /* some encoders generate an incorrect size for this
1501
                   part. We must go back into the data */
1502
                s_index -= 4;
1503
                skip_bits_long(&s->gb, last_pos - pos);
1504
                av_log(s->avctx, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos);
1505
                if(s->error_recognition >= FF_ER_COMPLIANT)
1506
                    s_index=0;
1507
                break;
1508
            }
1509
//                av_log(NULL, AV_LOG_ERROR, "pos2: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);
1510
            switch_buffer(s, &pos, &end_pos, &end_pos2);
1511
//                av_log(NULL, AV_LOG_ERROR, "new pos2: %d %d %d\n", pos, end_pos, s_index);
1512
            if(pos >= end_pos)
1513
                break;
1514
        }
1515
        last_pos= pos;
1516

    
1517
        code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);
1518
        dprintf(s->avctx, "t=%d code=%d\n", g->count1table_select, code);
1519
        g->sb_hybrid[s_index+0]=
1520
        g->sb_hybrid[s_index+1]=
1521
        g->sb_hybrid[s_index+2]=
1522
        g->sb_hybrid[s_index+3]= 0;
1523
        while(code){
1524
            static const int idxtab[16]={3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0};
1525
            int v;
1526
            int pos= s_index+idxtab[code];
1527
            code ^= 8>>idxtab[code];
1528
            v = exp_table[ exponents[pos] ];
1529
//            v = exp_table[ (exponents[pos]&3) ] >> FFMIN(0 - (exponents[pos]>>2), 31);
1530
            if(get_bits1(&s->gb))
1531
                v = -v;
1532
            g->sb_hybrid[pos] = v;
1533
        }
1534
        s_index+=4;
1535
    }
1536
    /* skip extension bits */
1537
    bits_left = end_pos2 - get_bits_count(&s->gb);
1538
//av_log(NULL, AV_LOG_ERROR, "left:%d buf:%p\n", bits_left, s->in_gb.buffer);
1539
    if (bits_left < 0 && s->error_recognition >= FF_ER_COMPLIANT) {
1540
        av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
1541
        s_index=0;
1542
    }else if(bits_left > 0 && s->error_recognition >= FF_ER_AGGRESSIVE){
1543
        av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
1544
        s_index=0;
1545
    }
1546
    memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*(576 - s_index));
1547
    skip_bits_long(&s->gb, bits_left);
1548

    
1549
    i= get_bits_count(&s->gb);
1550
    switch_buffer(s, &i, &end_pos, &end_pos2);
1551

    
1552
    return 0;
1553
}
1554

    
1555
/* Reorder short blocks from bitstream order to interleaved order. It
1556
   would be faster to do it in parsing, but the code would be far more
1557
   complicated */
1558
static void reorder_block(MPADecodeContext *s, GranuleDef *g)
1559
{
1560
    int i, j, len;
1561
    int32_t *ptr, *dst, *ptr1;
1562
    int32_t tmp[576];
1563

    
1564
    if (g->block_type != 2)
1565
        return;
1566

    
1567
    if (g->switch_point) {
1568
        if (s->sample_rate_index != 8) {
1569
            ptr = g->sb_hybrid + 36;
1570
        } else {
1571
            ptr = g->sb_hybrid + 48;
1572
        }
1573
    } else {
1574
        ptr = g->sb_hybrid;
1575
    }
1576

    
1577
    for(i=g->short_start;i<13;i++) {
1578
        len = band_size_short[s->sample_rate_index][i];
1579
        ptr1 = ptr;
1580
        dst = tmp;
1581
        for(j=len;j>0;j--) {
1582
            *dst++ = ptr[0*len];
1583
            *dst++ = ptr[1*len];
1584
            *dst++ = ptr[2*len];
1585
            ptr++;
1586
        }
1587
        ptr+=2*len;
1588
        memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));
1589
    }
1590
}
1591

    
1592
#define ISQRT2 FIXR(0.70710678118654752440)
1593

    
1594
static void compute_stereo(MPADecodeContext *s,
1595
                           GranuleDef *g0, GranuleDef *g1)
1596
{
1597
    int i, j, k, l;
1598
    int32_t v1, v2;
1599
    int sf_max, tmp0, tmp1, sf, len, non_zero_found;
1600
    int32_t (*is_tab)[16];
1601
    int32_t *tab0, *tab1;
1602
    int non_zero_found_short[3];
1603

    
1604
    /* intensity stereo */
1605
    if (s->mode_ext & MODE_EXT_I_STEREO) {
1606
        if (!s->lsf) {
1607
            is_tab = is_table;
1608
            sf_max = 7;
1609
        } else {
1610
            is_tab = is_table_lsf[g1->scalefac_compress & 1];
1611
            sf_max = 16;
1612
        }
1613

    
1614
        tab0 = g0->sb_hybrid + 576;
1615
        tab1 = g1->sb_hybrid + 576;
1616

    
1617
        non_zero_found_short[0] = 0;
1618
        non_zero_found_short[1] = 0;
1619
        non_zero_found_short[2] = 0;
1620
        k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1621
        for(i = 12;i >= g1->short_start;i--) {
1622
            /* for last band, use previous scale factor */
1623
            if (i != 11)
1624
                k -= 3;
1625
            len = band_size_short[s->sample_rate_index][i];
1626
            for(l=2;l>=0;l--) {
1627
                tab0 -= len;
1628
                tab1 -= len;
1629
                if (!non_zero_found_short[l]) {
1630
                    /* test if non zero band. if so, stop doing i-stereo */
1631
                    for(j=0;j<len;j++) {
1632
                        if (tab1[j] != 0) {
1633
                            non_zero_found_short[l] = 1;
1634
                            goto found1;
1635
                        }
1636
                    }
1637
                    sf = g1->scale_factors[k + l];
1638
                    if (sf >= sf_max)
1639
                        goto found1;
1640

    
1641
                    v1 = is_tab[0][sf];
1642
                    v2 = is_tab[1][sf];
1643
                    for(j=0;j<len;j++) {
1644
                        tmp0 = tab0[j];
1645
                        tab0[j] = MULL(tmp0, v1, FRAC_BITS);
1646
                        tab1[j] = MULL(tmp0, v2, FRAC_BITS);
1647
                    }
1648
                } else {
1649
                found1:
1650
                    if (s->mode_ext & MODE_EXT_MS_STEREO) {
1651
                        /* lower part of the spectrum : do ms stereo
1652
                           if enabled */
1653
                        for(j=0;j<len;j++) {
1654
                            tmp0 = tab0[j];
1655
                            tmp1 = tab1[j];
1656
                            tab0[j] = MULL(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1657
                            tab1[j] = MULL(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1658
                        }
1659
                    }
1660
                }
1661
            }
1662
        }
1663

    
1664
        non_zero_found = non_zero_found_short[0] |
1665
            non_zero_found_short[1] |
1666
            non_zero_found_short[2];
1667

    
1668
        for(i = g1->long_end - 1;i >= 0;i--) {
1669
            len = band_size_long[s->sample_rate_index][i];
1670
            tab0 -= len;
1671
            tab1 -= len;
1672
            /* test if non zero band. if so, stop doing i-stereo */
1673
            if (!non_zero_found) {
1674
                for(j=0;j<len;j++) {
1675
                    if (tab1[j] != 0) {
1676
                        non_zero_found = 1;
1677
                        goto found2;
1678
                    }
1679
                }
1680
                /* for last band, use previous scale factor */
1681
                k = (i == 21) ? 20 : i;
1682
                sf = g1->scale_factors[k];
1683
                if (sf >= sf_max)
1684
                    goto found2;
1685
                v1 = is_tab[0][sf];
1686
                v2 = is_tab[1][sf];
1687
                for(j=0;j<len;j++) {
1688
                    tmp0 = tab0[j];
1689
                    tab0[j] = MULL(tmp0, v1, FRAC_BITS);
1690
                    tab1[j] = MULL(tmp0, v2, FRAC_BITS);
1691
                }
1692
            } else {
1693
            found2:
1694
                if (s->mode_ext & MODE_EXT_MS_STEREO) {
1695
                    /* lower part of the spectrum : do ms stereo
1696
                       if enabled */
1697
                    for(j=0;j<len;j++) {
1698
                        tmp0 = tab0[j];
1699
                        tmp1 = tab1[j];
1700
                        tab0[j] = MULL(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1701
                        tab1[j] = MULL(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1702
                    }
1703
                }
1704
            }
1705
        }
1706
    } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1707
        /* ms stereo ONLY */
1708
        /* NOTE: the 1/sqrt(2) normalization factor is included in the
1709
           global gain */
1710
        tab0 = g0->sb_hybrid;
1711
        tab1 = g1->sb_hybrid;
1712
        for(i=0;i<576;i++) {
1713
            tmp0 = tab0[i];
1714
            tmp1 = tab1[i];
1715
            tab0[i] = tmp0 + tmp1;
1716
            tab1[i] = tmp0 - tmp1;
1717
        }
1718
    }
1719
}
1720

    
1721
static void compute_antialias_integer(MPADecodeContext *s,
1722
                              GranuleDef *g)
1723
{
1724
    int32_t *ptr, *csa;
1725
    int n, i;
1726

    
1727
    /* we antialias only "long" bands */
1728
    if (g->block_type == 2) {
1729
        if (!g->switch_point)
1730
            return;
1731
        /* XXX: check this for 8000Hz case */
1732
        n = 1;
1733
    } else {
1734
        n = SBLIMIT - 1;
1735
    }
1736

    
1737
    ptr = g->sb_hybrid + 18;
1738
    for(i = n;i > 0;i--) {
1739
        int tmp0, tmp1, tmp2;
1740
        csa = &csa_table[0][0];
1741
#define INT_AA(j) \
1742
            tmp0 = ptr[-1-j];\
1743
            tmp1 = ptr[   j];\
1744
            tmp2= MULH(tmp0 + tmp1, csa[0+4*j]);\
1745
            ptr[-1-j] = 4*(tmp2 - MULH(tmp1, csa[2+4*j]));\
1746
            ptr[   j] = 4*(tmp2 + MULH(tmp0, csa[3+4*j]));
1747

    
1748
        INT_AA(0)
1749
        INT_AA(1)
1750
        INT_AA(2)
1751
        INT_AA(3)
1752
        INT_AA(4)
1753
        INT_AA(5)
1754
        INT_AA(6)
1755
        INT_AA(7)
1756

    
1757
        ptr += 18;
1758
    }
1759
}
1760

    
1761
static void compute_antialias_float(MPADecodeContext *s,
1762
                              GranuleDef *g)
1763
{
1764
    int32_t *ptr;
1765
    int n, i;
1766

    
1767
    /* we antialias only "long" bands */
1768
    if (g->block_type == 2) {
1769
        if (!g->switch_point)
1770
            return;
1771
        /* XXX: check this for 8000Hz case */
1772
        n = 1;
1773
    } else {
1774
        n = SBLIMIT - 1;
1775
    }
1776

    
1777
    ptr = g->sb_hybrid + 18;
1778
    for(i = n;i > 0;i--) {
1779
        float tmp0, tmp1;
1780
        float *csa = &csa_table_float[0][0];
1781
#define FLOAT_AA(j)\
1782
        tmp0= ptr[-1-j];\
1783
        tmp1= ptr[   j];\
1784
        ptr[-1-j] = lrintf(tmp0 * csa[0+4*j] - tmp1 * csa[1+4*j]);\
1785
        ptr[   j] = lrintf(tmp0 * csa[1+4*j] + tmp1 * csa[0+4*j]);
1786

    
1787
        FLOAT_AA(0)
1788
        FLOAT_AA(1)
1789
        FLOAT_AA(2)
1790
        FLOAT_AA(3)
1791
        FLOAT_AA(4)
1792
        FLOAT_AA(5)
1793
        FLOAT_AA(6)
1794
        FLOAT_AA(7)
1795

    
1796
        ptr += 18;
1797
    }
1798
}
1799

    
1800
static void compute_imdct(MPADecodeContext *s,
1801
                          GranuleDef *g,
1802
                          int32_t *sb_samples,
1803
                          int32_t *mdct_buf)
1804
{
1805
    int32_t *ptr, *win, *win1, *buf, *out_ptr, *ptr1;
1806
    int32_t out2[12];
1807
    int i, j, mdct_long_end, v, sblimit;
1808

    
1809
    /* find last non zero block */
1810
    ptr = g->sb_hybrid + 576;
1811
    ptr1 = g->sb_hybrid + 2 * 18;
1812
    while (ptr >= ptr1) {
1813
        ptr -= 6;
1814
        v = ptr[0] | ptr[1] | ptr[2] | ptr[3] | ptr[4] | ptr[5];
1815
        if (v != 0)
1816
            break;
1817
    }
1818
    sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
1819

    
1820
    if (g->block_type == 2) {
1821
        /* XXX: check for 8000 Hz */
1822
        if (g->switch_point)
1823
            mdct_long_end = 2;
1824
        else
1825
            mdct_long_end = 0;
1826
    } else {
1827
        mdct_long_end = sblimit;
1828
    }
1829

    
1830
    buf = mdct_buf;
1831
    ptr = g->sb_hybrid;
1832
    for(j=0;j<mdct_long_end;j++) {
1833
        /* apply window & overlap with previous buffer */
1834
        out_ptr = sb_samples + j;
1835
        /* select window */
1836
        if (g->switch_point && j < 2)
1837
            win1 = mdct_win[0];
1838
        else
1839
            win1 = mdct_win[g->block_type];
1840
        /* select frequency inversion */
1841
        win = win1 + ((4 * 36) & -(j & 1));
1842
        imdct36(out_ptr, buf, ptr, win);
1843
        out_ptr += 18*SBLIMIT;
1844
        ptr += 18;
1845
        buf += 18;
1846
    }
1847
    for(j=mdct_long_end;j<sblimit;j++) {
1848
        /* select frequency inversion */
1849
        win = mdct_win[2] + ((4 * 36) & -(j & 1));
1850
        out_ptr = sb_samples + j;
1851

    
1852
        for(i=0; i<6; i++){
1853
            *out_ptr = buf[i];
1854
            out_ptr += SBLIMIT;
1855
        }
1856
        imdct12(out2, ptr + 0);
1857
        for(i=0;i<6;i++) {
1858
            *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*1];
1859
            buf[i + 6*2] = MULH(out2[i + 6], win[i + 6]);
1860
            out_ptr += SBLIMIT;
1861
        }
1862
        imdct12(out2, ptr + 1);
1863
        for(i=0;i<6;i++) {
1864
            *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*2];
1865
            buf[i + 6*0] = MULH(out2[i + 6], win[i + 6]);
1866
            out_ptr += SBLIMIT;
1867
        }
1868
        imdct12(out2, ptr + 2);
1869
        for(i=0;i<6;i++) {
1870
            buf[i + 6*0] = MULH(out2[i], win[i]) + buf[i + 6*0];
1871
            buf[i + 6*1] = MULH(out2[i + 6], win[i + 6]);
1872
            buf[i + 6*2] = 0;
1873
        }
1874
        ptr += 18;
1875
        buf += 18;
1876
    }
1877
    /* zero bands */
1878
    for(j=sblimit;j<SBLIMIT;j++) {
1879
        /* overlap */
1880
        out_ptr = sb_samples + j;
1881
        for(i=0;i<18;i++) {
1882
            *out_ptr = buf[i];
1883
            buf[i] = 0;
1884
            out_ptr += SBLIMIT;
1885
        }
1886
        buf += 18;
1887
    }
1888
}
1889

    
1890
/* main layer3 decoding function */
1891
static int mp_decode_layer3(MPADecodeContext *s)
1892
{
1893
    int nb_granules, main_data_begin, private_bits;
1894
    int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;
1895
    GranuleDef *g;
1896
    int16_t exponents[576];
1897

    
1898
    /* read side info */
1899
    if (s->lsf) {
1900
        main_data_begin = get_bits(&s->gb, 8);
1901
        private_bits = get_bits(&s->gb, s->nb_channels);
1902
        nb_granules = 1;
1903
    } else {
1904
        main_data_begin = get_bits(&s->gb, 9);
1905
        if (s->nb_channels == 2)
1906
            private_bits = get_bits(&s->gb, 3);
1907
        else
1908
            private_bits = get_bits(&s->gb, 5);
1909
        nb_granules = 2;
1910
        for(ch=0;ch<s->nb_channels;ch++) {
1911
            s->granules[ch][0].scfsi = 0;/* all scale factors are transmitted */
1912
            s->granules[ch][1].scfsi = get_bits(&s->gb, 4);
1913
        }
1914
    }
1915

    
1916
    for(gr=0;gr<nb_granules;gr++) {
1917
        for(ch=0;ch<s->nb_channels;ch++) {
1918
            dprintf(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);
1919
            g = &s->granules[ch][gr];
1920
            g->part2_3_length = get_bits(&s->gb, 12);
1921
            g->big_values = get_bits(&s->gb, 9);
1922
            if(g->big_values > 288){
1923
                av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");
1924
                return -1;
1925
            }
1926

    
1927
            g->global_gain = get_bits(&s->gb, 8);
1928
            /* if MS stereo only is selected, we precompute the
1929
               1/sqrt(2) renormalization factor */
1930
            if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
1931
                MODE_EXT_MS_STEREO)
1932
                g->global_gain -= 2;
1933
            if (s->lsf)
1934
                g->scalefac_compress = get_bits(&s->gb, 9);
1935
            else
1936
                g->scalefac_compress = get_bits(&s->gb, 4);
1937
            blocksplit_flag = get_bits1(&s->gb);
1938
            if (blocksplit_flag) {
1939
                g->block_type = get_bits(&s->gb, 2);
1940
                if (g->block_type == 0){
1941
                    av_log(s->avctx, AV_LOG_ERROR, "invalid block type\n");
1942
                    return -1;
1943
                }
1944
                g->switch_point = get_bits1(&s->gb);
1945
                for(i=0;i<2;i++)
1946
                    g->table_select[i] = get_bits(&s->gb, 5);
1947
                for(i=0;i<3;i++)
1948
                    g->subblock_gain[i] = get_bits(&s->gb, 3);
1949
                ff_init_short_region(s, g);
1950
            } else {
1951
                int region_address1, region_address2;
1952
                g->block_type = 0;
1953
                g->switch_point = 0;
1954
                for(i=0;i<3;i++)
1955
                    g->table_select[i] = get_bits(&s->gb, 5);
1956
                /* compute huffman coded region sizes */
1957
                region_address1 = get_bits(&s->gb, 4);
1958
                region_address2 = get_bits(&s->gb, 3);
1959
                dprintf(s->avctx, "region1=%d region2=%d\n",
1960
                        region_address1, region_address2);
1961
                ff_init_long_region(s, g, region_address1, region_address2);
1962
            }
1963
            ff_region_offset2size(g);
1964
            ff_compute_band_indexes(s, g);
1965

    
1966
            g->preflag = 0;
1967
            if (!s->lsf)
1968
                g->preflag = get_bits1(&s->gb);
1969
            g->scalefac_scale = get_bits1(&s->gb);
1970
            g->count1table_select = get_bits1(&s->gb);
1971
            dprintf(s->avctx, "block_type=%d switch_point=%d\n",
1972
                    g->block_type, g->switch_point);
1973
        }
1974
    }
1975

    
1976
  if (!s->adu_mode) {
1977
    const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);
1978
    assert((get_bits_count(&s->gb) & 7) == 0);
1979
    /* now we get bits from the main_data_begin offset */
1980
    dprintf(s->avctx, "seekback: %d\n", main_data_begin);
1981
//av_log(NULL, AV_LOG_ERROR, "backstep:%d, lastbuf:%d\n", main_data_begin, s->last_buf_size);
1982

    
1983
    memcpy(s->last_buf + s->last_buf_size, ptr, EXTRABYTES);
1984
    s->in_gb= s->gb;
1985
        init_get_bits(&s->gb, s->last_buf, s->last_buf_size*8);
1986
        skip_bits_long(&s->gb, 8*(s->last_buf_size - main_data_begin));
1987
  }
1988

    
1989
    for(gr=0;gr<nb_granules;gr++) {
1990
        for(ch=0;ch<s->nb_channels;ch++) {
1991
            g = &s->granules[ch][gr];
1992
            if(get_bits_count(&s->gb)<0){
1993
                av_log(s->avctx, AV_LOG_DEBUG, "mdb:%d, lastbuf:%d skipping granule %d\n",
1994
                                            main_data_begin, s->last_buf_size, gr);
1995
                skip_bits_long(&s->gb, g->part2_3_length);
1996
                memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));
1997
                if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->in_gb.buffer){
1998
                    skip_bits_long(&s->in_gb, get_bits_count(&s->gb) - s->gb.size_in_bits);
1999
                    s->gb= s->in_gb;
2000
                    s->in_gb.buffer=NULL;
2001
                }
2002
                continue;
2003
            }
2004

    
2005
            bits_pos = get_bits_count(&s->gb);
2006

    
2007
            if (!s->lsf) {
2008
                uint8_t *sc;
2009
                int slen, slen1, slen2;
2010

    
2011
                /* MPEG1 scale factors */
2012
                slen1 = slen_table[0][g->scalefac_compress];
2013
                slen2 = slen_table[1][g->scalefac_compress];
2014
                dprintf(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);
2015
                if (g->block_type == 2) {
2016
                    n = g->switch_point ? 17 : 18;
2017
                    j = 0;
2018
                    if(slen1){
2019
                        for(i=0;i<n;i++)
2020
                            g->scale_factors[j++] = get_bits(&s->gb, slen1);
2021
                    }else{
2022
                        for(i=0;i<n;i++)
2023
                            g->scale_factors[j++] = 0;
2024
                    }
2025
                    if(slen2){
2026
                        for(i=0;i<18;i++)
2027
                            g->scale_factors[j++] = get_bits(&s->gb, slen2);
2028
                        for(i=0;i<3;i++)
2029
                            g->scale_factors[j++] = 0;
2030
                    }else{
2031
                        for(i=0;i<21;i++)
2032
                            g->scale_factors[j++] = 0;
2033
                    }
2034
                } else {
2035
                    sc = s->granules[ch][0].scale_factors;
2036
                    j = 0;
2037
                    for(k=0;k<4;k++) {
2038
                        n = (k == 0 ? 6 : 5);
2039
                        if ((g->scfsi & (0x8 >> k)) == 0) {
2040
                            slen = (k < 2) ? slen1 : slen2;
2041
                            if(slen){
2042
                                for(i=0;i<n;i++)
2043
                                    g->scale_factors[j++] = get_bits(&s->gb, slen);
2044
                            }else{
2045
                                for(i=0;i<n;i++)
2046
                                    g->scale_factors[j++] = 0;
2047
                            }
2048
                        } else {
2049
                            /* simply copy from last granule */
2050
                            for(i=0;i<n;i++) {
2051
                                g->scale_factors[j] = sc[j];
2052
                                j++;
2053
                            }
2054
                        }
2055
                    }
2056
                    g->scale_factors[j++] = 0;
2057
                }
2058
            } else {
2059
                int tindex, tindex2, slen[4], sl, sf;
2060

    
2061
                /* LSF scale factors */
2062
                if (g->block_type == 2) {
2063
                    tindex = g->switch_point ? 2 : 1;
2064
                } else {
2065
                    tindex = 0;
2066
                }
2067
                sf = g->scalefac_compress;
2068
                if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
2069
                    /* intensity stereo case */
2070
                    sf >>= 1;
2071
                    if (sf < 180) {
2072
                        lsf_sf_expand(slen, sf, 6, 6, 0);
2073
                        tindex2 = 3;
2074
                    } else if (sf < 244) {
2075
                        lsf_sf_expand(slen, sf - 180, 4, 4, 0);
2076
                        tindex2 = 4;
2077
                    } else {
2078
                        lsf_sf_expand(slen, sf - 244, 3, 0, 0);
2079
                        tindex2 = 5;
2080
                    }
2081
                } else {
2082
                    /* normal case */
2083
                    if (sf < 400) {
2084
                        lsf_sf_expand(slen, sf, 5, 4, 4);
2085
                        tindex2 = 0;
2086
                    } else if (sf < 500) {
2087
                        lsf_sf_expand(slen, sf - 400, 5, 4, 0);
2088
                        tindex2 = 1;
2089
                    } else {
2090
                        lsf_sf_expand(slen, sf - 500, 3, 0, 0);
2091
                        tindex2 = 2;
2092
                        g->preflag = 1;
2093
                    }
2094
                }
2095

    
2096
                j = 0;
2097
                for(k=0;k<4;k++) {
2098
                    n = lsf_nsf_table[tindex2][tindex][k];
2099
                    sl = slen[k];
2100
                    if(sl){
2101
                        for(i=0;i<n;i++)
2102
                            g->scale_factors[j++] = get_bits(&s->gb, sl);
2103
                    }else{
2104
                        for(i=0;i<n;i++)
2105
                            g->scale_factors[j++] = 0;
2106
                    }
2107
                }
2108
                /* XXX: should compute exact size */
2109
                for(;j<40;j++)
2110
                    g->scale_factors[j] = 0;
2111
            }
2112

    
2113
            exponents_from_scale_factors(s, g, exponents);
2114

    
2115
            /* read Huffman coded residue */
2116
            huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);
2117
        } /* ch */
2118

    
2119
        if (s->nb_channels == 2)
2120
            compute_stereo(s, &s->granules[0][gr], &s->granules[1][gr]);
2121

    
2122
        for(ch=0;ch<s->nb_channels;ch++) {
2123
            g = &s->granules[ch][gr];
2124

    
2125
            reorder_block(s, g);
2126
            s->compute_antialias(s, g);
2127
            compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
2128
        }
2129
    } /* gr */
2130
    if(get_bits_count(&s->gb)<0)
2131
        skip_bits_long(&s->gb, -get_bits_count(&s->gb));
2132
    return nb_granules * 18;
2133
}
2134

    
2135
static int mp_decode_frame(MPADecodeContext *s,
2136
                           OUT_INT *samples, const uint8_t *buf, int buf_size)
2137
{
2138
    int i, nb_frames, ch;
2139
    OUT_INT *samples_ptr;
2140

    
2141
    init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE)*8);
2142

    
2143
    /* skip error protection field */
2144
    if (s->error_protection)
2145
        skip_bits(&s->gb, 16);
2146

    
2147
    dprintf(s->avctx, "frame %d:\n", s->frame_count);
2148
    switch(s->layer) {
2149
    case 1:
2150
        s->avctx->frame_size = 384;
2151
        nb_frames = mp_decode_layer1(s);
2152
        break;
2153
    case 2:
2154
        s->avctx->frame_size = 1152;
2155
        nb_frames = mp_decode_layer2(s);
2156
        break;
2157
    case 3:
2158
        s->avctx->frame_size = s->lsf ? 576 : 1152;
2159
    default:
2160
        nb_frames = mp_decode_layer3(s);
2161

    
2162
        s->last_buf_size=0;
2163
        if(s->in_gb.buffer){
2164
            align_get_bits(&s->gb);
2165
            i= get_bits_left(&s->gb)>>3;
2166
            if(i >= 0 && i <= BACKSTEP_SIZE){
2167
                memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);
2168
                s->last_buf_size=i;
2169
            }else
2170
                av_log(s->avctx, AV_LOG_ERROR, "invalid old backstep %d\n", i);
2171
            s->gb= s->in_gb;
2172
            s->in_gb.buffer= NULL;
2173
        }
2174

    
2175
        align_get_bits(&s->gb);
2176
        assert((get_bits_count(&s->gb) & 7) == 0);
2177
        i= get_bits_left(&s->gb)>>3;
2178

    
2179
        if(i<0 || i > BACKSTEP_SIZE || nb_frames<0){
2180
            if(i<0)
2181
                av_log(s->avctx, AV_LOG_ERROR, "invalid new backstep %d\n", i);
2182
            i= FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);
2183
        }
2184
        assert(i <= buf_size - HEADER_SIZE && i>= 0);
2185
        memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);
2186
        s->last_buf_size += i;
2187

    
2188
        break;
2189
    }
2190

    
2191
    /* apply the synthesis filter */
2192
    for(ch=0;ch<s->nb_channels;ch++) {
2193
        samples_ptr = samples + ch;
2194
        for(i=0;i<nb_frames;i++) {
2195
            ff_mpa_synth_filter(s->synth_buf[ch], &(s->synth_buf_offset[ch]),
2196
                         ff_mpa_synth_window, &s->dither_state,
2197
                         samples_ptr, s->nb_channels,
2198
                         s->sb_samples[ch][i]);
2199
            samples_ptr += 32 * s->nb_channels;
2200
        }
2201
    }
2202

    
2203
    return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
2204
}
2205

    
2206
static int decode_frame(AVCodecContext * avctx,
2207
                        void *data, int *data_size,
2208
                        AVPacket *avpkt)
2209
{
2210
    const uint8_t *buf = avpkt->data;
2211
    int buf_size = avpkt->size;
2212
    MPADecodeContext *s = avctx->priv_data;
2213
    uint32_t header;
2214
    int out_size;
2215
    OUT_INT *out_samples = data;
2216

    
2217
    if(buf_size < HEADER_SIZE)
2218
        return -1;
2219

    
2220
    header = AV_RB32(buf);
2221
    if(ff_mpa_check_header(header) < 0){
2222
        av_log(avctx, AV_LOG_ERROR, "Header missing\n");
2223
        return -1;
2224
    }
2225

    
2226
    if (ff_mpegaudio_decode_header((MPADecodeHeader *)s, header) == 1) {
2227
        /* free format: prepare to compute frame size */
2228
        s->frame_size = -1;
2229
        return -1;
2230
    }
2231
    /* update codec info */
2232
    avctx->channels = s->nb_channels;
2233
    avctx->bit_rate = s->bit_rate;
2234
    avctx->sub_id = s->layer;
2235

    
2236
    if(*data_size < 1152*avctx->channels*sizeof(OUT_INT))
2237
        return -1;
2238
    *data_size = 0;
2239

    
2240
    if(s->frame_size<=0 || s->frame_size > buf_size){
2241
        av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
2242
        return -1;
2243
    }else if(s->frame_size < buf_size){
2244
        av_log(avctx, AV_LOG_ERROR, "incorrect frame size\n");
2245
        buf_size= s->frame_size;
2246
    }
2247

    
2248
    out_size = mp_decode_frame(s, out_samples, buf, buf_size);
2249
    if(out_size>=0){
2250
        *data_size = out_size;
2251
        avctx->sample_rate = s->sample_rate;
2252
        //FIXME maybe move the other codec info stuff from above here too
2253
    }else
2254
        av_log(avctx, AV_LOG_DEBUG, "Error while decoding MPEG audio frame.\n"); //FIXME return -1 / but also return the number of bytes consumed
2255
    s->frame_size = 0;
2256
    return buf_size;
2257
}
2258

    
2259
static void flush(AVCodecContext *avctx){
2260
    MPADecodeContext *s = avctx->priv_data;
2261
    memset(s->synth_buf, 0, sizeof(s->synth_buf));
2262
    s->last_buf_size= 0;
2263
}
2264

    
2265
#if CONFIG_MP3ADU_DECODER
2266
static int decode_frame_adu(AVCodecContext * avctx,
2267
                        void *data, int *data_size,
2268
                        AVPacket *avpkt)
2269
{
2270
    const uint8_t *buf = avpkt->data;
2271
    int buf_size = avpkt->size;
2272
    MPADecodeContext *s = avctx->priv_data;
2273
    uint32_t header;
2274
    int len, out_size;
2275
    OUT_INT *out_samples = data;
2276

    
2277
    len = buf_size;
2278

    
2279
    // Discard too short frames
2280
    if (buf_size < HEADER_SIZE) {
2281
        *data_size = 0;
2282
        return buf_size;
2283
    }
2284

    
2285

    
2286
    if (len > MPA_MAX_CODED_FRAME_SIZE)
2287
        len = MPA_MAX_CODED_FRAME_SIZE;
2288

    
2289
    // Get header and restore sync word
2290
    header = AV_RB32(buf) | 0xffe00000;
2291

    
2292
    if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame
2293
        *data_size = 0;
2294
        return buf_size;
2295
    }
2296

    
2297
    ff_mpegaudio_decode_header((MPADecodeHeader *)s, header);
2298
    /* update codec info */
2299
    avctx->sample_rate = s->sample_rate;
2300
    avctx->channels = s->nb_channels;
2301
    avctx->bit_rate = s->bit_rate;
2302
    avctx->sub_id = s->layer;
2303

    
2304
    s->frame_size = len;
2305

    
2306
    if (avctx->parse_only) {
2307
        out_size = buf_size;
2308
    } else {
2309
        out_size = mp_decode_frame(s, out_samples, buf, buf_size);
2310
    }
2311

    
2312
    *data_size = out_size;
2313
    return buf_size;
2314
}
2315
#endif /* CONFIG_MP3ADU_DECODER */
2316

    
2317
#if CONFIG_MP3ON4_DECODER
2318

    
2319
/**
2320
 * Context for MP3On4 decoder
2321
 */
2322
typedef struct MP3On4DecodeContext {
2323
    int frames;   ///< number of mp3 frames per block (number of mp3 decoder instances)
2324
    int syncword; ///< syncword patch
2325
    const uint8_t *coff; ///< channels offsets in output buffer
2326
    MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance
2327
} MP3On4DecodeContext;
2328

    
2329
#include "mpeg4audio.h"
2330

    
2331
/* Next 3 arrays are indexed by channel config number (passed via codecdata) */
2332
static const uint8_t mp3Frames[8] = {0,1,1,2,3,3,4,5};   /* number of mp3 decoder instances */
2333
/* offsets into output buffer, assume output order is FL FR BL BR C LFE */
2334
static const uint8_t chan_offset[8][5] = {
2335
    {0},
2336
    {0},            // C
2337
    {0},            // FLR
2338
    {2,0},          // C FLR
2339
    {2,0,3},        // C FLR BS
2340
    {4,0,2},        // C FLR BLRS
2341
    {4,0,2,5},      // C FLR BLRS LFE
2342
    {4,0,2,6,5},    // C FLR BLRS BLR LFE
2343
};
2344

    
2345

    
2346
static int decode_init_mp3on4(AVCodecContext * avctx)
2347
{
2348
    MP3On4DecodeContext *s = avctx->priv_data;
2349
    MPEG4AudioConfig cfg;
2350
    int i;
2351

    
2352
    if ((avctx->extradata_size < 2) || (avctx->extradata == NULL)) {
2353
        av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
2354
        return -1;
2355
    }
2356

    
2357
    ff_mpeg4audio_get_config(&cfg, avctx->extradata, avctx->extradata_size);
2358
    if (!cfg.chan_config || cfg.chan_config > 7) {
2359
        av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
2360
        return -1;
2361
    }
2362
    s->frames = mp3Frames[cfg.chan_config];
2363
    s->coff = chan_offset[cfg.chan_config];
2364
    avctx->channels = ff_mpeg4audio_channels[cfg.chan_config];
2365

    
2366
    if (cfg.sample_rate < 16000)
2367
        s->syncword = 0xffe00000;
2368
    else
2369
        s->syncword = 0xfff00000;
2370

    
2371
    /* Init the first mp3 decoder in standard way, so that all tables get builded
2372
     * We replace avctx->priv_data with the context of the first decoder so that
2373
     * decode_init() does not have to be changed.
2374
     * Other decoders will be initialized here copying data from the first context
2375
     */
2376
    // Allocate zeroed memory for the first decoder context
2377
    s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
2378
    // Put decoder context in place to make init_decode() happy
2379
    avctx->priv_data = s->mp3decctx[0];
2380
    decode_init(avctx);
2381
    // Restore mp3on4 context pointer
2382
    avctx->priv_data = s;
2383
    s->mp3decctx[0]->adu_mode = 1; // Set adu mode
2384

    
2385
    /* Create a separate codec/context for each frame (first is already ok).
2386
     * Each frame is 1 or 2 channels - up to 5 frames allowed
2387
     */
2388
    for (i = 1; i < s->frames; i++) {
2389
        s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
2390
        s->mp3decctx[i]->compute_antialias = s->mp3decctx[0]->compute_antialias;
2391
        s->mp3decctx[i]->adu_mode = 1;
2392
        s->mp3decctx[i]->avctx = avctx;
2393
    }
2394

    
2395
    return 0;
2396
}
2397

    
2398

    
2399
static av_cold int decode_close_mp3on4(AVCodecContext * avctx)
2400
{
2401
    MP3On4DecodeContext *s = avctx->priv_data;
2402
    int i;
2403

    
2404
    for (i = 0; i < s->frames; i++)
2405
        if (s->mp3decctx[i])
2406
            av_free(s->mp3decctx[i]);
2407

    
2408
    return 0;
2409
}
2410

    
2411

    
2412
static int decode_frame_mp3on4(AVCodecContext * avctx,
2413
                        void *data, int *data_size,
2414
                        AVPacket *avpkt)
2415
{
2416
    const uint8_t *buf = avpkt->data;
2417
    int buf_size = avpkt->size;
2418
    MP3On4DecodeContext *s = avctx->priv_data;
2419
    MPADecodeContext *m;
2420
    int fsize, len = buf_size, out_size = 0;
2421
    uint32_t header;
2422
    OUT_INT *out_samples = data;
2423
    OUT_INT decoded_buf[MPA_FRAME_SIZE * MPA_MAX_CHANNELS];
2424
    OUT_INT *outptr, *bp;
2425
    int fr, j, n;
2426

    
2427
    if(*data_size < MPA_FRAME_SIZE * MPA_MAX_CHANNELS * s->frames * sizeof(OUT_INT))
2428
        return -1;
2429

    
2430
    *data_size = 0;
2431
    // Discard too short frames
2432
    if (buf_size < HEADER_SIZE)
2433
        return -1;
2434

    
2435
    // If only one decoder interleave is not needed
2436
    outptr = s->frames == 1 ? out_samples : decoded_buf;
2437

    
2438
    avctx->bit_rate = 0;
2439

    
2440
    for (fr = 0; fr < s->frames; fr++) {
2441
        fsize = AV_RB16(buf) >> 4;
2442
        fsize = FFMIN3(fsize, len, MPA_MAX_CODED_FRAME_SIZE);
2443
        m = s->mp3decctx[fr];
2444
        assert (m != NULL);
2445

    
2446
        header = (AV_RB32(buf) & 0x000fffff) | s->syncword; // patch header
2447

    
2448
        if (ff_mpa_check_header(header) < 0) // Bad header, discard block
2449
            break;
2450

    
2451
        ff_mpegaudio_decode_header((MPADecodeHeader *)m, header);
2452
        out_size += mp_decode_frame(m, outptr, buf, fsize);
2453
        buf += fsize;
2454
        len -= fsize;
2455

    
2456
        if(s->frames > 1) {
2457
            n = m->avctx->frame_size*m->nb_channels;
2458
            /* interleave output data */
2459
            bp = out_samples + s->coff[fr];
2460
            if(m->nb_channels == 1) {
2461
                for(j = 0; j < n; j++) {
2462
                    *bp = decoded_buf[j];
2463
                    bp += avctx->channels;
2464
                }
2465
            } else {
2466
                for(j = 0; j < n; j++) {
2467
                    bp[0] = decoded_buf[j++];
2468
                    bp[1] = decoded_buf[j];
2469
                    bp += avctx->channels;
2470
                }
2471
            }
2472
        }
2473
        avctx->bit_rate += m->bit_rate;
2474
    }
2475

    
2476
    /* update codec info */
2477
    avctx->sample_rate = s->mp3decctx[0]->sample_rate;
2478

    
2479
    *data_size = out_size;
2480
    return buf_size;
2481
}
2482
#endif /* CONFIG_MP3ON4_DECODER */
2483

    
2484
#if CONFIG_MP1_DECODER
2485
AVCodec mp1_decoder =
2486
{
2487
    "mp1",
2488
    CODEC_TYPE_AUDIO,
2489
    CODEC_ID_MP1,
2490
    sizeof(MPADecodeContext),
2491
    decode_init,
2492
    NULL,
2493
    NULL,
2494
    decode_frame,
2495
    CODEC_CAP_PARSE_ONLY,
2496
    .flush= flush,
2497
    .long_name= NULL_IF_CONFIG_SMALL("MP1 (MPEG audio layer 1)"),
2498
};
2499
#endif
2500
#if CONFIG_MP2_DECODER
2501
AVCodec mp2_decoder =
2502
{
2503
    "mp2",
2504
    CODEC_TYPE_AUDIO,
2505
    CODEC_ID_MP2,
2506
    sizeof(MPADecodeContext),
2507
    decode_init,
2508
    NULL,
2509
    NULL,
2510
    decode_frame,
2511
    CODEC_CAP_PARSE_ONLY,
2512
    .flush= flush,
2513
    .long_name= NULL_IF_CONFIG_SMALL("MP2 (MPEG audio layer 2)"),
2514
};
2515
#endif
2516
#if CONFIG_MP3_DECODER
2517
AVCodec mp3_decoder =
2518
{
2519
    "mp3",
2520
    CODEC_TYPE_AUDIO,
2521
    CODEC_ID_MP3,
2522
    sizeof(MPADecodeContext),
2523
    decode_init,
2524
    NULL,
2525
    NULL,
2526
    decode_frame,
2527
    CODEC_CAP_PARSE_ONLY,
2528
    .flush= flush,
2529
    .long_name= NULL_IF_CONFIG_SMALL("MP3 (MPEG audio layer 3)"),
2530
};
2531
#endif
2532
#if CONFIG_MP3ADU_DECODER
2533
AVCodec mp3adu_decoder =
2534
{
2535
    "mp3adu",
2536
    CODEC_TYPE_AUDIO,
2537
    CODEC_ID_MP3ADU,
2538
    sizeof(MPADecodeContext),
2539
    decode_init,
2540
    NULL,
2541
    NULL,
2542
    decode_frame_adu,
2543
    CODEC_CAP_PARSE_ONLY,
2544
    .flush= flush,
2545
    .long_name= NULL_IF_CONFIG_SMALL("ADU (Application Data Unit) MP3 (MPEG audio layer 3)"),
2546
};
2547
#endif
2548
#if CONFIG_MP3ON4_DECODER
2549
AVCodec mp3on4_decoder =
2550
{
2551
    "mp3on4",
2552
    CODEC_TYPE_AUDIO,
2553
    CODEC_ID_MP3ON4,
2554
    sizeof(MP3On4DecodeContext),
2555
    decode_init_mp3on4,
2556
    NULL,
2557
    decode_close_mp3on4,
2558
    decode_frame_mp3on4,
2559
    .flush= flush,
2560
    .long_name= NULL_IF_CONFIG_SMALL("MP3onMP4"),
2561
};
2562
#endif