Statistics
| Branch: | Revision:

ffmpeg / libavcodec / mpegaudiodec.c @ 5509bffa

History | View | Annotate | Download (85.4 KB)

1
/*
2
 * MPEG Audio decoder
3
 * Copyright (c) 2001, 2002 Fabrice Bellard.
4
 *
5
 * This library is free software; you can redistribute it and/or
6
 * modify it under the terms of the GNU Lesser General Public
7
 * License as published by the Free Software Foundation; either
8
 * version 2 of the License, or (at your option) any later version.
9
 *
10
 * This library is distributed in the hope that it will be useful,
11
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13
 * Lesser General Public License for more details.
14
 *
15
 * You should have received a copy of the GNU Lesser General Public
16
 * License along with this library; if not, write to the Free Software
17
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
18
 */
19

    
20
/**
21
 * @file mpegaudiodec.c
22
 * MPEG Audio decoder.
23
 */
24

    
25
//#define DEBUG
26
#include "avcodec.h"
27
#include "bitstream.h"
28
#include "dsputil.h"
29

    
30
/*
31
 * TODO:
32
 *  - in low precision mode, use more 16 bit multiplies in synth filter
33
 *  - test lsf / mpeg25 extensively.
34
 */
35

    
36
/* define USE_HIGHPRECISION to have a bit exact (but slower) mpeg
37
   audio decoder */
38
#ifdef CONFIG_MPEGAUDIO_HP
39
#define USE_HIGHPRECISION
40
#endif
41

    
42
#include "mpegaudio.h"
43

    
44
#define FRAC_ONE    (1 << FRAC_BITS)
45

    
46
#define MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> FRAC_BITS)
47
#define MUL64(a,b) ((int64_t)(a) * (int64_t)(b))
48
#define FIX(a)   ((int)((a) * FRAC_ONE))
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
//#define MULH(a,b) (((int64_t)(a) * (int64_t)(b))>>32) //gcc 3.4 creates an incredibly bloated mess out of this
55
static always_inline int MULH(int a, int b){
56
    return ((int64_t)(a) * (int64_t)(b))>>32;
57
}
58

    
59
/****************/
60

    
61
#define HEADER_SIZE 4
62
#define BACKSTEP_SIZE 512
63

    
64
struct GranuleDef;
65

    
66
typedef struct MPADecodeContext {
67
    uint8_t inbuf1[2][MPA_MAX_CODED_FRAME_SIZE + BACKSTEP_SIZE];        /* input buffer */
68
    int inbuf_index;
69
    uint8_t *inbuf_ptr, *inbuf;
70
    int frame_size;
71
    int free_format_frame_size; /* frame size in case of free format
72
                                   (zero if currently unknown) */
73
    /* next header (used in free format parsing) */
74
    uint32_t free_format_next_header;
75
    int error_protection;
76
    int layer;
77
    int sample_rate;
78
    int sample_rate_index; /* between 0 and 8 */
79
    int bit_rate;
80
    int old_frame_size;
81
    GetBitContext gb;
82
    int nb_channels;
83
    int mode;
84
    int mode_ext;
85
    int lsf;
86
    MPA_INT synth_buf[MPA_MAX_CHANNELS][512 * 2] __attribute__((aligned(16)));
87
    int synth_buf_offset[MPA_MAX_CHANNELS];
88
    int32_t sb_samples[MPA_MAX_CHANNELS][36][SBLIMIT] __attribute__((aligned(16)));
89
    int32_t mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */
90
#ifdef DEBUG
91
    int frame_count;
92
#endif
93
    void (*compute_antialias)(struct MPADecodeContext *s, struct GranuleDef *g);
94
    int adu_mode; ///< 0 for standard mp3, 1 for adu formatted mp3
95
    unsigned int dither_state;
96
} MPADecodeContext;
97

    
98
/**
99
 * Context for MP3On4 decoder
100
 */
101
typedef struct MP3On4DecodeContext {
102
    int frames;   ///< number of mp3 frames per block (number of mp3 decoder instances)
103
    int chan_cfg; ///< channel config number
104
    MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance
105
} MP3On4DecodeContext;
106

    
107
/* layer 3 "granule" */
108
typedef struct GranuleDef {
109
    uint8_t scfsi;
110
    int part2_3_length;
111
    int big_values;
112
    int global_gain;
113
    int scalefac_compress;
114
    uint8_t block_type;
115
    uint8_t switch_point;
116
    int table_select[3];
117
    int subblock_gain[3];
118
    uint8_t scalefac_scale;
119
    uint8_t count1table_select;
120
    int region_size[3]; /* number of huffman codes in each region */
121
    int preflag;
122
    int short_start, long_end; /* long/short band indexes */
123
    uint8_t scale_factors[40];
124
    int32_t sb_hybrid[SBLIMIT * 18]; /* 576 samples */
125
} GranuleDef;
126

    
127
#define MODE_EXT_MS_STEREO 2
128
#define MODE_EXT_I_STEREO  1
129

    
130
/* layer 3 huffman tables */
131
typedef struct HuffTable {
132
    int xsize;
133
    const uint8_t *bits;
134
    const uint16_t *codes;
135
} HuffTable;
136

    
137
#include "mpegaudiodectab.h"
138

    
139
static void compute_antialias_integer(MPADecodeContext *s, GranuleDef *g);
140
static void compute_antialias_float(MPADecodeContext *s, GranuleDef *g);
141

    
142
/* vlc structure for decoding layer 3 huffman tables */
143
static VLC huff_vlc[16];
144
static uint8_t *huff_code_table[16];
145
static VLC huff_quad_vlc[2];
146
/* computed from band_size_long */
147
static uint16_t band_index_long[9][23];
148
/* XXX: free when all decoders are closed */
149
#define TABLE_4_3_SIZE (8191 + 16)*4
150
static int8_t  *table_4_3_exp;
151
static uint32_t *table_4_3_value;
152
/* intensity stereo coef table */
153
static int32_t is_table[2][16];
154
static int32_t is_table_lsf[2][2][16];
155
static int32_t csa_table[8][4];
156
static float csa_table_float[8][4];
157
static int32_t mdct_win[8][36];
158

    
159
/* lower 2 bits: modulo 3, higher bits: shift */
160
static uint16_t scale_factor_modshift[64];
161
/* [i][j]:  2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
162
static int32_t scale_factor_mult[15][3];
163
/* mult table for layer 2 group quantization */
164

    
165
#define SCALE_GEN(v) \
166
{ FIXR(1.0 * (v)), FIXR(0.7937005259 * (v)), FIXR(0.6299605249 * (v)) }
167

    
168
static const int32_t scale_factor_mult2[3][3] = {
169
    SCALE_GEN(4.0 / 3.0), /* 3 steps */
170
    SCALE_GEN(4.0 / 5.0), /* 5 steps */
171
    SCALE_GEN(4.0 / 9.0), /* 9 steps */
172
};
173

    
174
void ff_mpa_synth_init(MPA_INT *window);
175
static MPA_INT window[512] __attribute__((aligned(16)));
176

    
177
/* layer 1 unscaling */
178
/* n = number of bits of the mantissa minus 1 */
179
static inline int l1_unscale(int n, int mant, int scale_factor)
180
{
181
    int shift, mod;
182
    int64_t val;
183

    
184
    shift = scale_factor_modshift[scale_factor];
185
    mod = shift & 3;
186
    shift >>= 2;
187
    val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);
188
    shift += n;
189
    /* NOTE: at this point, 1 <= shift >= 21 + 15 */
190
    return (int)((val + (1LL << (shift - 1))) >> shift);
191
}
192

    
193
static inline int l2_unscale_group(int steps, int mant, int scale_factor)
194
{
195
    int shift, mod, val;
196

    
197
    shift = scale_factor_modshift[scale_factor];
198
    mod = shift & 3;
199
    shift >>= 2;
200

    
201
    val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
202
    /* NOTE: at this point, 0 <= shift <= 21 */
203
    if (shift > 0)
204
        val = (val + (1 << (shift - 1))) >> shift;
205
    return val;
206
}
207

    
208
/* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
209
static inline int l3_unscale(int value, int exponent)
210
{
211
    unsigned int m;
212
    int e;
213

    
214
    e = table_4_3_exp  [4*value + (exponent&3)];
215
    m = table_4_3_value[4*value + (exponent&3)];
216
    e -= (exponent >> 2);
217
    assert(e>=1);
218
    if (e > 31)
219
        return 0;
220
    m = (m + (1 << (e-1))) >> e;
221

    
222
    return m;
223
}
224

    
225
/* all integer n^(4/3) computation code */
226
#define DEV_ORDER 13
227

    
228
#define POW_FRAC_BITS 24
229
#define POW_FRAC_ONE    (1 << POW_FRAC_BITS)
230
#define POW_FIX(a)   ((int)((a) * POW_FRAC_ONE))
231
#define POW_MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> POW_FRAC_BITS)
232

    
233
static int dev_4_3_coefs[DEV_ORDER];
234

    
235
#if 0 /* unused */
236
static int pow_mult3[3] = {
237
    POW_FIX(1.0),
238
    POW_FIX(1.25992104989487316476),
239
    POW_FIX(1.58740105196819947474),
240
};
241
#endif
242

    
243
static void int_pow_init(void)
244
{
245
    int i, a;
246

    
247
    a = POW_FIX(1.0);
248
    for(i=0;i<DEV_ORDER;i++) {
249
        a = POW_MULL(a, POW_FIX(4.0 / 3.0) - i * POW_FIX(1.0)) / (i + 1);
250
        dev_4_3_coefs[i] = a;
251
    }
252
}
253

    
254
#if 0 /* unused, remove? */
255
/* return the mantissa and the binary exponent */
256
static int int_pow(int i, int *exp_ptr)
257
{
258
    int e, er, eq, j;
259
    int a, a1;
260

261
    /* renormalize */
262
    a = i;
263
    e = POW_FRAC_BITS;
264
    while (a < (1 << (POW_FRAC_BITS - 1))) {
265
        a = a << 1;
266
        e--;
267
    }
268
    a -= (1 << POW_FRAC_BITS);
269
    a1 = 0;
270
    for(j = DEV_ORDER - 1; j >= 0; j--)
271
        a1 = POW_MULL(a, dev_4_3_coefs[j] + a1);
272
    a = (1 << POW_FRAC_BITS) + a1;
273
    /* exponent compute (exact) */
274
    e = e * 4;
275
    er = e % 3;
276
    eq = e / 3;
277
    a = POW_MULL(a, pow_mult3[er]);
278
    while (a >= 2 * POW_FRAC_ONE) {
279
        a = a >> 1;
280
        eq++;
281
    }
282
    /* convert to float */
283
    while (a < POW_FRAC_ONE) {
284
        a = a << 1;
285
        eq--;
286
    }
287
    /* now POW_FRAC_ONE <= a < 2 * POW_FRAC_ONE */
288
#if POW_FRAC_BITS > FRAC_BITS
289
    a = (a + (1 << (POW_FRAC_BITS - FRAC_BITS - 1))) >> (POW_FRAC_BITS - FRAC_BITS);
290
    /* correct overflow */
291
    if (a >= 2 * (1 << FRAC_BITS)) {
292
        a = a >> 1;
293
        eq++;
294
    }
295
#endif
296
    *exp_ptr = eq;
297
    return a;
298
}
299
#endif
300

    
301
static int decode_init(AVCodecContext * avctx)
302
{
303
    MPADecodeContext *s = avctx->priv_data;
304
    static int init=0;
305
    int i, j, k;
306

    
307
#if defined(USE_HIGHPRECISION) && defined(CONFIG_AUDIO_NONSHORT)
308
    avctx->sample_fmt= SAMPLE_FMT_S32;
309
#else
310
    avctx->sample_fmt= SAMPLE_FMT_S16;
311
#endif
312

    
313
    if(avctx->antialias_algo != FF_AA_FLOAT)
314
        s->compute_antialias= compute_antialias_integer;
315
    else
316
        s->compute_antialias= compute_antialias_float;
317

    
318
    if (!init && !avctx->parse_only) {
319
        /* scale factors table for layer 1/2 */
320
        for(i=0;i<64;i++) {
321
            int shift, mod;
322
            /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */
323
            shift = (i / 3);
324
            mod = i % 3;
325
            scale_factor_modshift[i] = mod | (shift << 2);
326
        }
327

    
328
        /* scale factor multiply for layer 1 */
329
        for(i=0;i<15;i++) {
330
            int n, norm;
331
            n = i + 2;
332
            norm = ((int64_t_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
333
            scale_factor_mult[i][0] = MULL(FIXR(1.0 * 2.0), norm);
334
            scale_factor_mult[i][1] = MULL(FIXR(0.7937005259 * 2.0), norm);
335
            scale_factor_mult[i][2] = MULL(FIXR(0.6299605249 * 2.0), norm);
336
            dprintf("%d: norm=%x s=%x %x %x\n",
337
                    i, norm,
338
                    scale_factor_mult[i][0],
339
                    scale_factor_mult[i][1],
340
                    scale_factor_mult[i][2]);
341
        }
342

    
343
        ff_mpa_synth_init(window);
344

    
345
        /* huffman decode tables */
346
        huff_code_table[0] = NULL;
347
        for(i=1;i<16;i++) {
348
            const HuffTable *h = &mpa_huff_tables[i];
349
            int xsize, x, y;
350
            unsigned int n;
351
            uint8_t *code_table;
352

    
353
            xsize = h->xsize;
354
            n = xsize * xsize;
355
            /* XXX: fail test */
356
            init_vlc(&huff_vlc[i], 8, n,
357
                     h->bits, 1, 1, h->codes, 2, 2, 1);
358

    
359
            code_table = av_mallocz(n);
360
            j = 0;
361
            for(x=0;x<xsize;x++) {
362
                for(y=0;y<xsize;y++)
363
                    code_table[j++] = (x << 4) | y;
364
            }
365
            huff_code_table[i] = code_table;
366
        }
367
        for(i=0;i<2;i++) {
368
            init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
369
                     mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1, 1);
370
        }
371

    
372
        for(i=0;i<9;i++) {
373
            k = 0;
374
            for(j=0;j<22;j++) {
375
                band_index_long[i][j] = k;
376
                k += band_size_long[i][j];
377
            }
378
            band_index_long[i][22] = k;
379
        }
380

    
381
        /* compute n ^ (4/3) and store it in mantissa/exp format */
382
        table_4_3_exp= av_mallocz_static(TABLE_4_3_SIZE * sizeof(table_4_3_exp[0]));
383
        if(!table_4_3_exp)
384
            return -1;
385
        table_4_3_value= av_mallocz_static(TABLE_4_3_SIZE * sizeof(table_4_3_value[0]));
386
        if(!table_4_3_value)
387
            return -1;
388

    
389
        int_pow_init();
390
        for(i=1;i<TABLE_4_3_SIZE;i++) {
391
            double f, fm;
392
            int e, m;
393
            f = pow((double)(i/4), 4.0 / 3.0) * pow(2, (i&3)*0.25);
394
            fm = frexp(f, &e);
395
            m = (uint32_t)(fm*(1LL<<31) + 0.5);
396
            e+= FRAC_BITS - 31 + 5;
397

    
398
            /* normalized to FRAC_BITS */
399
            table_4_3_value[i] = m;
400
//            av_log(NULL, AV_LOG_DEBUG, "%d %d %f\n", i, m, pow((double)i, 4.0 / 3.0));
401
            table_4_3_exp[i] = -e;
402
        }
403

    
404
        for(i=0;i<7;i++) {
405
            float f;
406
            int v;
407
            if (i != 6) {
408
                f = tan((double)i * M_PI / 12.0);
409
                v = FIXR(f / (1.0 + f));
410
            } else {
411
                v = FIXR(1.0);
412
            }
413
            is_table[0][i] = v;
414
            is_table[1][6 - i] = v;
415
        }
416
        /* invalid values */
417
        for(i=7;i<16;i++)
418
            is_table[0][i] = is_table[1][i] = 0.0;
419

    
420
        for(i=0;i<16;i++) {
421
            double f;
422
            int e, k;
423

    
424
            for(j=0;j<2;j++) {
425
                e = -(j + 1) * ((i + 1) >> 1);
426
                f = pow(2.0, e / 4.0);
427
                k = i & 1;
428
                is_table_lsf[j][k ^ 1][i] = FIXR(f);
429
                is_table_lsf[j][k][i] = FIXR(1.0);
430
                dprintf("is_table_lsf %d %d: %x %x\n",
431
                        i, j, is_table_lsf[j][0][i], is_table_lsf[j][1][i]);
432
            }
433
        }
434

    
435
        for(i=0;i<8;i++) {
436
            float ci, cs, ca;
437
            ci = ci_table[i];
438
            cs = 1.0 / sqrt(1.0 + ci * ci);
439
            ca = cs * ci;
440
            csa_table[i][0] = FIXHR(cs/4);
441
            csa_table[i][1] = FIXHR(ca/4);
442
            csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);
443
            csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4);
444
            csa_table_float[i][0] = cs;
445
            csa_table_float[i][1] = ca;
446
            csa_table_float[i][2] = ca + cs;
447
            csa_table_float[i][3] = ca - cs;
448
//            printf("%d %d %d %d\n", FIX(cs), FIX(cs-1), FIX(ca), FIX(cs)-FIX(ca));
449
//            av_log(NULL, AV_LOG_DEBUG,"%f %f %f %f\n", cs, ca, ca+cs, ca-cs);
450
        }
451

    
452
        /* compute mdct windows */
453
        for(i=0;i<36;i++) {
454
            for(j=0; j<4; j++){
455
                double d;
456

    
457
                if(j==2 && i%3 != 1)
458
                    continue;
459

    
460
                d= sin(M_PI * (i + 0.5) / 36.0);
461
                if(j==1){
462
                    if     (i>=30) d= 0;
463
                    else if(i>=24) d= sin(M_PI * (i - 18 + 0.5) / 12.0);
464
                    else if(i>=18) d= 1;
465
                }else if(j==3){
466
                    if     (i<  6) d= 0;
467
                    else if(i< 12) d= sin(M_PI * (i -  6 + 0.5) / 12.0);
468
                    else if(i< 18) d= 1;
469
                }
470
                //merge last stage of imdct into the window coefficients
471
                d*= 0.5 / cos(M_PI*(2*i + 19)/72);
472

    
473
                if(j==2)
474
                    mdct_win[j][i/3] = FIXHR((d / (1<<5)));
475
                else
476
                    mdct_win[j][i  ] = FIXHR((d / (1<<5)));
477
//                av_log(NULL, AV_LOG_DEBUG, "%2d %d %f\n", i,j,d / (1<<5));
478
            }
479
        }
480

    
481
        /* NOTE: we do frequency inversion adter the MDCT by changing
482
           the sign of the right window coefs */
483
        for(j=0;j<4;j++) {
484
            for(i=0;i<36;i+=2) {
485
                mdct_win[j + 4][i] = mdct_win[j][i];
486
                mdct_win[j + 4][i + 1] = -mdct_win[j][i + 1];
487
            }
488
        }
489

    
490
#if defined(DEBUG)
491
        for(j=0;j<8;j++) {
492
            printf("win%d=\n", j);
493
            for(i=0;i<36;i++)
494
                printf("%f, ", (double)mdct_win[j][i] / FRAC_ONE);
495
            printf("\n");
496
        }
497
#endif
498
        init = 1;
499
    }
500

    
501
    s->inbuf_index = 0;
502
    s->inbuf = &s->inbuf1[s->inbuf_index][BACKSTEP_SIZE];
503
    s->inbuf_ptr = s->inbuf;
504
#ifdef DEBUG
505
    s->frame_count = 0;
506
#endif
507
    if (avctx->codec_id == CODEC_ID_MP3ADU)
508
        s->adu_mode = 1;
509
    return 0;
510
}
511

    
512
/* tab[i][j] = 1.0 / (2.0 * cos(pi*(2*k+1) / 2^(6 - j))) */
513

    
514
/* cos(i*pi/64) */
515

    
516
#define COS0_0  FIXR(0.50060299823519630134)
517
#define COS0_1  FIXR(0.50547095989754365998)
518
#define COS0_2  FIXR(0.51544730992262454697)
519
#define COS0_3  FIXR(0.53104259108978417447)
520
#define COS0_4  FIXR(0.55310389603444452782)
521
#define COS0_5  FIXR(0.58293496820613387367)
522
#define COS0_6  FIXR(0.62250412303566481615)
523
#define COS0_7  FIXR(0.67480834145500574602)
524
#define COS0_8  FIXR(0.74453627100229844977)
525
#define COS0_9  FIXR(0.83934964541552703873)
526
#define COS0_10 FIXR(0.97256823786196069369)
527
#define COS0_11 FIXR(1.16943993343288495515)
528
#define COS0_12 FIXR(1.48416461631416627724)
529
#define COS0_13 FIXR(2.05778100995341155085)
530
#define COS0_14 FIXR(3.40760841846871878570)
531
#define COS0_15 FIXR(10.19000812354805681150)
532

    
533
#define COS1_0 FIXR(0.50241928618815570551)
534
#define COS1_1 FIXR(0.52249861493968888062)
535
#define COS1_2 FIXR(0.56694403481635770368)
536
#define COS1_3 FIXR(0.64682178335999012954)
537
#define COS1_4 FIXR(0.78815462345125022473)
538
#define COS1_5 FIXR(1.06067768599034747134)
539
#define COS1_6 FIXR(1.72244709823833392782)
540
#define COS1_7 FIXR(5.10114861868916385802)
541

    
542
#define COS2_0 FIXR(0.50979557910415916894)
543
#define COS2_1 FIXR(0.60134488693504528054)
544
#define COS2_2 FIXR(0.89997622313641570463)
545
#define COS2_3 FIXR(2.56291544774150617881)
546

    
547
#define COS3_0 FIXR(0.54119610014619698439)
548
#define COS3_1 FIXR(1.30656296487637652785)
549

    
550
#define COS4_0 FIXR(0.70710678118654752439)
551

    
552
/* butterfly operator */
553
#define BF(a, b, c)\
554
{\
555
    tmp0 = tab[a] + tab[b];\
556
    tmp1 = tab[a] - tab[b];\
557
    tab[a] = tmp0;\
558
    tab[b] = MULL(tmp1, c);\
559
}
560

    
561
#define BF1(a, b, c, d)\
562
{\
563
    BF(a, b, COS4_0);\
564
    BF(c, d, -COS4_0);\
565
    tab[c] += tab[d];\
566
}
567

    
568
#define BF2(a, b, c, d)\
569
{\
570
    BF(a, b, COS4_0);\
571
    BF(c, d, -COS4_0);\
572
    tab[c] += tab[d];\
573
    tab[a] += tab[c];\
574
    tab[c] += tab[b];\
575
    tab[b] += tab[d];\
576
}
577

    
578
#define ADD(a, b) tab[a] += tab[b]
579

    
580
/* DCT32 without 1/sqrt(2) coef zero scaling. */
581
static void dct32(int32_t *out, int32_t *tab)
582
{
583
    int tmp0, tmp1;
584

    
585
    /* pass 1 */
586
    BF(0, 31, COS0_0);
587
    BF(1, 30, COS0_1);
588
    BF(2, 29, COS0_2);
589
    BF(3, 28, COS0_3);
590
    BF(4, 27, COS0_4);
591
    BF(5, 26, COS0_5);
592
    BF(6, 25, COS0_6);
593
    BF(7, 24, COS0_7);
594
    BF(8, 23, COS0_8);
595
    BF(9, 22, COS0_9);
596
    BF(10, 21, COS0_10);
597
    BF(11, 20, COS0_11);
598
    BF(12, 19, COS0_12);
599
    BF(13, 18, COS0_13);
600
    BF(14, 17, COS0_14);
601
    BF(15, 16, COS0_15);
602

    
603
    /* pass 2 */
604
    BF(0, 15, COS1_0);
605
    BF(1, 14, COS1_1);
606
    BF(2, 13, COS1_2);
607
    BF(3, 12, COS1_3);
608
    BF(4, 11, COS1_4);
609
    BF(5, 10, COS1_5);
610
    BF(6,  9, COS1_6);
611
    BF(7,  8, COS1_7);
612

    
613
    BF(16, 31, -COS1_0);
614
    BF(17, 30, -COS1_1);
615
    BF(18, 29, -COS1_2);
616
    BF(19, 28, -COS1_3);
617
    BF(20, 27, -COS1_4);
618
    BF(21, 26, -COS1_5);
619
    BF(22, 25, -COS1_6);
620
    BF(23, 24, -COS1_7);
621

    
622
    /* pass 3 */
623
    BF(0, 7, COS2_0);
624
    BF(1, 6, COS2_1);
625
    BF(2, 5, COS2_2);
626
    BF(3, 4, COS2_3);
627

    
628
    BF(8, 15, -COS2_0);
629
    BF(9, 14, -COS2_1);
630
    BF(10, 13, -COS2_2);
631
    BF(11, 12, -COS2_3);
632

    
633
    BF(16, 23, COS2_0);
634
    BF(17, 22, COS2_1);
635
    BF(18, 21, COS2_2);
636
    BF(19, 20, COS2_3);
637

    
638
    BF(24, 31, -COS2_0);
639
    BF(25, 30, -COS2_1);
640
    BF(26, 29, -COS2_2);
641
    BF(27, 28, -COS2_3);
642

    
643
    /* pass 4 */
644
    BF(0, 3, COS3_0);
645
    BF(1, 2, COS3_1);
646

    
647
    BF(4, 7, -COS3_0);
648
    BF(5, 6, -COS3_1);
649

    
650
    BF(8, 11, COS3_0);
651
    BF(9, 10, COS3_1);
652

    
653
    BF(12, 15, -COS3_0);
654
    BF(13, 14, -COS3_1);
655

    
656
    BF(16, 19, COS3_0);
657
    BF(17, 18, COS3_1);
658

    
659
    BF(20, 23, -COS3_0);
660
    BF(21, 22, -COS3_1);
661

    
662
    BF(24, 27, COS3_0);
663
    BF(25, 26, COS3_1);
664

    
665
    BF(28, 31, -COS3_0);
666
    BF(29, 30, -COS3_1);
667

    
668
    /* pass 5 */
669
    BF1(0, 1, 2, 3);
670
    BF2(4, 5, 6, 7);
671
    BF1(8, 9, 10, 11);
672
    BF2(12, 13, 14, 15);
673
    BF1(16, 17, 18, 19);
674
    BF2(20, 21, 22, 23);
675
    BF1(24, 25, 26, 27);
676
    BF2(28, 29, 30, 31);
677

    
678
    /* pass 6 */
679

    
680
    ADD( 8, 12);
681
    ADD(12, 10);
682
    ADD(10, 14);
683
    ADD(14,  9);
684
    ADD( 9, 13);
685
    ADD(13, 11);
686
    ADD(11, 15);
687

    
688
    out[ 0] = tab[0];
689
    out[16] = tab[1];
690
    out[ 8] = tab[2];
691
    out[24] = tab[3];
692
    out[ 4] = tab[4];
693
    out[20] = tab[5];
694
    out[12] = tab[6];
695
    out[28] = tab[7];
696
    out[ 2] = tab[8];
697
    out[18] = tab[9];
698
    out[10] = tab[10];
699
    out[26] = tab[11];
700
    out[ 6] = tab[12];
701
    out[22] = tab[13];
702
    out[14] = tab[14];
703
    out[30] = tab[15];
704

    
705
    ADD(24, 28);
706
    ADD(28, 26);
707
    ADD(26, 30);
708
    ADD(30, 25);
709
    ADD(25, 29);
710
    ADD(29, 27);
711
    ADD(27, 31);
712

    
713
    out[ 1] = tab[16] + tab[24];
714
    out[17] = tab[17] + tab[25];
715
    out[ 9] = tab[18] + tab[26];
716
    out[25] = tab[19] + tab[27];
717
    out[ 5] = tab[20] + tab[28];
718
    out[21] = tab[21] + tab[29];
719
    out[13] = tab[22] + tab[30];
720
    out[29] = tab[23] + tab[31];
721
    out[ 3] = tab[24] + tab[20];
722
    out[19] = tab[25] + tab[21];
723
    out[11] = tab[26] + tab[22];
724
    out[27] = tab[27] + tab[23];
725
    out[ 7] = tab[28] + tab[18];
726
    out[23] = tab[29] + tab[19];
727
    out[15] = tab[30] + tab[17];
728
    out[31] = tab[31];
729
}
730

    
731
#if FRAC_BITS <= 15
732

    
733
static inline int round_sample(int *sum)
734
{
735
    int sum1;
736
    sum1 = (*sum) >> OUT_SHIFT;
737
    *sum &= (1<<OUT_SHIFT)-1;
738
    if (sum1 < OUT_MIN)
739
        sum1 = OUT_MIN;
740
    else if (sum1 > OUT_MAX)
741
        sum1 = OUT_MAX;
742
    return sum1;
743
}
744

    
745
#if defined(ARCH_POWERPC_405)
746

    
747
/* signed 16x16 -> 32 multiply add accumulate */
748
#define MACS(rt, ra, rb) \
749
    asm ("maclhw %0, %2, %3" : "=r" (rt) : "0" (rt), "r" (ra), "r" (rb));
750

    
751
/* signed 16x16 -> 32 multiply */
752
#define MULS(ra, rb) \
753
    ({ int __rt; asm ("mullhw %0, %1, %2" : "=r" (__rt) : "r" (ra), "r" (rb)); __rt; })
754

    
755
#else
756

    
757
/* signed 16x16 -> 32 multiply add accumulate */
758
#define MACS(rt, ra, rb) rt += (ra) * (rb)
759

    
760
/* signed 16x16 -> 32 multiply */
761
#define MULS(ra, rb) ((ra) * (rb))
762

    
763
#endif
764

    
765
#else
766

    
767
static inline int round_sample(int64_t *sum)
768
{
769
    int sum1;
770
    sum1 = (int)((*sum) >> OUT_SHIFT);
771
    *sum &= (1<<OUT_SHIFT)-1;
772
    if (sum1 < OUT_MIN)
773
        sum1 = OUT_MIN;
774
    else if (sum1 > OUT_MAX)
775
        sum1 = OUT_MAX;
776
    return sum1;
777
}
778

    
779
#define MULS(ra, rb) MUL64(ra, rb)
780

    
781
#endif
782

    
783
#define SUM8(sum, op, w, p) \
784
{                                               \
785
    sum op MULS((w)[0 * 64], p[0 * 64]);\
786
    sum op MULS((w)[1 * 64], p[1 * 64]);\
787
    sum op MULS((w)[2 * 64], p[2 * 64]);\
788
    sum op MULS((w)[3 * 64], p[3 * 64]);\
789
    sum op MULS((w)[4 * 64], p[4 * 64]);\
790
    sum op MULS((w)[5 * 64], p[5 * 64]);\
791
    sum op MULS((w)[6 * 64], p[6 * 64]);\
792
    sum op MULS((w)[7 * 64], p[7 * 64]);\
793
}
794

    
795
#define SUM8P2(sum1, op1, sum2, op2, w1, w2, p) \
796
{                                               \
797
    int tmp;\
798
    tmp = p[0 * 64];\
799
    sum1 op1 MULS((w1)[0 * 64], tmp);\
800
    sum2 op2 MULS((w2)[0 * 64], tmp);\
801
    tmp = p[1 * 64];\
802
    sum1 op1 MULS((w1)[1 * 64], tmp);\
803
    sum2 op2 MULS((w2)[1 * 64], tmp);\
804
    tmp = p[2 * 64];\
805
    sum1 op1 MULS((w1)[2 * 64], tmp);\
806
    sum2 op2 MULS((w2)[2 * 64], tmp);\
807
    tmp = p[3 * 64];\
808
    sum1 op1 MULS((w1)[3 * 64], tmp);\
809
    sum2 op2 MULS((w2)[3 * 64], tmp);\
810
    tmp = p[4 * 64];\
811
    sum1 op1 MULS((w1)[4 * 64], tmp);\
812
    sum2 op2 MULS((w2)[4 * 64], tmp);\
813
    tmp = p[5 * 64];\
814
    sum1 op1 MULS((w1)[5 * 64], tmp);\
815
    sum2 op2 MULS((w2)[5 * 64], tmp);\
816
    tmp = p[6 * 64];\
817
    sum1 op1 MULS((w1)[6 * 64], tmp);\
818
    sum2 op2 MULS((w2)[6 * 64], tmp);\
819
    tmp = p[7 * 64];\
820
    sum1 op1 MULS((w1)[7 * 64], tmp);\
821
    sum2 op2 MULS((w2)[7 * 64], tmp);\
822
}
823

    
824
void ff_mpa_synth_init(MPA_INT *window)
825
{
826
    int i;
827

    
828
    /* max = 18760, max sum over all 16 coefs : 44736 */
829
    for(i=0;i<257;i++) {
830
        int v;
831
        v = mpa_enwindow[i];
832
#if WFRAC_BITS < 16
833
        v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS);
834
#endif
835
        window[i] = v;
836
        if ((i & 63) != 0)
837
            v = -v;
838
        if (i != 0)
839
            window[512 - i] = v;
840
    }
841
}
842

    
843
/* 32 sub band synthesis filter. Input: 32 sub band samples, Output:
844
   32 samples. */
845
/* XXX: optimize by avoiding ring buffer usage */
846
void ff_mpa_synth_filter(MPA_INT *synth_buf_ptr, int *synth_buf_offset,
847
                         MPA_INT *window, int *dither_state,
848
                         OUT_INT *samples, int incr,
849
                         int32_t sb_samples[SBLIMIT])
850
{
851
    int32_t tmp[32];
852
    register MPA_INT *synth_buf;
853
    register const MPA_INT *w, *w2, *p;
854
    int j, offset, v;
855
    OUT_INT *samples2;
856
#if FRAC_BITS <= 15
857
    int sum, sum2;
858
#else
859
    int64_t sum, sum2;
860
#endif
861

    
862
    dct32(tmp, sb_samples);
863

    
864
    offset = *synth_buf_offset;
865
    synth_buf = synth_buf_ptr + offset;
866

    
867
    for(j=0;j<32;j++) {
868
        v = tmp[j];
869
#if FRAC_BITS <= 15
870
        /* NOTE: can cause a loss in precision if very high amplitude
871
           sound */
872
        if (v > 32767)
873
            v = 32767;
874
        else if (v < -32768)
875
            v = -32768;
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
/* 12 points IMDCT. We compute it "by hand" by factorizing obvious
938
   cases. */
939
static void imdct12(int *out, int *in)
940
{
941
    int in0, in1, in2, in3, in4, in5, t1, t2;
942

    
943
    in0= in[0*3];
944
    in1= in[1*3] + in[0*3];
945
    in2= in[2*3] + in[1*3];
946
    in3= in[3*3] + in[2*3];
947
    in4= in[4*3] + in[3*3];
948
    in5= in[5*3] + in[4*3];
949
    in5 += in3;
950
    in3 += in1;
951

    
952
    in2= MULH(2*in2, C3);
953
    in3= MULH(2*in3, C3);
954

    
955
    t1 = in0 - in4;
956
    t2 = MULL(in1 - in5, icos36[4]);
957

    
958
    out[ 7]=
959
    out[10]= t1 + t2;
960
    out[ 1]=
961
    out[ 4]= t1 - t2;
962

    
963
    in0 += in4>>1;
964
    in4 = in0 + in2;
965
    in1 += in5>>1;
966
    in5 = MULL(in1 + in3, icos36[1]);
967
    out[ 8]=
968
    out[ 9]= in4 + in5;
969
    out[ 2]=
970
    out[ 3]= in4 - in5;
971

    
972
    in0 -= in2;
973
    in1 = MULL(in1 - in3, icos36[7]);
974
    out[ 0]=
975
    out[ 5]= in0 - in1;
976
    out[ 6]=
977
    out[11]= in0 + in1;
978
}
979

    
980
/* cos(pi*i/18) */
981
#define C1 FIXHR(0.98480775301220805936/2)
982
#define C2 FIXHR(0.93969262078590838405/2)
983
#define C3 FIXHR(0.86602540378443864676/2)
984
#define C4 FIXHR(0.76604444311897803520/2)
985
#define C5 FIXHR(0.64278760968653932632/2)
986
#define C6 FIXHR(0.5/2)
987
#define C7 FIXHR(0.34202014332566873304/2)
988
#define C8 FIXHR(0.17364817766693034885/2)
989

    
990

    
991
/* using Lee like decomposition followed by hand coded 9 points DCT */
992
static void imdct36(int *out, int *buf, int *in, int *win)
993
{
994
    int i, j, t0, t1, t2, t3, s0, s1, s2, s3;
995
    int tmp[18], *tmp1, *in1;
996

    
997
    for(i=17;i>=1;i--)
998
        in[i] += in[i-1];
999
    for(i=17;i>=3;i-=2)
1000
        in[i] += in[i-2];
1001

    
1002
    for(j=0;j<2;j++) {
1003
        tmp1 = tmp + j;
1004
        in1 = in + j;
1005
#if 0
1006
//more accurate but slower
1007
        int64_t t0, t1, t2, t3;
1008
        t2 = in1[2*4] + in1[2*8] - in1[2*2];
1009

1010
        t3 = (in1[2*0] + (int64_t)(in1[2*6]>>1))<<32;
1011
        t1 = in1[2*0] - in1[2*6];
1012
        tmp1[ 6] = t1 - (t2>>1);
1013
        tmp1[16] = t1 + t2;
1014

1015
        t0 = MUL64(2*(in1[2*2] + in1[2*4]),    C2);
1016
        t1 = MUL64(   in1[2*4] - in1[2*8] , -2*C8);
1017
        t2 = MUL64(2*(in1[2*2] + in1[2*8]),   -C4);
1018

1019
        tmp1[10] = (t3 - t0 - t2) >> 32;
1020
        tmp1[ 2] = (t3 + t0 + t1) >> 32;
1021
        tmp1[14] = (t3 + t2 - t1) >> 32;
1022

1023
        tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
1024
        t2 = MUL64(2*(in1[2*1] + in1[2*5]),    C1);
1025
        t3 = MUL64(   in1[2*5] - in1[2*7] , -2*C7);
1026
        t0 = MUL64(2*in1[2*3], C3);
1027

1028
        t1 = MUL64(2*(in1[2*1] + in1[2*7]),   -C5);
1029

1030
        tmp1[ 0] = (t2 + t3 + t0) >> 32;
1031
        tmp1[12] = (t2 + t1 - t0) >> 32;
1032
        tmp1[ 8] = (t3 - t1 - t0) >> 32;
1033
#else
1034
        t2 = in1[2*4] + in1[2*8] - in1[2*2];
1035

    
1036
        t3 = in1[2*0] + (in1[2*6]>>1);
1037
        t1 = in1[2*0] - in1[2*6];
1038
        tmp1[ 6] = t1 - (t2>>1);
1039
        tmp1[16] = t1 + t2;
1040

    
1041
        t0 = MULH(2*(in1[2*2] + in1[2*4]),    C2);
1042
        t1 = MULH(   in1[2*4] - in1[2*8] , -2*C8);
1043
        t2 = MULH(2*(in1[2*2] + in1[2*8]),   -C4);
1044

    
1045
        tmp1[10] = t3 - t0 - t2;
1046
        tmp1[ 2] = t3 + t0 + t1;
1047
        tmp1[14] = t3 + t2 - t1;
1048

    
1049
        tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
1050
        t2 = MULH(2*(in1[2*1] + in1[2*5]),    C1);
1051
        t3 = MULH(   in1[2*5] - in1[2*7] , -2*C7);
1052
        t0 = MULH(2*in1[2*3], C3);
1053

    
1054
        t1 = MULH(2*(in1[2*1] + in1[2*7]),   -C5);
1055

    
1056
        tmp1[ 0] = t2 + t3 + t0;
1057
        tmp1[12] = t2 + t1 - t0;
1058
        tmp1[ 8] = t3 - t1 - t0;
1059
#endif
1060
    }
1061

    
1062
    i = 0;
1063
    for(j=0;j<4;j++) {
1064
        t0 = tmp[i];
1065
        t1 = tmp[i + 2];
1066
        s0 = t1 + t0;
1067
        s2 = t1 - t0;
1068

    
1069
        t2 = tmp[i + 1];
1070
        t3 = tmp[i + 3];
1071
        s1 = MULL(t3 + t2, icos36[j]);
1072
        s3 = MULL(t3 - t2, icos36[8 - j]);
1073

    
1074
        t0 = s0 + s1;
1075
        t1 = s0 - s1;
1076
        out[(9 + j)*SBLIMIT] =  MULH(t1, win[9 + j]) + buf[9 + j];
1077
        out[(8 - j)*SBLIMIT] =  MULH(t1, win[8 - j]) + buf[8 - j];
1078
        buf[9 + j] = MULH(t0, win[18 + 9 + j]);
1079
        buf[8 - j] = MULH(t0, win[18 + 8 - j]);
1080

    
1081
        t0 = s2 + s3;
1082
        t1 = s2 - s3;
1083
        out[(9 + 8 - j)*SBLIMIT] =  MULH(t1, win[9 + 8 - j]) + buf[9 + 8 - j];
1084
        out[(        j)*SBLIMIT] =  MULH(t1, win[        j]) + buf[        j];
1085
        buf[9 + 8 - j] = MULH(t0, win[18 + 9 + 8 - j]);
1086
        buf[      + j] = MULH(t0, win[18         + j]);
1087
        i += 4;
1088
    }
1089

    
1090
    s0 = tmp[16];
1091
    s1 = MULL(tmp[17], icos36[4]);
1092
    t0 = s0 + s1;
1093
    t1 = s0 - s1;
1094
    out[(9 + 4)*SBLIMIT] =  MULH(t1, win[9 + 4]) + buf[9 + 4];
1095
    out[(8 - 4)*SBLIMIT] =  MULH(t1, win[8 - 4]) + buf[8 - 4];
1096
    buf[9 + 4] = MULH(t0, win[18 + 9 + 4]);
1097
    buf[8 - 4] = MULH(t0, win[18 + 8 - 4]);
1098
}
1099

    
1100
/* header decoding. MUST check the header before because no
1101
   consistency check is done there. Return 1 if free format found and
1102
   that the frame size must be computed externally */
1103
static int decode_header(MPADecodeContext *s, uint32_t header)
1104
{
1105
    int sample_rate, frame_size, mpeg25, padding;
1106
    int sample_rate_index, bitrate_index;
1107
    if (header & (1<<20)) {
1108
        s->lsf = (header & (1<<19)) ? 0 : 1;
1109
        mpeg25 = 0;
1110
    } else {
1111
        s->lsf = 1;
1112
        mpeg25 = 1;
1113
    }
1114

    
1115
    s->layer = 4 - ((header >> 17) & 3);
1116
    /* extract frequency */
1117
    sample_rate_index = (header >> 10) & 3;
1118
    sample_rate = mpa_freq_tab[sample_rate_index] >> (s->lsf + mpeg25);
1119
    sample_rate_index += 3 * (s->lsf + mpeg25);
1120
    s->sample_rate_index = sample_rate_index;
1121
    s->error_protection = ((header >> 16) & 1) ^ 1;
1122
    s->sample_rate = sample_rate;
1123

    
1124
    bitrate_index = (header >> 12) & 0xf;
1125
    padding = (header >> 9) & 1;
1126
    //extension = (header >> 8) & 1;
1127
    s->mode = (header >> 6) & 3;
1128
    s->mode_ext = (header >> 4) & 3;
1129
    //copyright = (header >> 3) & 1;
1130
    //original = (header >> 2) & 1;
1131
    //emphasis = header & 3;
1132

    
1133
    if (s->mode == MPA_MONO)
1134
        s->nb_channels = 1;
1135
    else
1136
        s->nb_channels = 2;
1137

    
1138
    if (bitrate_index != 0) {
1139
        frame_size = mpa_bitrate_tab[s->lsf][s->layer - 1][bitrate_index];
1140
        s->bit_rate = frame_size * 1000;
1141
        switch(s->layer) {
1142
        case 1:
1143
            frame_size = (frame_size * 12000) / sample_rate;
1144
            frame_size = (frame_size + padding) * 4;
1145
            break;
1146
        case 2:
1147
            frame_size = (frame_size * 144000) / sample_rate;
1148
            frame_size += padding;
1149
            break;
1150
        default:
1151
        case 3:
1152
            frame_size = (frame_size * 144000) / (sample_rate << s->lsf);
1153
            frame_size += padding;
1154
            break;
1155
        }
1156
        s->frame_size = frame_size;
1157
    } else {
1158
        /* if no frame size computed, signal it */
1159
        if (!s->free_format_frame_size)
1160
            return 1;
1161
        /* free format: compute bitrate and real frame size from the
1162
           frame size we extracted by reading the bitstream */
1163
        s->frame_size = s->free_format_frame_size;
1164
        switch(s->layer) {
1165
        case 1:
1166
            s->frame_size += padding  * 4;
1167
            s->bit_rate = (s->frame_size * sample_rate) / 48000;
1168
            break;
1169
        case 2:
1170
            s->frame_size += padding;
1171
            s->bit_rate = (s->frame_size * sample_rate) / 144000;
1172
            break;
1173
        default:
1174
        case 3:
1175
            s->frame_size += padding;
1176
            s->bit_rate = (s->frame_size * (sample_rate << s->lsf)) / 144000;
1177
            break;
1178
        }
1179
    }
1180

    
1181
#if defined(DEBUG)
1182
    printf("layer%d, %d Hz, %d kbits/s, ",
1183
           s->layer, s->sample_rate, s->bit_rate);
1184
    if (s->nb_channels == 2) {
1185
        if (s->layer == 3) {
1186
            if (s->mode_ext & MODE_EXT_MS_STEREO)
1187
                printf("ms-");
1188
            if (s->mode_ext & MODE_EXT_I_STEREO)
1189
                printf("i-");
1190
        }
1191
        printf("stereo");
1192
    } else {
1193
        printf("mono");
1194
    }
1195
    printf("\n");
1196
#endif
1197
    return 0;
1198
}
1199

    
1200
/* useful helper to get mpeg audio stream infos. Return -1 if error in
1201
   header, otherwise the coded frame size in bytes */
1202
int mpa_decode_header(AVCodecContext *avctx, uint32_t head)
1203
{
1204
    MPADecodeContext s1, *s = &s1;
1205
    memset( s, 0, sizeof(MPADecodeContext) );
1206

    
1207
    if (ff_mpa_check_header(head) != 0)
1208
        return -1;
1209

    
1210
    if (decode_header(s, head) != 0) {
1211
        return -1;
1212
    }
1213

    
1214
    switch(s->layer) {
1215
    case 1:
1216
        avctx->frame_size = 384;
1217
        break;
1218
    case 2:
1219
        avctx->frame_size = 1152;
1220
        break;
1221
    default:
1222
    case 3:
1223
        if (s->lsf)
1224
            avctx->frame_size = 576;
1225
        else
1226
            avctx->frame_size = 1152;
1227
        break;
1228
    }
1229

    
1230
    avctx->sample_rate = s->sample_rate;
1231
    avctx->channels = s->nb_channels;
1232
    avctx->bit_rate = s->bit_rate;
1233
    avctx->sub_id = s->layer;
1234
    return s->frame_size;
1235
}
1236

    
1237
/* return the number of decoded frames */
1238
static int mp_decode_layer1(MPADecodeContext *s)
1239
{
1240
    int bound, i, v, n, ch, j, mant;
1241
    uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
1242
    uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
1243

    
1244
    if (s->mode == MPA_JSTEREO)
1245
        bound = (s->mode_ext + 1) * 4;
1246
    else
1247
        bound = SBLIMIT;
1248

    
1249
    /* allocation bits */
1250
    for(i=0;i<bound;i++) {
1251
        for(ch=0;ch<s->nb_channels;ch++) {
1252
            allocation[ch][i] = get_bits(&s->gb, 4);
1253
        }
1254
    }
1255
    for(i=bound;i<SBLIMIT;i++) {
1256
        allocation[0][i] = get_bits(&s->gb, 4);
1257
    }
1258

    
1259
    /* scale factors */
1260
    for(i=0;i<bound;i++) {
1261
        for(ch=0;ch<s->nb_channels;ch++) {
1262
            if (allocation[ch][i])
1263
                scale_factors[ch][i] = get_bits(&s->gb, 6);
1264
        }
1265
    }
1266
    for(i=bound;i<SBLIMIT;i++) {
1267
        if (allocation[0][i]) {
1268
            scale_factors[0][i] = get_bits(&s->gb, 6);
1269
            scale_factors[1][i] = get_bits(&s->gb, 6);
1270
        }
1271
    }
1272

    
1273
    /* compute samples */
1274
    for(j=0;j<12;j++) {
1275
        for(i=0;i<bound;i++) {
1276
            for(ch=0;ch<s->nb_channels;ch++) {
1277
                n = allocation[ch][i];
1278
                if (n) {
1279
                    mant = get_bits(&s->gb, n + 1);
1280
                    v = l1_unscale(n, mant, scale_factors[ch][i]);
1281
                } else {
1282
                    v = 0;
1283
                }
1284
                s->sb_samples[ch][j][i] = v;
1285
            }
1286
        }
1287
        for(i=bound;i<SBLIMIT;i++) {
1288
            n = allocation[0][i];
1289
            if (n) {
1290
                mant = get_bits(&s->gb, n + 1);
1291
                v = l1_unscale(n, mant, scale_factors[0][i]);
1292
                s->sb_samples[0][j][i] = v;
1293
                v = l1_unscale(n, mant, scale_factors[1][i]);
1294
                s->sb_samples[1][j][i] = v;
1295
            } else {
1296
                s->sb_samples[0][j][i] = 0;
1297
                s->sb_samples[1][j][i] = 0;
1298
            }
1299
        }
1300
    }
1301
    return 12;
1302
}
1303

    
1304
/* bitrate is in kb/s */
1305
int l2_select_table(int bitrate, int nb_channels, int freq, int lsf)
1306
{
1307
    int ch_bitrate, table;
1308

    
1309
    ch_bitrate = bitrate / nb_channels;
1310
    if (!lsf) {
1311
        if ((freq == 48000 && ch_bitrate >= 56) ||
1312
            (ch_bitrate >= 56 && ch_bitrate <= 80))
1313
            table = 0;
1314
        else if (freq != 48000 && ch_bitrate >= 96)
1315
            table = 1;
1316
        else if (freq != 32000 && ch_bitrate <= 48)
1317
            table = 2;
1318
        else
1319
            table = 3;
1320
    } else {
1321
        table = 4;
1322
    }
1323
    return table;
1324
}
1325

    
1326
static int mp_decode_layer2(MPADecodeContext *s)
1327
{
1328
    int sblimit; /* number of used subbands */
1329
    const unsigned char *alloc_table;
1330
    int table, bit_alloc_bits, i, j, ch, bound, v;
1331
    unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
1332
    unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
1333
    unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
1334
    int scale, qindex, bits, steps, k, l, m, b;
1335

    
1336
    /* select decoding table */
1337
    table = l2_select_table(s->bit_rate / 1000, s->nb_channels,
1338
                            s->sample_rate, s->lsf);
1339
    sblimit = sblimit_table[table];
1340
    alloc_table = alloc_tables[table];
1341

    
1342
    if (s->mode == MPA_JSTEREO)
1343
        bound = (s->mode_ext + 1) * 4;
1344
    else
1345
        bound = sblimit;
1346

    
1347
    dprintf("bound=%d sblimit=%d\n", bound, sblimit);
1348

    
1349
    /* sanity check */
1350
    if( bound > sblimit ) bound = sblimit;
1351

    
1352
    /* parse bit allocation */
1353
    j = 0;
1354
    for(i=0;i<bound;i++) {
1355
        bit_alloc_bits = alloc_table[j];
1356
        for(ch=0;ch<s->nb_channels;ch++) {
1357
            bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
1358
        }
1359
        j += 1 << bit_alloc_bits;
1360
    }
1361
    for(i=bound;i<sblimit;i++) {
1362
        bit_alloc_bits = alloc_table[j];
1363
        v = get_bits(&s->gb, bit_alloc_bits);
1364
        bit_alloc[0][i] = v;
1365
        bit_alloc[1][i] = v;
1366
        j += 1 << bit_alloc_bits;
1367
    }
1368

    
1369
#ifdef DEBUG
1370
    {
1371
        for(ch=0;ch<s->nb_channels;ch++) {
1372
            for(i=0;i<sblimit;i++)
1373
                printf(" %d", bit_alloc[ch][i]);
1374
            printf("\n");
1375
        }
1376
    }
1377
#endif
1378

    
1379
    /* scale codes */
1380
    for(i=0;i<sblimit;i++) {
1381
        for(ch=0;ch<s->nb_channels;ch++) {
1382
            if (bit_alloc[ch][i])
1383
                scale_code[ch][i] = get_bits(&s->gb, 2);
1384
        }
1385
    }
1386

    
1387
    /* scale factors */
1388
    for(i=0;i<sblimit;i++) {
1389
        for(ch=0;ch<s->nb_channels;ch++) {
1390
            if (bit_alloc[ch][i]) {
1391
                sf = scale_factors[ch][i];
1392
                switch(scale_code[ch][i]) {
1393
                default:
1394
                case 0:
1395
                    sf[0] = get_bits(&s->gb, 6);
1396
                    sf[1] = get_bits(&s->gb, 6);
1397
                    sf[2] = get_bits(&s->gb, 6);
1398
                    break;
1399
                case 2:
1400
                    sf[0] = get_bits(&s->gb, 6);
1401
                    sf[1] = sf[0];
1402
                    sf[2] = sf[0];
1403
                    break;
1404
                case 1:
1405
                    sf[0] = get_bits(&s->gb, 6);
1406
                    sf[2] = get_bits(&s->gb, 6);
1407
                    sf[1] = sf[0];
1408
                    break;
1409
                case 3:
1410
                    sf[0] = get_bits(&s->gb, 6);
1411
                    sf[2] = get_bits(&s->gb, 6);
1412
                    sf[1] = sf[2];
1413
                    break;
1414
                }
1415
            }
1416
        }
1417
    }
1418

    
1419
#ifdef DEBUG
1420
    for(ch=0;ch<s->nb_channels;ch++) {
1421
        for(i=0;i<sblimit;i++) {
1422
            if (bit_alloc[ch][i]) {
1423
                sf = scale_factors[ch][i];
1424
                printf(" %d %d %d", sf[0], sf[1], sf[2]);
1425
            } else {
1426
                printf(" -");
1427
            }
1428
        }
1429
        printf("\n");
1430
    }
1431
#endif
1432

    
1433
    /* samples */
1434
    for(k=0;k<3;k++) {
1435
        for(l=0;l<12;l+=3) {
1436
            j = 0;
1437
            for(i=0;i<bound;i++) {
1438
                bit_alloc_bits = alloc_table[j];
1439
                for(ch=0;ch<s->nb_channels;ch++) {
1440
                    b = bit_alloc[ch][i];
1441
                    if (b) {
1442
                        scale = scale_factors[ch][i][k];
1443
                        qindex = alloc_table[j+b];
1444
                        bits = quant_bits[qindex];
1445
                        if (bits < 0) {
1446
                            /* 3 values at the same time */
1447
                            v = get_bits(&s->gb, -bits);
1448
                            steps = quant_steps[qindex];
1449
                            s->sb_samples[ch][k * 12 + l + 0][i] =
1450
                                l2_unscale_group(steps, v % steps, scale);
1451
                            v = v / steps;
1452
                            s->sb_samples[ch][k * 12 + l + 1][i] =
1453
                                l2_unscale_group(steps, v % steps, scale);
1454
                            v = v / steps;
1455
                            s->sb_samples[ch][k * 12 + l + 2][i] =
1456
                                l2_unscale_group(steps, v, scale);
1457
                        } else {
1458
                            for(m=0;m<3;m++) {
1459
                                v = get_bits(&s->gb, bits);
1460
                                v = l1_unscale(bits - 1, v, scale);
1461
                                s->sb_samples[ch][k * 12 + l + m][i] = v;
1462
                            }
1463
                        }
1464
                    } else {
1465
                        s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1466
                        s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1467
                        s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1468
                    }
1469
                }
1470
                /* next subband in alloc table */
1471
                j += 1 << bit_alloc_bits;
1472
            }
1473
            /* XXX: find a way to avoid this duplication of code */
1474
            for(i=bound;i<sblimit;i++) {
1475
                bit_alloc_bits = alloc_table[j];
1476
                b = bit_alloc[0][i];
1477
                if (b) {
1478
                    int mant, scale0, scale1;
1479
                    scale0 = scale_factors[0][i][k];
1480
                    scale1 = scale_factors[1][i][k];
1481
                    qindex = alloc_table[j+b];
1482
                    bits = quant_bits[qindex];
1483
                    if (bits < 0) {
1484
                        /* 3 values at the same time */
1485
                        v = get_bits(&s->gb, -bits);
1486
                        steps = quant_steps[qindex];
1487
                        mant = v % steps;
1488
                        v = v / steps;
1489
                        s->sb_samples[0][k * 12 + l + 0][i] =
1490
                            l2_unscale_group(steps, mant, scale0);
1491
                        s->sb_samples[1][k * 12 + l + 0][i] =
1492
                            l2_unscale_group(steps, mant, scale1);
1493
                        mant = v % steps;
1494
                        v = v / steps;
1495
                        s->sb_samples[0][k * 12 + l + 1][i] =
1496
                            l2_unscale_group(steps, mant, scale0);
1497
                        s->sb_samples[1][k * 12 + l + 1][i] =
1498
                            l2_unscale_group(steps, mant, scale1);
1499
                        s->sb_samples[0][k * 12 + l + 2][i] =
1500
                            l2_unscale_group(steps, v, scale0);
1501
                        s->sb_samples[1][k * 12 + l + 2][i] =
1502
                            l2_unscale_group(steps, v, scale1);
1503
                    } else {
1504
                        for(m=0;m<3;m++) {
1505
                            mant = get_bits(&s->gb, bits);
1506
                            s->sb_samples[0][k * 12 + l + m][i] =
1507
                                l1_unscale(bits - 1, mant, scale0);
1508
                            s->sb_samples[1][k * 12 + l + m][i] =
1509
                                l1_unscale(bits - 1, mant, scale1);
1510
                        }
1511
                    }
1512
                } else {
1513
                    s->sb_samples[0][k * 12 + l + 0][i] = 0;
1514
                    s->sb_samples[0][k * 12 + l + 1][i] = 0;
1515
                    s->sb_samples[0][k * 12 + l + 2][i] = 0;
1516
                    s->sb_samples[1][k * 12 + l + 0][i] = 0;
1517
                    s->sb_samples[1][k * 12 + l + 1][i] = 0;
1518
                    s->sb_samples[1][k * 12 + l + 2][i] = 0;
1519
                }
1520
                /* next subband in alloc table */
1521
                j += 1 << bit_alloc_bits;
1522
            }
1523
            /* fill remaining samples to zero */
1524
            for(i=sblimit;i<SBLIMIT;i++) {
1525
                for(ch=0;ch<s->nb_channels;ch++) {
1526
                    s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1527
                    s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1528
                    s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1529
                }
1530
            }
1531
        }
1532
    }
1533
    return 3 * 12;
1534
}
1535

    
1536
/*
1537
 * Seek back in the stream for backstep bytes (at most 511 bytes)
1538
 */
1539
static void seek_to_maindata(MPADecodeContext *s, unsigned int backstep)
1540
{
1541
    uint8_t *ptr;
1542

    
1543
    /* compute current position in stream */
1544
    ptr = (uint8_t *)(s->gb.buffer + (get_bits_count(&s->gb)>>3));
1545

    
1546
    /* copy old data before current one */
1547
    ptr -= backstep;
1548
    memcpy(ptr, s->inbuf1[s->inbuf_index ^ 1] +
1549
           BACKSTEP_SIZE + s->old_frame_size - backstep, backstep);
1550
    /* init get bits again */
1551
    init_get_bits(&s->gb, ptr, (s->frame_size + backstep)*8);
1552

    
1553
    /* prepare next buffer */
1554
    s->inbuf_index ^= 1;
1555
    s->inbuf = &s->inbuf1[s->inbuf_index][BACKSTEP_SIZE];
1556
    s->old_frame_size = s->frame_size;
1557
}
1558

    
1559
static inline void lsf_sf_expand(int *slen,
1560
                                 int sf, int n1, int n2, int n3)
1561
{
1562
    if (n3) {
1563
        slen[3] = sf % n3;
1564
        sf /= n3;
1565
    } else {
1566
        slen[3] = 0;
1567
    }
1568
    if (n2) {
1569
        slen[2] = sf % n2;
1570
        sf /= n2;
1571
    } else {
1572
        slen[2] = 0;
1573
    }
1574
    slen[1] = sf % n1;
1575
    sf /= n1;
1576
    slen[0] = sf;
1577
}
1578

    
1579
static void exponents_from_scale_factors(MPADecodeContext *s,
1580
                                         GranuleDef *g,
1581
                                         int16_t *exponents)
1582
{
1583
    const uint8_t *bstab, *pretab;
1584
    int len, i, j, k, l, v0, shift, gain, gains[3];
1585
    int16_t *exp_ptr;
1586

    
1587
    exp_ptr = exponents;
1588
    gain = g->global_gain - 210;
1589
    shift = g->scalefac_scale + 1;
1590

    
1591
    bstab = band_size_long[s->sample_rate_index];
1592
    pretab = mpa_pretab[g->preflag];
1593
    for(i=0;i<g->long_end;i++) {
1594
        v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift);
1595
        len = bstab[i];
1596
        for(j=len;j>0;j--)
1597
            *exp_ptr++ = v0;
1598
    }
1599

    
1600
    if (g->short_start < 13) {
1601
        bstab = band_size_short[s->sample_rate_index];
1602
        gains[0] = gain - (g->subblock_gain[0] << 3);
1603
        gains[1] = gain - (g->subblock_gain[1] << 3);
1604
        gains[2] = gain - (g->subblock_gain[2] << 3);
1605
        k = g->long_end;
1606
        for(i=g->short_start;i<13;i++) {
1607
            len = bstab[i];
1608
            for(l=0;l<3;l++) {
1609
                v0 = gains[l] - (g->scale_factors[k++] << shift);
1610
                for(j=len;j>0;j--)
1611
                *exp_ptr++ = v0;
1612
            }
1613
        }
1614
    }
1615
}
1616

    
1617
/* handle n = 0 too */
1618
static inline int get_bitsz(GetBitContext *s, int n)
1619
{
1620
    if (n == 0)
1621
        return 0;
1622
    else
1623
        return get_bits(s, n);
1624
}
1625

    
1626
static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
1627
                          int16_t *exponents, int end_pos)
1628
{
1629
    int s_index;
1630
    int linbits, code, x, y, l, v, i, j, k, pos;
1631
    GetBitContext last_gb;
1632
    VLC *vlc;
1633
    uint8_t *code_table;
1634

    
1635
    /* low frequencies (called big values) */
1636
    s_index = 0;
1637
    for(i=0;i<3;i++) {
1638
        j = g->region_size[i];
1639
        if (j == 0)
1640
            continue;
1641
        /* select vlc table */
1642
        k = g->table_select[i];
1643
        l = mpa_huff_data[k][0];
1644
        linbits = mpa_huff_data[k][1];
1645
        vlc = &huff_vlc[l];
1646
        code_table = huff_code_table[l];
1647

    
1648
        /* read huffcode and compute each couple */
1649
        for(;j>0;j--) {
1650
            if (get_bits_count(&s->gb) >= end_pos)
1651
                break;
1652
            if (code_table) {
1653
                code = get_vlc2(&s->gb, vlc->table, 8, 2);
1654
                if (code < 0)
1655
                    return -1;
1656
                y = code_table[code];
1657
                x = y >> 4;
1658
                y = y & 0x0f;
1659
            } else {
1660
                x = 0;
1661
                y = 0;
1662
            }
1663
            dprintf("region=%d n=%d x=%d y=%d exp=%d\n",
1664
                    i, g->region_size[i] - j, x, y, exponents[s_index]);
1665
            if (x) {
1666
                if (x == 15)
1667
                    x += get_bitsz(&s->gb, linbits);
1668
                v = l3_unscale(x, exponents[s_index]);
1669
                if (get_bits1(&s->gb))
1670
                    v = -v;
1671
            } else {
1672
                v = 0;
1673
            }
1674
            g->sb_hybrid[s_index++] = v;
1675
            if (y) {
1676
                if (y == 15)
1677
                    y += get_bitsz(&s->gb, linbits);
1678
                v = l3_unscale(y, exponents[s_index]);
1679
                if (get_bits1(&s->gb))
1680
                    v = -v;
1681
            } else {
1682
                v = 0;
1683
            }
1684
            g->sb_hybrid[s_index++] = v;
1685
        }
1686
    }
1687

    
1688
    /* high frequencies */
1689
    vlc = &huff_quad_vlc[g->count1table_select];
1690
    last_gb.buffer = NULL;
1691
    while (s_index <= 572) {
1692
        pos = get_bits_count(&s->gb);
1693
        if (pos >= end_pos) {
1694
            if (pos > end_pos && last_gb.buffer != NULL) {
1695
                /* some encoders generate an incorrect size for this
1696
                   part. We must go back into the data */
1697
                s_index -= 4;
1698
                s->gb = last_gb;
1699
            }
1700
            break;
1701
        }
1702
        last_gb= s->gb;
1703

    
1704
        code = get_vlc2(&s->gb, vlc->table, vlc->bits, 2);
1705
        dprintf("t=%d code=%d\n", g->count1table_select, code);
1706
        if (code < 0)
1707
            return -1;
1708
        for(i=0;i<4;i++) {
1709
            if (code & (8 >> i)) {
1710
                /* non zero value. Could use a hand coded function for
1711
                   'one' value */
1712
                v = l3_unscale(1, exponents[s_index]);
1713
                if(get_bits1(&s->gb))
1714
                    v = -v;
1715
            } else {
1716
                v = 0;
1717
            }
1718
            g->sb_hybrid[s_index++] = v;
1719
        }
1720
    }
1721
    while (s_index < 576)
1722
        g->sb_hybrid[s_index++] = 0;
1723
    return 0;
1724
}
1725

    
1726
/* Reorder short blocks from bitstream order to interleaved order. It
1727
   would be faster to do it in parsing, but the code would be far more
1728
   complicated */
1729
static void reorder_block(MPADecodeContext *s, GranuleDef *g)
1730
{
1731
    int i, j, k, len;
1732
    int32_t *ptr, *dst, *ptr1;
1733
    int32_t tmp[576];
1734

    
1735
    if (g->block_type != 2)
1736
        return;
1737

    
1738
    if (g->switch_point) {
1739
        if (s->sample_rate_index != 8) {
1740
            ptr = g->sb_hybrid + 36;
1741
        } else {
1742
            ptr = g->sb_hybrid + 48;
1743
        }
1744
    } else {
1745
        ptr = g->sb_hybrid;
1746
    }
1747

    
1748
    for(i=g->short_start;i<13;i++) {
1749
        len = band_size_short[s->sample_rate_index][i];
1750
        ptr1 = ptr;
1751
        for(k=0;k<3;k++) {
1752
            dst = tmp + k;
1753
            for(j=len;j>0;j--) {
1754
                *dst = *ptr++;
1755
                dst += 3;
1756
            }
1757
        }
1758
        memcpy(ptr1, tmp, len * 3 * sizeof(int32_t));
1759
    }
1760
}
1761

    
1762
#define ISQRT2 FIXR(0.70710678118654752440)
1763

    
1764
static void compute_stereo(MPADecodeContext *s,
1765
                           GranuleDef *g0, GranuleDef *g1)
1766
{
1767
    int i, j, k, l;
1768
    int32_t v1, v2;
1769
    int sf_max, tmp0, tmp1, sf, len, non_zero_found;
1770
    int32_t (*is_tab)[16];
1771
    int32_t *tab0, *tab1;
1772
    int non_zero_found_short[3];
1773

    
1774
    /* intensity stereo */
1775
    if (s->mode_ext & MODE_EXT_I_STEREO) {
1776
        if (!s->lsf) {
1777
            is_tab = is_table;
1778
            sf_max = 7;
1779
        } else {
1780
            is_tab = is_table_lsf[g1->scalefac_compress & 1];
1781
            sf_max = 16;
1782
        }
1783

    
1784
        tab0 = g0->sb_hybrid + 576;
1785
        tab1 = g1->sb_hybrid + 576;
1786

    
1787
        non_zero_found_short[0] = 0;
1788
        non_zero_found_short[1] = 0;
1789
        non_zero_found_short[2] = 0;
1790
        k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1791
        for(i = 12;i >= g1->short_start;i--) {
1792
            /* for last band, use previous scale factor */
1793
            if (i != 11)
1794
                k -= 3;
1795
            len = band_size_short[s->sample_rate_index][i];
1796
            for(l=2;l>=0;l--) {
1797
                tab0 -= len;
1798
                tab1 -= len;
1799
                if (!non_zero_found_short[l]) {
1800
                    /* test if non zero band. if so, stop doing i-stereo */
1801
                    for(j=0;j<len;j++) {
1802
                        if (tab1[j] != 0) {
1803
                            non_zero_found_short[l] = 1;
1804
                            goto found1;
1805
                        }
1806
                    }
1807
                    sf = g1->scale_factors[k + l];
1808
                    if (sf >= sf_max)
1809
                        goto found1;
1810

    
1811
                    v1 = is_tab[0][sf];
1812
                    v2 = is_tab[1][sf];
1813
                    for(j=0;j<len;j++) {
1814
                        tmp0 = tab0[j];
1815
                        tab0[j] = MULL(tmp0, v1);
1816
                        tab1[j] = MULL(tmp0, v2);
1817
                    }
1818
                } else {
1819
                found1:
1820
                    if (s->mode_ext & MODE_EXT_MS_STEREO) {
1821
                        /* lower part of the spectrum : do ms stereo
1822
                           if enabled */
1823
                        for(j=0;j<len;j++) {
1824
                            tmp0 = tab0[j];
1825
                            tmp1 = tab1[j];
1826
                            tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
1827
                            tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
1828
                        }
1829
                    }
1830
                }
1831
            }
1832
        }
1833

    
1834
        non_zero_found = non_zero_found_short[0] |
1835
            non_zero_found_short[1] |
1836
            non_zero_found_short[2];
1837

    
1838
        for(i = g1->long_end - 1;i >= 0;i--) {
1839
            len = band_size_long[s->sample_rate_index][i];
1840
            tab0 -= len;
1841
            tab1 -= len;
1842
            /* test if non zero band. if so, stop doing i-stereo */
1843
            if (!non_zero_found) {
1844
                for(j=0;j<len;j++) {
1845
                    if (tab1[j] != 0) {
1846
                        non_zero_found = 1;
1847
                        goto found2;
1848
                    }
1849
                }
1850
                /* for last band, use previous scale factor */
1851
                k = (i == 21) ? 20 : i;
1852
                sf = g1->scale_factors[k];
1853
                if (sf >= sf_max)
1854
                    goto found2;
1855
                v1 = is_tab[0][sf];
1856
                v2 = is_tab[1][sf];
1857
                for(j=0;j<len;j++) {
1858
                    tmp0 = tab0[j];
1859
                    tab0[j] = MULL(tmp0, v1);
1860
                    tab1[j] = MULL(tmp0, v2);
1861
                }
1862
            } else {
1863
            found2:
1864
                if (s->mode_ext & MODE_EXT_MS_STEREO) {
1865
                    /* lower part of the spectrum : do ms stereo
1866
                       if enabled */
1867
                    for(j=0;j<len;j++) {
1868
                        tmp0 = tab0[j];
1869
                        tmp1 = tab1[j];
1870
                        tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
1871
                        tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
1872
                    }
1873
                }
1874
            }
1875
        }
1876
    } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1877
        /* ms stereo ONLY */
1878
        /* NOTE: the 1/sqrt(2) normalization factor is included in the
1879
           global gain */
1880
        tab0 = g0->sb_hybrid;
1881
        tab1 = g1->sb_hybrid;
1882
        for(i=0;i<576;i++) {
1883
            tmp0 = tab0[i];
1884
            tmp1 = tab1[i];
1885
            tab0[i] = tmp0 + tmp1;
1886
            tab1[i] = tmp0 - tmp1;
1887
        }
1888
    }
1889
}
1890

    
1891
static void compute_antialias_integer(MPADecodeContext *s,
1892
                              GranuleDef *g)
1893
{
1894
    int32_t *ptr, *csa;
1895
    int n, i;
1896

    
1897
    /* we antialias only "long" bands */
1898
    if (g->block_type == 2) {
1899
        if (!g->switch_point)
1900
            return;
1901
        /* XXX: check this for 8000Hz case */
1902
        n = 1;
1903
    } else {
1904
        n = SBLIMIT - 1;
1905
    }
1906

    
1907
    ptr = g->sb_hybrid + 18;
1908
    for(i = n;i > 0;i--) {
1909
        int tmp0, tmp1, tmp2;
1910
        csa = &csa_table[0][0];
1911
#define INT_AA(j) \
1912
            tmp0 = ptr[-1-j];\
1913
            tmp1 = ptr[   j];\
1914
            tmp2= MULH(tmp0 + tmp1, csa[0+4*j]);\
1915
            ptr[-1-j] = 4*(tmp2 - MULH(tmp1, csa[2+4*j]));\
1916
            ptr[   j] = 4*(tmp2 + MULH(tmp0, csa[3+4*j]));
1917

    
1918
        INT_AA(0)
1919
        INT_AA(1)
1920
        INT_AA(2)
1921
        INT_AA(3)
1922
        INT_AA(4)
1923
        INT_AA(5)
1924
        INT_AA(6)
1925
        INT_AA(7)
1926

    
1927
        ptr += 18;
1928
    }
1929
}
1930

    
1931
static void compute_antialias_float(MPADecodeContext *s,
1932
                              GranuleDef *g)
1933
{
1934
    int32_t *ptr;
1935
    int n, i;
1936

    
1937
    /* we antialias only "long" bands */
1938
    if (g->block_type == 2) {
1939
        if (!g->switch_point)
1940
            return;
1941
        /* XXX: check this for 8000Hz case */
1942
        n = 1;
1943
    } else {
1944
        n = SBLIMIT - 1;
1945
    }
1946

    
1947
    ptr = g->sb_hybrid + 18;
1948
    for(i = n;i > 0;i--) {
1949
        float tmp0, tmp1;
1950
        float *csa = &csa_table_float[0][0];
1951
#define FLOAT_AA(j)\
1952
        tmp0= ptr[-1-j];\
1953
        tmp1= ptr[   j];\
1954
        ptr[-1-j] = lrintf(tmp0 * csa[0+4*j] - tmp1 * csa[1+4*j]);\
1955
        ptr[   j] = lrintf(tmp0 * csa[1+4*j] + tmp1 * csa[0+4*j]);
1956

    
1957
        FLOAT_AA(0)
1958
        FLOAT_AA(1)
1959
        FLOAT_AA(2)
1960
        FLOAT_AA(3)
1961
        FLOAT_AA(4)
1962
        FLOAT_AA(5)
1963
        FLOAT_AA(6)
1964
        FLOAT_AA(7)
1965

    
1966
        ptr += 18;
1967
    }
1968
}
1969

    
1970
static void compute_imdct(MPADecodeContext *s,
1971
                          GranuleDef *g,
1972
                          int32_t *sb_samples,
1973
                          int32_t *mdct_buf)
1974
{
1975
    int32_t *ptr, *win, *win1, *buf, *out_ptr, *ptr1;
1976
    int32_t out2[12];
1977
    int i, j, mdct_long_end, v, sblimit;
1978

    
1979
    /* find last non zero block */
1980
    ptr = g->sb_hybrid + 576;
1981
    ptr1 = g->sb_hybrid + 2 * 18;
1982
    while (ptr >= ptr1) {
1983
        ptr -= 6;
1984
        v = ptr[0] | ptr[1] | ptr[2] | ptr[3] | ptr[4] | ptr[5];
1985
        if (v != 0)
1986
            break;
1987
    }
1988
    sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
1989

    
1990
    if (g->block_type == 2) {
1991
        /* XXX: check for 8000 Hz */
1992
        if (g->switch_point)
1993
            mdct_long_end = 2;
1994
        else
1995
            mdct_long_end = 0;
1996
    } else {
1997
        mdct_long_end = sblimit;
1998
    }
1999

    
2000
    buf = mdct_buf;
2001
    ptr = g->sb_hybrid;
2002
    for(j=0;j<mdct_long_end;j++) {
2003
        /* apply window & overlap with previous buffer */
2004
        out_ptr = sb_samples + j;
2005
        /* select window */
2006
        if (g->switch_point && j < 2)
2007
            win1 = mdct_win[0];
2008
        else
2009
            win1 = mdct_win[g->block_type];
2010
        /* select frequency inversion */
2011
        win = win1 + ((4 * 36) & -(j & 1));
2012
        imdct36(out_ptr, buf, ptr, win);
2013
        out_ptr += 18*SBLIMIT;
2014
        ptr += 18;
2015
        buf += 18;
2016
    }
2017
    for(j=mdct_long_end;j<sblimit;j++) {
2018
        /* select frequency inversion */
2019
        win = mdct_win[2] + ((4 * 36) & -(j & 1));
2020
        out_ptr = sb_samples + j;
2021

    
2022
        for(i=0; i<6; i++){
2023
            *out_ptr = buf[i];
2024
            out_ptr += SBLIMIT;
2025
        }
2026
        imdct12(out2, ptr + 0);
2027
        for(i=0;i<6;i++) {
2028
            *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*1];
2029
            buf[i + 6*2] = MULH(out2[i + 6], win[i + 6]);
2030
            out_ptr += SBLIMIT;
2031
        }
2032
        imdct12(out2, ptr + 1);
2033
        for(i=0;i<6;i++) {
2034
            *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*2];
2035
            buf[i + 6*0] = MULH(out2[i + 6], win[i + 6]);
2036
            out_ptr += SBLIMIT;
2037
        }
2038
        imdct12(out2, ptr + 2);
2039
        for(i=0;i<6;i++) {
2040
            buf[i + 6*0] = MULH(out2[i], win[i]) + buf[i + 6*0];
2041
            buf[i + 6*1] = MULH(out2[i + 6], win[i + 6]);
2042
            buf[i + 6*2] = 0;
2043
        }
2044
        ptr += 18;
2045
        buf += 18;
2046
    }
2047
    /* zero bands */
2048
    for(j=sblimit;j<SBLIMIT;j++) {
2049
        /* overlap */
2050
        out_ptr = sb_samples + j;
2051
        for(i=0;i<18;i++) {
2052
            *out_ptr = buf[i];
2053
            buf[i] = 0;
2054
            out_ptr += SBLIMIT;
2055
        }
2056
        buf += 18;
2057
    }
2058
}
2059

    
2060
#if defined(DEBUG)
2061
void sample_dump(int fnum, int32_t *tab, int n)
2062
{
2063
    static FILE *files[16], *f;
2064
    char buf[512];
2065
    int i;
2066
    int32_t v;
2067

    
2068
    f = files[fnum];
2069
    if (!f) {
2070
        snprintf(buf, sizeof(buf), "/tmp/out%d.%s.pcm",
2071
                fnum,
2072
#ifdef USE_HIGHPRECISION
2073
                "hp"
2074
#else
2075
                "lp"
2076
#endif
2077
                );
2078
        f = fopen(buf, "w");
2079
        if (!f)
2080
            return;
2081
        files[fnum] = f;
2082
    }
2083

    
2084
    if (fnum == 0) {
2085
        static int pos = 0;
2086
        av_log(NULL, AV_LOG_DEBUG, "pos=%d\n", pos);
2087
        for(i=0;i<n;i++) {
2088
            av_log(NULL, AV_LOG_DEBUG, " %0.4f", (double)tab[i] / FRAC_ONE);
2089
            if ((i % 18) == 17)
2090
                av_log(NULL, AV_LOG_DEBUG, "\n");
2091
        }
2092
        pos += n;
2093
    }
2094
    for(i=0;i<n;i++) {
2095
        /* normalize to 23 frac bits */
2096
        v = tab[i] << (23 - FRAC_BITS);
2097
        fwrite(&v, 1, sizeof(int32_t), f);
2098
    }
2099
}
2100
#endif
2101

    
2102

    
2103
/* main layer3 decoding function */
2104
static int mp_decode_layer3(MPADecodeContext *s)
2105
{
2106
    int nb_granules, main_data_begin, private_bits;
2107
    int gr, ch, blocksplit_flag, i, j, k, n, bits_pos, bits_left;
2108
    GranuleDef granules[2][2], *g;
2109
    int16_t exponents[576];
2110

    
2111
    /* read side info */
2112
    if (s->lsf) {
2113
        main_data_begin = get_bits(&s->gb, 8);
2114
        if (s->nb_channels == 2)
2115
            private_bits = get_bits(&s->gb, 2);
2116
        else
2117
            private_bits = get_bits(&s->gb, 1);
2118
        nb_granules = 1;
2119
    } else {
2120
        main_data_begin = get_bits(&s->gb, 9);
2121
        if (s->nb_channels == 2)
2122
            private_bits = get_bits(&s->gb, 3);
2123
        else
2124
            private_bits = get_bits(&s->gb, 5);
2125
        nb_granules = 2;
2126
        for(ch=0;ch<s->nb_channels;ch++) {
2127
            granules[ch][0].scfsi = 0; /* all scale factors are transmitted */
2128
            granules[ch][1].scfsi = get_bits(&s->gb, 4);
2129
        }
2130
    }
2131

    
2132
    for(gr=0;gr<nb_granules;gr++) {
2133
        for(ch=0;ch<s->nb_channels;ch++) {
2134
            dprintf("gr=%d ch=%d: side_info\n", gr, ch);
2135
            g = &granules[ch][gr];
2136
            g->part2_3_length = get_bits(&s->gb, 12);
2137
            g->big_values = get_bits(&s->gb, 9);
2138
            g->global_gain = get_bits(&s->gb, 8);
2139
            /* if MS stereo only is selected, we precompute the
2140
               1/sqrt(2) renormalization factor */
2141
            if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
2142
                MODE_EXT_MS_STEREO)
2143
                g->global_gain -= 2;
2144
            if (s->lsf)
2145
                g->scalefac_compress = get_bits(&s->gb, 9);
2146
            else
2147
                g->scalefac_compress = get_bits(&s->gb, 4);
2148
            blocksplit_flag = get_bits(&s->gb, 1);
2149
            if (blocksplit_flag) {
2150
                g->block_type = get_bits(&s->gb, 2);
2151
                if (g->block_type == 0)
2152
                    return -1;
2153
                g->switch_point = get_bits(&s->gb, 1);
2154
                for(i=0;i<2;i++)
2155
                    g->table_select[i] = get_bits(&s->gb, 5);
2156
                for(i=0;i<3;i++)
2157
                    g->subblock_gain[i] = get_bits(&s->gb, 3);
2158
                /* compute huffman coded region sizes */
2159
                if (g->block_type == 2)
2160
                    g->region_size[0] = (36 / 2);
2161
                else {
2162
                    if (s->sample_rate_index <= 2)
2163
                        g->region_size[0] = (36 / 2);
2164
                    else if (s->sample_rate_index != 8)
2165
                        g->region_size[0] = (54 / 2);
2166
                    else
2167
                        g->region_size[0] = (108 / 2);
2168
                }
2169
                g->region_size[1] = (576 / 2);
2170
            } else {
2171
                int region_address1, region_address2, l;
2172
                g->block_type = 0;
2173
                g->switch_point = 0;
2174
                for(i=0;i<3;i++)
2175
                    g->table_select[i] = get_bits(&s->gb, 5);
2176
                /* compute huffman coded region sizes */
2177
                region_address1 = get_bits(&s->gb, 4);
2178
                region_address2 = get_bits(&s->gb, 3);
2179
                dprintf("region1=%d region2=%d\n",
2180
                        region_address1, region_address2);
2181
                g->region_size[0] =
2182
                    band_index_long[s->sample_rate_index][region_address1 + 1] >> 1;
2183
                l = region_address1 + region_address2 + 2;
2184
                /* should not overflow */
2185
                if (l > 22)
2186
                    l = 22;
2187
                g->region_size[1] =
2188
                    band_index_long[s->sample_rate_index][l] >> 1;
2189
            }
2190
            /* convert region offsets to region sizes and truncate
2191
               size to big_values */
2192
            g->region_size[2] = (576 / 2);
2193
            j = 0;
2194
            for(i=0;i<3;i++) {
2195
                k = g->region_size[i];
2196
                if (k > g->big_values)
2197
                    k = g->big_values;
2198
                g->region_size[i] = k - j;
2199
                j = k;
2200
            }
2201

    
2202
            /* compute band indexes */
2203
            if (g->block_type == 2) {
2204
                if (g->switch_point) {
2205
                    /* if switched mode, we handle the 36 first samples as
2206
                       long blocks.  For 8000Hz, we handle the 48 first
2207
                       exponents as long blocks (XXX: check this!) */
2208
                    if (s->sample_rate_index <= 2)
2209
                        g->long_end = 8;
2210
                    else if (s->sample_rate_index != 8)
2211
                        g->long_end = 6;
2212
                    else
2213
                        g->long_end = 4; /* 8000 Hz */
2214

    
2215
                    if (s->sample_rate_index != 8)
2216
                        g->short_start = 3;
2217
                    else
2218
                        g->short_start = 2;
2219
                } else {
2220
                    g->long_end = 0;
2221
                    g->short_start = 0;
2222
                }
2223
            } else {
2224
                g->short_start = 13;
2225
                g->long_end = 22;
2226
            }
2227

    
2228
            g->preflag = 0;
2229
            if (!s->lsf)
2230
                g->preflag = get_bits(&s->gb, 1);
2231
            g->scalefac_scale = get_bits(&s->gb, 1);
2232
            g->count1table_select = get_bits(&s->gb, 1);
2233
            dprintf("block_type=%d switch_point=%d\n",
2234
                    g->block_type, g->switch_point);
2235
        }
2236
    }
2237

    
2238
  if (!s->adu_mode) {
2239
    /* now we get bits from the main_data_begin offset */
2240
    dprintf("seekback: %d\n", main_data_begin);
2241
    seek_to_maindata(s, main_data_begin);
2242
  }
2243

    
2244
    for(gr=0;gr<nb_granules;gr++) {
2245
        for(ch=0;ch<s->nb_channels;ch++) {
2246
            g = &granules[ch][gr];
2247

    
2248
            bits_pos = get_bits_count(&s->gb);
2249

    
2250
            if (!s->lsf) {
2251
                uint8_t *sc;
2252
                int slen, slen1, slen2;
2253

    
2254
                /* MPEG1 scale factors */
2255
                slen1 = slen_table[0][g->scalefac_compress];
2256
                slen2 = slen_table[1][g->scalefac_compress];
2257
                dprintf("slen1=%d slen2=%d\n", slen1, slen2);
2258
                if (g->block_type == 2) {
2259
                    n = g->switch_point ? 17 : 18;
2260
                    j = 0;
2261
                    for(i=0;i<n;i++)
2262
                        g->scale_factors[j++] = get_bitsz(&s->gb, slen1);
2263
                    for(i=0;i<18;i++)
2264
                        g->scale_factors[j++] = get_bitsz(&s->gb, slen2);
2265
                    for(i=0;i<3;i++)
2266
                        g->scale_factors[j++] = 0;
2267
                } else {
2268
                    sc = granules[ch][0].scale_factors;
2269
                    j = 0;
2270
                    for(k=0;k<4;k++) {
2271
                        n = (k == 0 ? 6 : 5);
2272
                        if ((g->scfsi & (0x8 >> k)) == 0) {
2273
                            slen = (k < 2) ? slen1 : slen2;
2274
                            for(i=0;i<n;i++)
2275
                                g->scale_factors[j++] = get_bitsz(&s->gb, slen);
2276
                        } else {
2277
                            /* simply copy from last granule */
2278
                            for(i=0;i<n;i++) {
2279
                                g->scale_factors[j] = sc[j];
2280
                                j++;
2281
                            }
2282
                        }
2283
                    }
2284
                    g->scale_factors[j++] = 0;
2285
                }
2286
#if defined(DEBUG)
2287
                {
2288
                    printf("scfsi=%x gr=%d ch=%d scale_factors:\n",
2289
                           g->scfsi, gr, ch);
2290
                    for(i=0;i<j;i++)
2291
                        printf(" %d", g->scale_factors[i]);
2292
                    printf("\n");
2293
                }
2294
#endif
2295
            } else {
2296
                int tindex, tindex2, slen[4], sl, sf;
2297

    
2298
                /* LSF scale factors */
2299
                if (g->block_type == 2) {
2300
                    tindex = g->switch_point ? 2 : 1;
2301
                } else {
2302
                    tindex = 0;
2303
                }
2304
                sf = g->scalefac_compress;
2305
                if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
2306
                    /* intensity stereo case */
2307
                    sf >>= 1;
2308
                    if (sf < 180) {
2309
                        lsf_sf_expand(slen, sf, 6, 6, 0);
2310
                        tindex2 = 3;
2311
                    } else if (sf < 244) {
2312
                        lsf_sf_expand(slen, sf - 180, 4, 4, 0);
2313
                        tindex2 = 4;
2314
                    } else {
2315
                        lsf_sf_expand(slen, sf - 244, 3, 0, 0);
2316
                        tindex2 = 5;
2317
                    }
2318
                } else {
2319
                    /* normal case */
2320
                    if (sf < 400) {
2321
                        lsf_sf_expand(slen, sf, 5, 4, 4);
2322
                        tindex2 = 0;
2323
                    } else if (sf < 500) {
2324
                        lsf_sf_expand(slen, sf - 400, 5, 4, 0);
2325
                        tindex2 = 1;
2326
                    } else {
2327
                        lsf_sf_expand(slen, sf - 500, 3, 0, 0);
2328
                        tindex2 = 2;
2329
                        g->preflag = 1;
2330
                    }
2331
                }
2332

    
2333
                j = 0;
2334
                for(k=0;k<4;k++) {
2335
                    n = lsf_nsf_table[tindex2][tindex][k];
2336
                    sl = slen[k];
2337
                    for(i=0;i<n;i++)
2338
                        g->scale_factors[j++] = get_bitsz(&s->gb, sl);
2339
                }
2340
                /* XXX: should compute exact size */
2341
                for(;j<40;j++)
2342
                    g->scale_factors[j] = 0;
2343
#if defined(DEBUG)
2344
                {
2345
                    printf("gr=%d ch=%d scale_factors:\n",
2346
                           gr, ch);
2347
                    for(i=0;i<40;i++)
2348
                        printf(" %d", g->scale_factors[i]);
2349
                    printf("\n");
2350
                }
2351
#endif
2352
            }
2353

    
2354
            exponents_from_scale_factors(s, g, exponents);
2355

    
2356
            /* read Huffman coded residue */
2357
            if (huffman_decode(s, g, exponents,
2358
                               bits_pos + g->part2_3_length) < 0)
2359
                return -1;
2360
#if defined(DEBUG)
2361
            sample_dump(0, g->sb_hybrid, 576);
2362
#endif
2363

    
2364
            /* skip extension bits */
2365
            bits_left = g->part2_3_length - (get_bits_count(&s->gb) - bits_pos);
2366
            if (bits_left < 0) {
2367
                dprintf("bits_left=%d\n", bits_left);
2368
                return -1;
2369
            }
2370
            while (bits_left >= 16) {
2371
                skip_bits(&s->gb, 16);
2372
                bits_left -= 16;
2373
            }
2374
            if (bits_left > 0)
2375
                skip_bits(&s->gb, bits_left);
2376
        } /* ch */
2377

    
2378
        if (s->nb_channels == 2)
2379
            compute_stereo(s, &granules[0][gr], &granules[1][gr]);
2380

    
2381
        for(ch=0;ch<s->nb_channels;ch++) {
2382
            g = &granules[ch][gr];
2383

    
2384
            reorder_block(s, g);
2385
#if defined(DEBUG)
2386
            sample_dump(0, g->sb_hybrid, 576);
2387
#endif
2388
            s->compute_antialias(s, g);
2389
#if defined(DEBUG)
2390
            sample_dump(1, g->sb_hybrid, 576);
2391
#endif
2392
            compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
2393
#if defined(DEBUG)
2394
            sample_dump(2, &s->sb_samples[ch][18 * gr][0], 576);
2395
#endif
2396
        }
2397
    } /* gr */
2398
    return nb_granules * 18;
2399
}
2400

    
2401
static int mp_decode_frame(MPADecodeContext *s,
2402
                           OUT_INT *samples)
2403
{
2404
    int i, nb_frames, ch;
2405
    OUT_INT *samples_ptr;
2406

    
2407
    init_get_bits(&s->gb, s->inbuf + HEADER_SIZE,
2408
                  (s->inbuf_ptr - s->inbuf - HEADER_SIZE)*8);
2409

    
2410
    /* skip error protection field */
2411
    if (s->error_protection)
2412
        get_bits(&s->gb, 16);
2413

    
2414
    dprintf("frame %d:\n", s->frame_count);
2415
    switch(s->layer) {
2416
    case 1:
2417
        nb_frames = mp_decode_layer1(s);
2418
        break;
2419
    case 2:
2420
        nb_frames = mp_decode_layer2(s);
2421
        break;
2422
    case 3:
2423
    default:
2424
        nb_frames = mp_decode_layer3(s);
2425
        break;
2426
    }
2427
#if defined(DEBUG)
2428
    for(i=0;i<nb_frames;i++) {
2429
        for(ch=0;ch<s->nb_channels;ch++) {
2430
            int j;
2431
            printf("%d-%d:", i, ch);
2432
            for(j=0;j<SBLIMIT;j++)
2433
                printf(" %0.6f", (double)s->sb_samples[ch][i][j] / FRAC_ONE);
2434
            printf("\n");
2435
        }
2436
    }
2437
#endif
2438
    /* apply the synthesis filter */
2439
    for(ch=0;ch<s->nb_channels;ch++) {
2440
        samples_ptr = samples + ch;
2441
        for(i=0;i<nb_frames;i++) {
2442
            ff_mpa_synth_filter(s->synth_buf[ch], &(s->synth_buf_offset[ch]),
2443
                         window, &s->dither_state,
2444
                         samples_ptr, s->nb_channels,
2445
                         s->sb_samples[ch][i]);
2446
            samples_ptr += 32 * s->nb_channels;
2447
        }
2448
    }
2449
#ifdef DEBUG
2450
    s->frame_count++;
2451
#endif
2452
    return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
2453
}
2454

    
2455
static int decode_frame(AVCodecContext * avctx,
2456
                        void *data, int *data_size,
2457
                        uint8_t * buf, int buf_size)
2458
{
2459
    MPADecodeContext *s = avctx->priv_data;
2460
    uint32_t header;
2461
    uint8_t *buf_ptr;
2462
    int len, out_size;
2463
    OUT_INT *out_samples = data;
2464

    
2465
    buf_ptr = buf;
2466
    while (buf_size > 0) {
2467
        len = s->inbuf_ptr - s->inbuf;
2468
        if (s->frame_size == 0) {
2469
            /* special case for next header for first frame in free
2470
               format case (XXX: find a simpler method) */
2471
            if (s->free_format_next_header != 0) {
2472
                s->inbuf[0] = s->free_format_next_header >> 24;
2473
                s->inbuf[1] = s->free_format_next_header >> 16;
2474
                s->inbuf[2] = s->free_format_next_header >> 8;
2475
                s->inbuf[3] = s->free_format_next_header;
2476
                s->inbuf_ptr = s->inbuf + 4;
2477
                s->free_format_next_header = 0;
2478
                goto got_header;
2479
            }
2480
            /* no header seen : find one. We need at least HEADER_SIZE
2481
               bytes to parse it */
2482
            len = HEADER_SIZE - len;
2483
            if (len > buf_size)
2484
                len = buf_size;
2485
            if (len > 0) {
2486
                memcpy(s->inbuf_ptr, buf_ptr, len);
2487
                buf_ptr += len;
2488
                buf_size -= len;
2489
                s->inbuf_ptr += len;
2490
            }
2491
            if ((s->inbuf_ptr - s->inbuf) >= HEADER_SIZE) {
2492
            got_header:
2493
                header = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |
2494
                    (s->inbuf[2] << 8) | s->inbuf[3];
2495

    
2496
                if (ff_mpa_check_header(header) < 0) {
2497
                    /* no sync found : move by one byte (inefficient, but simple!) */
2498
                    memmove(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1);
2499
                    s->inbuf_ptr--;
2500
                    dprintf("skip %x\n", header);
2501
                    /* reset free format frame size to give a chance
2502
                       to get a new bitrate */
2503
                    s->free_format_frame_size = 0;
2504
                } else {
2505
                    if (decode_header(s, header) == 1) {
2506
                        /* free format: prepare to compute frame size */
2507
                        s->frame_size = -1;
2508
                    }
2509
                    /* update codec info */
2510
                    avctx->sample_rate = s->sample_rate;
2511
                    avctx->channels = s->nb_channels;
2512
                    avctx->bit_rate = s->bit_rate;
2513
                    avctx->sub_id = s->layer;
2514
                    switch(s->layer) {
2515
                    case 1:
2516
                        avctx->frame_size = 384;
2517
                        break;
2518
                    case 2:
2519
                        avctx->frame_size = 1152;
2520
                        break;
2521
                    case 3:
2522
                        if (s->lsf)
2523
                            avctx->frame_size = 576;
2524
                        else
2525
                            avctx->frame_size = 1152;
2526
                        break;
2527
                    }
2528
                }
2529
            }
2530
        } else if (s->frame_size == -1) {
2531
            /* free format : find next sync to compute frame size */
2532
            len = MPA_MAX_CODED_FRAME_SIZE - len;
2533
            if (len > buf_size)
2534
                len = buf_size;
2535
            if (len == 0) {
2536
                /* frame too long: resync */
2537
                s->frame_size = 0;
2538
                memmove(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1);
2539
                s->inbuf_ptr--;
2540
            } else {
2541
                uint8_t *p, *pend;
2542
                uint32_t header1;
2543
                int padding;
2544

    
2545
                memcpy(s->inbuf_ptr, buf_ptr, len);
2546
                /* check for header */
2547
                p = s->inbuf_ptr - 3;
2548
                pend = s->inbuf_ptr + len - 4;
2549
                while (p <= pend) {
2550
                    header = (p[0] << 24) | (p[1] << 16) |
2551
                        (p[2] << 8) | p[3];
2552
                    header1 = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |
2553
                        (s->inbuf[2] << 8) | s->inbuf[3];
2554
                    /* check with high probability that we have a
2555
                       valid header */
2556
                    if ((header & SAME_HEADER_MASK) ==
2557
                        (header1 & SAME_HEADER_MASK)) {
2558
                        /* header found: update pointers */
2559
                        len = (p + 4) - s->inbuf_ptr;
2560
                        buf_ptr += len;
2561
                        buf_size -= len;
2562
                        s->inbuf_ptr = p;
2563
                        /* compute frame size */
2564
                        s->free_format_next_header = header;
2565
                        s->free_format_frame_size = s->inbuf_ptr - s->inbuf;
2566
                        padding = (header1 >> 9) & 1;
2567
                        if (s->layer == 1)
2568
                            s->free_format_frame_size -= padding * 4;
2569
                        else
2570
                            s->free_format_frame_size -= padding;
2571
                        dprintf("free frame size=%d padding=%d\n",
2572
                                s->free_format_frame_size, padding);
2573
                        decode_header(s, header1);
2574
                        goto next_data;
2575
                    }
2576
                    p++;
2577
                }
2578
                /* not found: simply increase pointers */
2579
                buf_ptr += len;
2580
                s->inbuf_ptr += len;
2581
                buf_size -= len;
2582
            }
2583
        } else if (len < s->frame_size) {
2584
            if (s->frame_size > MPA_MAX_CODED_FRAME_SIZE)
2585
                s->frame_size = MPA_MAX_CODED_FRAME_SIZE;
2586
            len = s->frame_size - len;
2587
            if (len > buf_size)
2588
                len = buf_size;
2589
            memcpy(s->inbuf_ptr, buf_ptr, len);
2590
            buf_ptr += len;
2591
            s->inbuf_ptr += len;
2592
            buf_size -= len;
2593
        }
2594
    next_data:
2595
        if (s->frame_size > 0 &&
2596
            (s->inbuf_ptr - s->inbuf) >= s->frame_size) {
2597
            if (avctx->parse_only) {
2598
                /* simply return the frame data */
2599
                *(uint8_t **)data = s->inbuf;
2600
                out_size = s->inbuf_ptr - s->inbuf;
2601
            } else {
2602
                out_size = mp_decode_frame(s, out_samples);
2603
            }
2604
            s->inbuf_ptr = s->inbuf;
2605
            s->frame_size = 0;
2606
            if(out_size>=0)
2607
                *data_size = out_size;
2608
            else
2609
                av_log(avctx, AV_LOG_DEBUG, "Error while decoding mpeg audio frame\n"); //FIXME return -1 / but also return the number of bytes consumed
2610
            break;
2611
        }
2612
    }
2613
    return buf_ptr - buf;
2614
}
2615

    
2616

    
2617
static int decode_frame_adu(AVCodecContext * avctx,
2618
                        void *data, int *data_size,
2619
                        uint8_t * buf, int buf_size)
2620
{
2621
    MPADecodeContext *s = avctx->priv_data;
2622
    uint32_t header;
2623
    int len, out_size;
2624
    OUT_INT *out_samples = data;
2625

    
2626
    len = buf_size;
2627

    
2628
    // Discard too short frames
2629
    if (buf_size < HEADER_SIZE) {
2630
        *data_size = 0;
2631
        return buf_size;
2632
    }
2633

    
2634

    
2635
    if (len > MPA_MAX_CODED_FRAME_SIZE)
2636
        len = MPA_MAX_CODED_FRAME_SIZE;
2637

    
2638
    memcpy(s->inbuf, buf, len);
2639
    s->inbuf_ptr = s->inbuf + len;
2640

    
2641
    // Get header and restore sync word
2642
    header = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |
2643
              (s->inbuf[2] << 8) | s->inbuf[3] | 0xffe00000;
2644

    
2645
    if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame
2646
        *data_size = 0;
2647
        return buf_size;
2648
    }
2649

    
2650
    decode_header(s, header);
2651
    /* update codec info */
2652
    avctx->sample_rate = s->sample_rate;
2653
    avctx->channels = s->nb_channels;
2654
    avctx->bit_rate = s->bit_rate;
2655
    avctx->sub_id = s->layer;
2656

    
2657
    avctx->frame_size=s->frame_size = len;
2658

    
2659
    if (avctx->parse_only) {
2660
        /* simply return the frame data */
2661
        *(uint8_t **)data = s->inbuf;
2662
        out_size = s->inbuf_ptr - s->inbuf;
2663
    } else {
2664
        out_size = mp_decode_frame(s, out_samples);
2665
    }
2666

    
2667
    *data_size = out_size;
2668
    return buf_size;
2669
}
2670

    
2671

    
2672
/* Next 3 arrays are indexed by channel config number (passed via codecdata) */
2673
static int mp3Frames[16] = {0,1,1,2,3,3,4,5,2};   /* number of mp3 decoder instances */
2674
static int mp3Channels[16] = {0,1,2,3,4,5,6,8,4}; /* total output channels */
2675
/* offsets into output buffer, assume output order is FL FR BL BR C LFE */
2676
static int chan_offset[9][5] = {
2677
    {0},
2678
    {0},            // C
2679
    {0},            // FLR
2680
    {2,0},          // C FLR
2681
    {2,0,3},        // C FLR BS
2682
    {4,0,2},        // C FLR BLRS
2683
    {4,0,2,5},      // C FLR BLRS LFE
2684
    {4,0,2,6,5},    // C FLR BLRS BLR LFE
2685
    {0,2}           // FLR BLRS
2686
};
2687

    
2688

    
2689
static int decode_init_mp3on4(AVCodecContext * avctx)
2690
{
2691
    MP3On4DecodeContext *s = avctx->priv_data;
2692
    int i;
2693

    
2694
    if ((avctx->extradata_size < 2) || (avctx->extradata == NULL)) {
2695
        av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
2696
        return -1;
2697
    }
2698

    
2699
    s->chan_cfg = (((unsigned char *)avctx->extradata)[1] >> 3) & 0x0f;
2700
    s->frames = mp3Frames[s->chan_cfg];
2701
    if(!s->frames) {
2702
        av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
2703
        return -1;
2704
    }
2705
    avctx->channels = mp3Channels[s->chan_cfg];
2706

    
2707
    /* Init the first mp3 decoder in standard way, so that all tables get builded
2708
     * We replace avctx->priv_data with the context of the first decoder so that
2709
     * decode_init() does not have to be changed.
2710
     * Other decoders will be inited here copying data from the first context
2711
     */
2712
    // Allocate zeroed memory for the first decoder context
2713
    s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
2714
    // Put decoder context in place to make init_decode() happy
2715
    avctx->priv_data = s->mp3decctx[0];
2716
    decode_init(avctx);
2717
    // Restore mp3on4 context pointer
2718
    avctx->priv_data = s;
2719
    s->mp3decctx[0]->adu_mode = 1; // Set adu mode
2720

    
2721
    /* Create a separate codec/context for each frame (first is already ok).
2722
     * Each frame is 1 or 2 channels - up to 5 frames allowed
2723
     */
2724
    for (i = 1; i < s->frames; i++) {
2725
        s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
2726
        s->mp3decctx[i]->compute_antialias = s->mp3decctx[0]->compute_antialias;
2727
        s->mp3decctx[i]->inbuf = &s->mp3decctx[i]->inbuf1[0][BACKSTEP_SIZE];
2728
        s->mp3decctx[i]->inbuf_ptr = s->mp3decctx[i]->inbuf;
2729
        s->mp3decctx[i]->adu_mode = 1;
2730
    }
2731

    
2732
    return 0;
2733
}
2734

    
2735

    
2736
static int decode_close_mp3on4(AVCodecContext * avctx)
2737
{
2738
    MP3On4DecodeContext *s = avctx->priv_data;
2739
    int i;
2740

    
2741
    for (i = 0; i < s->frames; i++)
2742
        if (s->mp3decctx[i])
2743
            av_free(s->mp3decctx[i]);
2744

    
2745
    return 0;
2746
}
2747

    
2748

    
2749
static int decode_frame_mp3on4(AVCodecContext * avctx,
2750
                        void *data, int *data_size,
2751
                        uint8_t * buf, int buf_size)
2752
{
2753
    MP3On4DecodeContext *s = avctx->priv_data;
2754
    MPADecodeContext *m;
2755
    int len, out_size = 0;
2756
    uint32_t header;
2757
    OUT_INT *out_samples = data;
2758
    OUT_INT decoded_buf[MPA_FRAME_SIZE * MPA_MAX_CHANNELS];
2759
    OUT_INT *outptr, *bp;
2760
    int fsize;
2761
    unsigned char *start2 = buf, *start;
2762
    int fr, i, j, n;
2763
    int off = avctx->channels;
2764
    int *coff = chan_offset[s->chan_cfg];
2765

    
2766
    len = buf_size;
2767

    
2768
    // Discard too short frames
2769
    if (buf_size < HEADER_SIZE) {
2770
        *data_size = 0;
2771
        return buf_size;
2772
    }
2773

    
2774
    // If only one decoder interleave is not needed
2775
    outptr = s->frames == 1 ? out_samples : decoded_buf;
2776

    
2777
    for (fr = 0; fr < s->frames; fr++) {
2778
        start = start2;
2779
        fsize = (start[0] << 4) | (start[1] >> 4);
2780
        start2 += fsize;
2781
        if (fsize > len)
2782
            fsize = len;
2783
        len -= fsize;
2784
        if (fsize > MPA_MAX_CODED_FRAME_SIZE)
2785
            fsize = MPA_MAX_CODED_FRAME_SIZE;
2786
        m = s->mp3decctx[fr];
2787
        assert (m != NULL);
2788
        /* copy original to new */
2789
        m->inbuf_ptr = m->inbuf + fsize;
2790
        memcpy(m->inbuf, start, fsize);
2791

    
2792
        // Get header
2793
        header = (m->inbuf[0] << 24) | (m->inbuf[1] << 16) |
2794
                  (m->inbuf[2] << 8) | m->inbuf[3] | 0xfff00000;
2795

    
2796
        if (ff_mpa_check_header(header) < 0) { // Bad header, discard block
2797
            *data_size = 0;
2798
            return buf_size;
2799
        }
2800

    
2801
        decode_header(m, header);
2802
        mp_decode_frame(m, decoded_buf);
2803

    
2804
        n = MPA_FRAME_SIZE * m->nb_channels;
2805
        out_size += n * sizeof(OUT_INT);
2806
        if(s->frames > 1) {
2807
            /* interleave output data */
2808
            bp = out_samples + coff[fr];
2809
            if(m->nb_channels == 1) {
2810
                for(j = 0; j < n; j++) {
2811
                    *bp = decoded_buf[j];
2812
                    bp += off;
2813
                }
2814
            } else {
2815
                for(j = 0; j < n; j++) {
2816
                    bp[0] = decoded_buf[j++];
2817
                    bp[1] = decoded_buf[j];
2818
                    bp += off;
2819
                }
2820
            }
2821
        }
2822
    }
2823

    
2824
    /* update codec info */
2825
    avctx->sample_rate = s->mp3decctx[0]->sample_rate;
2826
    avctx->frame_size= buf_size;
2827
    avctx->bit_rate = 0;
2828
    for (i = 0; i < s->frames; i++)
2829
        avctx->bit_rate += s->mp3decctx[i]->bit_rate;
2830

    
2831
    *data_size = out_size;
2832
    return buf_size;
2833
}
2834

    
2835

    
2836
AVCodec mp2_decoder =
2837
{
2838
    "mp2",
2839
    CODEC_TYPE_AUDIO,
2840
    CODEC_ID_MP2,
2841
    sizeof(MPADecodeContext),
2842
    decode_init,
2843
    NULL,
2844
    NULL,
2845
    decode_frame,
2846
    CODEC_CAP_PARSE_ONLY,
2847
};
2848

    
2849
AVCodec mp3_decoder =
2850
{
2851
    "mp3",
2852
    CODEC_TYPE_AUDIO,
2853
    CODEC_ID_MP3,
2854
    sizeof(MPADecodeContext),
2855
    decode_init,
2856
    NULL,
2857
    NULL,
2858
    decode_frame,
2859
    CODEC_CAP_PARSE_ONLY,
2860
};
2861

    
2862
AVCodec mp3adu_decoder =
2863
{
2864
    "mp3adu",
2865
    CODEC_TYPE_AUDIO,
2866
    CODEC_ID_MP3ADU,
2867
    sizeof(MPADecodeContext),
2868
    decode_init,
2869
    NULL,
2870
    NULL,
2871
    decode_frame_adu,
2872
    CODEC_CAP_PARSE_ONLY,
2873
};
2874

    
2875
AVCodec mp3on4_decoder =
2876
{
2877
    "mp3on4",
2878
    CODEC_TYPE_AUDIO,
2879
    CODEC_ID_MP3ON4,
2880
    sizeof(MP3On4DecodeContext),
2881
    decode_init_mp3on4,
2882
    NULL,
2883
    decode_close_mp3on4,
2884
    decode_frame_mp3on4,
2885
    0
2886
};