Statistics
| Branch: | Revision:

ffmpeg / libavcodec / mpegaudiodec.c @ ce4a29c0

History | View | Annotate | Download (81 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., 59 Temple Place, Suite 330, Boston, MA  02111-1307  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 "mpegaudio.h"
29
#include "dsputil.h"
30

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

    
37
/* define USE_HIGHPRECISION to have a bit exact (but slower) mpeg
38
   audio decoder */
39
#ifdef CONFIG_MPEGAUDIO_HP
40
#define USE_HIGHPRECISION
41
#endif
42

    
43
#ifdef USE_HIGHPRECISION
44
#define FRAC_BITS   23   /* fractional bits for sb_samples and dct */
45
#define WFRAC_BITS  16   /* fractional bits for window */
46
#else
47
#define FRAC_BITS   15   /* fractional bits for sb_samples and dct */
48
#define WFRAC_BITS  14   /* fractional bits for window */
49
#endif
50

    
51
#if defined(USE_HIGHPRECISION) && defined(CONFIG_AUDIO_NONSHORT)
52
typedef int32_t OUT_INT;
53
#define OUT_MAX INT32_MAX
54
#define OUT_MIN INT32_MIN
55
#define OUT_SHIFT (WFRAC_BITS + FRAC_BITS - 31)
56
#else
57
typedef int16_t OUT_INT;
58
#define OUT_MAX INT16_MAX
59
#define OUT_MIN INT16_MIN
60
#define OUT_SHIFT (WFRAC_BITS + FRAC_BITS - 15)
61
#endif
62

    
63
#define FRAC_ONE    (1 << FRAC_BITS)
64

    
65
#define MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> FRAC_BITS)
66
#define MUL64(a,b) ((int64_t)(a) * (int64_t)(b))
67
#define FIX(a)   ((int)((a) * FRAC_ONE))
68
/* WARNING: only correct for posititive numbers */
69
#define FIXR(a)   ((int)((a) * FRAC_ONE + 0.5))
70
#define FRAC_RND(a) (((a) + (FRAC_ONE/2)) >> FRAC_BITS)
71

    
72
#define FIXHR(a) ((int)((a) * (1LL<<32) + 0.5))
73
//#define MULH(a,b) (((int64_t)(a) * (int64_t)(b))>>32) //gcc 3.4 creates an incredibly bloated mess out of this
74
static always_inline int MULH(int a, int b){
75
    return ((int64_t)(a) * (int64_t)(b))>>32;
76
}
77

    
78
#if FRAC_BITS <= 15
79
typedef int16_t MPA_INT;
80
#else
81
typedef int32_t MPA_INT;
82
#endif
83

    
84
/****************/
85

    
86
#define HEADER_SIZE 4
87
#define BACKSTEP_SIZE 512
88

    
89
struct GranuleDef;
90

    
91
typedef struct MPADecodeContext {
92
    uint8_t inbuf1[2][MPA_MAX_CODED_FRAME_SIZE + BACKSTEP_SIZE];        /* input buffer */
93
    int inbuf_index;
94
    uint8_t *inbuf_ptr, *inbuf;
95
    int frame_size;
96
    int free_format_frame_size; /* frame size in case of free format
97
                                   (zero if currently unknown) */
98
    /* next header (used in free format parsing) */
99
    uint32_t free_format_next_header; 
100
    int error_protection;
101
    int layer;
102
    int sample_rate;
103
    int sample_rate_index; /* between 0 and 8 */
104
    int bit_rate;
105
    int old_frame_size;
106
    GetBitContext gb;
107
    int nb_channels;
108
    int mode;
109
    int mode_ext;
110
    int lsf;
111
    MPA_INT synth_buf[MPA_MAX_CHANNELS][512 * 2] __attribute__((aligned(16)));
112
    int synth_buf_offset[MPA_MAX_CHANNELS];
113
    int32_t sb_samples[MPA_MAX_CHANNELS][36][SBLIMIT] __attribute__((aligned(16)));
114
    int32_t mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */
115
#ifdef DEBUG
116
    int frame_count;
117
#endif
118
    void (*compute_antialias)(struct MPADecodeContext *s, struct GranuleDef *g);
119
    int adu_mode; ///< 0 for standard mp3, 1 for adu formatted mp3
120
    unsigned int dither_state;
121
} MPADecodeContext;
122

    
123
/* layer 3 "granule" */
124
typedef struct GranuleDef {
125
    uint8_t scfsi;
126
    int part2_3_length;
127
    int big_values;
128
    int global_gain;
129
    int scalefac_compress;
130
    uint8_t block_type;
131
    uint8_t switch_point;
132
    int table_select[3];
133
    int subblock_gain[3];
134
    uint8_t scalefac_scale;
135
    uint8_t count1table_select;
136
    int region_size[3]; /* number of huffman codes in each region */
137
    int preflag;
138
    int short_start, long_end; /* long/short band indexes */
139
    uint8_t scale_factors[40];
140
    int32_t sb_hybrid[SBLIMIT * 18]; /* 576 samples */
141
} GranuleDef;
142

    
143
#define MODE_EXT_MS_STEREO 2
144
#define MODE_EXT_I_STEREO  1
145

    
146
/* layer 3 huffman tables */
147
typedef struct HuffTable {
148
    int xsize;
149
    const uint8_t *bits;
150
    const uint16_t *codes;
151
} HuffTable;
152

    
153
#include "mpegaudiodectab.h"
154

    
155
static void compute_antialias_integer(MPADecodeContext *s, GranuleDef *g);
156
static void compute_antialias_float(MPADecodeContext *s, GranuleDef *g);
157

    
158
/* vlc structure for decoding layer 3 huffman tables */
159
static VLC huff_vlc[16]; 
160
static uint8_t *huff_code_table[16];
161
static VLC huff_quad_vlc[2];
162
/* computed from band_size_long */
163
static uint16_t band_index_long[9][23];
164
/* XXX: free when all decoders are closed */
165
#define TABLE_4_3_SIZE (8191 + 16)*4
166
static int8_t  *table_4_3_exp;
167
static uint32_t *table_4_3_value;
168
/* intensity stereo coef table */
169
static int32_t is_table[2][16];
170
static int32_t is_table_lsf[2][2][16];
171
static int32_t csa_table[8][4];
172
static float csa_table_float[8][4];
173
static int32_t mdct_win[8][36];
174

    
175
/* lower 2 bits: modulo 3, higher bits: shift */
176
static uint16_t scale_factor_modshift[64];
177
/* [i][j]:  2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
178
static int32_t scale_factor_mult[15][3];
179
/* mult table for layer 2 group quantization */
180

    
181
#define SCALE_GEN(v) \
182
{ FIXR(1.0 * (v)), FIXR(0.7937005259 * (v)), FIXR(0.6299605249 * (v)) }
183

    
184
static int32_t scale_factor_mult2[3][3] = {
185
    SCALE_GEN(4.0 / 3.0), /* 3 steps */
186
    SCALE_GEN(4.0 / 5.0), /* 5 steps */
187
    SCALE_GEN(4.0 / 9.0), /* 9 steps */
188
};
189

    
190
void ff_mpa_synth_init(MPA_INT *window);
191
static MPA_INT window[512] __attribute__((aligned(16)));
192
    
193
/* layer 1 unscaling */
194
/* n = number of bits of the mantissa minus 1 */
195
static inline int l1_unscale(int n, int mant, int scale_factor)
196
{
197
    int shift, mod;
198
    int64_t val;
199

    
200
    shift = scale_factor_modshift[scale_factor];
201
    mod = shift & 3;
202
    shift >>= 2;
203
    val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);
204
    shift += n;
205
    /* NOTE: at this point, 1 <= shift >= 21 + 15 */
206
    return (int)((val + (1LL << (shift - 1))) >> shift);
207
}
208

    
209
static inline int l2_unscale_group(int steps, int mant, int scale_factor)
210
{
211
    int shift, mod, val;
212

    
213
    shift = scale_factor_modshift[scale_factor];
214
    mod = shift & 3;
215
    shift >>= 2;
216

    
217
    val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
218
    /* NOTE: at this point, 0 <= shift <= 21 */
219
    if (shift > 0)
220
        val = (val + (1 << (shift - 1))) >> shift;
221
    return val;
222
}
223

    
224
/* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
225
static inline int l3_unscale(int value, int exponent)
226
{
227
    unsigned int m;
228
    int e;
229

    
230
    e = table_4_3_exp  [4*value + (exponent&3)];
231
    m = table_4_3_value[4*value + (exponent&3)];
232
    e -= (exponent >> 2);
233
    assert(e>=1);
234
    if (e > 31)
235
        return 0;
236
    m = (m + (1 << (e-1))) >> e;
237

    
238
    return m;
239
}
240

    
241
/* all integer n^(4/3) computation code */
242
#define DEV_ORDER 13
243

    
244
#define POW_FRAC_BITS 24
245
#define POW_FRAC_ONE    (1 << POW_FRAC_BITS)
246
#define POW_FIX(a)   ((int)((a) * POW_FRAC_ONE))
247
#define POW_MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> POW_FRAC_BITS)
248

    
249
static int dev_4_3_coefs[DEV_ORDER];
250

    
251
static int pow_mult3[3] = {
252
    POW_FIX(1.0),
253
    POW_FIX(1.25992104989487316476),
254
    POW_FIX(1.58740105196819947474),
255
};
256

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

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

    
268
/* return the mantissa and the binary exponent */
269
static int int_pow(int i, int *exp_ptr)
270
{
271
    int e, er, eq, j;
272
    int a, a1;
273
    
274
    /* renormalize */
275
    a = i;
276
    e = POW_FRAC_BITS;
277
    while (a < (1 << (POW_FRAC_BITS - 1))) {
278
        a = a << 1;
279
        e--;
280
    }
281
    a -= (1 << POW_FRAC_BITS);
282
    a1 = 0;
283
    for(j = DEV_ORDER - 1; j >= 0; j--)
284
        a1 = POW_MULL(a, dev_4_3_coefs[j] + a1);
285
    a = (1 << POW_FRAC_BITS) + a1;
286
    /* exponent compute (exact) */
287
    e = e * 4;
288
    er = e % 3;
289
    eq = e / 3;
290
    a = POW_MULL(a, pow_mult3[er]);
291
    while (a >= 2 * POW_FRAC_ONE) {
292
        a = a >> 1;
293
        eq++;
294
    }
295
    /* convert to float */
296
    while (a < POW_FRAC_ONE) {
297
        a = a << 1;
298
        eq--;
299
    }
300
    /* now POW_FRAC_ONE <= a < 2 * POW_FRAC_ONE */
301
#if POW_FRAC_BITS > FRAC_BITS
302
    a = (a + (1 << (POW_FRAC_BITS - FRAC_BITS - 1))) >> (POW_FRAC_BITS - FRAC_BITS);
303
    /* correct overflow */
304
    if (a >= 2 * (1 << FRAC_BITS)) {
305
        a = a >> 1;
306
        eq++;
307
    }
308
#endif
309
    *exp_ptr = eq;
310
    return a;
311
}
312

    
313
static int decode_init(AVCodecContext * avctx)
314
{
315
    MPADecodeContext *s = avctx->priv_data;
316
    static int init=0;
317
    int i, j, k;
318

    
319
#if defined(USE_HIGHPRECISION) && defined(CONFIG_AUDIO_NONSHORT)
320
    avctx->sample_fmt= SAMPLE_FMT_S32;
321
#else
322
    avctx->sample_fmt= SAMPLE_FMT_S16;
323
#endif    
324
    
325
    if(avctx->antialias_algo != FF_AA_FLOAT)
326
        s->compute_antialias= compute_antialias_integer;
327
    else
328
        s->compute_antialias= compute_antialias_float;
329

    
330
    if (!init && !avctx->parse_only) {
331
        /* scale factors table for layer 1/2 */
332
        for(i=0;i<64;i++) {
333
            int shift, mod;
334
            /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */
335
            shift = (i / 3);
336
            mod = i % 3;
337
            scale_factor_modshift[i] = mod | (shift << 2);
338
        }
339

    
340
        /* scale factor multiply for layer 1 */
341
        for(i=0;i<15;i++) {
342
            int n, norm;
343
            n = i + 2;
344
            norm = ((int64_t_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
345
            scale_factor_mult[i][0] = MULL(FIXR(1.0 * 2.0), norm);
346
            scale_factor_mult[i][1] = MULL(FIXR(0.7937005259 * 2.0), norm);
347
            scale_factor_mult[i][2] = MULL(FIXR(0.6299605249 * 2.0), norm);
348
            dprintf("%d: norm=%x s=%x %x %x\n",
349
                    i, norm, 
350
                    scale_factor_mult[i][0],
351
                    scale_factor_mult[i][1],
352
                    scale_factor_mult[i][2]);
353
        }
354
        
355
        ff_mpa_synth_init(window);
356
        
357
        /* huffman decode tables */
358
        huff_code_table[0] = NULL;
359
        for(i=1;i<16;i++) {
360
            const HuffTable *h = &mpa_huff_tables[i];
361
            int xsize, x, y;
362
            unsigned int n;
363
            uint8_t *code_table;
364

    
365
            xsize = h->xsize;
366
            n = xsize * xsize;
367
            /* XXX: fail test */
368
            init_vlc(&huff_vlc[i], 8, n, 
369
                     h->bits, 1, 1, h->codes, 2, 2, 1);
370
            
371
            code_table = av_mallocz(n);
372
            j = 0;
373
            for(x=0;x<xsize;x++) {
374
                for(y=0;y<xsize;y++)
375
                    code_table[j++] = (x << 4) | y;
376
            }
377
            huff_code_table[i] = code_table;
378
        }
379
        for(i=0;i<2;i++) {
380
            init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16, 
381
                     mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1, 1);
382
        }
383

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

    
393
        /* compute n ^ (4/3) and store it in mantissa/exp format */
394
        table_4_3_exp= av_mallocz_static(TABLE_4_3_SIZE * sizeof(table_4_3_exp[0]));
395
        if(!table_4_3_exp)
396
            return -1;
397
        table_4_3_value= av_mallocz_static(TABLE_4_3_SIZE * sizeof(table_4_3_value[0]));
398
        if(!table_4_3_value)
399
            return -1;
400
        
401
        int_pow_init();
402
        for(i=1;i<TABLE_4_3_SIZE;i++) {
403
            double f, fm;
404
            int e, m;
405
            f = pow((double)(i/4), 4.0 / 3.0) * pow(2, (i&3)*0.25);
406
            fm = frexp(f, &e);
407
            m = FIXHR(fm*0.5);
408
            e+= FRAC_BITS - 31;
409

    
410
            /* normalized to FRAC_BITS */
411
            table_4_3_value[i] = m;
412
//            av_log(NULL, AV_LOG_DEBUG, "%d %d %f\n", i, m, pow((double)i, 4.0 / 3.0));
413
            table_4_3_exp[i] = -e;
414
        }
415
        
416
        for(i=0;i<7;i++) {
417
            float f;
418
            int v;
419
            if (i != 6) {
420
                f = tan((double)i * M_PI / 12.0);
421
                v = FIXR(f / (1.0 + f));
422
            } else {
423
                v = FIXR(1.0);
424
            }
425
            is_table[0][i] = v;
426
            is_table[1][6 - i] = v;
427
        }
428
        /* invalid values */
429
        for(i=7;i<16;i++)
430
            is_table[0][i] = is_table[1][i] = 0.0;
431

    
432
        for(i=0;i<16;i++) {
433
            double f;
434
            int e, k;
435

    
436
            for(j=0;j<2;j++) {
437
                e = -(j + 1) * ((i + 1) >> 1);
438
                f = pow(2.0, e / 4.0);
439
                k = i & 1;
440
                is_table_lsf[j][k ^ 1][i] = FIXR(f);
441
                is_table_lsf[j][k][i] = FIXR(1.0);
442
                dprintf("is_table_lsf %d %d: %x %x\n", 
443
                        i, j, is_table_lsf[j][0][i], is_table_lsf[j][1][i]);
444
            }
445
        }
446

    
447
        for(i=0;i<8;i++) {
448
            float ci, cs, ca;
449
            ci = ci_table[i];
450
            cs = 1.0 / sqrt(1.0 + ci * ci);
451
            ca = cs * ci;
452
            csa_table[i][0] = FIXHR(cs/4);
453
            csa_table[i][1] = FIXHR(ca/4);
454
            csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);
455
            csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4); 
456
            csa_table_float[i][0] = cs;
457
            csa_table_float[i][1] = ca;
458
            csa_table_float[i][2] = ca + cs;
459
            csa_table_float[i][3] = ca - cs; 
460
//            printf("%d %d %d %d\n", FIX(cs), FIX(cs-1), FIX(ca), FIX(cs)-FIX(ca));
461
//            av_log(NULL, AV_LOG_DEBUG,"%f %f %f %f\n", cs, ca, ca+cs, ca-cs);
462
        }
463

    
464
        /* compute mdct windows */
465
        for(i=0;i<36;i++) {
466
            for(j=0; j<4; j++){
467
                double d;
468
                if(j==2) continue;
469
                
470
                d= sin(M_PI * (i + 0.5) / 36.0);
471
                if(j==1){
472
                    if     (i>=30) d= 0;
473
                    else if(i>=24) d= sin(M_PI * (i - 18 + 0.5) / 12.0);
474
                    else if(i>=18) d= 1;
475
                }else if(j==3){
476
                    if     (i<  6) d= 0;
477
                    else if(i< 12) d= sin(M_PI * (i -  6 + 0.5) / 12.0);
478
                    else if(i< 18) d= 1;
479
                }
480
                //merge last stage of imdct into the window coefficients
481
                if     (i/9 == 0) d*= 0.5 / cos(M_PI*(2*(     i) +19)/72);
482
                else if(i/9 == 1) d*= 0.5 / cos(M_PI*(2*(17 - i) +19)/72);
483
                else if(i/9 == 2) d*= 0.5 / cos(M_PI*(2*(     i) +19)/72);
484
                else              d*=-0.5 / cos(M_PI*(2*(17 - i) +19)/72);
485
                mdct_win[j][i] = FIXHR((d / (1<<5)));
486
//                av_log(NULL, AV_LOG_DEBUG, "%2d %d %f\n", i,j,d / (1<<5));
487
            }
488
        }
489

    
490
        for(i=0;i<12;i++)
491
            mdct_win[2][i] = FIXR(sin(M_PI * (i + 0.5) / 12.0));
492
        
493
        /* NOTE: we do frequency inversion adter the MDCT by changing
494
           the sign of the right window coefs */
495
        for(j=0;j<4;j++) {
496
            for(i=0;i<36;i+=2) {
497
                mdct_win[j + 4][i] = mdct_win[j][i];
498
                mdct_win[j + 4][i + 1] = -mdct_win[j][i + 1];
499
            }
500
        }
501

    
502
#if defined(DEBUG)
503
        for(j=0;j<8;j++) {
504
            printf("win%d=\n", j);
505
            for(i=0;i<36;i++)
506
                printf("%f, ", (double)mdct_win[j][i] / FRAC_ONE);
507
            printf("\n");
508
        }
509
#endif
510
        init = 1;
511
    }
512

    
513
    s->inbuf_index = 0;
514
    s->inbuf = &s->inbuf1[s->inbuf_index][BACKSTEP_SIZE];
515
    s->inbuf_ptr = s->inbuf;
516
#ifdef DEBUG
517
    s->frame_count = 0;
518
#endif
519
    if (avctx->codec_id == CODEC_ID_MP3ADU)
520
        s->adu_mode = 1;
521
    return 0;
522
}
523

    
524
/* tab[i][j] = 1.0 / (2.0 * cos(pi*(2*k+1) / 2^(6 - j))) */
525

    
526
/* cos(i*pi/64) */
527

    
528
#define COS0_0  FIXR(0.50060299823519630134)
529
#define COS0_1  FIXR(0.50547095989754365998)
530
#define COS0_2  FIXR(0.51544730992262454697)
531
#define COS0_3  FIXR(0.53104259108978417447)
532
#define COS0_4  FIXR(0.55310389603444452782)
533
#define COS0_5  FIXR(0.58293496820613387367)
534
#define COS0_6  FIXR(0.62250412303566481615)
535
#define COS0_7  FIXR(0.67480834145500574602)
536
#define COS0_8  FIXR(0.74453627100229844977)
537
#define COS0_9  FIXR(0.83934964541552703873)
538
#define COS0_10 FIXR(0.97256823786196069369)
539
#define COS0_11 FIXR(1.16943993343288495515)
540
#define COS0_12 FIXR(1.48416461631416627724)
541
#define COS0_13 FIXR(2.05778100995341155085)
542
#define COS0_14 FIXR(3.40760841846871878570)
543
#define COS0_15 FIXR(10.19000812354805681150)
544

    
545
#define COS1_0 FIXR(0.50241928618815570551)
546
#define COS1_1 FIXR(0.52249861493968888062)
547
#define COS1_2 FIXR(0.56694403481635770368)
548
#define COS1_3 FIXR(0.64682178335999012954)
549
#define COS1_4 FIXR(0.78815462345125022473)
550
#define COS1_5 FIXR(1.06067768599034747134)
551
#define COS1_6 FIXR(1.72244709823833392782)
552
#define COS1_7 FIXR(5.10114861868916385802)
553

    
554
#define COS2_0 FIXR(0.50979557910415916894)
555
#define COS2_1 FIXR(0.60134488693504528054)
556
#define COS2_2 FIXR(0.89997622313641570463)
557
#define COS2_3 FIXR(2.56291544774150617881)
558

    
559
#define COS3_0 FIXR(0.54119610014619698439)
560
#define COS3_1 FIXR(1.30656296487637652785)
561

    
562
#define COS4_0 FIXR(0.70710678118654752439)
563

    
564
/* butterfly operator */
565
#define BF(a, b, c)\
566
{\
567
    tmp0 = tab[a] + tab[b];\
568
    tmp1 = tab[a] - tab[b];\
569
    tab[a] = tmp0;\
570
    tab[b] = MULL(tmp1, c);\
571
}
572

    
573
#define BF1(a, b, c, d)\
574
{\
575
    BF(a, b, COS4_0);\
576
    BF(c, d, -COS4_0);\
577
    tab[c] += tab[d];\
578
}
579

    
580
#define BF2(a, b, c, d)\
581
{\
582
    BF(a, b, COS4_0);\
583
    BF(c, d, -COS4_0);\
584
    tab[c] += tab[d];\
585
    tab[a] += tab[c];\
586
    tab[c] += tab[b];\
587
    tab[b] += tab[d];\
588
}
589

    
590
#define ADD(a, b) tab[a] += tab[b]
591

    
592
/* DCT32 without 1/sqrt(2) coef zero scaling. */
593
static void dct32(int32_t *out, int32_t *tab)
594
{
595
    int tmp0, tmp1;
596

    
597
    /* pass 1 */
598
    BF(0, 31, COS0_0);
599
    BF(1, 30, COS0_1);
600
    BF(2, 29, COS0_2);
601
    BF(3, 28, COS0_3);
602
    BF(4, 27, COS0_4);
603
    BF(5, 26, COS0_5);
604
    BF(6, 25, COS0_6);
605
    BF(7, 24, COS0_7);
606
    BF(8, 23, COS0_8);
607
    BF(9, 22, COS0_9);
608
    BF(10, 21, COS0_10);
609
    BF(11, 20, COS0_11);
610
    BF(12, 19, COS0_12);
611
    BF(13, 18, COS0_13);
612
    BF(14, 17, COS0_14);
613
    BF(15, 16, COS0_15);
614

    
615
    /* pass 2 */
616
    BF(0, 15, COS1_0);
617
    BF(1, 14, COS1_1);
618
    BF(2, 13, COS1_2);
619
    BF(3, 12, COS1_3);
620
    BF(4, 11, COS1_4);
621
    BF(5, 10, COS1_5);
622
    BF(6,  9, COS1_6);
623
    BF(7,  8, COS1_7);
624
    
625
    BF(16, 31, -COS1_0);
626
    BF(17, 30, -COS1_1);
627
    BF(18, 29, -COS1_2);
628
    BF(19, 28, -COS1_3);
629
    BF(20, 27, -COS1_4);
630
    BF(21, 26, -COS1_5);
631
    BF(22, 25, -COS1_6);
632
    BF(23, 24, -COS1_7);
633
    
634
    /* pass 3 */
635
    BF(0, 7, COS2_0);
636
    BF(1, 6, COS2_1);
637
    BF(2, 5, COS2_2);
638
    BF(3, 4, COS2_3);
639
    
640
    BF(8, 15, -COS2_0);
641
    BF(9, 14, -COS2_1);
642
    BF(10, 13, -COS2_2);
643
    BF(11, 12, -COS2_3);
644
    
645
    BF(16, 23, COS2_0);
646
    BF(17, 22, COS2_1);
647
    BF(18, 21, COS2_2);
648
    BF(19, 20, COS2_3);
649
    
650
    BF(24, 31, -COS2_0);
651
    BF(25, 30, -COS2_1);
652
    BF(26, 29, -COS2_2);
653
    BF(27, 28, -COS2_3);
654

    
655
    /* pass 4 */
656
    BF(0, 3, COS3_0);
657
    BF(1, 2, COS3_1);
658
    
659
    BF(4, 7, -COS3_0);
660
    BF(5, 6, -COS3_1);
661
    
662
    BF(8, 11, COS3_0);
663
    BF(9, 10, COS3_1);
664
    
665
    BF(12, 15, -COS3_0);
666
    BF(13, 14, -COS3_1);
667
    
668
    BF(16, 19, COS3_0);
669
    BF(17, 18, COS3_1);
670
    
671
    BF(20, 23, -COS3_0);
672
    BF(21, 22, -COS3_1);
673
    
674
    BF(24, 27, COS3_0);
675
    BF(25, 26, COS3_1);
676
    
677
    BF(28, 31, -COS3_0);
678
    BF(29, 30, -COS3_1);
679
    
680
    /* pass 5 */
681
    BF1(0, 1, 2, 3);
682
    BF2(4, 5, 6, 7);
683
    BF1(8, 9, 10, 11);
684
    BF2(12, 13, 14, 15);
685
    BF1(16, 17, 18, 19);
686
    BF2(20, 21, 22, 23);
687
    BF1(24, 25, 26, 27);
688
    BF2(28, 29, 30, 31);
689
    
690
    /* pass 6 */
691
    
692
    ADD( 8, 12);
693
    ADD(12, 10);
694
    ADD(10, 14);
695
    ADD(14,  9);
696
    ADD( 9, 13);
697
    ADD(13, 11);
698
    ADD(11, 15);
699

    
700
    out[ 0] = tab[0];
701
    out[16] = tab[1];
702
    out[ 8] = tab[2];
703
    out[24] = tab[3];
704
    out[ 4] = tab[4];
705
    out[20] = tab[5];
706
    out[12] = tab[6];
707
    out[28] = tab[7];
708
    out[ 2] = tab[8];
709
    out[18] = tab[9];
710
    out[10] = tab[10];
711
    out[26] = tab[11];
712
    out[ 6] = tab[12];
713
    out[22] = tab[13];
714
    out[14] = tab[14];
715
    out[30] = tab[15];
716
    
717
    ADD(24, 28);
718
    ADD(28, 26);
719
    ADD(26, 30);
720
    ADD(30, 25);
721
    ADD(25, 29);
722
    ADD(29, 27);
723
    ADD(27, 31);
724

    
725
    out[ 1] = tab[16] + tab[24];
726
    out[17] = tab[17] + tab[25];
727
    out[ 9] = tab[18] + tab[26];
728
    out[25] = tab[19] + tab[27];
729
    out[ 5] = tab[20] + tab[28];
730
    out[21] = tab[21] + tab[29];
731
    out[13] = tab[22] + tab[30];
732
    out[29] = tab[23] + tab[31];
733
    out[ 3] = tab[24] + tab[20];
734
    out[19] = tab[25] + tab[21];
735
    out[11] = tab[26] + tab[22];
736
    out[27] = tab[27] + tab[23];
737
    out[ 7] = tab[28] + tab[18];
738
    out[23] = tab[29] + tab[19];
739
    out[15] = tab[30] + tab[17];
740
    out[31] = tab[31];
741
}
742

    
743
#if FRAC_BITS <= 15
744

    
745
static inline int round_sample(int *sum)
746
{
747
    int sum1;
748
    sum1 = (*sum) >> OUT_SHIFT;
749
    *sum &= (1<<OUT_SHIFT)-1;
750
    if (sum1 < OUT_MIN)
751
        sum1 = OUT_MIN;
752
    else if (sum1 > OUT_MAX)
753
        sum1 = OUT_MAX;
754
    return sum1;
755
}
756

    
757
#if defined(ARCH_POWERPC_405)
758

    
759
/* signed 16x16 -> 32 multiply add accumulate */
760
#define MACS(rt, ra, rb) \
761
    asm ("maclhw %0, %2, %3" : "=r" (rt) : "0" (rt), "r" (ra), "r" (rb));
762

    
763
/* signed 16x16 -> 32 multiply */
764
#define MULS(ra, rb) \
765
    ({ int __rt; asm ("mullhw %0, %1, %2" : "=r" (__rt) : "r" (ra), "r" (rb)); __rt; })
766

    
767
#else
768

    
769
/* signed 16x16 -> 32 multiply add accumulate */
770
#define MACS(rt, ra, rb) rt += (ra) * (rb)
771

    
772
/* signed 16x16 -> 32 multiply */
773
#define MULS(ra, rb) ((ra) * (rb))
774

    
775
#endif
776

    
777
#else
778

    
779
static inline int round_sample(int64_t *sum) 
780
{
781
    int sum1;
782
    sum1 = (int)((*sum) >> OUT_SHIFT);
783
    *sum &= (1<<OUT_SHIFT)-1;
784
    if (sum1 < OUT_MIN)
785
        sum1 = OUT_MIN;
786
    else if (sum1 > OUT_MAX)
787
        sum1 = OUT_MAX;
788
    return sum1;
789
}
790

    
791
#define MULS(ra, rb) MUL64(ra, rb)
792

    
793
#endif
794

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

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

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

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

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

    
874
    dct32(tmp, sb_samples);
875
    
876
    offset = *synth_buf_offset;
877
    synth_buf = synth_buf_ptr + offset;
878

    
879
    for(j=0;j<32;j++) {
880
        v = tmp[j];
881
#if FRAC_BITS <= 15
882
        /* NOTE: can cause a loss in precision if very high amplitude
883
           sound */
884
        if (v > 32767)
885
            v = 32767;
886
        else if (v < -32768)
887
            v = -32768;
888
#endif
889
        synth_buf[j] = v;
890
    }
891
    /* copy to avoid wrap */
892
    memcpy(synth_buf + 512, synth_buf, 32 * sizeof(MPA_INT));
893

    
894
    samples2 = samples + 31 * incr;
895
    w = window;
896
    w2 = window + 31;
897

    
898
    sum = *dither_state;
899
    p = synth_buf + 16;
900
    SUM8(sum, +=, w, p);
901
    p = synth_buf + 48;
902
    SUM8(sum, -=, w + 32, p);
903
    *samples = round_sample(&sum);
904
    samples += incr;
905
    w++;
906

    
907
    /* we calculate two samples at the same time to avoid one memory
908
       access per two sample */
909
    for(j=1;j<16;j++) {
910
        sum2 = 0;
911
        p = synth_buf + 16 + j;
912
        SUM8P2(sum, +=, sum2, -=, w, w2, p);
913
        p = synth_buf + 48 - j;
914
        SUM8P2(sum, -=, sum2, -=, w + 32, w2 + 32, p);
915

    
916
        *samples = round_sample(&sum);
917
        samples += incr;
918
        sum += sum2;
919
        *samples2 = round_sample(&sum);
920
        samples2 -= incr;
921
        w++;
922
        w2--;
923
    }
924
    
925
    p = synth_buf + 32;
926
    SUM8(sum, -=, w + 32, p);
927
    *samples = round_sample(&sum);
928
    *dither_state= sum;
929

    
930
    offset = (offset - 32) & 511;
931
    *synth_buf_offset = offset;
932
}
933

    
934
/* cos(pi*i/24) */
935
#define C1  FIXR(0.99144486137381041114)
936
#define C3  FIXR(0.92387953251128675612)
937
#define C5  FIXR(0.79335334029123516458)
938
#define C7  FIXR(0.60876142900872063941)
939
#define C9  FIXR(0.38268343236508977173)
940
#define C11 FIXR(0.13052619222005159154)
941

    
942
/* 12 points IMDCT. We compute it "by hand" by factorizing obvious
943
   cases. */
944
static void imdct12(int *out, int *in)
945
{
946
    int tmp;
947
    int64_t in1_3, in1_9, in4_3, in4_9;
948

    
949
    in1_3 = MUL64(in[1], C3);
950
    in1_9 = MUL64(in[1], C9);
951
    in4_3 = MUL64(in[4], C3);
952
    in4_9 = MUL64(in[4], C9);
953
    
954
    tmp = FRAC_RND(MUL64(in[0], C7) - in1_3 - MUL64(in[2], C11) + 
955
                   MUL64(in[3], C1) - in4_9 - MUL64(in[5], C5));
956
    out[0] = tmp;
957
    out[5] = -tmp;
958
    tmp = FRAC_RND(MUL64(in[0] - in[3], C9) - in1_3 + 
959
                   MUL64(in[2] + in[5], C3) - in4_9);
960
    out[1] = tmp;
961
    out[4] = -tmp;
962
    tmp = FRAC_RND(MUL64(in[0], C11) - in1_9 + MUL64(in[2], C7) -
963
                   MUL64(in[3], C5) + in4_3 - MUL64(in[5], C1));
964
    out[2] = tmp;
965
    out[3] = -tmp;
966
    tmp = FRAC_RND(MUL64(-in[0], C5) + in1_9 + MUL64(in[2], C1) + 
967
                   MUL64(in[3], C11) - in4_3 - MUL64(in[5], C7));
968
    out[6] = tmp;
969
    out[11] = tmp;
970
    tmp = FRAC_RND(MUL64(-in[0] + in[3], C3) - in1_9 + 
971
                   MUL64(in[2] + in[5], C9) + in4_3);
972
    out[7] = tmp;
973
    out[10] = tmp;
974
    tmp = FRAC_RND(-MUL64(in[0], C1) - in1_3 - MUL64(in[2], C5) -
975
                   MUL64(in[3], C7) - in4_9 - MUL64(in[5], C11));
976
    out[8] = tmp;
977
    out[9] = tmp;
978
}
979

    
980
#undef C1
981
#undef C3
982
#undef C5
983
#undef C7
984
#undef C9
985
#undef C11
986

    
987
/* cos(pi*i/18) */
988
#define C1 FIXHR(0.98480775301220805936/2)
989
#define C2 FIXHR(0.93969262078590838405/2)
990
#define C3 FIXHR(0.86602540378443864676/2)
991
#define C4 FIXHR(0.76604444311897803520/2)
992
#define C5 FIXHR(0.64278760968653932632/2)
993
#define C6 FIXHR(0.5/2)
994
#define C7 FIXHR(0.34202014332566873304/2)
995
#define C8 FIXHR(0.17364817766693034885/2)
996

    
997

    
998
/* 0.5 / cos(pi*(2*i+1)/36) */
999
static const int icos36[9] = {
1000
    FIXR(0.50190991877167369479),
1001
    FIXR(0.51763809020504152469),
1002
    FIXR(0.55168895948124587824),
1003
    FIXR(0.61038729438072803416),
1004
    FIXR(0.70710678118654752439),
1005
    FIXR(0.87172339781054900991),
1006
    FIXR(1.18310079157624925896),
1007
    FIXR(1.93185165257813657349),
1008
    FIXR(5.73685662283492756461),
1009
};
1010
/* using Lee like decomposition followed by hand coded 9 points DCT */
1011
static void imdct36(int *out, int *buf, int *in, int *win)
1012
{
1013
    int i, j, t0, t1, t2, t3, s0, s1, s2, s3;
1014
    int tmp[18], *tmp1, *in1;
1015

    
1016
    for(i=17;i>=1;i--)
1017
        in[i] += in[i-1];
1018
    for(i=17;i>=3;i-=2)
1019
        in[i] += in[i-2];
1020

    
1021
    for(j=0;j<2;j++) {
1022
        tmp1 = tmp + j;
1023
        in1 = in + j;
1024
#if 0
1025
//more accurate but slower
1026
        int64_t t0, t1, t2, t3;
1027
        t2 = in1[2*4] + in1[2*8] - in1[2*2];
1028
        
1029
        t3 = (in1[2*0] + (int64_t)(in1[2*6]>>1))<<32;
1030
        t1 = in1[2*0] - in1[2*6];
1031
        tmp1[ 6] = t1 - (t2>>1);
1032
        tmp1[16] = t1 + t2;
1033

1034
        t0 = MUL64(2*(in1[2*2] + in1[2*4]),    C2);
1035
        t1 = MUL64(   in1[2*4] - in1[2*8] , -2*C8);
1036
        t2 = MUL64(2*(in1[2*2] + in1[2*8]),   -C4);
1037
        
1038
        tmp1[10] = (t3 - t0 - t2) >> 32;
1039
        tmp1[ 2] = (t3 + t0 + t1) >> 32;
1040
        tmp1[14] = (t3 + t2 - t1) >> 32;
1041
        
1042
        tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
1043
        t2 = MUL64(2*(in1[2*1] + in1[2*5]),    C1);
1044
        t3 = MUL64(   in1[2*5] - in1[2*7] , -2*C7);
1045
        t0 = MUL64(2*in1[2*3], C3);
1046

1047
        t1 = MUL64(2*(in1[2*1] + in1[2*7]),   -C5);
1048

1049
        tmp1[ 0] = (t2 + t3 + t0) >> 32;
1050
        tmp1[12] = (t2 + t1 - t0) >> 32;
1051
        tmp1[ 8] = (t3 - t1 - t0) >> 32;
1052
#else
1053
        t2 = in1[2*4] + in1[2*8] - in1[2*2];
1054
        
1055
        t3 = in1[2*0] + (in1[2*6]>>1);
1056
        t1 = in1[2*0] - in1[2*6];
1057
        tmp1[ 6] = t1 - (t2>>1);
1058
        tmp1[16] = t1 + t2;
1059

    
1060
        t0 = MULH(2*(in1[2*2] + in1[2*4]),    C2);
1061
        t1 = MULH(   in1[2*4] - in1[2*8] , -2*C8);
1062
        t2 = MULH(2*(in1[2*2] + in1[2*8]),   -C4);
1063
        
1064
        tmp1[10] = t3 - t0 - t2;
1065
        tmp1[ 2] = t3 + t0 + t1;
1066
        tmp1[14] = t3 + t2 - t1;
1067
        
1068
        tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
1069
        t2 = MULH(2*(in1[2*1] + in1[2*5]),    C1);
1070
        t3 = MULH(   in1[2*5] - in1[2*7] , -2*C7);
1071
        t0 = MULH(2*in1[2*3], C3);
1072

    
1073
        t1 = MULH(2*(in1[2*1] + in1[2*7]),   -C5);
1074

    
1075
        tmp1[ 0] = t2 + t3 + t0;
1076
        tmp1[12] = t2 + t1 - t0;
1077
        tmp1[ 8] = t3 - t1 - t0;
1078
#endif
1079
    }
1080

    
1081
    i = 0;
1082
    for(j=0;j<4;j++) {
1083
        t0 = tmp[i];
1084
        t1 = tmp[i + 2];
1085
        s0 = t1 + t0;
1086
        s2 = t1 - t0;
1087

    
1088
        t2 = tmp[i + 1];
1089
        t3 = tmp[i + 3];
1090
        s1 = MULL(t3 + t2, icos36[j]);
1091
        s3 = MULL(t3 - t2, icos36[8 - j]);
1092
        
1093
        t0 = (s0 + s1) << 5;
1094
        t1 = (s0 - s1) << 5;
1095
        out[(9 + j)*SBLIMIT] = -MULH(t1, win[9 + j]) + buf[9 + j];
1096
        out[(8 - j)*SBLIMIT] =  MULH(t1, win[8 - j]) + buf[8 - j];
1097
        buf[9 + j] = MULH(t0, win[18 + 9 + j]);
1098
        buf[8 - j] = MULH(t0, win[18 + 8 - j]);
1099
        
1100
        t0 = (s2 + s3) << 5;
1101
        t1 = (s2 - s3) << 5;
1102
        out[(9 + 8 - j)*SBLIMIT] = -MULH(t1, win[9 + 8 - j]) + buf[9 + 8 - j];
1103
        out[(        j)*SBLIMIT] =  MULH(t1, win[        j]) + buf[        j];
1104
        buf[9 + 8 - j] = MULH(t0, win[18 + 9 + 8 - j]);
1105
        buf[      + j] = MULH(t0, win[18         + j]);
1106
        i += 4;
1107
    }
1108

    
1109
    s0 = tmp[16];
1110
    s1 = MULL(tmp[17], icos36[4]);
1111
    t0 = (s0 + s1) << 5;
1112
    t1 = (s0 - s1) << 5;
1113
    out[(9 + 4)*SBLIMIT] = -MULH(t1, win[9 + 4]) + buf[9 + 4];
1114
    out[(8 - 4)*SBLIMIT] =  MULH(t1, win[8 - 4]) + buf[8 - 4];
1115
    buf[9 + 4] = MULH(t0, win[18 + 9 + 4]);
1116
    buf[8 - 4] = MULH(t0, win[18 + 8 - 4]);
1117
}
1118

    
1119
/* header decoding. MUST check the header before because no
1120
   consistency check is done there. Return 1 if free format found and
1121
   that the frame size must be computed externally */
1122
static int decode_header(MPADecodeContext *s, uint32_t header)
1123
{
1124
    int sample_rate, frame_size, mpeg25, padding;
1125
    int sample_rate_index, bitrate_index;
1126
    if (header & (1<<20)) {
1127
        s->lsf = (header & (1<<19)) ? 0 : 1;
1128
        mpeg25 = 0;
1129
    } else {
1130
        s->lsf = 1;
1131
        mpeg25 = 1;
1132
    }
1133
    
1134
    s->layer = 4 - ((header >> 17) & 3);
1135
    /* extract frequency */
1136
    sample_rate_index = (header >> 10) & 3;
1137
    sample_rate = mpa_freq_tab[sample_rate_index] >> (s->lsf + mpeg25);
1138
    sample_rate_index += 3 * (s->lsf + mpeg25);
1139
    s->sample_rate_index = sample_rate_index;
1140
    s->error_protection = ((header >> 16) & 1) ^ 1;
1141
    s->sample_rate = sample_rate;
1142

    
1143
    bitrate_index = (header >> 12) & 0xf;
1144
    padding = (header >> 9) & 1;
1145
    //extension = (header >> 8) & 1;
1146
    s->mode = (header >> 6) & 3;
1147
    s->mode_ext = (header >> 4) & 3;
1148
    //copyright = (header >> 3) & 1;
1149
    //original = (header >> 2) & 1;
1150
    //emphasis = header & 3;
1151

    
1152
    if (s->mode == MPA_MONO)
1153
        s->nb_channels = 1;
1154
    else
1155
        s->nb_channels = 2;
1156
    
1157
    if (bitrate_index != 0) {
1158
        frame_size = mpa_bitrate_tab[s->lsf][s->layer - 1][bitrate_index];
1159
        s->bit_rate = frame_size * 1000;
1160
        switch(s->layer) {
1161
        case 1:
1162
            frame_size = (frame_size * 12000) / sample_rate;
1163
            frame_size = (frame_size + padding) * 4;
1164
            break;
1165
        case 2:
1166
            frame_size = (frame_size * 144000) / sample_rate;
1167
            frame_size += padding;
1168
            break;
1169
        default:
1170
        case 3:
1171
            frame_size = (frame_size * 144000) / (sample_rate << s->lsf);
1172
            frame_size += padding;
1173
            break;
1174
        }
1175
        s->frame_size = frame_size;
1176
    } else {
1177
        /* if no frame size computed, signal it */
1178
        if (!s->free_format_frame_size)
1179
            return 1;
1180
        /* free format: compute bitrate and real frame size from the
1181
           frame size we extracted by reading the bitstream */
1182
        s->frame_size = s->free_format_frame_size;
1183
        switch(s->layer) {
1184
        case 1:
1185
            s->frame_size += padding  * 4;
1186
            s->bit_rate = (s->frame_size * sample_rate) / 48000;
1187
            break;
1188
        case 2:
1189
            s->frame_size += padding;
1190
            s->bit_rate = (s->frame_size * sample_rate) / 144000;
1191
            break;
1192
        default:
1193
        case 3:
1194
            s->frame_size += padding;
1195
            s->bit_rate = (s->frame_size * (sample_rate << s->lsf)) / 144000;
1196
            break;
1197
        }
1198
    }
1199
    
1200
#if defined(DEBUG)
1201
    printf("layer%d, %d Hz, %d kbits/s, ",
1202
           s->layer, s->sample_rate, s->bit_rate);
1203
    if (s->nb_channels == 2) {
1204
        if (s->layer == 3) {
1205
            if (s->mode_ext & MODE_EXT_MS_STEREO)
1206
                printf("ms-");
1207
            if (s->mode_ext & MODE_EXT_I_STEREO)
1208
                printf("i-");
1209
        }
1210
        printf("stereo");
1211
    } else {
1212
        printf("mono");
1213
    }
1214
    printf("\n");
1215
#endif
1216
    return 0;
1217
}
1218

    
1219
/* useful helper to get mpeg audio stream infos. Return -1 if error in
1220
   header, otherwise the coded frame size in bytes */
1221
int mpa_decode_header(AVCodecContext *avctx, uint32_t head)
1222
{
1223
    MPADecodeContext s1, *s = &s1;
1224
    memset( s, 0, sizeof(MPADecodeContext) );
1225

    
1226
    if (ff_mpa_check_header(head) != 0)
1227
        return -1;
1228

    
1229
    if (decode_header(s, head) != 0) {
1230
        return -1;
1231
    }
1232

    
1233
    switch(s->layer) {
1234
    case 1:
1235
        avctx->frame_size = 384;
1236
        break;
1237
    case 2:
1238
        avctx->frame_size = 1152;
1239
        break;
1240
    default:
1241
    case 3:
1242
        if (s->lsf)
1243
            avctx->frame_size = 576;
1244
        else
1245
            avctx->frame_size = 1152;
1246
        break;
1247
    }
1248

    
1249
    avctx->sample_rate = s->sample_rate;
1250
    avctx->channels = s->nb_channels;
1251
    avctx->bit_rate = s->bit_rate;
1252
    avctx->sub_id = s->layer;
1253
    return s->frame_size;
1254
}
1255

    
1256
/* return the number of decoded frames */
1257
static int mp_decode_layer1(MPADecodeContext *s)
1258
{
1259
    int bound, i, v, n, ch, j, mant;
1260
    uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
1261
    uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
1262

    
1263
    if (s->mode == MPA_JSTEREO) 
1264
        bound = (s->mode_ext + 1) * 4;
1265
    else
1266
        bound = SBLIMIT;
1267

    
1268
    /* allocation bits */
1269
    for(i=0;i<bound;i++) {
1270
        for(ch=0;ch<s->nb_channels;ch++) {
1271
            allocation[ch][i] = get_bits(&s->gb, 4);
1272
        }
1273
    }
1274
    for(i=bound;i<SBLIMIT;i++) {
1275
        allocation[0][i] = get_bits(&s->gb, 4);
1276
    }
1277

    
1278
    /* scale factors */
1279
    for(i=0;i<bound;i++) {
1280
        for(ch=0;ch<s->nb_channels;ch++) {
1281
            if (allocation[ch][i])
1282
                scale_factors[ch][i] = get_bits(&s->gb, 6);
1283
        }
1284
    }
1285
    for(i=bound;i<SBLIMIT;i++) {
1286
        if (allocation[0][i]) {
1287
            scale_factors[0][i] = get_bits(&s->gb, 6);
1288
            scale_factors[1][i] = get_bits(&s->gb, 6);
1289
        }
1290
    }
1291
    
1292
    /* compute samples */
1293
    for(j=0;j<12;j++) {
1294
        for(i=0;i<bound;i++) {
1295
            for(ch=0;ch<s->nb_channels;ch++) {
1296
                n = allocation[ch][i];
1297
                if (n) {
1298
                    mant = get_bits(&s->gb, n + 1);
1299
                    v = l1_unscale(n, mant, scale_factors[ch][i]);
1300
                } else {
1301
                    v = 0;
1302
                }
1303
                s->sb_samples[ch][j][i] = v;
1304
            }
1305
        }
1306
        for(i=bound;i<SBLIMIT;i++) {
1307
            n = allocation[0][i];
1308
            if (n) {
1309
                mant = get_bits(&s->gb, n + 1);
1310
                v = l1_unscale(n, mant, scale_factors[0][i]);
1311
                s->sb_samples[0][j][i] = v;
1312
                v = l1_unscale(n, mant, scale_factors[1][i]);
1313
                s->sb_samples[1][j][i] = v;
1314
            } else {
1315
                s->sb_samples[0][j][i] = 0;
1316
                s->sb_samples[1][j][i] = 0;
1317
            }
1318
        }
1319
    }
1320
    return 12;
1321
}
1322

    
1323
/* bitrate is in kb/s */
1324
int l2_select_table(int bitrate, int nb_channels, int freq, int lsf)
1325
{
1326
    int ch_bitrate, table;
1327
    
1328
    ch_bitrate = bitrate / nb_channels;
1329
    if (!lsf) {
1330
        if ((freq == 48000 && ch_bitrate >= 56) ||
1331
            (ch_bitrate >= 56 && ch_bitrate <= 80)) 
1332
            table = 0;
1333
        else if (freq != 48000 && ch_bitrate >= 96) 
1334
            table = 1;
1335
        else if (freq != 32000 && ch_bitrate <= 48) 
1336
            table = 2;
1337
        else 
1338
            table = 3;
1339
    } else {
1340
        table = 4;
1341
    }
1342
    return table;
1343
}
1344

    
1345
static int mp_decode_layer2(MPADecodeContext *s)
1346
{
1347
    int sblimit; /* number of used subbands */
1348
    const unsigned char *alloc_table;
1349
    int table, bit_alloc_bits, i, j, ch, bound, v;
1350
    unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
1351
    unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
1352
    unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
1353
    int scale, qindex, bits, steps, k, l, m, b;
1354

    
1355
    /* select decoding table */
1356
    table = l2_select_table(s->bit_rate / 1000, s->nb_channels, 
1357
                            s->sample_rate, s->lsf);
1358
    sblimit = sblimit_table[table];
1359
    alloc_table = alloc_tables[table];
1360

    
1361
    if (s->mode == MPA_JSTEREO) 
1362
        bound = (s->mode_ext + 1) * 4;
1363
    else
1364
        bound = sblimit;
1365

    
1366
    dprintf("bound=%d sblimit=%d\n", bound, sblimit);
1367

    
1368
    /* sanity check */
1369
    if( bound > sblimit ) bound = sblimit;
1370

    
1371
    /* parse bit allocation */
1372
    j = 0;
1373
    for(i=0;i<bound;i++) {
1374
        bit_alloc_bits = alloc_table[j];
1375
        for(ch=0;ch<s->nb_channels;ch++) {
1376
            bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
1377
        }
1378
        j += 1 << bit_alloc_bits;
1379
    }
1380
    for(i=bound;i<sblimit;i++) {
1381
        bit_alloc_bits = alloc_table[j];
1382
        v = get_bits(&s->gb, bit_alloc_bits);
1383
        bit_alloc[0][i] = v;
1384
        bit_alloc[1][i] = v;
1385
        j += 1 << bit_alloc_bits;
1386
    }
1387

    
1388
#ifdef DEBUG
1389
    {
1390
        for(ch=0;ch<s->nb_channels;ch++) {
1391
            for(i=0;i<sblimit;i++)
1392
                printf(" %d", bit_alloc[ch][i]);
1393
            printf("\n");
1394
        }
1395
    }
1396
#endif
1397

    
1398
    /* scale codes */
1399
    for(i=0;i<sblimit;i++) {
1400
        for(ch=0;ch<s->nb_channels;ch++) {
1401
            if (bit_alloc[ch][i]) 
1402
                scale_code[ch][i] = get_bits(&s->gb, 2);
1403
        }
1404
    }
1405
    
1406
    /* scale factors */
1407
    for(i=0;i<sblimit;i++) {
1408
        for(ch=0;ch<s->nb_channels;ch++) {
1409
            if (bit_alloc[ch][i]) {
1410
                sf = scale_factors[ch][i];
1411
                switch(scale_code[ch][i]) {
1412
                default:
1413
                case 0:
1414
                    sf[0] = get_bits(&s->gb, 6);
1415
                    sf[1] = get_bits(&s->gb, 6);
1416
                    sf[2] = get_bits(&s->gb, 6);
1417
                    break;
1418
                case 2:
1419
                    sf[0] = get_bits(&s->gb, 6);
1420
                    sf[1] = sf[0];
1421
                    sf[2] = sf[0];
1422
                    break;
1423
                case 1:
1424
                    sf[0] = get_bits(&s->gb, 6);
1425
                    sf[2] = get_bits(&s->gb, 6);
1426
                    sf[1] = sf[0];
1427
                    break;
1428
                case 3:
1429
                    sf[0] = get_bits(&s->gb, 6);
1430
                    sf[2] = get_bits(&s->gb, 6);
1431
                    sf[1] = sf[2];
1432
                    break;
1433
                }
1434
            }
1435
        }
1436
    }
1437

    
1438
#ifdef DEBUG
1439
    for(ch=0;ch<s->nb_channels;ch++) {
1440
        for(i=0;i<sblimit;i++) {
1441
            if (bit_alloc[ch][i]) {
1442
                sf = scale_factors[ch][i];
1443
                printf(" %d %d %d", sf[0], sf[1], sf[2]);
1444
            } else {
1445
                printf(" -");
1446
            }
1447
        }
1448
        printf("\n");
1449
    }
1450
#endif
1451

    
1452
    /* samples */
1453
    for(k=0;k<3;k++) {
1454
        for(l=0;l<12;l+=3) {
1455
            j = 0;
1456
            for(i=0;i<bound;i++) {
1457
                bit_alloc_bits = alloc_table[j];
1458
                for(ch=0;ch<s->nb_channels;ch++) {
1459
                    b = bit_alloc[ch][i];
1460
                    if (b) {
1461
                        scale = scale_factors[ch][i][k];
1462
                        qindex = alloc_table[j+b];
1463
                        bits = quant_bits[qindex];
1464
                        if (bits < 0) {
1465
                            /* 3 values at the same time */
1466
                            v = get_bits(&s->gb, -bits);
1467
                            steps = quant_steps[qindex];
1468
                            s->sb_samples[ch][k * 12 + l + 0][i] = 
1469
                                l2_unscale_group(steps, v % steps, scale);
1470
                            v = v / steps;
1471
                            s->sb_samples[ch][k * 12 + l + 1][i] = 
1472
                                l2_unscale_group(steps, v % steps, scale);
1473
                            v = v / steps;
1474
                            s->sb_samples[ch][k * 12 + l + 2][i] = 
1475
                                l2_unscale_group(steps, v, scale);
1476
                        } else {
1477
                            for(m=0;m<3;m++) {
1478
                                v = get_bits(&s->gb, bits);
1479
                                v = l1_unscale(bits - 1, v, scale);
1480
                                s->sb_samples[ch][k * 12 + l + m][i] = v;
1481
                            }
1482
                        }
1483
                    } else {
1484
                        s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1485
                        s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1486
                        s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1487
                    }
1488
                }
1489
                /* next subband in alloc table */
1490
                j += 1 << bit_alloc_bits; 
1491
            }
1492
            /* XXX: find a way to avoid this duplication of code */
1493
            for(i=bound;i<sblimit;i++) {
1494
                bit_alloc_bits = alloc_table[j];
1495
                b = bit_alloc[0][i];
1496
                if (b) {
1497
                    int mant, scale0, scale1;
1498
                    scale0 = scale_factors[0][i][k];
1499
                    scale1 = scale_factors[1][i][k];
1500
                    qindex = alloc_table[j+b];
1501
                    bits = quant_bits[qindex];
1502
                    if (bits < 0) {
1503
                        /* 3 values at the same time */
1504
                        v = get_bits(&s->gb, -bits);
1505
                        steps = quant_steps[qindex];
1506
                        mant = v % steps;
1507
                        v = v / steps;
1508
                        s->sb_samples[0][k * 12 + l + 0][i] = 
1509
                            l2_unscale_group(steps, mant, scale0);
1510
                        s->sb_samples[1][k * 12 + l + 0][i] = 
1511
                            l2_unscale_group(steps, mant, scale1);
1512
                        mant = v % steps;
1513
                        v = v / steps;
1514
                        s->sb_samples[0][k * 12 + l + 1][i] = 
1515
                            l2_unscale_group(steps, mant, scale0);
1516
                        s->sb_samples[1][k * 12 + l + 1][i] = 
1517
                            l2_unscale_group(steps, mant, scale1);
1518
                        s->sb_samples[0][k * 12 + l + 2][i] = 
1519
                            l2_unscale_group(steps, v, scale0);
1520
                        s->sb_samples[1][k * 12 + l + 2][i] = 
1521
                            l2_unscale_group(steps, v, scale1);
1522
                    } else {
1523
                        for(m=0;m<3;m++) {
1524
                            mant = get_bits(&s->gb, bits);
1525
                            s->sb_samples[0][k * 12 + l + m][i] = 
1526
                                l1_unscale(bits - 1, mant, scale0);
1527
                            s->sb_samples[1][k * 12 + l + m][i] = 
1528
                                l1_unscale(bits - 1, mant, scale1);
1529
                        }
1530
                    }
1531
                } else {
1532
                    s->sb_samples[0][k * 12 + l + 0][i] = 0;
1533
                    s->sb_samples[0][k * 12 + l + 1][i] = 0;
1534
                    s->sb_samples[0][k * 12 + l + 2][i] = 0;
1535
                    s->sb_samples[1][k * 12 + l + 0][i] = 0;
1536
                    s->sb_samples[1][k * 12 + l + 1][i] = 0;
1537
                    s->sb_samples[1][k * 12 + l + 2][i] = 0;
1538
                }
1539
                /* next subband in alloc table */
1540
                j += 1 << bit_alloc_bits; 
1541
            }
1542
            /* fill remaining samples to zero */
1543
            for(i=sblimit;i<SBLIMIT;i++) {
1544
                for(ch=0;ch<s->nb_channels;ch++) {
1545
                    s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1546
                    s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1547
                    s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1548
                }
1549
            }
1550
        }
1551
    }
1552
    return 3 * 12;
1553
}
1554

    
1555
/*
1556
 * Seek back in the stream for backstep bytes (at most 511 bytes)
1557
 */
1558
static void seek_to_maindata(MPADecodeContext *s, unsigned int backstep)
1559
{
1560
    uint8_t *ptr;
1561

    
1562
    /* compute current position in stream */
1563
    ptr = (uint8_t *)(s->gb.buffer + (get_bits_count(&s->gb)>>3));
1564

    
1565
    /* copy old data before current one */
1566
    ptr -= backstep;
1567
    memcpy(ptr, s->inbuf1[s->inbuf_index ^ 1] + 
1568
           BACKSTEP_SIZE + s->old_frame_size - backstep, backstep);
1569
    /* init get bits again */
1570
    init_get_bits(&s->gb, ptr, (s->frame_size + backstep)*8);
1571

    
1572
    /* prepare next buffer */
1573
    s->inbuf_index ^= 1;
1574
    s->inbuf = &s->inbuf1[s->inbuf_index][BACKSTEP_SIZE];
1575
    s->old_frame_size = s->frame_size;
1576
}
1577

    
1578
static inline void lsf_sf_expand(int *slen,
1579
                                 int sf, int n1, int n2, int n3)
1580
{
1581
    if (n3) {
1582
        slen[3] = sf % n3;
1583
        sf /= n3;
1584
    } else {
1585
        slen[3] = 0;
1586
    }
1587
    if (n2) {
1588
        slen[2] = sf % n2;
1589
        sf /= n2;
1590
    } else {
1591
        slen[2] = 0;
1592
    }
1593
    slen[1] = sf % n1;
1594
    sf /= n1;
1595
    slen[0] = sf;
1596
}
1597

    
1598
static void exponents_from_scale_factors(MPADecodeContext *s, 
1599
                                         GranuleDef *g,
1600
                                         int16_t *exponents)
1601
{
1602
    const uint8_t *bstab, *pretab;
1603
    int len, i, j, k, l, v0, shift, gain, gains[3];
1604
    int16_t *exp_ptr;
1605

    
1606
    exp_ptr = exponents;
1607
    gain = g->global_gain - 210;
1608
    shift = g->scalefac_scale + 1;
1609

    
1610
    bstab = band_size_long[s->sample_rate_index];
1611
    pretab = mpa_pretab[g->preflag];
1612
    for(i=0;i<g->long_end;i++) {
1613
        v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift);
1614
        len = bstab[i];
1615
        for(j=len;j>0;j--)
1616
            *exp_ptr++ = v0;
1617
    }
1618

    
1619
    if (g->short_start < 13) {
1620
        bstab = band_size_short[s->sample_rate_index];
1621
        gains[0] = gain - (g->subblock_gain[0] << 3);
1622
        gains[1] = gain - (g->subblock_gain[1] << 3);
1623
        gains[2] = gain - (g->subblock_gain[2] << 3);
1624
        k = g->long_end;
1625
        for(i=g->short_start;i<13;i++) {
1626
            len = bstab[i];
1627
            for(l=0;l<3;l++) {
1628
                v0 = gains[l] - (g->scale_factors[k++] << shift);
1629
                for(j=len;j>0;j--)
1630
                *exp_ptr++ = v0;
1631
            }
1632
        }
1633
    }
1634
}
1635

    
1636
/* handle n = 0 too */
1637
static inline int get_bitsz(GetBitContext *s, int n)
1638
{
1639
    if (n == 0)
1640
        return 0;
1641
    else
1642
        return get_bits(s, n);
1643
}
1644

    
1645
static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
1646
                          int16_t *exponents, int end_pos)
1647
{
1648
    int s_index;
1649
    int linbits, code, x, y, l, v, i, j, k, pos;
1650
    GetBitContext last_gb;
1651
    VLC *vlc;
1652
    uint8_t *code_table;
1653

    
1654
    /* low frequencies (called big values) */
1655
    s_index = 0;
1656
    for(i=0;i<3;i++) {
1657
        j = g->region_size[i];
1658
        if (j == 0)
1659
            continue;
1660
        /* select vlc table */
1661
        k = g->table_select[i];
1662
        l = mpa_huff_data[k][0];
1663
        linbits = mpa_huff_data[k][1];
1664
        vlc = &huff_vlc[l];
1665
        code_table = huff_code_table[l];
1666

    
1667
        /* read huffcode and compute each couple */
1668
        for(;j>0;j--) {
1669
            if (get_bits_count(&s->gb) >= end_pos)
1670
                break;
1671
            if (code_table) {
1672
                code = get_vlc(&s->gb, vlc);
1673
                if (code < 0)
1674
                    return -1;
1675
                y = code_table[code];
1676
                x = y >> 4;
1677
                y = y & 0x0f;
1678
            } else {
1679
                x = 0;
1680
                y = 0;
1681
            }
1682
            dprintf("region=%d n=%d x=%d y=%d exp=%d\n", 
1683
                    i, g->region_size[i] - j, x, y, exponents[s_index]);
1684
            if (x) {
1685
                if (x == 15)
1686
                    x += get_bitsz(&s->gb, linbits);
1687
                v = l3_unscale(x, exponents[s_index]);
1688
                if (get_bits1(&s->gb))
1689
                    v = -v;
1690
            } else {
1691
                v = 0;
1692
            }
1693
            g->sb_hybrid[s_index++] = v;
1694
            if (y) {
1695
                if (y == 15)
1696
                    y += get_bitsz(&s->gb, linbits);
1697
                v = l3_unscale(y, exponents[s_index]);
1698
                if (get_bits1(&s->gb))
1699
                    v = -v;
1700
            } else {
1701
                v = 0;
1702
            }
1703
            g->sb_hybrid[s_index++] = v;
1704
        }
1705
    }
1706
            
1707
    /* high frequencies */
1708
    vlc = &huff_quad_vlc[g->count1table_select];
1709
    last_gb.buffer = NULL;
1710
    while (s_index <= 572) {
1711
        pos = get_bits_count(&s->gb);
1712
        if (pos >= end_pos) {
1713
            if (pos > end_pos && last_gb.buffer != NULL) {
1714
                /* some encoders generate an incorrect size for this
1715
                   part. We must go back into the data */
1716
                s_index -= 4;
1717
                s->gb = last_gb;
1718
            }
1719
            break;
1720
        }
1721
        last_gb= s->gb;
1722

    
1723
        code = get_vlc(&s->gb, vlc);
1724
        dprintf("t=%d code=%d\n", g->count1table_select, code);
1725
        if (code < 0)
1726
            return -1;
1727
        for(i=0;i<4;i++) {
1728
            if (code & (8 >> i)) {
1729
                /* non zero value. Could use a hand coded function for
1730
                   'one' value */
1731
                v = l3_unscale(1, exponents[s_index]);
1732
                if(get_bits1(&s->gb))
1733
                    v = -v;
1734
            } else {
1735
                v = 0;
1736
            }
1737
            g->sb_hybrid[s_index++] = v;
1738
        }
1739
    }
1740
    while (s_index < 576)
1741
        g->sb_hybrid[s_index++] = 0;
1742
    return 0;
1743
}
1744

    
1745
/* Reorder short blocks from bitstream order to interleaved order. It
1746
   would be faster to do it in parsing, but the code would be far more
1747
   complicated */
1748
static void reorder_block(MPADecodeContext *s, GranuleDef *g)
1749
{
1750
    int i, j, k, len;
1751
    int32_t *ptr, *dst, *ptr1;
1752
    int32_t tmp[576];
1753

    
1754
    if (g->block_type != 2)
1755
        return;
1756

    
1757
    if (g->switch_point) {
1758
        if (s->sample_rate_index != 8) {
1759
            ptr = g->sb_hybrid + 36;
1760
        } else {
1761
            ptr = g->sb_hybrid + 48;
1762
        }
1763
    } else {
1764
        ptr = g->sb_hybrid;
1765
    }
1766
    
1767
    for(i=g->short_start;i<13;i++) {
1768
        len = band_size_short[s->sample_rate_index][i];
1769
        ptr1 = ptr;
1770
        for(k=0;k<3;k++) {
1771
            dst = tmp + k;
1772
            for(j=len;j>0;j--) {
1773
                *dst = *ptr++;
1774
                dst += 3;
1775
            }
1776
        }
1777
        memcpy(ptr1, tmp, len * 3 * sizeof(int32_t));
1778
    }
1779
}
1780

    
1781
#define ISQRT2 FIXR(0.70710678118654752440)
1782

    
1783
static void compute_stereo(MPADecodeContext *s,
1784
                           GranuleDef *g0, GranuleDef *g1)
1785
{
1786
    int i, j, k, l;
1787
    int32_t v1, v2;
1788
    int sf_max, tmp0, tmp1, sf, len, non_zero_found;
1789
    int32_t (*is_tab)[16];
1790
    int32_t *tab0, *tab1;
1791
    int non_zero_found_short[3];
1792

    
1793
    /* intensity stereo */
1794
    if (s->mode_ext & MODE_EXT_I_STEREO) {
1795
        if (!s->lsf) {
1796
            is_tab = is_table;
1797
            sf_max = 7;
1798
        } else {
1799
            is_tab = is_table_lsf[g1->scalefac_compress & 1];
1800
            sf_max = 16;
1801
        }
1802
            
1803
        tab0 = g0->sb_hybrid + 576;
1804
        tab1 = g1->sb_hybrid + 576;
1805

    
1806
        non_zero_found_short[0] = 0;
1807
        non_zero_found_short[1] = 0;
1808
        non_zero_found_short[2] = 0;
1809
        k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1810
        for(i = 12;i >= g1->short_start;i--) {
1811
            /* for last band, use previous scale factor */
1812
            if (i != 11)
1813
                k -= 3;
1814
            len = band_size_short[s->sample_rate_index][i];
1815
            for(l=2;l>=0;l--) {
1816
                tab0 -= len;
1817
                tab1 -= len;
1818
                if (!non_zero_found_short[l]) {
1819
                    /* test if non zero band. if so, stop doing i-stereo */
1820
                    for(j=0;j<len;j++) {
1821
                        if (tab1[j] != 0) {
1822
                            non_zero_found_short[l] = 1;
1823
                            goto found1;
1824
                        }
1825
                    }
1826
                    sf = g1->scale_factors[k + l];
1827
                    if (sf >= sf_max)
1828
                        goto found1;
1829

    
1830
                    v1 = is_tab[0][sf];
1831
                    v2 = is_tab[1][sf];
1832
                    for(j=0;j<len;j++) {
1833
                        tmp0 = tab0[j];
1834
                        tab0[j] = MULL(tmp0, v1);
1835
                        tab1[j] = MULL(tmp0, v2);
1836
                    }
1837
                } else {
1838
                found1:
1839
                    if (s->mode_ext & MODE_EXT_MS_STEREO) {
1840
                        /* lower part of the spectrum : do ms stereo
1841
                           if enabled */
1842
                        for(j=0;j<len;j++) {
1843
                            tmp0 = tab0[j];
1844
                            tmp1 = tab1[j];
1845
                            tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
1846
                            tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
1847
                        }
1848
                    }
1849
                }
1850
            }
1851
        }
1852

    
1853
        non_zero_found = non_zero_found_short[0] | 
1854
            non_zero_found_short[1] | 
1855
            non_zero_found_short[2];
1856

    
1857
        for(i = g1->long_end - 1;i >= 0;i--) {
1858
            len = band_size_long[s->sample_rate_index][i];
1859
            tab0 -= len;
1860
            tab1 -= len;
1861
            /* test if non zero band. if so, stop doing i-stereo */
1862
            if (!non_zero_found) {
1863
                for(j=0;j<len;j++) {
1864
                    if (tab1[j] != 0) {
1865
                        non_zero_found = 1;
1866
                        goto found2;
1867
                    }
1868
                }
1869
                /* for last band, use previous scale factor */
1870
                k = (i == 21) ? 20 : i;
1871
                sf = g1->scale_factors[k];
1872
                if (sf >= sf_max)
1873
                    goto found2;
1874
                v1 = is_tab[0][sf];
1875
                v2 = is_tab[1][sf];
1876
                for(j=0;j<len;j++) {
1877
                    tmp0 = tab0[j];
1878
                    tab0[j] = MULL(tmp0, v1);
1879
                    tab1[j] = MULL(tmp0, v2);
1880
                }
1881
            } else {
1882
            found2:
1883
                if (s->mode_ext & MODE_EXT_MS_STEREO) {
1884
                    /* lower part of the spectrum : do ms stereo
1885
                       if enabled */
1886
                    for(j=0;j<len;j++) {
1887
                        tmp0 = tab0[j];
1888
                        tmp1 = tab1[j];
1889
                        tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
1890
                        tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
1891
                    }
1892
                }
1893
            }
1894
        }
1895
    } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1896
        /* ms stereo ONLY */
1897
        /* NOTE: the 1/sqrt(2) normalization factor is included in the
1898
           global gain */
1899
        tab0 = g0->sb_hybrid;
1900
        tab1 = g1->sb_hybrid;
1901
        for(i=0;i<576;i++) {
1902
            tmp0 = tab0[i];
1903
            tmp1 = tab1[i];
1904
            tab0[i] = tmp0 + tmp1;
1905
            tab1[i] = tmp0 - tmp1;
1906
        }
1907
    }
1908
}
1909

    
1910
static void compute_antialias_integer(MPADecodeContext *s,
1911
                              GranuleDef *g)
1912
{
1913
    int32_t *ptr, *csa;
1914
    int n, i;
1915

    
1916
    /* we antialias only "long" bands */
1917
    if (g->block_type == 2) {
1918
        if (!g->switch_point)
1919
            return;
1920
        /* XXX: check this for 8000Hz case */
1921
        n = 1;
1922
    } else {
1923
        n = SBLIMIT - 1;
1924
    }
1925
    
1926
    ptr = g->sb_hybrid + 18;
1927
    for(i = n;i > 0;i--) {
1928
        int tmp0, tmp1, tmp2;
1929
        csa = &csa_table[0][0];
1930
#define INT_AA(j) \
1931
            tmp0 = 4*(ptr[-1-j]);\
1932
            tmp1 = 4*(ptr[   j]);\
1933
            tmp2= MULH(tmp0 + tmp1, csa[0+4*j]);\
1934
            ptr[-1-j] = tmp2 - MULH(tmp1, csa[2+4*j]);\
1935
            ptr[   j] = tmp2 + MULH(tmp0, csa[3+4*j]);
1936

    
1937
        INT_AA(0)
1938
        INT_AA(1)
1939
        INT_AA(2)
1940
        INT_AA(3)
1941
        INT_AA(4)
1942
        INT_AA(5)
1943
        INT_AA(6)
1944
        INT_AA(7)
1945
            
1946
        ptr += 18;       
1947
    }
1948
}
1949

    
1950
static void compute_antialias_float(MPADecodeContext *s,
1951
                              GranuleDef *g)
1952
{
1953
    int32_t *ptr;
1954
    int n, i;
1955

    
1956
    /* we antialias only "long" bands */
1957
    if (g->block_type == 2) {
1958
        if (!g->switch_point)
1959
            return;
1960
        /* XXX: check this for 8000Hz case */
1961
        n = 1;
1962
    } else {
1963
        n = SBLIMIT - 1;
1964
    }
1965
    
1966
    ptr = g->sb_hybrid + 18;
1967
    for(i = n;i > 0;i--) {
1968
        float tmp0, tmp1;
1969
        float *csa = &csa_table_float[0][0];       
1970
#define FLOAT_AA(j)\
1971
        tmp0= ptr[-1-j];\
1972
        tmp1= ptr[   j];\
1973
        ptr[-1-j] = lrintf(tmp0 * csa[0+4*j] - tmp1 * csa[1+4*j]);\
1974
        ptr[   j] = lrintf(tmp0 * csa[1+4*j] + tmp1 * csa[0+4*j]);
1975
        
1976
        FLOAT_AA(0)
1977
        FLOAT_AA(1)
1978
        FLOAT_AA(2)
1979
        FLOAT_AA(3)
1980
        FLOAT_AA(4)
1981
        FLOAT_AA(5)
1982
        FLOAT_AA(6)
1983
        FLOAT_AA(7)
1984

    
1985
        ptr += 18;       
1986
    }
1987
}
1988

    
1989
static void compute_imdct(MPADecodeContext *s,
1990
                          GranuleDef *g, 
1991
                          int32_t *sb_samples,
1992
                          int32_t *mdct_buf)
1993
{
1994
    int32_t *ptr, *win, *win1, *buf, *buf2, *out_ptr, *ptr1;
1995
    int32_t in[6];
1996
    int32_t out[36];
1997
    int32_t out2[12];
1998
    int i, j, k, mdct_long_end, v, sblimit;
1999

    
2000
    /* find last non zero block */
2001
    ptr = g->sb_hybrid + 576;
2002
    ptr1 = g->sb_hybrid + 2 * 18;
2003
    while (ptr >= ptr1) {
2004
        ptr -= 6;
2005
        v = ptr[0] | ptr[1] | ptr[2] | ptr[3] | ptr[4] | ptr[5];
2006
        if (v != 0)
2007
            break;
2008
    }
2009
    sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
2010

    
2011
    if (g->block_type == 2) {
2012
        /* XXX: check for 8000 Hz */
2013
        if (g->switch_point)
2014
            mdct_long_end = 2;
2015
        else
2016
            mdct_long_end = 0;
2017
    } else {
2018
        mdct_long_end = sblimit;
2019
    }
2020

    
2021
    buf = mdct_buf;
2022
    ptr = g->sb_hybrid;
2023
    for(j=0;j<mdct_long_end;j++) {
2024
        /* apply window & overlap with previous buffer */
2025
        out_ptr = sb_samples + j;
2026
        /* select window */
2027
        if (g->switch_point && j < 2)
2028
            win1 = mdct_win[0];
2029
        else
2030
            win1 = mdct_win[g->block_type];
2031
        /* select frequency inversion */
2032
        win = win1 + ((4 * 36) & -(j & 1));
2033
        imdct36(out_ptr, buf, ptr, win);
2034
        out_ptr += 18*SBLIMIT;
2035
        ptr += 18;
2036
        buf += 18;
2037
    }
2038
    for(j=mdct_long_end;j<sblimit;j++) {
2039
        for(i=0;i<6;i++) {
2040
            out[i] = 0;
2041
            out[6 + i] = 0;
2042
            out[30+i] = 0;
2043
        }
2044
        /* select frequency inversion */
2045
        win = mdct_win[2] + ((4 * 36) & -(j & 1));
2046
        buf2 = out + 6;
2047
        for(k=0;k<3;k++) {
2048
            /* reorder input for short mdct */
2049
            ptr1 = ptr + k;
2050
            for(i=0;i<6;i++) {
2051
                in[i] = *ptr1;
2052
                ptr1 += 3;
2053
            }
2054
            imdct12(out2, in);
2055
            /* apply 12 point window and do small overlap */
2056
            for(i=0;i<6;i++) {
2057
                buf2[i] = MULL(out2[i], win[i]) + buf2[i];
2058
                buf2[i + 6] = MULL(out2[i + 6], win[i + 6]);
2059
            }
2060
            buf2 += 6;
2061
        }
2062
        /* overlap */
2063
        out_ptr = sb_samples + j;
2064
        for(i=0;i<18;i++) {
2065
            *out_ptr = out[i] + buf[i];
2066
            buf[i] = out[i + 18];
2067
            out_ptr += SBLIMIT;
2068
        }
2069
        ptr += 18;
2070
        buf += 18;
2071
    }
2072
    /* zero bands */
2073
    for(j=sblimit;j<SBLIMIT;j++) {
2074
        /* overlap */
2075
        out_ptr = sb_samples + j;
2076
        for(i=0;i<18;i++) {
2077
            *out_ptr = buf[i];
2078
            buf[i] = 0;
2079
            out_ptr += SBLIMIT;
2080
        }
2081
        buf += 18;
2082
    }
2083
}
2084

    
2085
#if defined(DEBUG)
2086
void sample_dump(int fnum, int32_t *tab, int n)
2087
{
2088
    static FILE *files[16], *f;
2089
    char buf[512];
2090
    int i;
2091
    int32_t v;
2092
    
2093
    f = files[fnum];
2094
    if (!f) {
2095
        snprintf(buf, sizeof(buf), "/tmp/out%d.%s.pcm", 
2096
                fnum, 
2097
#ifdef USE_HIGHPRECISION
2098
                "hp"
2099
#else
2100
                "lp"
2101
#endif
2102
                );
2103
        f = fopen(buf, "w");
2104
        if (!f)
2105
            return;
2106
        files[fnum] = f;
2107
    }
2108
    
2109
    if (fnum == 0) {
2110
        static int pos = 0;
2111
        av_log(NULL, AV_LOG_DEBUG, "pos=%d\n", pos);
2112
        for(i=0;i<n;i++) {
2113
            av_log(NULL, AV_LOG_DEBUG, " %0.4f", (double)tab[i] / FRAC_ONE);
2114
            if ((i % 18) == 17)
2115
                av_log(NULL, AV_LOG_DEBUG, "\n");
2116
        }
2117
        pos += n;
2118
    }
2119
    for(i=0;i<n;i++) {
2120
        /* normalize to 23 frac bits */
2121
        v = tab[i] << (23 - FRAC_BITS);
2122
        fwrite(&v, 1, sizeof(int32_t), f);
2123
    }
2124
}
2125
#endif
2126

    
2127

    
2128
/* main layer3 decoding function */
2129
static int mp_decode_layer3(MPADecodeContext *s)
2130
{
2131
    int nb_granules, main_data_begin, private_bits;
2132
    int gr, ch, blocksplit_flag, i, j, k, n, bits_pos, bits_left;
2133
    GranuleDef granules[2][2], *g;
2134
    int16_t exponents[576];
2135

    
2136
    /* read side info */
2137
    if (s->lsf) {
2138
        main_data_begin = get_bits(&s->gb, 8);
2139
        if (s->nb_channels == 2)
2140
            private_bits = get_bits(&s->gb, 2);
2141
        else
2142
            private_bits = get_bits(&s->gb, 1);
2143
        nb_granules = 1;
2144
    } else {
2145
        main_data_begin = get_bits(&s->gb, 9);
2146
        if (s->nb_channels == 2)
2147
            private_bits = get_bits(&s->gb, 3);
2148
        else
2149
            private_bits = get_bits(&s->gb, 5);
2150
        nb_granules = 2;
2151
        for(ch=0;ch<s->nb_channels;ch++) {
2152
            granules[ch][0].scfsi = 0; /* all scale factors are transmitted */
2153
            granules[ch][1].scfsi = get_bits(&s->gb, 4);
2154
        }
2155
    }
2156
    
2157
    for(gr=0;gr<nb_granules;gr++) {
2158
        for(ch=0;ch<s->nb_channels;ch++) {
2159
            dprintf("gr=%d ch=%d: side_info\n", gr, ch);
2160
            g = &granules[ch][gr];
2161
            g->part2_3_length = get_bits(&s->gb, 12);
2162
            g->big_values = get_bits(&s->gb, 9);
2163
            g->global_gain = get_bits(&s->gb, 8);
2164
            /* if MS stereo only is selected, we precompute the
2165
               1/sqrt(2) renormalization factor */
2166
            if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) == 
2167
                MODE_EXT_MS_STEREO)
2168
                g->global_gain -= 2;
2169
            if (s->lsf)
2170
                g->scalefac_compress = get_bits(&s->gb, 9);
2171
            else
2172
                g->scalefac_compress = get_bits(&s->gb, 4);
2173
            blocksplit_flag = get_bits(&s->gb, 1);
2174
            if (blocksplit_flag) {
2175
                g->block_type = get_bits(&s->gb, 2);
2176
                if (g->block_type == 0)
2177
                    return -1;
2178
                g->switch_point = get_bits(&s->gb, 1);
2179
                for(i=0;i<2;i++)
2180
                    g->table_select[i] = get_bits(&s->gb, 5);
2181
                for(i=0;i<3;i++) 
2182
                    g->subblock_gain[i] = get_bits(&s->gb, 3);
2183
                /* compute huffman coded region sizes */
2184
                if (g->block_type == 2)
2185
                    g->region_size[0] = (36 / 2);
2186
                else {
2187
                    if (s->sample_rate_index <= 2) 
2188
                        g->region_size[0] = (36 / 2);
2189
                    else if (s->sample_rate_index != 8) 
2190
                        g->region_size[0] = (54 / 2);
2191
                    else
2192
                        g->region_size[0] = (108 / 2);
2193
                }
2194
                g->region_size[1] = (576 / 2);
2195
            } else {
2196
                int region_address1, region_address2, l;
2197
                g->block_type = 0;
2198
                g->switch_point = 0;
2199
                for(i=0;i<3;i++)
2200
                    g->table_select[i] = get_bits(&s->gb, 5);
2201
                /* compute huffman coded region sizes */
2202
                region_address1 = get_bits(&s->gb, 4);
2203
                region_address2 = get_bits(&s->gb, 3);
2204
                dprintf("region1=%d region2=%d\n", 
2205
                        region_address1, region_address2);
2206
                g->region_size[0] = 
2207
                    band_index_long[s->sample_rate_index][region_address1 + 1] >> 1;
2208
                l = region_address1 + region_address2 + 2;
2209
                /* should not overflow */
2210
                if (l > 22)
2211
                    l = 22;
2212
                g->region_size[1] = 
2213
                    band_index_long[s->sample_rate_index][l] >> 1;
2214
            }
2215
            /* convert region offsets to region sizes and truncate
2216
               size to big_values */
2217
            g->region_size[2] = (576 / 2);
2218
            j = 0;
2219
            for(i=0;i<3;i++) {
2220
                k = g->region_size[i];
2221
                if (k > g->big_values)
2222
                    k = g->big_values;
2223
                g->region_size[i] = k - j;
2224
                j = k;
2225
            }
2226

    
2227
            /* compute band indexes */
2228
            if (g->block_type == 2) {
2229
                if (g->switch_point) {
2230
                    /* if switched mode, we handle the 36 first samples as
2231
                       long blocks.  For 8000Hz, we handle the 48 first
2232
                       exponents as long blocks (XXX: check this!) */
2233
                    if (s->sample_rate_index <= 2)
2234
                        g->long_end = 8;
2235
                    else if (s->sample_rate_index != 8)
2236
                        g->long_end = 6;
2237
                    else
2238
                        g->long_end = 4; /* 8000 Hz */
2239
                    
2240
                    if (s->sample_rate_index != 8)
2241
                        g->short_start = 3;
2242
                    else
2243
                        g->short_start = 2; 
2244
                } else {
2245
                    g->long_end = 0;
2246
                    g->short_start = 0;
2247
                }
2248
            } else {
2249
                g->short_start = 13;
2250
                g->long_end = 22;
2251
            }
2252
            
2253
            g->preflag = 0;
2254
            if (!s->lsf)
2255
                g->preflag = get_bits(&s->gb, 1);
2256
            g->scalefac_scale = get_bits(&s->gb, 1);
2257
            g->count1table_select = get_bits(&s->gb, 1);
2258
            dprintf("block_type=%d switch_point=%d\n",
2259
                    g->block_type, g->switch_point);
2260
        }
2261
    }
2262

    
2263
  if (!s->adu_mode) {
2264
    /* now we get bits from the main_data_begin offset */
2265
    dprintf("seekback: %d\n", main_data_begin);
2266
    seek_to_maindata(s, main_data_begin);
2267
  }
2268

    
2269
    for(gr=0;gr<nb_granules;gr++) {
2270
        for(ch=0;ch<s->nb_channels;ch++) {
2271
            g = &granules[ch][gr];
2272
            
2273
            bits_pos = get_bits_count(&s->gb);
2274
            
2275
            if (!s->lsf) {
2276
                uint8_t *sc;
2277
                int slen, slen1, slen2;
2278

    
2279
                /* MPEG1 scale factors */
2280
                slen1 = slen_table[0][g->scalefac_compress];
2281
                slen2 = slen_table[1][g->scalefac_compress];
2282
                dprintf("slen1=%d slen2=%d\n", slen1, slen2);
2283
                if (g->block_type == 2) {
2284
                    n = g->switch_point ? 17 : 18;
2285
                    j = 0;
2286
                    for(i=0;i<n;i++)
2287
                        g->scale_factors[j++] = get_bitsz(&s->gb, slen1);
2288
                    for(i=0;i<18;i++)
2289
                        g->scale_factors[j++] = get_bitsz(&s->gb, slen2);
2290
                    for(i=0;i<3;i++)
2291
                        g->scale_factors[j++] = 0;
2292
                } else {
2293
                    sc = granules[ch][0].scale_factors;
2294
                    j = 0;
2295
                    for(k=0;k<4;k++) {
2296
                        n = (k == 0 ? 6 : 5);
2297
                        if ((g->scfsi & (0x8 >> k)) == 0) {
2298
                            slen = (k < 2) ? slen1 : slen2;
2299
                            for(i=0;i<n;i++)
2300
                                g->scale_factors[j++] = get_bitsz(&s->gb, slen);
2301
                        } else {
2302
                            /* simply copy from last granule */
2303
                            for(i=0;i<n;i++) {
2304
                                g->scale_factors[j] = sc[j];
2305
                                j++;
2306
                            }
2307
                        }
2308
                    }
2309
                    g->scale_factors[j++] = 0;
2310
                }
2311
#if defined(DEBUG)
2312
                {
2313
                    printf("scfsi=%x gr=%d ch=%d scale_factors:\n", 
2314
                           g->scfsi, gr, ch);
2315
                    for(i=0;i<j;i++)
2316
                        printf(" %d", g->scale_factors[i]);
2317
                    printf("\n");
2318
                }
2319
#endif
2320
            } else {
2321
                int tindex, tindex2, slen[4], sl, sf;
2322

    
2323
                /* LSF scale factors */
2324
                if (g->block_type == 2) {
2325
                    tindex = g->switch_point ? 2 : 1;
2326
                } else {
2327
                    tindex = 0;
2328
                }
2329
                sf = g->scalefac_compress;
2330
                if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
2331
                    /* intensity stereo case */
2332
                    sf >>= 1;
2333
                    if (sf < 180) {
2334
                        lsf_sf_expand(slen, sf, 6, 6, 0);
2335
                        tindex2 = 3;
2336
                    } else if (sf < 244) {
2337
                        lsf_sf_expand(slen, sf - 180, 4, 4, 0);
2338
                        tindex2 = 4;
2339
                    } else {
2340
                        lsf_sf_expand(slen, sf - 244, 3, 0, 0);
2341
                        tindex2 = 5;
2342
                    }
2343
                } else {
2344
                    /* normal case */
2345
                    if (sf < 400) {
2346
                        lsf_sf_expand(slen, sf, 5, 4, 4);
2347
                        tindex2 = 0;
2348
                    } else if (sf < 500) {
2349
                        lsf_sf_expand(slen, sf - 400, 5, 4, 0);
2350
                        tindex2 = 1;
2351
                    } else {
2352
                        lsf_sf_expand(slen, sf - 500, 3, 0, 0);
2353
                        tindex2 = 2;
2354
                        g->preflag = 1;
2355
                    }
2356
                }
2357

    
2358
                j = 0;
2359
                for(k=0;k<4;k++) {
2360
                    n = lsf_nsf_table[tindex2][tindex][k];
2361
                    sl = slen[k];
2362
                    for(i=0;i<n;i++)
2363
                        g->scale_factors[j++] = get_bitsz(&s->gb, sl);
2364
                }
2365
                /* XXX: should compute exact size */
2366
                for(;j<40;j++)
2367
                    g->scale_factors[j] = 0;
2368
#if defined(DEBUG)
2369
                {
2370
                    printf("gr=%d ch=%d scale_factors:\n", 
2371
                           gr, ch);
2372
                    for(i=0;i<40;i++)
2373
                        printf(" %d", g->scale_factors[i]);
2374
                    printf("\n");
2375
                }
2376
#endif
2377
            }
2378

    
2379
            exponents_from_scale_factors(s, g, exponents);
2380

    
2381
            /* read Huffman coded residue */
2382
            if (huffman_decode(s, g, exponents,
2383
                               bits_pos + g->part2_3_length) < 0)
2384
                return -1;
2385
#if defined(DEBUG)
2386
            sample_dump(0, g->sb_hybrid, 576);
2387
#endif
2388

    
2389
            /* skip extension bits */
2390
            bits_left = g->part2_3_length - (get_bits_count(&s->gb) - bits_pos);
2391
            if (bits_left < 0) {
2392
                dprintf("bits_left=%d\n", bits_left);
2393
                return -1;
2394
            }
2395
            while (bits_left >= 16) {
2396
                skip_bits(&s->gb, 16);
2397
                bits_left -= 16;
2398
            }
2399
            if (bits_left > 0)
2400
                skip_bits(&s->gb, bits_left);
2401
        } /* ch */
2402

    
2403
        if (s->nb_channels == 2)
2404
            compute_stereo(s, &granules[0][gr], &granules[1][gr]);
2405

    
2406
        for(ch=0;ch<s->nb_channels;ch++) {
2407
            g = &granules[ch][gr];
2408

    
2409
            reorder_block(s, g);
2410
#if defined(DEBUG)
2411
            sample_dump(0, g->sb_hybrid, 576);
2412
#endif
2413
            s->compute_antialias(s, g);
2414
#if defined(DEBUG)
2415
            sample_dump(1, g->sb_hybrid, 576);
2416
#endif
2417
            compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]); 
2418
#if defined(DEBUG)
2419
            sample_dump(2, &s->sb_samples[ch][18 * gr][0], 576);
2420
#endif
2421
        }
2422
    } /* gr */
2423
    return nb_granules * 18;
2424
}
2425

    
2426
static int mp_decode_frame(MPADecodeContext *s, 
2427
                           OUT_INT *samples)
2428
{
2429
    int i, nb_frames, ch;
2430
    OUT_INT *samples_ptr;
2431

    
2432
    init_get_bits(&s->gb, s->inbuf + HEADER_SIZE, 
2433
                  (s->inbuf_ptr - s->inbuf - HEADER_SIZE)*8);
2434
    
2435
    /* skip error protection field */
2436
    if (s->error_protection)
2437
        get_bits(&s->gb, 16);
2438

    
2439
    dprintf("frame %d:\n", s->frame_count);
2440
    switch(s->layer) {
2441
    case 1:
2442
        nb_frames = mp_decode_layer1(s);
2443
        break;
2444
    case 2:
2445
        nb_frames = mp_decode_layer2(s);
2446
        break;
2447
    case 3:
2448
    default:
2449
        nb_frames = mp_decode_layer3(s);
2450
        break;
2451
    }
2452
#if defined(DEBUG)
2453
    for(i=0;i<nb_frames;i++) {
2454
        for(ch=0;ch<s->nb_channels;ch++) {
2455
            int j;
2456
            printf("%d-%d:", i, ch);
2457
            for(j=0;j<SBLIMIT;j++)
2458
                printf(" %0.6f", (double)s->sb_samples[ch][i][j] / FRAC_ONE);
2459
            printf("\n");
2460
        }
2461
    }
2462
#endif
2463
    /* apply the synthesis filter */
2464
    for(ch=0;ch<s->nb_channels;ch++) {
2465
        samples_ptr = samples + ch;
2466
        for(i=0;i<nb_frames;i++) {
2467
            ff_mpa_synth_filter(s->synth_buf[ch], &(s->synth_buf_offset[ch]),
2468
                         window, &s->dither_state,
2469
                         samples_ptr, s->nb_channels,
2470
                         s->sb_samples[ch][i]);
2471
            samples_ptr += 32 * s->nb_channels;
2472
        }
2473
    }
2474
#ifdef DEBUG
2475
    s->frame_count++;        
2476
#endif
2477
    return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
2478
}
2479

    
2480
static int decode_frame(AVCodecContext * avctx,
2481
                        void *data, int *data_size,
2482
                        uint8_t * buf, int buf_size)
2483
{
2484
    MPADecodeContext *s = avctx->priv_data;
2485
    uint32_t header;
2486
    uint8_t *buf_ptr;
2487
    int len, out_size;
2488
    OUT_INT *out_samples = data;
2489

    
2490
    buf_ptr = buf;
2491
    while (buf_size > 0) {
2492
        len = s->inbuf_ptr - s->inbuf;
2493
        if (s->frame_size == 0) {
2494
            /* special case for next header for first frame in free
2495
               format case (XXX: find a simpler method) */
2496
            if (s->free_format_next_header != 0) {
2497
                s->inbuf[0] = s->free_format_next_header >> 24;
2498
                s->inbuf[1] = s->free_format_next_header >> 16;
2499
                s->inbuf[2] = s->free_format_next_header >> 8;
2500
                s->inbuf[3] = s->free_format_next_header;
2501
                s->inbuf_ptr = s->inbuf + 4;
2502
                s->free_format_next_header = 0;
2503
                goto got_header;
2504
            }
2505
            /* no header seen : find one. We need at least HEADER_SIZE
2506
               bytes to parse it */
2507
            len = HEADER_SIZE - len;
2508
            if (len > buf_size)
2509
                len = buf_size;
2510
            if (len > 0) {
2511
                memcpy(s->inbuf_ptr, buf_ptr, len);
2512
                buf_ptr += len;
2513
                buf_size -= len;
2514
                s->inbuf_ptr += len;
2515
            }
2516
            if ((s->inbuf_ptr - s->inbuf) >= HEADER_SIZE) {
2517
            got_header:
2518
                header = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |
2519
                    (s->inbuf[2] << 8) | s->inbuf[3];
2520

    
2521
                if (ff_mpa_check_header(header) < 0) {
2522
                    /* no sync found : move by one byte (inefficient, but simple!) */
2523
                    memmove(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1);
2524
                    s->inbuf_ptr--;
2525
                    dprintf("skip %x\n", header);
2526
                    /* reset free format frame size to give a chance
2527
                       to get a new bitrate */
2528
                    s->free_format_frame_size = 0;
2529
                } else {
2530
                    if (decode_header(s, header) == 1) {
2531
                        /* free format: prepare to compute frame size */
2532
                        s->frame_size = -1;
2533
                    }
2534
                    /* update codec info */
2535
                    avctx->sample_rate = s->sample_rate;
2536
                    avctx->channels = s->nb_channels;
2537
                    avctx->bit_rate = s->bit_rate;
2538
                    avctx->sub_id = s->layer;
2539
                    switch(s->layer) {
2540
                    case 1:
2541
                        avctx->frame_size = 384;
2542
                        break;
2543
                    case 2:
2544
                        avctx->frame_size = 1152;
2545
                        break;
2546
                    case 3:
2547
                        if (s->lsf)
2548
                            avctx->frame_size = 576;
2549
                        else
2550
                            avctx->frame_size = 1152;
2551
                        break;
2552
                    }
2553
                }
2554
            }
2555
        } else if (s->frame_size == -1) {
2556
            /* free format : find next sync to compute frame size */
2557
            len = MPA_MAX_CODED_FRAME_SIZE - len;
2558
            if (len > buf_size)
2559
                len = buf_size;
2560
            if (len == 0) {
2561
                /* frame too long: resync */
2562
                s->frame_size = 0;
2563
                memmove(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1);
2564
                s->inbuf_ptr--;
2565
            } else {
2566
                uint8_t *p, *pend;
2567
                uint32_t header1;
2568
                int padding;
2569

    
2570
                memcpy(s->inbuf_ptr, buf_ptr, len);
2571
                /* check for header */
2572
                p = s->inbuf_ptr - 3;
2573
                pend = s->inbuf_ptr + len - 4;
2574
                while (p <= pend) {
2575
                    header = (p[0] << 24) | (p[1] << 16) |
2576
                        (p[2] << 8) | p[3];
2577
                    header1 = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |
2578
                        (s->inbuf[2] << 8) | s->inbuf[3];
2579
                    /* check with high probability that we have a
2580
                       valid header */
2581
                    if ((header & SAME_HEADER_MASK) ==
2582
                        (header1 & SAME_HEADER_MASK)) {
2583
                        /* header found: update pointers */
2584
                        len = (p + 4) - s->inbuf_ptr;
2585
                        buf_ptr += len;
2586
                        buf_size -= len;
2587
                        s->inbuf_ptr = p;
2588
                        /* compute frame size */
2589
                        s->free_format_next_header = header;
2590
                        s->free_format_frame_size = s->inbuf_ptr - s->inbuf;
2591
                        padding = (header1 >> 9) & 1;
2592
                        if (s->layer == 1)
2593
                            s->free_format_frame_size -= padding * 4;
2594
                        else
2595
                            s->free_format_frame_size -= padding;
2596
                        dprintf("free frame size=%d padding=%d\n", 
2597
                                s->free_format_frame_size, padding);
2598
                        decode_header(s, header1);
2599
                        goto next_data;
2600
                    }
2601
                    p++;
2602
                }
2603
                /* not found: simply increase pointers */
2604
                buf_ptr += len;
2605
                s->inbuf_ptr += len;
2606
                buf_size -= len;
2607
            }
2608
        } else if (len < s->frame_size) {
2609
            if (s->frame_size > MPA_MAX_CODED_FRAME_SIZE)
2610
                s->frame_size = MPA_MAX_CODED_FRAME_SIZE;
2611
            len = s->frame_size - len;
2612
            if (len > buf_size)
2613
                len = buf_size;
2614
            memcpy(s->inbuf_ptr, buf_ptr, len);
2615
            buf_ptr += len;
2616
            s->inbuf_ptr += len;
2617
            buf_size -= len;
2618
        }
2619
    next_data:
2620
        if (s->frame_size > 0 && 
2621
            (s->inbuf_ptr - s->inbuf) >= s->frame_size) {
2622
            if (avctx->parse_only) {
2623
                /* simply return the frame data */
2624
                *(uint8_t **)data = s->inbuf;
2625
                out_size = s->inbuf_ptr - s->inbuf;
2626
            } else {
2627
                out_size = mp_decode_frame(s, out_samples);
2628
            }
2629
            s->inbuf_ptr = s->inbuf;
2630
            s->frame_size = 0;
2631
            *data_size = out_size;
2632
            break;
2633
        }
2634
    }
2635
    return buf_ptr - buf;
2636
}
2637

    
2638

    
2639
static int decode_frame_adu(AVCodecContext * avctx,
2640
                        void *data, int *data_size,
2641
                        uint8_t * buf, int buf_size)
2642
{
2643
    MPADecodeContext *s = avctx->priv_data;
2644
    uint32_t header;
2645
    int len, out_size;
2646
    OUT_INT *out_samples = data;
2647

    
2648
    len = buf_size;
2649

    
2650
    // Discard too short frames
2651
    if (buf_size < HEADER_SIZE) {
2652
        *data_size = 0;
2653
        return buf_size;
2654
    }
2655

    
2656

    
2657
    if (len > MPA_MAX_CODED_FRAME_SIZE)
2658
        len = MPA_MAX_CODED_FRAME_SIZE;
2659

    
2660
    memcpy(s->inbuf, buf, len);
2661
    s->inbuf_ptr = s->inbuf + len;
2662

    
2663
    // Get header and restore sync word
2664
    header = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |
2665
              (s->inbuf[2] << 8) | s->inbuf[3] | 0xffe00000;
2666

    
2667
    if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame
2668
        *data_size = 0;
2669
        return buf_size;
2670
    }
2671

    
2672
    decode_header(s, header);
2673
    /* update codec info */
2674
    avctx->sample_rate = s->sample_rate;
2675
    avctx->channels = s->nb_channels;
2676
    avctx->bit_rate = s->bit_rate;
2677
    avctx->sub_id = s->layer;
2678

    
2679
    avctx->frame_size=s->frame_size = len;
2680

    
2681
    if (avctx->parse_only) {
2682
        /* simply return the frame data */
2683
        *(uint8_t **)data = s->inbuf;
2684
        out_size = s->inbuf_ptr - s->inbuf;
2685
    } else {
2686
        out_size = mp_decode_frame(s, out_samples);
2687
    }
2688

    
2689
    *data_size = out_size;
2690
    return buf_size;
2691
}
2692

    
2693

    
2694
AVCodec mp2_decoder =
2695
{
2696
    "mp2",
2697
    CODEC_TYPE_AUDIO,
2698
    CODEC_ID_MP2,
2699
    sizeof(MPADecodeContext),
2700
    decode_init,
2701
    NULL,
2702
    NULL,
2703
    decode_frame,
2704
    CODEC_CAP_PARSE_ONLY,
2705
};
2706

    
2707
AVCodec mp3_decoder =
2708
{
2709
    "mp3",
2710
    CODEC_TYPE_AUDIO,
2711
    CODEC_ID_MP3,
2712
    sizeof(MPADecodeContext),
2713
    decode_init,
2714
    NULL,
2715
    NULL,
2716
    decode_frame,
2717
    CODEC_CAP_PARSE_ONLY,
2718
};
2719

    
2720
AVCodec mp3adu_decoder =
2721
{
2722
    "mp3adu",
2723
    CODEC_TYPE_AUDIO,
2724
    CODEC_ID_MP3ADU,
2725
    sizeof(MPADecodeContext),
2726
    decode_init,
2727
    NULL,
2728
    NULL,
2729
    decode_frame_adu,
2730
    CODEC_CAP_PARSE_ONLY,
2731
};