Statistics
| Branch: | Revision:

ffmpeg / libavcodec / mpegaudiodec.c @ 4d49a5a7

History | View | Annotate | Download (78.8 KB)

1
/*
2
 * MPEG Audio decoder
3
 * Copyright (c) 2001, 2002 Fabrice Bellard
4
 *
5
 * This file is part of FFmpeg.
6
 *
7
 * FFmpeg is free software; you can redistribute it and/or
8
 * modify it under the terms of the GNU Lesser General Public
9
 * License as published by the Free Software Foundation; either
10
 * version 2.1 of the License, or (at your option) any later version.
11
 *
12
 * FFmpeg is distributed in the hope that it will be useful,
13
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15
 * Lesser General Public License for more details.
16
 *
17
 * You should have received a copy of the GNU Lesser General Public
18
 * License along with FFmpeg; if not, write to the Free Software
19
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20
 */
21

    
22
/**
23
 * @file
24
 * MPEG Audio decoder.
25
 */
26

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

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

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

    
40
#include "mathops.h"
41

    
42
#if CONFIG_FLOAT
43
#   define SHR(a,b)       ((a)*(1.0f/(1<<(b))))
44
#   define compute_antialias compute_antialias_float
45
#   define FIXR_OLD(a)    ((int)((a) * FRAC_ONE + 0.5))
46
#   define FIXR(x)        ((float)(x))
47
#   define FIXHR(x)       ((float)(x))
48
#   define MULH3(x, y, s) ((s)*(y)*(x))
49
#   define MULLx(x, y, s) ((y)*(x))
50
#   define RENAME(a) a ## _float
51
#else
52
#   define SHR(a,b)       ((a)>>(b))
53
#   define compute_antialias compute_antialias_integer
54
/* WARNING: only correct for posititive numbers */
55
#   define FIXR_OLD(a)    ((int)((a) * FRAC_ONE + 0.5))
56
#   define FIXR(a)        ((int)((a) * FRAC_ONE + 0.5))
57
#   define FIXHR(a)       ((int)((a) * (1LL<<32) + 0.5))
58
#   define MULH3(x, y, s) MULH((s)*(x), y)
59
#   define MULLx(x, y, s) MULL(x,y,s)
60
#   define RENAME(a)      a
61
#endif
62

    
63
/****************/
64

    
65
#define HEADER_SIZE 4
66

    
67
#include "mpegaudiodata.h"
68
#include "mpegaudiodectab.h"
69

    
70
static void compute_antialias_integer(MPADecodeContext *s, GranuleDef *g);
71
static void compute_antialias_float(MPADecodeContext *s, GranuleDef *g);
72
static void apply_window_mp3_c(MPA_INT *synth_buf, MPA_INT *window,
73
                               int *dither_state, OUT_INT *samples, int incr);
74

    
75
/* vlc structure for decoding layer 3 huffman tables */
76
static VLC huff_vlc[16];
77
static VLC_TYPE huff_vlc_tables[
78
  0+128+128+128+130+128+154+166+
79
  142+204+190+170+542+460+662+414
80
  ][2];
81
static const int huff_vlc_tables_sizes[16] = {
82
  0, 128, 128, 128, 130, 128, 154, 166,
83
  142, 204, 190, 170, 542, 460, 662, 414
84
};
85
static VLC huff_quad_vlc[2];
86
static VLC_TYPE huff_quad_vlc_tables[128+16][2];
87
static const int huff_quad_vlc_tables_sizes[2] = {
88
  128, 16
89
};
90
/* computed from band_size_long */
91
static uint16_t band_index_long[9][23];
92
#include "mpegaudio_tablegen.h"
93
/* intensity stereo coef table */
94
static INTFLOAT is_table[2][16];
95
static INTFLOAT is_table_lsf[2][2][16];
96
static int32_t csa_table[8][4];
97
static float csa_table_float[8][4];
98
static INTFLOAT mdct_win[8][36];
99

    
100
/* lower 2 bits: modulo 3, higher bits: shift */
101
static uint16_t scale_factor_modshift[64];
102
/* [i][j]:  2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
103
static int32_t scale_factor_mult[15][3];
104
/* mult table for layer 2 group quantization */
105

    
106
#define SCALE_GEN(v) \
107
{ FIXR_OLD(1.0 * (v)), FIXR_OLD(0.7937005259 * (v)), FIXR_OLD(0.6299605249 * (v)) }
108

    
109
static const int32_t scale_factor_mult2[3][3] = {
110
    SCALE_GEN(4.0 / 3.0), /* 3 steps */
111
    SCALE_GEN(4.0 / 5.0), /* 5 steps */
112
    SCALE_GEN(4.0 / 9.0), /* 9 steps */
113
};
114

    
115
DECLARE_ALIGNED(16, MPA_INT, RENAME(ff_mpa_synth_window))[512];
116

    
117
/**
118
 * Convert region offsets to region sizes and truncate
119
 * size to big_values.
120
 */
121
static void ff_region_offset2size(GranuleDef *g){
122
    int i, k, j=0;
123
    g->region_size[2] = (576 / 2);
124
    for(i=0;i<3;i++) {
125
        k = FFMIN(g->region_size[i], g->big_values);
126
        g->region_size[i] = k - j;
127
        j = k;
128
    }
129
}
130

    
131
static void ff_init_short_region(MPADecodeContext *s, GranuleDef *g){
132
    if (g->block_type == 2)
133
        g->region_size[0] = (36 / 2);
134
    else {
135
        if (s->sample_rate_index <= 2)
136
            g->region_size[0] = (36 / 2);
137
        else if (s->sample_rate_index != 8)
138
            g->region_size[0] = (54 / 2);
139
        else
140
            g->region_size[0] = (108 / 2);
141
    }
142
    g->region_size[1] = (576 / 2);
143
}
144

    
145
static void ff_init_long_region(MPADecodeContext *s, GranuleDef *g, int ra1, int ra2){
146
    int l;
147
    g->region_size[0] =
148
        band_index_long[s->sample_rate_index][ra1 + 1] >> 1;
149
    /* should not overflow */
150
    l = FFMIN(ra1 + ra2 + 2, 22);
151
    g->region_size[1] =
152
        band_index_long[s->sample_rate_index][l] >> 1;
153
}
154

    
155
static void ff_compute_band_indexes(MPADecodeContext *s, GranuleDef *g){
156
    if (g->block_type == 2) {
157
        if (g->switch_point) {
158
            /* if switched mode, we handle the 36 first samples as
159
                long blocks.  For 8000Hz, we handle the 48 first
160
                exponents as long blocks (XXX: check this!) */
161
            if (s->sample_rate_index <= 2)
162
                g->long_end = 8;
163
            else if (s->sample_rate_index != 8)
164
                g->long_end = 6;
165
            else
166
                g->long_end = 4; /* 8000 Hz */
167

    
168
            g->short_start = 2 + (s->sample_rate_index != 8);
169
        } else {
170
            g->long_end = 0;
171
            g->short_start = 0;
172
        }
173
    } else {
174
        g->short_start = 13;
175
        g->long_end = 22;
176
    }
177
}
178

    
179
/* layer 1 unscaling */
180
/* n = number of bits of the mantissa minus 1 */
181
static inline int l1_unscale(int n, int mant, int scale_factor)
182
{
183
    int shift, mod;
184
    int64_t val;
185

    
186
    shift = scale_factor_modshift[scale_factor];
187
    mod = shift & 3;
188
    shift >>= 2;
189
    val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);
190
    shift += n;
191
    /* NOTE: at this point, 1 <= shift >= 21 + 15 */
192
    return (int)((val + (1LL << (shift - 1))) >> shift);
193
}
194

    
195
static inline int l2_unscale_group(int steps, int mant, int scale_factor)
196
{
197
    int shift, mod, val;
198

    
199
    shift = scale_factor_modshift[scale_factor];
200
    mod = shift & 3;
201
    shift >>= 2;
202

    
203
    val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
204
    /* NOTE: at this point, 0 <= shift <= 21 */
205
    if (shift > 0)
206
        val = (val + (1 << (shift - 1))) >> shift;
207
    return val;
208
}
209

    
210
/* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
211
static inline int l3_unscale(int value, int exponent)
212
{
213
    unsigned int m;
214
    int e;
215

    
216
    e = table_4_3_exp  [4*value + (exponent&3)];
217
    m = table_4_3_value[4*value + (exponent&3)];
218
    e -= (exponent >> 2);
219
    assert(e>=1);
220
    if (e > 31)
221
        return 0;
222
    m = (m + (1 << (e-1))) >> e;
223

    
224
    return m;
225
}
226

    
227
/* all integer n^(4/3) computation code */
228
#define DEV_ORDER 13
229

    
230
#define POW_FRAC_BITS 24
231
#define POW_FRAC_ONE    (1 << POW_FRAC_BITS)
232
#define POW_FIX(a)   ((int)((a) * POW_FRAC_ONE))
233
#define POW_MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> POW_FRAC_BITS)
234

    
235
static int dev_4_3_coefs[DEV_ORDER];
236

    
237
#if 0 /* unused */
238
static int pow_mult3[3] = {
239
    POW_FIX(1.0),
240
    POW_FIX(1.25992104989487316476),
241
    POW_FIX(1.58740105196819947474),
242
};
243
#endif
244

    
245
static av_cold void int_pow_init(void)
246
{
247
    int i, a;
248

    
249
    a = POW_FIX(1.0);
250
    for(i=0;i<DEV_ORDER;i++) {
251
        a = POW_MULL(a, POW_FIX(4.0 / 3.0) - i * POW_FIX(1.0)) / (i + 1);
252
        dev_4_3_coefs[i] = a;
253
    }
254
}
255

    
256
#if 0 /* unused, remove? */
257
/* return the mantissa and the binary exponent */
258
static int int_pow(int i, int *exp_ptr)
259
{
260
    int e, er, eq, j;
261
    int a, a1;
262

263
    /* renormalize */
264
    a = i;
265
    e = POW_FRAC_BITS;
266
    while (a < (1 << (POW_FRAC_BITS - 1))) {
267
        a = a << 1;
268
        e--;
269
    }
270
    a -= (1 << POW_FRAC_BITS);
271
    a1 = 0;
272
    for(j = DEV_ORDER - 1; j >= 0; j--)
273
        a1 = POW_MULL(a, dev_4_3_coefs[j] + a1);
274
    a = (1 << POW_FRAC_BITS) + a1;
275
    /* exponent compute (exact) */
276
    e = e * 4;
277
    er = e % 3;
278
    eq = e / 3;
279
    a = POW_MULL(a, pow_mult3[er]);
280
    while (a >= 2 * POW_FRAC_ONE) {
281
        a = a >> 1;
282
        eq++;
283
    }
284
    /* convert to float */
285
    while (a < POW_FRAC_ONE) {
286
        a = a << 1;
287
        eq--;
288
    }
289
    /* now POW_FRAC_ONE <= a < 2 * POW_FRAC_ONE */
290
#if POW_FRAC_BITS > FRAC_BITS
291
    a = (a + (1 << (POW_FRAC_BITS - FRAC_BITS - 1))) >> (POW_FRAC_BITS - FRAC_BITS);
292
    /* correct overflow */
293
    if (a >= 2 * (1 << FRAC_BITS)) {
294
        a = a >> 1;
295
        eq++;
296
    }
297
#endif
298
    *exp_ptr = eq;
299
    return a;
300
}
301
#endif
302

    
303
static av_cold int decode_init(AVCodecContext * avctx)
304
{
305
    MPADecodeContext *s = avctx->priv_data;
306
    static int init=0;
307
    int i, j, k;
308

    
309
    s->avctx = avctx;
310
    s->apply_window_mp3 = apply_window_mp3_c;
311

    
312
    avctx->sample_fmt= OUT_FMT;
313
    s->error_recognition= avctx->error_recognition;
314

    
315
    if (!init && !avctx->parse_only) {
316
        int offset;
317

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

    
327
        /* scale factor multiply for layer 1 */
328
        for(i=0;i<15;i++) {
329
            int n, norm;
330
            n = i + 2;
331
            norm = ((INT64_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
332
            scale_factor_mult[i][0] = MULLx(norm, FIXR(1.0          * 2.0), FRAC_BITS);
333
            scale_factor_mult[i][1] = MULLx(norm, FIXR(0.7937005259 * 2.0), FRAC_BITS);
334
            scale_factor_mult[i][2] = MULLx(norm, FIXR(0.6299605249 * 2.0), FRAC_BITS);
335
            dprintf(avctx, "%d: norm=%x s=%x %x %x\n",
336
                    i, norm,
337
                    scale_factor_mult[i][0],
338
                    scale_factor_mult[i][1],
339
                    scale_factor_mult[i][2]);
340
        }
341

    
342
        RENAME(ff_mpa_synth_init)(RENAME(ff_mpa_synth_window));
343

    
344
        /* huffman decode tables */
345
        offset = 0;
346
        for(i=1;i<16;i++) {
347
            const HuffTable *h = &mpa_huff_tables[i];
348
            int xsize, x, y;
349
            uint8_t  tmp_bits [512];
350
            uint16_t tmp_codes[512];
351

    
352
            memset(tmp_bits , 0, sizeof(tmp_bits ));
353
            memset(tmp_codes, 0, sizeof(tmp_codes));
354

    
355
            xsize = h->xsize;
356

    
357
            j = 0;
358
            for(x=0;x<xsize;x++) {
359
                for(y=0;y<xsize;y++){
360
                    tmp_bits [(x << 5) | y | ((x&&y)<<4)]= h->bits [j  ];
361
                    tmp_codes[(x << 5) | y | ((x&&y)<<4)]= h->codes[j++];
362
                }
363
            }
364

    
365
            /* XXX: fail test */
366
            huff_vlc[i].table = huff_vlc_tables+offset;
367
            huff_vlc[i].table_allocated = huff_vlc_tables_sizes[i];
368
            init_vlc(&huff_vlc[i], 7, 512,
369
                     tmp_bits, 1, 1, tmp_codes, 2, 2,
370
                     INIT_VLC_USE_NEW_STATIC);
371
            offset += huff_vlc_tables_sizes[i];
372
        }
373
        assert(offset == FF_ARRAY_ELEMS(huff_vlc_tables));
374

    
375
        offset = 0;
376
        for(i=0;i<2;i++) {
377
            huff_quad_vlc[i].table = huff_quad_vlc_tables+offset;
378
            huff_quad_vlc[i].table_allocated = huff_quad_vlc_tables_sizes[i];
379
            init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
380
                     mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1,
381
                     INIT_VLC_USE_NEW_STATIC);
382
            offset += huff_quad_vlc_tables_sizes[i];
383
        }
384
        assert(offset == FF_ARRAY_ELEMS(huff_quad_vlc_tables));
385

    
386
        for(i=0;i<9;i++) {
387
            k = 0;
388
            for(j=0;j<22;j++) {
389
                band_index_long[i][j] = k;
390
                k += band_size_long[i][j];
391
            }
392
            band_index_long[i][22] = k;
393
        }
394

    
395
        /* compute n ^ (4/3) and store it in mantissa/exp format */
396

    
397
        int_pow_init();
398
        mpegaudio_tableinit();
399

    
400
        for(i=0;i<7;i++) {
401
            float f;
402
            INTFLOAT v;
403
            if (i != 6) {
404
                f = tan((double)i * M_PI / 12.0);
405
                v = FIXR(f / (1.0 + f));
406
            } else {
407
                v = FIXR(1.0);
408
            }
409
            is_table[0][i] = v;
410
            is_table[1][6 - i] = v;
411
        }
412
        /* invalid values */
413
        for(i=7;i<16;i++)
414
            is_table[0][i] = is_table[1][i] = 0.0;
415

    
416
        for(i=0;i<16;i++) {
417
            double f;
418
            int e, k;
419

    
420
            for(j=0;j<2;j++) {
421
                e = -(j + 1) * ((i + 1) >> 1);
422
                f = pow(2.0, e / 4.0);
423
                k = i & 1;
424
                is_table_lsf[j][k ^ 1][i] = FIXR(f);
425
                is_table_lsf[j][k][i] = FIXR(1.0);
426
                dprintf(avctx, "is_table_lsf %d %d: %x %x\n",
427
                        i, j, is_table_lsf[j][0][i], is_table_lsf[j][1][i]);
428
            }
429
        }
430

    
431
        for(i=0;i<8;i++) {
432
            float ci, cs, ca;
433
            ci = ci_table[i];
434
            cs = 1.0 / sqrt(1.0 + ci * ci);
435
            ca = cs * ci;
436
            csa_table[i][0] = FIXHR(cs/4);
437
            csa_table[i][1] = FIXHR(ca/4);
438
            csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);
439
            csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4);
440
            csa_table_float[i][0] = cs;
441
            csa_table_float[i][1] = ca;
442
            csa_table_float[i][2] = ca + cs;
443
            csa_table_float[i][3] = ca - cs;
444
        }
445

    
446
        /* compute mdct windows */
447
        for(i=0;i<36;i++) {
448
            for(j=0; j<4; j++){
449
                double d;
450

    
451
                if(j==2 && i%3 != 1)
452
                    continue;
453

    
454
                d= sin(M_PI * (i + 0.5) / 36.0);
455
                if(j==1){
456
                    if     (i>=30) d= 0;
457
                    else if(i>=24) d= sin(M_PI * (i - 18 + 0.5) / 12.0);
458
                    else if(i>=18) d= 1;
459
                }else if(j==3){
460
                    if     (i<  6) d= 0;
461
                    else if(i< 12) d= sin(M_PI * (i -  6 + 0.5) / 12.0);
462
                    else if(i< 18) d= 1;
463
                }
464
                //merge last stage of imdct into the window coefficients
465
                d*= 0.5 / cos(M_PI*(2*i + 19)/72);
466

    
467
                if(j==2)
468
                    mdct_win[j][i/3] = FIXHR((d / (1<<5)));
469
                else
470
                    mdct_win[j][i  ] = FIXHR((d / (1<<5)));
471
            }
472
        }
473

    
474
        /* NOTE: we do frequency inversion adter the MDCT by changing
475
           the sign of the right window coefs */
476
        for(j=0;j<4;j++) {
477
            for(i=0;i<36;i+=2) {
478
                mdct_win[j + 4][i] = mdct_win[j][i];
479
                mdct_win[j + 4][i + 1] = -mdct_win[j][i + 1];
480
            }
481
        }
482

    
483
        init = 1;
484
    }
485

    
486
    if (avctx->codec_id == CODEC_ID_MP3ADU)
487
        s->adu_mode = 1;
488
    return 0;
489
}
490

    
491
/* tab[i][j] = 1.0 / (2.0 * cos(pi*(2*k+1) / 2^(6 - j))) */
492

    
493
/* cos(i*pi/64) */
494

    
495
#define COS0_0  FIXHR(0.50060299823519630134/2)
496
#define COS0_1  FIXHR(0.50547095989754365998/2)
497
#define COS0_2  FIXHR(0.51544730992262454697/2)
498
#define COS0_3  FIXHR(0.53104259108978417447/2)
499
#define COS0_4  FIXHR(0.55310389603444452782/2)
500
#define COS0_5  FIXHR(0.58293496820613387367/2)
501
#define COS0_6  FIXHR(0.62250412303566481615/2)
502
#define COS0_7  FIXHR(0.67480834145500574602/2)
503
#define COS0_8  FIXHR(0.74453627100229844977/2)
504
#define COS0_9  FIXHR(0.83934964541552703873/2)
505
#define COS0_10 FIXHR(0.97256823786196069369/2)
506
#define COS0_11 FIXHR(1.16943993343288495515/4)
507
#define COS0_12 FIXHR(1.48416461631416627724/4)
508
#define COS0_13 FIXHR(2.05778100995341155085/8)
509
#define COS0_14 FIXHR(3.40760841846871878570/8)
510
#define COS0_15 FIXHR(10.19000812354805681150/32)
511

    
512
#define COS1_0 FIXHR(0.50241928618815570551/2)
513
#define COS1_1 FIXHR(0.52249861493968888062/2)
514
#define COS1_2 FIXHR(0.56694403481635770368/2)
515
#define COS1_3 FIXHR(0.64682178335999012954/2)
516
#define COS1_4 FIXHR(0.78815462345125022473/2)
517
#define COS1_5 FIXHR(1.06067768599034747134/4)
518
#define COS1_6 FIXHR(1.72244709823833392782/4)
519
#define COS1_7 FIXHR(5.10114861868916385802/16)
520

    
521
#define COS2_0 FIXHR(0.50979557910415916894/2)
522
#define COS2_1 FIXHR(0.60134488693504528054/2)
523
#define COS2_2 FIXHR(0.89997622313641570463/2)
524
#define COS2_3 FIXHR(2.56291544774150617881/8)
525

    
526
#define COS3_0 FIXHR(0.54119610014619698439/2)
527
#define COS3_1 FIXHR(1.30656296487637652785/4)
528

    
529
#define COS4_0 FIXHR(0.70710678118654752439/2)
530

    
531
/* butterfly operator */
532
#define BF(a, b, c, s)\
533
{\
534
    tmp0 = val##a + val##b;\
535
    tmp1 = val##a - val##b;\
536
    val##a = tmp0;\
537
    val##b = MULH3(tmp1, c, 1<<(s));\
538
}
539

    
540
#define BF0(a, b, c, s)\
541
{\
542
    tmp0 = tab[a] + tab[b];\
543
    tmp1 = tab[a] - tab[b];\
544
    val##a = tmp0;\
545
    val##b = MULH3(tmp1, c, 1<<(s));\
546
}
547

    
548
#define BF1(a, b, c, d)\
549
{\
550
    BF(a, b, COS4_0, 1);\
551
    BF(c, d,-COS4_0, 1);\
552
    val##c += val##d;\
553
}
554

    
555
#define BF2(a, b, c, d)\
556
{\
557
    BF(a, b, COS4_0, 1);\
558
    BF(c, d,-COS4_0, 1);\
559
    val##c += val##d;\
560
    val##a += val##c;\
561
    val##c += val##b;\
562
    val##b += val##d;\
563
}
564

    
565
#define ADD(a, b) val##a += val##b
566

    
567
/* DCT32 without 1/sqrt(2) coef zero scaling. */
568
static void dct32(INTFLOAT *out, const INTFLOAT *tab)
569
{
570
    INTFLOAT tmp0, tmp1;
571

    
572
    INTFLOAT val0 , val1 , val2 , val3 , val4 , val5 , val6 , val7 ,
573
             val8 , val9 , val10, val11, val12, val13, val14, val15,
574
             val16, val17, val18, val19, val20, val21, val22, val23,
575
             val24, val25, val26, val27, val28, val29, val30, val31;
576

    
577
    /* pass 1 */
578
    BF0( 0, 31, COS0_0 , 1);
579
    BF0(15, 16, COS0_15, 5);
580
    /* pass 2 */
581
    BF( 0, 15, COS1_0 , 1);
582
    BF(16, 31,-COS1_0 , 1);
583
    /* pass 1 */
584
    BF0( 7, 24, COS0_7 , 1);
585
    BF0( 8, 23, COS0_8 , 1);
586
    /* pass 2 */
587
    BF( 7,  8, COS1_7 , 4);
588
    BF(23, 24,-COS1_7 , 4);
589
    /* pass 3 */
590
    BF( 0,  7, COS2_0 , 1);
591
    BF( 8, 15,-COS2_0 , 1);
592
    BF(16, 23, COS2_0 , 1);
593
    BF(24, 31,-COS2_0 , 1);
594
    /* pass 1 */
595
    BF0( 3, 28, COS0_3 , 1);
596
    BF0(12, 19, COS0_12, 2);
597
    /* pass 2 */
598
    BF( 3, 12, COS1_3 , 1);
599
    BF(19, 28,-COS1_3 , 1);
600
    /* pass 1 */
601
    BF0( 4, 27, COS0_4 , 1);
602
    BF0(11, 20, COS0_11, 2);
603
    /* pass 2 */
604
    BF( 4, 11, COS1_4 , 1);
605
    BF(20, 27,-COS1_4 , 1);
606
    /* pass 3 */
607
    BF( 3,  4, COS2_3 , 3);
608
    BF(11, 12,-COS2_3 , 3);
609
    BF(19, 20, COS2_3 , 3);
610
    BF(27, 28,-COS2_3 , 3);
611
    /* pass 4 */
612
    BF( 0,  3, COS3_0 , 1);
613
    BF( 4,  7,-COS3_0 , 1);
614
    BF( 8, 11, COS3_0 , 1);
615
    BF(12, 15,-COS3_0 , 1);
616
    BF(16, 19, COS3_0 , 1);
617
    BF(20, 23,-COS3_0 , 1);
618
    BF(24, 27, COS3_0 , 1);
619
    BF(28, 31,-COS3_0 , 1);
620

    
621

    
622

    
623
    /* pass 1 */
624
    BF0( 1, 30, COS0_1 , 1);
625
    BF0(14, 17, COS0_14, 3);
626
    /* pass 2 */
627
    BF( 1, 14, COS1_1 , 1);
628
    BF(17, 30,-COS1_1 , 1);
629
    /* pass 1 */
630
    BF0( 6, 25, COS0_6 , 1);
631
    BF0( 9, 22, COS0_9 , 1);
632
    /* pass 2 */
633
    BF( 6,  9, COS1_6 , 2);
634
    BF(22, 25,-COS1_6 , 2);
635
    /* pass 3 */
636
    BF( 1,  6, COS2_1 , 1);
637
    BF( 9, 14,-COS2_1 , 1);
638
    BF(17, 22, COS2_1 , 1);
639
    BF(25, 30,-COS2_1 , 1);
640

    
641
    /* pass 1 */
642
    BF0( 2, 29, COS0_2 , 1);
643
    BF0(13, 18, COS0_13, 3);
644
    /* pass 2 */
645
    BF( 2, 13, COS1_2 , 1);
646
    BF(18, 29,-COS1_2 , 1);
647
    /* pass 1 */
648
    BF0( 5, 26, COS0_5 , 1);
649
    BF0(10, 21, COS0_10, 1);
650
    /* pass 2 */
651
    BF( 5, 10, COS1_5 , 2);
652
    BF(21, 26,-COS1_5 , 2);
653
    /* pass 3 */
654
    BF( 2,  5, COS2_2 , 1);
655
    BF(10, 13,-COS2_2 , 1);
656
    BF(18, 21, COS2_2 , 1);
657
    BF(26, 29,-COS2_2 , 1);
658
    /* pass 4 */
659
    BF( 1,  2, COS3_1 , 2);
660
    BF( 5,  6,-COS3_1 , 2);
661
    BF( 9, 10, COS3_1 , 2);
662
    BF(13, 14,-COS3_1 , 2);
663
    BF(17, 18, COS3_1 , 2);
664
    BF(21, 22,-COS3_1 , 2);
665
    BF(25, 26, COS3_1 , 2);
666
    BF(29, 30,-COS3_1 , 2);
667

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

    
678
    /* pass 6 */
679

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

    
688
    out[ 0] = val0;
689
    out[16] = val1;
690
    out[ 8] = val2;
691
    out[24] = val3;
692
    out[ 4] = val4;
693
    out[20] = val5;
694
    out[12] = val6;
695
    out[28] = val7;
696
    out[ 2] = val8;
697
    out[18] = val9;
698
    out[10] = val10;
699
    out[26] = val11;
700
    out[ 6] = val12;
701
    out[22] = val13;
702
    out[14] = val14;
703
    out[30] = val15;
704

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

    
713
    out[ 1] = val16 + val24;
714
    out[17] = val17 + val25;
715
    out[ 9] = val18 + val26;
716
    out[25] = val19 + val27;
717
    out[ 5] = val20 + val28;
718
    out[21] = val21 + val29;
719
    out[13] = val22 + val30;
720
    out[29] = val23 + val31;
721
    out[ 3] = val24 + val20;
722
    out[19] = val25 + val21;
723
    out[11] = val26 + val22;
724
    out[27] = val27 + val23;
725
    out[ 7] = val28 + val18;
726
    out[23] = val29 + val19;
727
    out[15] = val30 + val17;
728
    out[31] = val31;
729
}
730

    
731
#if CONFIG_FLOAT
732
static inline float round_sample(float *sum)
733
{
734
    float sum1=*sum;
735
    *sum = 0;
736
    return sum1;
737
}
738

    
739
/* signed 16x16 -> 32 multiply add accumulate */
740
#define MACS(rt, ra, rb) rt+=(ra)*(rb)
741

    
742
/* signed 16x16 -> 32 multiply */
743
#define MULS(ra, rb) ((ra)*(rb))
744

    
745
#define MLSS(rt, ra, rb) rt-=(ra)*(rb)
746

    
747
#elif FRAC_BITS <= 15
748

    
749
static inline int round_sample(int *sum)
750
{
751
    int sum1;
752
    sum1 = (*sum) >> OUT_SHIFT;
753
    *sum &= (1<<OUT_SHIFT)-1;
754
    return av_clip(sum1, OUT_MIN, OUT_MAX);
755
}
756

    
757
/* signed 16x16 -> 32 multiply add accumulate */
758
#define MACS(rt, ra, rb) MAC16(rt, ra, rb)
759

    
760
/* signed 16x16 -> 32 multiply */
761
#define MULS(ra, rb) MUL16(ra, rb)
762

    
763
#define MLSS(rt, ra, rb) MLS16(rt, ra, rb)
764

    
765
#else
766

    
767
static inline int round_sample(int64_t *sum)
768
{
769
    int sum1;
770
    sum1 = (int)((*sum) >> OUT_SHIFT);
771
    *sum &= (1<<OUT_SHIFT)-1;
772
    return av_clip(sum1, OUT_MIN, OUT_MAX);
773
}
774

    
775
#   define MULS(ra, rb) MUL64(ra, rb)
776
#   define MACS(rt, ra, rb) MAC64(rt, ra, rb)
777
#   define MLSS(rt, ra, rb) MLS64(rt, ra, rb)
778
#endif
779

    
780
#define SUM8(op, sum, w, p)               \
781
{                                         \
782
    op(sum, (w)[0 * 64], (p)[0 * 64]);    \
783
    op(sum, (w)[1 * 64], (p)[1 * 64]);    \
784
    op(sum, (w)[2 * 64], (p)[2 * 64]);    \
785
    op(sum, (w)[3 * 64], (p)[3 * 64]);    \
786
    op(sum, (w)[4 * 64], (p)[4 * 64]);    \
787
    op(sum, (w)[5 * 64], (p)[5 * 64]);    \
788
    op(sum, (w)[6 * 64], (p)[6 * 64]);    \
789
    op(sum, (w)[7 * 64], (p)[7 * 64]);    \
790
}
791

    
792
#define SUM8P2(sum1, op1, sum2, op2, w1, w2, p) \
793
{                                               \
794
    INTFLOAT tmp;\
795
    tmp = p[0 * 64];\
796
    op1(sum1, (w1)[0 * 64], tmp);\
797
    op2(sum2, (w2)[0 * 64], tmp);\
798
    tmp = p[1 * 64];\
799
    op1(sum1, (w1)[1 * 64], tmp);\
800
    op2(sum2, (w2)[1 * 64], tmp);\
801
    tmp = p[2 * 64];\
802
    op1(sum1, (w1)[2 * 64], tmp);\
803
    op2(sum2, (w2)[2 * 64], tmp);\
804
    tmp = p[3 * 64];\
805
    op1(sum1, (w1)[3 * 64], tmp);\
806
    op2(sum2, (w2)[3 * 64], tmp);\
807
    tmp = p[4 * 64];\
808
    op1(sum1, (w1)[4 * 64], tmp);\
809
    op2(sum2, (w2)[4 * 64], tmp);\
810
    tmp = p[5 * 64];\
811
    op1(sum1, (w1)[5 * 64], tmp);\
812
    op2(sum2, (w2)[5 * 64], tmp);\
813
    tmp = p[6 * 64];\
814
    op1(sum1, (w1)[6 * 64], tmp);\
815
    op2(sum2, (w2)[6 * 64], tmp);\
816
    tmp = p[7 * 64];\
817
    op1(sum1, (w1)[7 * 64], tmp);\
818
    op2(sum2, (w2)[7 * 64], tmp);\
819
}
820

    
821
void av_cold RENAME(ff_mpa_synth_init)(MPA_INT *window)
822
{
823
    int i;
824

    
825
    /* max = 18760, max sum over all 16 coefs : 44736 */
826
    for(i=0;i<257;i++) {
827
        INTFLOAT v;
828
        v = ff_mpa_enwindow[i];
829
#if CONFIG_FLOAT
830
        v *= 1.0 / (1LL<<(16 + FRAC_BITS));
831
#elif WFRAC_BITS < 16
832
        v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS);
833
#endif
834
        window[i] = v;
835
        if ((i & 63) != 0)
836
            v = -v;
837
        if (i != 0)
838
            window[512 - i] = v;
839
    }
840
}
841

    
842
static void apply_window_mp3_c(MPA_INT *synth_buf, MPA_INT *window,
843
                               int *dither_state, OUT_INT *samples, int incr)
844
{
845
    register const MPA_INT *w, *w2, *p;
846
    int j;
847
    OUT_INT *samples2;
848
#if CONFIG_FLOAT
849
    float sum, sum2;
850
#elif FRAC_BITS <= 15
851
    int sum, sum2;
852
#else
853
    int64_t sum, sum2;
854
#endif
855

    
856
    /* copy to avoid wrap */
857
    memcpy(synth_buf + 512, synth_buf, 32 * sizeof(*synth_buf));
858

    
859
    samples2 = samples + 31 * incr;
860
    w = window;
861
    w2 = window + 31;
862

    
863
    sum = *dither_state;
864
    p = synth_buf + 16;
865
    SUM8(MACS, sum, w, p);
866
    p = synth_buf + 48;
867
    SUM8(MLSS, sum, w + 32, p);
868
    *samples = round_sample(&sum);
869
    samples += incr;
870
    w++;
871

    
872
    /* we calculate two samples at the same time to avoid one memory
873
       access per two sample */
874
    for(j=1;j<16;j++) {
875
        sum2 = 0;
876
        p = synth_buf + 16 + j;
877
        SUM8P2(sum, MACS, sum2, MLSS, w, w2, p);
878
        p = synth_buf + 48 - j;
879
        SUM8P2(sum, MLSS, sum2, MLSS, w + 32, w2 + 32, p);
880

    
881
        *samples = round_sample(&sum);
882
        samples += incr;
883
        sum += sum2;
884
        *samples2 = round_sample(&sum);
885
        samples2 -= incr;
886
        w++;
887
        w2--;
888
    }
889

    
890
    p = synth_buf + 32;
891
    SUM8(MLSS, sum, w + 32, p);
892
    *samples = round_sample(&sum);
893
    *dither_state= sum;
894
}
895

    
896

    
897
/* 32 sub band synthesis filter. Input: 32 sub band samples, Output:
898
   32 samples. */
899
/* XXX: optimize by avoiding ring buffer usage */
900
#if CONFIG_FLOAT
901
void ff_mpa_synth_filter_float(MPADecodeContext *s, float *synth_buf_ptr,
902
                               int *synth_buf_offset,
903
                               float *window, int *dither_state,
904
                               float *samples, int incr,
905
                               float sb_samples[SBLIMIT])
906
{
907
    float *synth_buf;
908
    int offset;
909

    
910
    offset = *synth_buf_offset;
911
    synth_buf = synth_buf_ptr + offset;
912

    
913
    dct32(synth_buf, sb_samples);
914
    s->apply_window_mp3(synth_buf, window, dither_state, samples, incr);
915

    
916
    offset = (offset - 32) & 511;
917
    *synth_buf_offset = offset;
918
}
919
#else
920
void ff_mpa_synth_filter(MPA_INT *synth_buf_ptr, int *synth_buf_offset,
921
                         MPA_INT *window, int *dither_state,
922
                         OUT_INT *samples, int incr,
923
                         INTFLOAT sb_samples[SBLIMIT])
924
{
925
    register MPA_INT *synth_buf;
926
    int offset;
927
#if FRAC_BITS <= 15
928
    int32_t tmp[32];
929
#endif
930

    
931
    offset = *synth_buf_offset;
932
    synth_buf = synth_buf_ptr + offset;
933

    
934
#if FRAC_BITS <= 15 && !CONFIG_FLOAT
935
    dct32(tmp, sb_samples);
936
    for(j=0;j<32;j++) {
937
        /* NOTE: can cause a loss in precision if very high amplitude
938
           sound */
939
        synth_buf[j] = av_clip_int16(tmp[j]);
940
    }
941
#else
942
    dct32(synth_buf, sb_samples);
943
#endif
944

    
945
    apply_window_mp3_c(synth_buf, window, dither_state, samples, incr);
946

    
947
    offset = (offset - 32) & 511;
948
    *synth_buf_offset = offset;
949
}
950
#endif
951

    
952
#define C3 FIXHR(0.86602540378443864676/2)
953

    
954
/* 0.5 / cos(pi*(2*i+1)/36) */
955
static const INTFLOAT icos36[9] = {
956
    FIXR(0.50190991877167369479),
957
    FIXR(0.51763809020504152469), //0
958
    FIXR(0.55168895948124587824),
959
    FIXR(0.61038729438072803416),
960
    FIXR(0.70710678118654752439), //1
961
    FIXR(0.87172339781054900991),
962
    FIXR(1.18310079157624925896),
963
    FIXR(1.93185165257813657349), //2
964
    FIXR(5.73685662283492756461),
965
};
966

    
967
/* 0.5 / cos(pi*(2*i+1)/36) */
968
static const INTFLOAT icos36h[9] = {
969
    FIXHR(0.50190991877167369479/2),
970
    FIXHR(0.51763809020504152469/2), //0
971
    FIXHR(0.55168895948124587824/2),
972
    FIXHR(0.61038729438072803416/2),
973
    FIXHR(0.70710678118654752439/2), //1
974
    FIXHR(0.87172339781054900991/2),
975
    FIXHR(1.18310079157624925896/4),
976
    FIXHR(1.93185165257813657349/4), //2
977
//    FIXHR(5.73685662283492756461),
978
};
979

    
980
/* 12 points IMDCT. We compute it "by hand" by factorizing obvious
981
   cases. */
982
static void imdct12(INTFLOAT *out, INTFLOAT *in)
983
{
984
    INTFLOAT in0, in1, in2, in3, in4, in5, t1, t2;
985

    
986
    in0= in[0*3];
987
    in1= in[1*3] + in[0*3];
988
    in2= in[2*3] + in[1*3];
989
    in3= in[3*3] + in[2*3];
990
    in4= in[4*3] + in[3*3];
991
    in5= in[5*3] + in[4*3];
992
    in5 += in3;
993
    in3 += in1;
994

    
995
    in2= MULH3(in2, C3, 2);
996
    in3= MULH3(in3, C3, 4);
997

    
998
    t1 = in0 - in4;
999
    t2 = MULH3(in1 - in5, icos36h[4], 2);
1000

    
1001
    out[ 7]=
1002
    out[10]= t1 + t2;
1003
    out[ 1]=
1004
    out[ 4]= t1 - t2;
1005

    
1006
    in0 += SHR(in4, 1);
1007
    in4 = in0 + in2;
1008
    in5 += 2*in1;
1009
    in1 = MULH3(in5 + in3, icos36h[1], 1);
1010
    out[ 8]=
1011
    out[ 9]= in4 + in1;
1012
    out[ 2]=
1013
    out[ 3]= in4 - in1;
1014

    
1015
    in0 -= in2;
1016
    in5 = MULH3(in5 - in3, icos36h[7], 2);
1017
    out[ 0]=
1018
    out[ 5]= in0 - in5;
1019
    out[ 6]=
1020
    out[11]= in0 + in5;
1021
}
1022

    
1023
/* cos(pi*i/18) */
1024
#define C1 FIXHR(0.98480775301220805936/2)
1025
#define C2 FIXHR(0.93969262078590838405/2)
1026
#define C3 FIXHR(0.86602540378443864676/2)
1027
#define C4 FIXHR(0.76604444311897803520/2)
1028
#define C5 FIXHR(0.64278760968653932632/2)
1029
#define C6 FIXHR(0.5/2)
1030
#define C7 FIXHR(0.34202014332566873304/2)
1031
#define C8 FIXHR(0.17364817766693034885/2)
1032

    
1033

    
1034
/* using Lee like decomposition followed by hand coded 9 points DCT */
1035
static void imdct36(INTFLOAT *out, INTFLOAT *buf, INTFLOAT *in, INTFLOAT *win)
1036
{
1037
    int i, j;
1038
    INTFLOAT t0, t1, t2, t3, s0, s1, s2, s3;
1039
    INTFLOAT tmp[18], *tmp1, *in1;
1040

    
1041
    for(i=17;i>=1;i--)
1042
        in[i] += in[i-1];
1043
    for(i=17;i>=3;i-=2)
1044
        in[i] += in[i-2];
1045

    
1046
    for(j=0;j<2;j++) {
1047
        tmp1 = tmp + j;
1048
        in1 = in + j;
1049

    
1050
        t2 = in1[2*4] + in1[2*8] - in1[2*2];
1051

    
1052
        t3 = in1[2*0] + SHR(in1[2*6],1);
1053
        t1 = in1[2*0] - in1[2*6];
1054
        tmp1[ 6] = t1 - SHR(t2,1);
1055
        tmp1[16] = t1 + t2;
1056

    
1057
        t0 = MULH3(in1[2*2] + in1[2*4] ,    C2, 2);
1058
        t1 = MULH3(in1[2*4] - in1[2*8] , -2*C8, 1);
1059
        t2 = MULH3(in1[2*2] + in1[2*8] ,   -C4, 2);
1060

    
1061
        tmp1[10] = t3 - t0 - t2;
1062
        tmp1[ 2] = t3 + t0 + t1;
1063
        tmp1[14] = t3 + t2 - t1;
1064

    
1065
        tmp1[ 4] = MULH3(in1[2*5] + in1[2*7] - in1[2*1], -C3, 2);
1066
        t2 = MULH3(in1[2*1] + in1[2*5],    C1, 2);
1067
        t3 = MULH3(in1[2*5] - in1[2*7], -2*C7, 1);
1068
        t0 = MULH3(in1[2*3], C3, 2);
1069

    
1070
        t1 = MULH3(in1[2*1] + in1[2*7],   -C5, 2);
1071

    
1072
        tmp1[ 0] = t2 + t3 + t0;
1073
        tmp1[12] = t2 + t1 - t0;
1074
        tmp1[ 8] = t3 - t1 - t0;
1075
    }
1076

    
1077
    i = 0;
1078
    for(j=0;j<4;j++) {
1079
        t0 = tmp[i];
1080
        t1 = tmp[i + 2];
1081
        s0 = t1 + t0;
1082
        s2 = t1 - t0;
1083

    
1084
        t2 = tmp[i + 1];
1085
        t3 = tmp[i + 3];
1086
        s1 = MULH3(t3 + t2, icos36h[j], 2);
1087
        s3 = MULLx(t3 - t2, icos36[8 - j], FRAC_BITS);
1088

    
1089
        t0 = s0 + s1;
1090
        t1 = s0 - s1;
1091
        out[(9 + j)*SBLIMIT] =  MULH3(t1, win[9 + j], 1) + buf[9 + j];
1092
        out[(8 - j)*SBLIMIT] =  MULH3(t1, win[8 - j], 1) + buf[8 - j];
1093
        buf[9 + j] = MULH3(t0, win[18 + 9 + j], 1);
1094
        buf[8 - j] = MULH3(t0, win[18 + 8 - j], 1);
1095

    
1096
        t0 = s2 + s3;
1097
        t1 = s2 - s3;
1098
        out[(9 + 8 - j)*SBLIMIT] =  MULH3(t1, win[9 + 8 - j], 1) + buf[9 + 8 - j];
1099
        out[(        j)*SBLIMIT] =  MULH3(t1, win[        j], 1) + buf[        j];
1100
        buf[9 + 8 - j] = MULH3(t0, win[18 + 9 + 8 - j], 1);
1101
        buf[      + j] = MULH3(t0, win[18         + j], 1);
1102
        i += 4;
1103
    }
1104

    
1105
    s0 = tmp[16];
1106
    s1 = MULH3(tmp[17], icos36h[4], 2);
1107
    t0 = s0 + s1;
1108
    t1 = s0 - s1;
1109
    out[(9 + 4)*SBLIMIT] =  MULH3(t1, win[9 + 4], 1) + buf[9 + 4];
1110
    out[(8 - 4)*SBLIMIT] =  MULH3(t1, win[8 - 4], 1) + buf[8 - 4];
1111
    buf[9 + 4] = MULH3(t0, win[18 + 9 + 4], 1);
1112
    buf[8 - 4] = MULH3(t0, win[18 + 8 - 4], 1);
1113
}
1114

    
1115
/* return the number of decoded frames */
1116
static int mp_decode_layer1(MPADecodeContext *s)
1117
{
1118
    int bound, i, v, n, ch, j, mant;
1119
    uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
1120
    uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
1121

    
1122
    if (s->mode == MPA_JSTEREO)
1123
        bound = (s->mode_ext + 1) * 4;
1124
    else
1125
        bound = SBLIMIT;
1126

    
1127
    /* allocation bits */
1128
    for(i=0;i<bound;i++) {
1129
        for(ch=0;ch<s->nb_channels;ch++) {
1130
            allocation[ch][i] = get_bits(&s->gb, 4);
1131
        }
1132
    }
1133
    for(i=bound;i<SBLIMIT;i++) {
1134
        allocation[0][i] = get_bits(&s->gb, 4);
1135
    }
1136

    
1137
    /* scale factors */
1138
    for(i=0;i<bound;i++) {
1139
        for(ch=0;ch<s->nb_channels;ch++) {
1140
            if (allocation[ch][i])
1141
                scale_factors[ch][i] = get_bits(&s->gb, 6);
1142
        }
1143
    }
1144
    for(i=bound;i<SBLIMIT;i++) {
1145
        if (allocation[0][i]) {
1146
            scale_factors[0][i] = get_bits(&s->gb, 6);
1147
            scale_factors[1][i] = get_bits(&s->gb, 6);
1148
        }
1149
    }
1150

    
1151
    /* compute samples */
1152
    for(j=0;j<12;j++) {
1153
        for(i=0;i<bound;i++) {
1154
            for(ch=0;ch<s->nb_channels;ch++) {
1155
                n = allocation[ch][i];
1156
                if (n) {
1157
                    mant = get_bits(&s->gb, n + 1);
1158
                    v = l1_unscale(n, mant, scale_factors[ch][i]);
1159
                } else {
1160
                    v = 0;
1161
                }
1162
                s->sb_samples[ch][j][i] = v;
1163
            }
1164
        }
1165
        for(i=bound;i<SBLIMIT;i++) {
1166
            n = allocation[0][i];
1167
            if (n) {
1168
                mant = get_bits(&s->gb, n + 1);
1169
                v = l1_unscale(n, mant, scale_factors[0][i]);
1170
                s->sb_samples[0][j][i] = v;
1171
                v = l1_unscale(n, mant, scale_factors[1][i]);
1172
                s->sb_samples[1][j][i] = v;
1173
            } else {
1174
                s->sb_samples[0][j][i] = 0;
1175
                s->sb_samples[1][j][i] = 0;
1176
            }
1177
        }
1178
    }
1179
    return 12;
1180
}
1181

    
1182
static int mp_decode_layer2(MPADecodeContext *s)
1183
{
1184
    int sblimit; /* number of used subbands */
1185
    const unsigned char *alloc_table;
1186
    int table, bit_alloc_bits, i, j, ch, bound, v;
1187
    unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
1188
    unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
1189
    unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
1190
    int scale, qindex, bits, steps, k, l, m, b;
1191

    
1192
    /* select decoding table */
1193
    table = ff_mpa_l2_select_table(s->bit_rate / 1000, s->nb_channels,
1194
                            s->sample_rate, s->lsf);
1195
    sblimit = ff_mpa_sblimit_table[table];
1196
    alloc_table = ff_mpa_alloc_tables[table];
1197

    
1198
    if (s->mode == MPA_JSTEREO)
1199
        bound = (s->mode_ext + 1) * 4;
1200
    else
1201
        bound = sblimit;
1202

    
1203
    dprintf(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit);
1204

    
1205
    /* sanity check */
1206
    if( bound > sblimit ) bound = sblimit;
1207

    
1208
    /* parse bit allocation */
1209
    j = 0;
1210
    for(i=0;i<bound;i++) {
1211
        bit_alloc_bits = alloc_table[j];
1212
        for(ch=0;ch<s->nb_channels;ch++) {
1213
            bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
1214
        }
1215
        j += 1 << bit_alloc_bits;
1216
    }
1217
    for(i=bound;i<sblimit;i++) {
1218
        bit_alloc_bits = alloc_table[j];
1219
        v = get_bits(&s->gb, bit_alloc_bits);
1220
        bit_alloc[0][i] = v;
1221
        bit_alloc[1][i] = v;
1222
        j += 1 << bit_alloc_bits;
1223
    }
1224

    
1225
    /* scale codes */
1226
    for(i=0;i<sblimit;i++) {
1227
        for(ch=0;ch<s->nb_channels;ch++) {
1228
            if (bit_alloc[ch][i])
1229
                scale_code[ch][i] = get_bits(&s->gb, 2);
1230
        }
1231
    }
1232

    
1233
    /* scale factors */
1234
    for(i=0;i<sblimit;i++) {
1235
        for(ch=0;ch<s->nb_channels;ch++) {
1236
            if (bit_alloc[ch][i]) {
1237
                sf = scale_factors[ch][i];
1238
                switch(scale_code[ch][i]) {
1239
                default:
1240
                case 0:
1241
                    sf[0] = get_bits(&s->gb, 6);
1242
                    sf[1] = get_bits(&s->gb, 6);
1243
                    sf[2] = get_bits(&s->gb, 6);
1244
                    break;
1245
                case 2:
1246
                    sf[0] = get_bits(&s->gb, 6);
1247
                    sf[1] = sf[0];
1248
                    sf[2] = sf[0];
1249
                    break;
1250
                case 1:
1251
                    sf[0] = get_bits(&s->gb, 6);
1252
                    sf[2] = get_bits(&s->gb, 6);
1253
                    sf[1] = sf[0];
1254
                    break;
1255
                case 3:
1256
                    sf[0] = get_bits(&s->gb, 6);
1257
                    sf[2] = get_bits(&s->gb, 6);
1258
                    sf[1] = sf[2];
1259
                    break;
1260
                }
1261
            }
1262
        }
1263
    }
1264

    
1265
    /* samples */
1266
    for(k=0;k<3;k++) {
1267
        for(l=0;l<12;l+=3) {
1268
            j = 0;
1269
            for(i=0;i<bound;i++) {
1270
                bit_alloc_bits = alloc_table[j];
1271
                for(ch=0;ch<s->nb_channels;ch++) {
1272
                    b = bit_alloc[ch][i];
1273
                    if (b) {
1274
                        scale = scale_factors[ch][i][k];
1275
                        qindex = alloc_table[j+b];
1276
                        bits = ff_mpa_quant_bits[qindex];
1277
                        if (bits < 0) {
1278
                            /* 3 values at the same time */
1279
                            v = get_bits(&s->gb, -bits);
1280
                            steps = ff_mpa_quant_steps[qindex];
1281
                            s->sb_samples[ch][k * 12 + l + 0][i] =
1282
                                l2_unscale_group(steps, v % steps, scale);
1283
                            v = v / steps;
1284
                            s->sb_samples[ch][k * 12 + l + 1][i] =
1285
                                l2_unscale_group(steps, v % steps, scale);
1286
                            v = v / steps;
1287
                            s->sb_samples[ch][k * 12 + l + 2][i] =
1288
                                l2_unscale_group(steps, v, scale);
1289
                        } else {
1290
                            for(m=0;m<3;m++) {
1291
                                v = get_bits(&s->gb, bits);
1292
                                v = l1_unscale(bits - 1, v, scale);
1293
                                s->sb_samples[ch][k * 12 + l + m][i] = v;
1294
                            }
1295
                        }
1296
                    } else {
1297
                        s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1298
                        s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1299
                        s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1300
                    }
1301
                }
1302
                /* next subband in alloc table */
1303
                j += 1 << bit_alloc_bits;
1304
            }
1305
            /* XXX: find a way to avoid this duplication of code */
1306
            for(i=bound;i<sblimit;i++) {
1307
                bit_alloc_bits = alloc_table[j];
1308
                b = bit_alloc[0][i];
1309
                if (b) {
1310
                    int mant, scale0, scale1;
1311
                    scale0 = scale_factors[0][i][k];
1312
                    scale1 = scale_factors[1][i][k];
1313
                    qindex = alloc_table[j+b];
1314
                    bits = ff_mpa_quant_bits[qindex];
1315
                    if (bits < 0) {
1316
                        /* 3 values at the same time */
1317
                        v = get_bits(&s->gb, -bits);
1318
                        steps = ff_mpa_quant_steps[qindex];
1319
                        mant = v % steps;
1320
                        v = v / steps;
1321
                        s->sb_samples[0][k * 12 + l + 0][i] =
1322
                            l2_unscale_group(steps, mant, scale0);
1323
                        s->sb_samples[1][k * 12 + l + 0][i] =
1324
                            l2_unscale_group(steps, mant, scale1);
1325
                        mant = v % steps;
1326
                        v = v / steps;
1327
                        s->sb_samples[0][k * 12 + l + 1][i] =
1328
                            l2_unscale_group(steps, mant, scale0);
1329
                        s->sb_samples[1][k * 12 + l + 1][i] =
1330
                            l2_unscale_group(steps, mant, scale1);
1331
                        s->sb_samples[0][k * 12 + l + 2][i] =
1332
                            l2_unscale_group(steps, v, scale0);
1333
                        s->sb_samples[1][k * 12 + l + 2][i] =
1334
                            l2_unscale_group(steps, v, scale1);
1335
                    } else {
1336
                        for(m=0;m<3;m++) {
1337
                            mant = get_bits(&s->gb, bits);
1338
                            s->sb_samples[0][k * 12 + l + m][i] =
1339
                                l1_unscale(bits - 1, mant, scale0);
1340
                            s->sb_samples[1][k * 12 + l + m][i] =
1341
                                l1_unscale(bits - 1, mant, scale1);
1342
                        }
1343
                    }
1344
                } else {
1345
                    s->sb_samples[0][k * 12 + l + 0][i] = 0;
1346
                    s->sb_samples[0][k * 12 + l + 1][i] = 0;
1347
                    s->sb_samples[0][k * 12 + l + 2][i] = 0;
1348
                    s->sb_samples[1][k * 12 + l + 0][i] = 0;
1349
                    s->sb_samples[1][k * 12 + l + 1][i] = 0;
1350
                    s->sb_samples[1][k * 12 + l + 2][i] = 0;
1351
                }
1352
                /* next subband in alloc table */
1353
                j += 1 << bit_alloc_bits;
1354
            }
1355
            /* fill remaining samples to zero */
1356
            for(i=sblimit;i<SBLIMIT;i++) {
1357
                for(ch=0;ch<s->nb_channels;ch++) {
1358
                    s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1359
                    s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1360
                    s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1361
                }
1362
            }
1363
        }
1364
    }
1365
    return 3 * 12;
1366
}
1367

    
1368
#define SPLIT(dst,sf,n)\
1369
    if(n==3){\
1370
        int m= (sf*171)>>9;\
1371
        dst= sf - 3*m;\
1372
        sf=m;\
1373
    }else if(n==4){\
1374
        dst= sf&3;\
1375
        sf>>=2;\
1376
    }else if(n==5){\
1377
        int m= (sf*205)>>10;\
1378
        dst= sf - 5*m;\
1379
        sf=m;\
1380
    }else if(n==6){\
1381
        int m= (sf*171)>>10;\
1382
        dst= sf - 6*m;\
1383
        sf=m;\
1384
    }else{\
1385
        dst=0;\
1386
    }
1387

    
1388
static av_always_inline void lsf_sf_expand(int *slen,
1389
                                 int sf, int n1, int n2, int n3)
1390
{
1391
    SPLIT(slen[3], sf, n3)
1392
    SPLIT(slen[2], sf, n2)
1393
    SPLIT(slen[1], sf, n1)
1394
    slen[0] = sf;
1395
}
1396

    
1397
static void exponents_from_scale_factors(MPADecodeContext *s,
1398
                                         GranuleDef *g,
1399
                                         int16_t *exponents)
1400
{
1401
    const uint8_t *bstab, *pretab;
1402
    int len, i, j, k, l, v0, shift, gain, gains[3];
1403
    int16_t *exp_ptr;
1404

    
1405
    exp_ptr = exponents;
1406
    gain = g->global_gain - 210;
1407
    shift = g->scalefac_scale + 1;
1408

    
1409
    bstab = band_size_long[s->sample_rate_index];
1410
    pretab = mpa_pretab[g->preflag];
1411
    for(i=0;i<g->long_end;i++) {
1412
        v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400;
1413
        len = bstab[i];
1414
        for(j=len;j>0;j--)
1415
            *exp_ptr++ = v0;
1416
    }
1417

    
1418
    if (g->short_start < 13) {
1419
        bstab = band_size_short[s->sample_rate_index];
1420
        gains[0] = gain - (g->subblock_gain[0] << 3);
1421
        gains[1] = gain - (g->subblock_gain[1] << 3);
1422
        gains[2] = gain - (g->subblock_gain[2] << 3);
1423
        k = g->long_end;
1424
        for(i=g->short_start;i<13;i++) {
1425
            len = bstab[i];
1426
            for(l=0;l<3;l++) {
1427
                v0 = gains[l] - (g->scale_factors[k++] << shift) + 400;
1428
                for(j=len;j>0;j--)
1429
                *exp_ptr++ = v0;
1430
            }
1431
        }
1432
    }
1433
}
1434

    
1435
/* handle n = 0 too */
1436
static inline int get_bitsz(GetBitContext *s, int n)
1437
{
1438
    if (n == 0)
1439
        return 0;
1440
    else
1441
        return get_bits(s, n);
1442
}
1443

    
1444

    
1445
static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos, int *end_pos2){
1446
    if(s->in_gb.buffer && *pos >= s->gb.size_in_bits){
1447
        s->gb= s->in_gb;
1448
        s->in_gb.buffer=NULL;
1449
        assert((get_bits_count(&s->gb) & 7) == 0);
1450
        skip_bits_long(&s->gb, *pos - *end_pos);
1451
        *end_pos2=
1452
        *end_pos= *end_pos2 + get_bits_count(&s->gb) - *pos;
1453
        *pos= get_bits_count(&s->gb);
1454
    }
1455
}
1456

    
1457
/* Following is a optimized code for
1458
            INTFLOAT v = *src
1459
            if(get_bits1(&s->gb))
1460
                v = -v;
1461
            *dst = v;
1462
*/
1463
#if CONFIG_FLOAT
1464
#define READ_FLIP_SIGN(dst,src)\
1465
            v = AV_RN32A(src) ^ (get_bits1(&s->gb)<<31);\
1466
            AV_WN32A(dst, v);
1467
#else
1468
#define READ_FLIP_SIGN(dst,src)\
1469
            v= -get_bits1(&s->gb);\
1470
            *(dst) = (*(src) ^ v) - v;
1471
#endif
1472

    
1473
static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
1474
                          int16_t *exponents, int end_pos2)
1475
{
1476
    int s_index;
1477
    int i;
1478
    int last_pos, bits_left;
1479
    VLC *vlc;
1480
    int end_pos= FFMIN(end_pos2, s->gb.size_in_bits);
1481

    
1482
    /* low frequencies (called big values) */
1483
    s_index = 0;
1484
    for(i=0;i<3;i++) {
1485
        int j, k, l, linbits;
1486
        j = g->region_size[i];
1487
        if (j == 0)
1488
            continue;
1489
        /* select vlc table */
1490
        k = g->table_select[i];
1491
        l = mpa_huff_data[k][0];
1492
        linbits = mpa_huff_data[k][1];
1493
        vlc = &huff_vlc[l];
1494

    
1495
        if(!l){
1496
            memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*2*j);
1497
            s_index += 2*j;
1498
            continue;
1499
        }
1500

    
1501
        /* read huffcode and compute each couple */
1502
        for(;j>0;j--) {
1503
            int exponent, x, y;
1504
            int v;
1505
            int pos= get_bits_count(&s->gb);
1506

    
1507
            if (pos >= end_pos){
1508
//                av_log(NULL, AV_LOG_ERROR, "pos: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);
1509
                switch_buffer(s, &pos, &end_pos, &end_pos2);
1510
//                av_log(NULL, AV_LOG_ERROR, "new pos: %d %d\n", pos, end_pos);
1511
                if(pos >= end_pos)
1512
                    break;
1513
            }
1514
            y = get_vlc2(&s->gb, vlc->table, 7, 3);
1515

    
1516
            if(!y){
1517
                g->sb_hybrid[s_index  ] =
1518
                g->sb_hybrid[s_index+1] = 0;
1519
                s_index += 2;
1520
                continue;
1521
            }
1522

    
1523
            exponent= exponents[s_index];
1524

    
1525
            dprintf(s->avctx, "region=%d n=%d x=%d y=%d exp=%d\n",
1526
                    i, g->region_size[i] - j, x, y, exponent);
1527
            if(y&16){
1528
                x = y >> 5;
1529
                y = y & 0x0f;
1530
                if (x < 15){
1531
                    READ_FLIP_SIGN(g->sb_hybrid+s_index, RENAME(expval_table)[ exponent ]+x)
1532
                }else{
1533
                    x += get_bitsz(&s->gb, linbits);
1534
                    v = l3_unscale(x, exponent);
1535
                    if (get_bits1(&s->gb))
1536
                        v = -v;
1537
                    g->sb_hybrid[s_index] = v;
1538
                }
1539
                if (y < 15){
1540
                    READ_FLIP_SIGN(g->sb_hybrid+s_index+1, RENAME(expval_table)[ exponent ]+y)
1541
                }else{
1542
                    y += get_bitsz(&s->gb, linbits);
1543
                    v = l3_unscale(y, exponent);
1544
                    if (get_bits1(&s->gb))
1545
                        v = -v;
1546
                    g->sb_hybrid[s_index+1] = v;
1547
                }
1548
            }else{
1549
                x = y >> 5;
1550
                y = y & 0x0f;
1551
                x += y;
1552
                if (x < 15){
1553
                    READ_FLIP_SIGN(g->sb_hybrid+s_index+!!y, RENAME(expval_table)[ exponent ]+x)
1554
                }else{
1555
                    x += get_bitsz(&s->gb, linbits);
1556
                    v = l3_unscale(x, exponent);
1557
                    if (get_bits1(&s->gb))
1558
                        v = -v;
1559
                    g->sb_hybrid[s_index+!!y] = v;
1560
                }
1561
                g->sb_hybrid[s_index+ !y] = 0;
1562
            }
1563
            s_index+=2;
1564
        }
1565
    }
1566

    
1567
    /* high frequencies */
1568
    vlc = &huff_quad_vlc[g->count1table_select];
1569
    last_pos=0;
1570
    while (s_index <= 572) {
1571
        int pos, code;
1572
        pos = get_bits_count(&s->gb);
1573
        if (pos >= end_pos) {
1574
            if (pos > end_pos2 && last_pos){
1575
                /* some encoders generate an incorrect size for this
1576
                   part. We must go back into the data */
1577
                s_index -= 4;
1578
                skip_bits_long(&s->gb, last_pos - pos);
1579
                av_log(s->avctx, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos);
1580
                if(s->error_recognition >= FF_ER_COMPLIANT)
1581
                    s_index=0;
1582
                break;
1583
            }
1584
//                av_log(NULL, AV_LOG_ERROR, "pos2: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);
1585
            switch_buffer(s, &pos, &end_pos, &end_pos2);
1586
//                av_log(NULL, AV_LOG_ERROR, "new pos2: %d %d %d\n", pos, end_pos, s_index);
1587
            if(pos >= end_pos)
1588
                break;
1589
        }
1590
        last_pos= pos;
1591

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

    
1620
    i= get_bits_count(&s->gb);
1621
    switch_buffer(s, &i, &end_pos, &end_pos2);
1622

    
1623
    return 0;
1624
}
1625

    
1626
/* Reorder short blocks from bitstream order to interleaved order. It
1627
   would be faster to do it in parsing, but the code would be far more
1628
   complicated */
1629
static void reorder_block(MPADecodeContext *s, GranuleDef *g)
1630
{
1631
    int i, j, len;
1632
    INTFLOAT *ptr, *dst, *ptr1;
1633
    INTFLOAT tmp[576];
1634

    
1635
    if (g->block_type != 2)
1636
        return;
1637

    
1638
    if (g->switch_point) {
1639
        if (s->sample_rate_index != 8) {
1640
            ptr = g->sb_hybrid + 36;
1641
        } else {
1642
            ptr = g->sb_hybrid + 48;
1643
        }
1644
    } else {
1645
        ptr = g->sb_hybrid;
1646
    }
1647

    
1648
    for(i=g->short_start;i<13;i++) {
1649
        len = band_size_short[s->sample_rate_index][i];
1650
        ptr1 = ptr;
1651
        dst = tmp;
1652
        for(j=len;j>0;j--) {
1653
            *dst++ = ptr[0*len];
1654
            *dst++ = ptr[1*len];
1655
            *dst++ = ptr[2*len];
1656
            ptr++;
1657
        }
1658
        ptr+=2*len;
1659
        memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));
1660
    }
1661
}
1662

    
1663
#define ISQRT2 FIXR(0.70710678118654752440)
1664

    
1665
static void compute_stereo(MPADecodeContext *s,
1666
                           GranuleDef *g0, GranuleDef *g1)
1667
{
1668
    int i, j, k, l;
1669
    int sf_max, sf, len, non_zero_found;
1670
    INTFLOAT (*is_tab)[16], *tab0, *tab1, tmp0, tmp1, v1, v2;
1671
    int non_zero_found_short[3];
1672

    
1673
    /* intensity stereo */
1674
    if (s->mode_ext & MODE_EXT_I_STEREO) {
1675
        if (!s->lsf) {
1676
            is_tab = is_table;
1677
            sf_max = 7;
1678
        } else {
1679
            is_tab = is_table_lsf[g1->scalefac_compress & 1];
1680
            sf_max = 16;
1681
        }
1682

    
1683
        tab0 = g0->sb_hybrid + 576;
1684
        tab1 = g1->sb_hybrid + 576;
1685

    
1686
        non_zero_found_short[0] = 0;
1687
        non_zero_found_short[1] = 0;
1688
        non_zero_found_short[2] = 0;
1689
        k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1690
        for(i = 12;i >= g1->short_start;i--) {
1691
            /* for last band, use previous scale factor */
1692
            if (i != 11)
1693
                k -= 3;
1694
            len = band_size_short[s->sample_rate_index][i];
1695
            for(l=2;l>=0;l--) {
1696
                tab0 -= len;
1697
                tab1 -= len;
1698
                if (!non_zero_found_short[l]) {
1699
                    /* test if non zero band. if so, stop doing i-stereo */
1700
                    for(j=0;j<len;j++) {
1701
                        if (tab1[j] != 0) {
1702
                            non_zero_found_short[l] = 1;
1703
                            goto found1;
1704
                        }
1705
                    }
1706
                    sf = g1->scale_factors[k + l];
1707
                    if (sf >= sf_max)
1708
                        goto found1;
1709

    
1710
                    v1 = is_tab[0][sf];
1711
                    v2 = is_tab[1][sf];
1712
                    for(j=0;j<len;j++) {
1713
                        tmp0 = tab0[j];
1714
                        tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1715
                        tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1716
                    }
1717
                } else {
1718
                found1:
1719
                    if (s->mode_ext & MODE_EXT_MS_STEREO) {
1720
                        /* lower part of the spectrum : do ms stereo
1721
                           if enabled */
1722
                        for(j=0;j<len;j++) {
1723
                            tmp0 = tab0[j];
1724
                            tmp1 = tab1[j];
1725
                            tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1726
                            tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1727
                        }
1728
                    }
1729
                }
1730
            }
1731
        }
1732

    
1733
        non_zero_found = non_zero_found_short[0] |
1734
            non_zero_found_short[1] |
1735
            non_zero_found_short[2];
1736

    
1737
        for(i = g1->long_end - 1;i >= 0;i--) {
1738
            len = band_size_long[s->sample_rate_index][i];
1739
            tab0 -= len;
1740
            tab1 -= len;
1741
            /* test if non zero band. if so, stop doing i-stereo */
1742
            if (!non_zero_found) {
1743
                for(j=0;j<len;j++) {
1744
                    if (tab1[j] != 0) {
1745
                        non_zero_found = 1;
1746
                        goto found2;
1747
                    }
1748
                }
1749
                /* for last band, use previous scale factor */
1750
                k = (i == 21) ? 20 : i;
1751
                sf = g1->scale_factors[k];
1752
                if (sf >= sf_max)
1753
                    goto found2;
1754
                v1 = is_tab[0][sf];
1755
                v2 = is_tab[1][sf];
1756
                for(j=0;j<len;j++) {
1757
                    tmp0 = tab0[j];
1758
                    tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1759
                    tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1760
                }
1761
            } else {
1762
            found2:
1763
                if (s->mode_ext & MODE_EXT_MS_STEREO) {
1764
                    /* lower part of the spectrum : do ms stereo
1765
                       if enabled */
1766
                    for(j=0;j<len;j++) {
1767
                        tmp0 = tab0[j];
1768
                        tmp1 = tab1[j];
1769
                        tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1770
                        tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1771
                    }
1772
                }
1773
            }
1774
        }
1775
    } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1776
        /* ms stereo ONLY */
1777
        /* NOTE: the 1/sqrt(2) normalization factor is included in the
1778
           global gain */
1779
        tab0 = g0->sb_hybrid;
1780
        tab1 = g1->sb_hybrid;
1781
        for(i=0;i<576;i++) {
1782
            tmp0 = tab0[i];
1783
            tmp1 = tab1[i];
1784
            tab0[i] = tmp0 + tmp1;
1785
            tab1[i] = tmp0 - tmp1;
1786
        }
1787
    }
1788
}
1789

    
1790
static void compute_antialias_integer(MPADecodeContext *s,
1791
                              GranuleDef *g)
1792
{
1793
    int32_t *ptr, *csa;
1794
    int n, i;
1795

    
1796
    /* we antialias only "long" bands */
1797
    if (g->block_type == 2) {
1798
        if (!g->switch_point)
1799
            return;
1800
        /* XXX: check this for 8000Hz case */
1801
        n = 1;
1802
    } else {
1803
        n = SBLIMIT - 1;
1804
    }
1805

    
1806
    ptr = g->sb_hybrid + 18;
1807
    for(i = n;i > 0;i--) {
1808
        int tmp0, tmp1, tmp2;
1809
        csa = &csa_table[0][0];
1810
#define INT_AA(j) \
1811
            tmp0 = ptr[-1-j];\
1812
            tmp1 = ptr[   j];\
1813
            tmp2= MULH(tmp0 + tmp1, csa[0+4*j]);\
1814
            ptr[-1-j] = 4*(tmp2 - MULH(tmp1, csa[2+4*j]));\
1815
            ptr[   j] = 4*(tmp2 + MULH(tmp0, csa[3+4*j]));
1816

    
1817
        INT_AA(0)
1818
        INT_AA(1)
1819
        INT_AA(2)
1820
        INT_AA(3)
1821
        INT_AA(4)
1822
        INT_AA(5)
1823
        INT_AA(6)
1824
        INT_AA(7)
1825

    
1826
        ptr += 18;
1827
    }
1828
}
1829

    
1830
static void compute_antialias_float(MPADecodeContext *s,
1831
                              GranuleDef *g)
1832
{
1833
    float *ptr;
1834
    int n, i;
1835

    
1836
    /* we antialias only "long" bands */
1837
    if (g->block_type == 2) {
1838
        if (!g->switch_point)
1839
            return;
1840
        /* XXX: check this for 8000Hz case */
1841
        n = 1;
1842
    } else {
1843
        n = SBLIMIT - 1;
1844
    }
1845

    
1846
    ptr = g->sb_hybrid + 18;
1847
    for(i = n;i > 0;i--) {
1848
        float tmp0, tmp1;
1849
        float *csa = &csa_table_float[0][0];
1850
#define FLOAT_AA(j)\
1851
        tmp0= ptr[-1-j];\
1852
        tmp1= ptr[   j];\
1853
        ptr[-1-j] = tmp0 * csa[0+4*j] - tmp1 * csa[1+4*j];\
1854
        ptr[   j] = tmp0 * csa[1+4*j] + tmp1 * csa[0+4*j];
1855

    
1856
        FLOAT_AA(0)
1857
        FLOAT_AA(1)
1858
        FLOAT_AA(2)
1859
        FLOAT_AA(3)
1860
        FLOAT_AA(4)
1861
        FLOAT_AA(5)
1862
        FLOAT_AA(6)
1863
        FLOAT_AA(7)
1864

    
1865
        ptr += 18;
1866
    }
1867
}
1868

    
1869
static void compute_imdct(MPADecodeContext *s,
1870
                          GranuleDef *g,
1871
                          INTFLOAT *sb_samples,
1872
                          INTFLOAT *mdct_buf)
1873
{
1874
    INTFLOAT *win, *win1, *out_ptr, *ptr, *buf, *ptr1;
1875
    INTFLOAT out2[12];
1876
    int i, j, mdct_long_end, sblimit;
1877

    
1878
    /* find last non zero block */
1879
    ptr = g->sb_hybrid + 576;
1880
    ptr1 = g->sb_hybrid + 2 * 18;
1881
    while (ptr >= ptr1) {
1882
        int32_t *p;
1883
        ptr -= 6;
1884
        p= (int32_t*)ptr;
1885
        if(p[0] | p[1] | p[2] | p[3] | p[4] | p[5])
1886
            break;
1887
    }
1888
    sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
1889

    
1890
    if (g->block_type == 2) {
1891
        /* XXX: check for 8000 Hz */
1892
        if (g->switch_point)
1893
            mdct_long_end = 2;
1894
        else
1895
            mdct_long_end = 0;
1896
    } else {
1897
        mdct_long_end = sblimit;
1898
    }
1899

    
1900
    buf = mdct_buf;
1901
    ptr = g->sb_hybrid;
1902
    for(j=0;j<mdct_long_end;j++) {
1903
        /* apply window & overlap with previous buffer */
1904
        out_ptr = sb_samples + j;
1905
        /* select window */
1906
        if (g->switch_point && j < 2)
1907
            win1 = mdct_win[0];
1908
        else
1909
            win1 = mdct_win[g->block_type];
1910
        /* select frequency inversion */
1911
        win = win1 + ((4 * 36) & -(j & 1));
1912
        imdct36(out_ptr, buf, ptr, win);
1913
        out_ptr += 18*SBLIMIT;
1914
        ptr += 18;
1915
        buf += 18;
1916
    }
1917
    for(j=mdct_long_end;j<sblimit;j++) {
1918
        /* select frequency inversion */
1919
        win = mdct_win[2] + ((4 * 36) & -(j & 1));
1920
        out_ptr = sb_samples + j;
1921

    
1922
        for(i=0; i<6; i++){
1923
            *out_ptr = buf[i];
1924
            out_ptr += SBLIMIT;
1925
        }
1926
        imdct12(out2, ptr + 0);
1927
        for(i=0;i<6;i++) {
1928
            *out_ptr     = MULH3(out2[i    ], win[i    ], 1) + buf[i + 6*1];
1929
            buf[i + 6*2] = MULH3(out2[i + 6], win[i + 6], 1);
1930
            out_ptr += SBLIMIT;
1931
        }
1932
        imdct12(out2, ptr + 1);
1933
        for(i=0;i<6;i++) {
1934
            *out_ptr     = MULH3(out2[i    ], win[i    ], 1) + buf[i + 6*2];
1935
            buf[i + 6*0] = MULH3(out2[i + 6], win[i + 6], 1);
1936
            out_ptr += SBLIMIT;
1937
        }
1938
        imdct12(out2, ptr + 2);
1939
        for(i=0;i<6;i++) {
1940
            buf[i + 6*0] = MULH3(out2[i    ], win[i    ], 1) + buf[i + 6*0];
1941
            buf[i + 6*1] = MULH3(out2[i + 6], win[i + 6], 1);
1942
            buf[i + 6*2] = 0;
1943
        }
1944
        ptr += 18;
1945
        buf += 18;
1946
    }
1947
    /* zero bands */
1948
    for(j=sblimit;j<SBLIMIT;j++) {
1949
        /* overlap */
1950
        out_ptr = sb_samples + j;
1951
        for(i=0;i<18;i++) {
1952
            *out_ptr = buf[i];
1953
            buf[i] = 0;
1954
            out_ptr += SBLIMIT;
1955
        }
1956
        buf += 18;
1957
    }
1958
}
1959

    
1960
/* main layer3 decoding function */
1961
static int mp_decode_layer3(MPADecodeContext *s)
1962
{
1963
    int nb_granules, main_data_begin, private_bits;
1964
    int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;
1965
    GranuleDef *g;
1966
    int16_t exponents[576]; //FIXME try INTFLOAT
1967

    
1968
    /* read side info */
1969
    if (s->lsf) {
1970
        main_data_begin = get_bits(&s->gb, 8);
1971
        private_bits = get_bits(&s->gb, s->nb_channels);
1972
        nb_granules = 1;
1973
    } else {
1974
        main_data_begin = get_bits(&s->gb, 9);
1975
        if (s->nb_channels == 2)
1976
            private_bits = get_bits(&s->gb, 3);
1977
        else
1978
            private_bits = get_bits(&s->gb, 5);
1979
        nb_granules = 2;
1980
        for(ch=0;ch<s->nb_channels;ch++) {
1981
            s->granules[ch][0].scfsi = 0;/* all scale factors are transmitted */
1982
            s->granules[ch][1].scfsi = get_bits(&s->gb, 4);
1983
        }
1984
    }
1985

    
1986
    for(gr=0;gr<nb_granules;gr++) {
1987
        for(ch=0;ch<s->nb_channels;ch++) {
1988
            dprintf(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);
1989
            g = &s->granules[ch][gr];
1990
            g->part2_3_length = get_bits(&s->gb, 12);
1991
            g->big_values = get_bits(&s->gb, 9);
1992
            if(g->big_values > 288){
1993
                av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");
1994
                return -1;
1995
            }
1996

    
1997
            g->global_gain = get_bits(&s->gb, 8);
1998
            /* if MS stereo only is selected, we precompute the
1999
               1/sqrt(2) renormalization factor */
2000
            if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
2001
                MODE_EXT_MS_STEREO)
2002
                g->global_gain -= 2;
2003
            if (s->lsf)
2004
                g->scalefac_compress = get_bits(&s->gb, 9);
2005
            else
2006
                g->scalefac_compress = get_bits(&s->gb, 4);
2007
            blocksplit_flag = get_bits1(&s->gb);
2008
            if (blocksplit_flag) {
2009
                g->block_type = get_bits(&s->gb, 2);
2010
                if (g->block_type == 0){
2011
                    av_log(s->avctx, AV_LOG_ERROR, "invalid block type\n");
2012
                    return -1;
2013
                }
2014
                g->switch_point = get_bits1(&s->gb);
2015
                for(i=0;i<2;i++)
2016
                    g->table_select[i] = get_bits(&s->gb, 5);
2017
                for(i=0;i<3;i++)
2018
                    g->subblock_gain[i] = get_bits(&s->gb, 3);
2019
                ff_init_short_region(s, g);
2020
            } else {
2021
                int region_address1, region_address2;
2022
                g->block_type = 0;
2023
                g->switch_point = 0;
2024
                for(i=0;i<3;i++)
2025
                    g->table_select[i] = get_bits(&s->gb, 5);
2026
                /* compute huffman coded region sizes */
2027
                region_address1 = get_bits(&s->gb, 4);
2028
                region_address2 = get_bits(&s->gb, 3);
2029
                dprintf(s->avctx, "region1=%d region2=%d\n",
2030
                        region_address1, region_address2);
2031
                ff_init_long_region(s, g, region_address1, region_address2);
2032
            }
2033
            ff_region_offset2size(g);
2034
            ff_compute_band_indexes(s, g);
2035

    
2036
            g->preflag = 0;
2037
            if (!s->lsf)
2038
                g->preflag = get_bits1(&s->gb);
2039
            g->scalefac_scale = get_bits1(&s->gb);
2040
            g->count1table_select = get_bits1(&s->gb);
2041
            dprintf(s->avctx, "block_type=%d switch_point=%d\n",
2042
                    g->block_type, g->switch_point);
2043
        }
2044
    }
2045

    
2046
  if (!s->adu_mode) {
2047
    const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);
2048
    assert((get_bits_count(&s->gb) & 7) == 0);
2049
    /* now we get bits from the main_data_begin offset */
2050
    dprintf(s->avctx, "seekback: %d\n", main_data_begin);
2051
//av_log(NULL, AV_LOG_ERROR, "backstep:%d, lastbuf:%d\n", main_data_begin, s->last_buf_size);
2052

    
2053
    memcpy(s->last_buf + s->last_buf_size, ptr, EXTRABYTES);
2054
    s->in_gb= s->gb;
2055
        init_get_bits(&s->gb, s->last_buf, s->last_buf_size*8);
2056
        skip_bits_long(&s->gb, 8*(s->last_buf_size - main_data_begin));
2057
  }
2058

    
2059
    for(gr=0;gr<nb_granules;gr++) {
2060
        for(ch=0;ch<s->nb_channels;ch++) {
2061
            g = &s->granules[ch][gr];
2062
            if(get_bits_count(&s->gb)<0){
2063
                av_log(s->avctx, AV_LOG_DEBUG, "mdb:%d, lastbuf:%d skipping granule %d\n",
2064
                                            main_data_begin, s->last_buf_size, gr);
2065
                skip_bits_long(&s->gb, g->part2_3_length);
2066
                memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));
2067
                if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->in_gb.buffer){
2068
                    skip_bits_long(&s->in_gb, get_bits_count(&s->gb) - s->gb.size_in_bits);
2069
                    s->gb= s->in_gb;
2070
                    s->in_gb.buffer=NULL;
2071
                }
2072
                continue;
2073
            }
2074

    
2075
            bits_pos = get_bits_count(&s->gb);
2076

    
2077
            if (!s->lsf) {
2078
                uint8_t *sc;
2079
                int slen, slen1, slen2;
2080

    
2081
                /* MPEG1 scale factors */
2082
                slen1 = slen_table[0][g->scalefac_compress];
2083
                slen2 = slen_table[1][g->scalefac_compress];
2084
                dprintf(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);
2085
                if (g->block_type == 2) {
2086
                    n = g->switch_point ? 17 : 18;
2087
                    j = 0;
2088
                    if(slen1){
2089
                        for(i=0;i<n;i++)
2090
                            g->scale_factors[j++] = get_bits(&s->gb, slen1);
2091
                    }else{
2092
                        for(i=0;i<n;i++)
2093
                            g->scale_factors[j++] = 0;
2094
                    }
2095
                    if(slen2){
2096
                        for(i=0;i<18;i++)
2097
                            g->scale_factors[j++] = get_bits(&s->gb, slen2);
2098
                        for(i=0;i<3;i++)
2099
                            g->scale_factors[j++] = 0;
2100
                    }else{
2101
                        for(i=0;i<21;i++)
2102
                            g->scale_factors[j++] = 0;
2103
                    }
2104
                } else {
2105
                    sc = s->granules[ch][0].scale_factors;
2106
                    j = 0;
2107
                    for(k=0;k<4;k++) {
2108
                        n = (k == 0 ? 6 : 5);
2109
                        if ((g->scfsi & (0x8 >> k)) == 0) {
2110
                            slen = (k < 2) ? slen1 : slen2;
2111
                            if(slen){
2112
                                for(i=0;i<n;i++)
2113
                                    g->scale_factors[j++] = get_bits(&s->gb, slen);
2114
                            }else{
2115
                                for(i=0;i<n;i++)
2116
                                    g->scale_factors[j++] = 0;
2117
                            }
2118
                        } else {
2119
                            /* simply copy from last granule */
2120
                            for(i=0;i<n;i++) {
2121
                                g->scale_factors[j] = sc[j];
2122
                                j++;
2123
                            }
2124
                        }
2125
                    }
2126
                    g->scale_factors[j++] = 0;
2127
                }
2128
            } else {
2129
                int tindex, tindex2, slen[4], sl, sf;
2130

    
2131
                /* LSF scale factors */
2132
                if (g->block_type == 2) {
2133
                    tindex = g->switch_point ? 2 : 1;
2134
                } else {
2135
                    tindex = 0;
2136
                }
2137
                sf = g->scalefac_compress;
2138
                if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
2139
                    /* intensity stereo case */
2140
                    sf >>= 1;
2141
                    if (sf < 180) {
2142
                        lsf_sf_expand(slen, sf, 6, 6, 0);
2143
                        tindex2 = 3;
2144
                    } else if (sf < 244) {
2145
                        lsf_sf_expand(slen, sf - 180, 4, 4, 0);
2146
                        tindex2 = 4;
2147
                    } else {
2148
                        lsf_sf_expand(slen, sf - 244, 3, 0, 0);
2149
                        tindex2 = 5;
2150
                    }
2151
                } else {
2152
                    /* normal case */
2153
                    if (sf < 400) {
2154
                        lsf_sf_expand(slen, sf, 5, 4, 4);
2155
                        tindex2 = 0;
2156
                    } else if (sf < 500) {
2157
                        lsf_sf_expand(slen, sf - 400, 5, 4, 0);
2158
                        tindex2 = 1;
2159
                    } else {
2160
                        lsf_sf_expand(slen, sf - 500, 3, 0, 0);
2161
                        tindex2 = 2;
2162
                        g->preflag = 1;
2163
                    }
2164
                }
2165

    
2166
                j = 0;
2167
                for(k=0;k<4;k++) {
2168
                    n = lsf_nsf_table[tindex2][tindex][k];
2169
                    sl = slen[k];
2170
                    if(sl){
2171
                        for(i=0;i<n;i++)
2172
                            g->scale_factors[j++] = get_bits(&s->gb, sl);
2173
                    }else{
2174
                        for(i=0;i<n;i++)
2175
                            g->scale_factors[j++] = 0;
2176
                    }
2177
                }
2178
                /* XXX: should compute exact size */
2179
                for(;j<40;j++)
2180
                    g->scale_factors[j] = 0;
2181
            }
2182

    
2183
            exponents_from_scale_factors(s, g, exponents);
2184

    
2185
            /* read Huffman coded residue */
2186
            huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);
2187
        } /* ch */
2188

    
2189
        if (s->nb_channels == 2)
2190
            compute_stereo(s, &s->granules[0][gr], &s->granules[1][gr]);
2191

    
2192
        for(ch=0;ch<s->nb_channels;ch++) {
2193
            g = &s->granules[ch][gr];
2194

    
2195
            reorder_block(s, g);
2196
            compute_antialias(s, g);
2197
            compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
2198
        }
2199
    } /* gr */
2200
    if(get_bits_count(&s->gb)<0)
2201
        skip_bits_long(&s->gb, -get_bits_count(&s->gb));
2202
    return nb_granules * 18;
2203
}
2204

    
2205
static int mp_decode_frame(MPADecodeContext *s,
2206
                           OUT_INT *samples, const uint8_t *buf, int buf_size)
2207
{
2208
    int i, nb_frames, ch;
2209
    OUT_INT *samples_ptr;
2210

    
2211
    init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE)*8);
2212

    
2213
    /* skip error protection field */
2214
    if (s->error_protection)
2215
        skip_bits(&s->gb, 16);
2216

    
2217
    dprintf(s->avctx, "frame %d:\n", s->frame_count);
2218
    switch(s->layer) {
2219
    case 1:
2220
        s->avctx->frame_size = 384;
2221
        nb_frames = mp_decode_layer1(s);
2222
        break;
2223
    case 2:
2224
        s->avctx->frame_size = 1152;
2225
        nb_frames = mp_decode_layer2(s);
2226
        break;
2227
    case 3:
2228
        s->avctx->frame_size = s->lsf ? 576 : 1152;
2229
    default:
2230
        nb_frames = mp_decode_layer3(s);
2231

    
2232
        s->last_buf_size=0;
2233
        if(s->in_gb.buffer){
2234
            align_get_bits(&s->gb);
2235
            i= get_bits_left(&s->gb)>>3;
2236
            if(i >= 0 && i <= BACKSTEP_SIZE){
2237
                memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);
2238
                s->last_buf_size=i;
2239
            }else
2240
                av_log(s->avctx, AV_LOG_ERROR, "invalid old backstep %d\n", i);
2241
            s->gb= s->in_gb;
2242
            s->in_gb.buffer= NULL;
2243
        }
2244

    
2245
        align_get_bits(&s->gb);
2246
        assert((get_bits_count(&s->gb) & 7) == 0);
2247
        i= get_bits_left(&s->gb)>>3;
2248

    
2249
        if(i<0 || i > BACKSTEP_SIZE || nb_frames<0){
2250
            if(i<0)
2251
                av_log(s->avctx, AV_LOG_ERROR, "invalid new backstep %d\n", i);
2252
            i= FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);
2253
        }
2254
        assert(i <= buf_size - HEADER_SIZE && i>= 0);
2255
        memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);
2256
        s->last_buf_size += i;
2257

    
2258
        break;
2259
    }
2260

    
2261
    /* apply the synthesis filter */
2262
    for(ch=0;ch<s->nb_channels;ch++) {
2263
        samples_ptr = samples + ch;
2264
        for(i=0;i<nb_frames;i++) {
2265
            RENAME(ff_mpa_synth_filter)(
2266
#if CONFIG_FLOAT
2267
                         s,
2268
#endif
2269
                         s->synth_buf[ch], &(s->synth_buf_offset[ch]),
2270
                         RENAME(ff_mpa_synth_window), &s->dither_state,
2271
                         samples_ptr, s->nb_channels,
2272
                         s->sb_samples[ch][i]);
2273
            samples_ptr += 32 * s->nb_channels;
2274
        }
2275
    }
2276

    
2277
    return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
2278
}
2279

    
2280
static int decode_frame(AVCodecContext * avctx,
2281
                        void *data, int *data_size,
2282
                        AVPacket *avpkt)
2283
{
2284
    const uint8_t *buf = avpkt->data;
2285
    int buf_size = avpkt->size;
2286
    MPADecodeContext *s = avctx->priv_data;
2287
    uint32_t header;
2288
    int out_size;
2289
    OUT_INT *out_samples = data;
2290

    
2291
    if(buf_size < HEADER_SIZE)
2292
        return -1;
2293

    
2294
    header = AV_RB32(buf);
2295
    if(ff_mpa_check_header(header) < 0){
2296
        av_log(avctx, AV_LOG_ERROR, "Header missing\n");
2297
        return -1;
2298
    }
2299

    
2300
    if (ff_mpegaudio_decode_header((MPADecodeHeader *)s, header) == 1) {
2301
        /* free format: prepare to compute frame size */
2302
        s->frame_size = -1;
2303
        return -1;
2304
    }
2305
    /* update codec info */
2306
    avctx->channels = s->nb_channels;
2307
    avctx->bit_rate = s->bit_rate;
2308
    avctx->sub_id = s->layer;
2309

    
2310
    if(*data_size < 1152*avctx->channels*sizeof(OUT_INT))
2311
        return -1;
2312
    *data_size = 0;
2313

    
2314
    if(s->frame_size<=0 || s->frame_size > buf_size){
2315
        av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
2316
        return -1;
2317
    }else if(s->frame_size < buf_size){
2318
        av_log(avctx, AV_LOG_ERROR, "incorrect frame size\n");
2319
        buf_size= s->frame_size;
2320
    }
2321

    
2322
    out_size = mp_decode_frame(s, out_samples, buf, buf_size);
2323
    if(out_size>=0){
2324
        *data_size = out_size;
2325
        avctx->sample_rate = s->sample_rate;
2326
        //FIXME maybe move the other codec info stuff from above here too
2327
    }else
2328
        av_log(avctx, AV_LOG_DEBUG, "Error while decoding MPEG audio frame.\n"); //FIXME return -1 / but also return the number of bytes consumed
2329
    s->frame_size = 0;
2330
    return buf_size;
2331
}
2332

    
2333
static void flush(AVCodecContext *avctx){
2334
    MPADecodeContext *s = avctx->priv_data;
2335
    memset(s->synth_buf, 0, sizeof(s->synth_buf));
2336
    s->last_buf_size= 0;
2337
}
2338

    
2339
#if CONFIG_MP3ADU_DECODER
2340
static int decode_frame_adu(AVCodecContext * avctx,
2341
                        void *data, int *data_size,
2342
                        AVPacket *avpkt)
2343
{
2344
    const uint8_t *buf = avpkt->data;
2345
    int buf_size = avpkt->size;
2346
    MPADecodeContext *s = avctx->priv_data;
2347
    uint32_t header;
2348
    int len, out_size;
2349
    OUT_INT *out_samples = data;
2350

    
2351
    len = buf_size;
2352

    
2353
    // Discard too short frames
2354
    if (buf_size < HEADER_SIZE) {
2355
        *data_size = 0;
2356
        return buf_size;
2357
    }
2358

    
2359

    
2360
    if (len > MPA_MAX_CODED_FRAME_SIZE)
2361
        len = MPA_MAX_CODED_FRAME_SIZE;
2362

    
2363
    // Get header and restore sync word
2364
    header = AV_RB32(buf) | 0xffe00000;
2365

    
2366
    if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame
2367
        *data_size = 0;
2368
        return buf_size;
2369
    }
2370

    
2371
    ff_mpegaudio_decode_header((MPADecodeHeader *)s, header);
2372
    /* update codec info */
2373
    avctx->sample_rate = s->sample_rate;
2374
    avctx->channels = s->nb_channels;
2375
    avctx->bit_rate = s->bit_rate;
2376
    avctx->sub_id = s->layer;
2377

    
2378
    s->frame_size = len;
2379

    
2380
    if (avctx->parse_only) {
2381
        out_size = buf_size;
2382
    } else {
2383
        out_size = mp_decode_frame(s, out_samples, buf, buf_size);
2384
    }
2385

    
2386
    *data_size = out_size;
2387
    return buf_size;
2388
}
2389
#endif /* CONFIG_MP3ADU_DECODER */
2390

    
2391
#if CONFIG_MP3ON4_DECODER
2392

    
2393
/**
2394
 * Context for MP3On4 decoder
2395
 */
2396
typedef struct MP3On4DecodeContext {
2397
    int frames;   ///< number of mp3 frames per block (number of mp3 decoder instances)
2398
    int syncword; ///< syncword patch
2399
    const uint8_t *coff; ///< channels offsets in output buffer
2400
    MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance
2401
} MP3On4DecodeContext;
2402

    
2403
#include "mpeg4audio.h"
2404

    
2405
/* Next 3 arrays are indexed by channel config number (passed via codecdata) */
2406
static const uint8_t mp3Frames[8] = {0,1,1,2,3,3,4,5};   /* number of mp3 decoder instances */
2407
/* offsets into output buffer, assume output order is FL FR BL BR C LFE */
2408
static const uint8_t chan_offset[8][5] = {
2409
    {0},
2410
    {0},            // C
2411
    {0},            // FLR
2412
    {2,0},          // C FLR
2413
    {2,0,3},        // C FLR BS
2414
    {4,0,2},        // C FLR BLRS
2415
    {4,0,2,5},      // C FLR BLRS LFE
2416
    {4,0,2,6,5},    // C FLR BLRS BLR LFE
2417
};
2418

    
2419

    
2420
static int decode_init_mp3on4(AVCodecContext * avctx)
2421
{
2422
    MP3On4DecodeContext *s = avctx->priv_data;
2423
    MPEG4AudioConfig cfg;
2424
    int i;
2425

    
2426
    if ((avctx->extradata_size < 2) || (avctx->extradata == NULL)) {
2427
        av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
2428
        return -1;
2429
    }
2430

    
2431
    ff_mpeg4audio_get_config(&cfg, avctx->extradata, avctx->extradata_size);
2432
    if (!cfg.chan_config || cfg.chan_config > 7) {
2433
        av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
2434
        return -1;
2435
    }
2436
    s->frames = mp3Frames[cfg.chan_config];
2437
    s->coff = chan_offset[cfg.chan_config];
2438
    avctx->channels = ff_mpeg4audio_channels[cfg.chan_config];
2439

    
2440
    if (cfg.sample_rate < 16000)
2441
        s->syncword = 0xffe00000;
2442
    else
2443
        s->syncword = 0xfff00000;
2444

    
2445
    /* Init the first mp3 decoder in standard way, so that all tables get builded
2446
     * We replace avctx->priv_data with the context of the first decoder so that
2447
     * decode_init() does not have to be changed.
2448
     * Other decoders will be initialized here copying data from the first context
2449
     */
2450
    // Allocate zeroed memory for the first decoder context
2451
    s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
2452
    // Put decoder context in place to make init_decode() happy
2453
    avctx->priv_data = s->mp3decctx[0];
2454
    decode_init(avctx);
2455
    // Restore mp3on4 context pointer
2456
    avctx->priv_data = s;
2457
    s->mp3decctx[0]->adu_mode = 1; // Set adu mode
2458

    
2459
    /* Create a separate codec/context for each frame (first is already ok).
2460
     * Each frame is 1 or 2 channels - up to 5 frames allowed
2461
     */
2462
    for (i = 1; i < s->frames; i++) {
2463
        s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
2464
        s->mp3decctx[i]->adu_mode = 1;
2465
        s->mp3decctx[i]->avctx = avctx;
2466
    }
2467

    
2468
    return 0;
2469
}
2470

    
2471

    
2472
static av_cold int decode_close_mp3on4(AVCodecContext * avctx)
2473
{
2474
    MP3On4DecodeContext *s = avctx->priv_data;
2475
    int i;
2476

    
2477
    for (i = 0; i < s->frames; i++)
2478
        if (s->mp3decctx[i])
2479
            av_free(s->mp3decctx[i]);
2480

    
2481
    return 0;
2482
}
2483

    
2484

    
2485
static int decode_frame_mp3on4(AVCodecContext * avctx,
2486
                        void *data, int *data_size,
2487
                        AVPacket *avpkt)
2488
{
2489
    const uint8_t *buf = avpkt->data;
2490
    int buf_size = avpkt->size;
2491
    MP3On4DecodeContext *s = avctx->priv_data;
2492
    MPADecodeContext *m;
2493
    int fsize, len = buf_size, out_size = 0;
2494
    uint32_t header;
2495
    OUT_INT *out_samples = data;
2496
    OUT_INT decoded_buf[MPA_FRAME_SIZE * MPA_MAX_CHANNELS];
2497
    OUT_INT *outptr, *bp;
2498
    int fr, j, n;
2499

    
2500
    if(*data_size < MPA_FRAME_SIZE * MPA_MAX_CHANNELS * s->frames * sizeof(OUT_INT))
2501
        return -1;
2502

    
2503
    *data_size = 0;
2504
    // Discard too short frames
2505
    if (buf_size < HEADER_SIZE)
2506
        return -1;
2507

    
2508
    // If only one decoder interleave is not needed
2509
    outptr = s->frames == 1 ? out_samples : decoded_buf;
2510

    
2511
    avctx->bit_rate = 0;
2512

    
2513
    for (fr = 0; fr < s->frames; fr++) {
2514
        fsize = AV_RB16(buf) >> 4;
2515
        fsize = FFMIN3(fsize, len, MPA_MAX_CODED_FRAME_SIZE);
2516
        m = s->mp3decctx[fr];
2517
        assert (m != NULL);
2518

    
2519
        header = (AV_RB32(buf) & 0x000fffff) | s->syncword; // patch header
2520

    
2521
        if (ff_mpa_check_header(header) < 0) // Bad header, discard block
2522
            break;
2523

    
2524
        ff_mpegaudio_decode_header((MPADecodeHeader *)m, header);
2525
        out_size += mp_decode_frame(m, outptr, buf, fsize);
2526
        buf += fsize;
2527
        len -= fsize;
2528

    
2529
        if(s->frames > 1) {
2530
            n = m->avctx->frame_size*m->nb_channels;
2531
            /* interleave output data */
2532
            bp = out_samples + s->coff[fr];
2533
            if(m->nb_channels == 1) {
2534
                for(j = 0; j < n; j++) {
2535
                    *bp = decoded_buf[j];
2536
                    bp += avctx->channels;
2537
                }
2538
            } else {
2539
                for(j = 0; j < n; j++) {
2540
                    bp[0] = decoded_buf[j++];
2541
                    bp[1] = decoded_buf[j];
2542
                    bp += avctx->channels;
2543
                }
2544
            }
2545
        }
2546
        avctx->bit_rate += m->bit_rate;
2547
    }
2548

    
2549
    /* update codec info */
2550
    avctx->sample_rate = s->mp3decctx[0]->sample_rate;
2551

    
2552
    *data_size = out_size;
2553
    return buf_size;
2554
}
2555
#endif /* CONFIG_MP3ON4_DECODER */
2556

    
2557
#if !CONFIG_FLOAT
2558
#if CONFIG_MP1_DECODER
2559
AVCodec mp1_decoder =
2560
{
2561
    "mp1",
2562
    AVMEDIA_TYPE_AUDIO,
2563
    CODEC_ID_MP1,
2564
    sizeof(MPADecodeContext),
2565
    decode_init,
2566
    NULL,
2567
    NULL,
2568
    decode_frame,
2569
    CODEC_CAP_PARSE_ONLY,
2570
    .flush= flush,
2571
    .long_name= NULL_IF_CONFIG_SMALL("MP1 (MPEG audio layer 1)"),
2572
};
2573
#endif
2574
#if CONFIG_MP2_DECODER
2575
AVCodec mp2_decoder =
2576
{
2577
    "mp2",
2578
    AVMEDIA_TYPE_AUDIO,
2579
    CODEC_ID_MP2,
2580
    sizeof(MPADecodeContext),
2581
    decode_init,
2582
    NULL,
2583
    NULL,
2584
    decode_frame,
2585
    CODEC_CAP_PARSE_ONLY,
2586
    .flush= flush,
2587
    .long_name= NULL_IF_CONFIG_SMALL("MP2 (MPEG audio layer 2)"),
2588
};
2589
#endif
2590
#if CONFIG_MP3_DECODER
2591
AVCodec mp3_decoder =
2592
{
2593
    "mp3",
2594
    AVMEDIA_TYPE_AUDIO,
2595
    CODEC_ID_MP3,
2596
    sizeof(MPADecodeContext),
2597
    decode_init,
2598
    NULL,
2599
    NULL,
2600
    decode_frame,
2601
    CODEC_CAP_PARSE_ONLY,
2602
    .flush= flush,
2603
    .long_name= NULL_IF_CONFIG_SMALL("MP3 (MPEG audio layer 3)"),
2604
};
2605
#endif
2606
#if CONFIG_MP3ADU_DECODER
2607
AVCodec mp3adu_decoder =
2608
{
2609
    "mp3adu",
2610
    AVMEDIA_TYPE_AUDIO,
2611
    CODEC_ID_MP3ADU,
2612
    sizeof(MPADecodeContext),
2613
    decode_init,
2614
    NULL,
2615
    NULL,
2616
    decode_frame_adu,
2617
    CODEC_CAP_PARSE_ONLY,
2618
    .flush= flush,
2619
    .long_name= NULL_IF_CONFIG_SMALL("ADU (Application Data Unit) MP3 (MPEG audio layer 3)"),
2620
};
2621
#endif
2622
#if CONFIG_MP3ON4_DECODER
2623
AVCodec mp3on4_decoder =
2624
{
2625
    "mp3on4",
2626
    AVMEDIA_TYPE_AUDIO,
2627
    CODEC_ID_MP3ON4,
2628
    sizeof(MP3On4DecodeContext),
2629
    decode_init_mp3on4,
2630
    NULL,
2631
    decode_close_mp3on4,
2632
    decode_frame_mp3on4,
2633
    .flush= flush,
2634
    .long_name= NULL_IF_CONFIG_SMALL("MP3onMP4"),
2635
};
2636
#endif
2637
#endif