Statistics
| Branch: | Revision:

ffmpeg / libavcodec / ac3enc.c @ 8dbcc9f2

History | View | Annotate | Download (42.9 KB)

1
/*
2
 * The simplest AC3 encoder
3
 * Copyright (c) 2000 Fabrice Bellard.
4
 *
5
 * This library is free software; you can redistribute it and/or
6
 * modify it under the terms of the GNU Lesser General Public
7
 * License as published by the Free Software Foundation; either
8
 * version 2 of the License, or (at your option) any later version.
9
 *
10
 * This library is distributed in the hope that it will be useful,
11
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13
 * Lesser General Public License for more details.
14
 *
15
 * You should have received a copy of the GNU Lesser General Public
16
 * License along with this library; if not, write to the Free Software
17
 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
18
 */
19

    
20
/**
21
 * @file ac3enc.c
22
 * The simplest AC3 encoder.
23
 */
24
//#define DEBUG
25
//#define DEBUG_BITALLOC
26
#include "avcodec.h"
27

    
28
#include "ac3.h"
29

    
30
typedef struct AC3EncodeContext {
31
    PutBitContext pb;
32
    int nb_channels;
33
    int nb_all_channels;
34
    int lfe_channel;
35
    int bit_rate;
36
    unsigned int sample_rate;
37
    unsigned int bsid;
38
    unsigned int frame_size_min; /* minimum frame size in case rounding is necessary */
39
    unsigned int frame_size; /* current frame size in words */
40
    int halfratecod;
41
    unsigned int frmsizecod;
42
    unsigned int fscod; /* frequency */
43
    unsigned int acmod;
44
    int lfe;
45
    unsigned int bsmod;
46
    short last_samples[AC3_MAX_CHANNELS][256];
47
    unsigned int chbwcod[AC3_MAX_CHANNELS];
48
    int nb_coefs[AC3_MAX_CHANNELS];
49
    
50
    /* bitrate allocation control */
51
    int sgaincod, sdecaycod, fdecaycod, dbkneecod, floorcod; 
52
    AC3BitAllocParameters bit_alloc;
53
    int csnroffst;
54
    int fgaincod[AC3_MAX_CHANNELS];
55
    int fsnroffst[AC3_MAX_CHANNELS];
56
    /* mantissa encoding */
57
    int mant1_cnt, mant2_cnt, mant4_cnt;
58
} AC3EncodeContext;
59

    
60
#include "ac3tab.h"
61

    
62
#define MDCT_NBITS 9
63
#define N         (1 << MDCT_NBITS)
64

    
65
/* new exponents are sent if their Norm 1 exceed this number */
66
#define EXP_DIFF_THRESHOLD 1000
67

    
68
static void fft_init(int ln);
69
static void ac3_crc_init(void);
70

    
71
static inline int16_t fix15(float a)
72
{
73
    int v;
74
    v = (int)(a * (float)(1 << 15));
75
    if (v < -32767)
76
        v = -32767;
77
    else if (v > 32767) 
78
        v = 32767;
79
    return v;
80
}
81

    
82
static inline int calc_lowcomp1(int a, int b0, int b1)
83
{
84
    if ((b0 + 256) == b1) {
85
        a = 384 ;
86
    } else if (b0 > b1) { 
87
        a = a - 64;
88
        if (a < 0) a=0;
89
    }
90
    return a;
91
}
92

    
93
static inline int calc_lowcomp(int a, int b0, int b1, int bin)
94
{
95
    if (bin < 7) {
96
        if ((b0 + 256) == b1) {
97
            a = 384 ;
98
        } else if (b0 > b1) { 
99
            a = a - 64;
100
            if (a < 0) a=0;
101
        }
102
    } else if (bin < 20) {
103
        if ((b0 + 256) == b1) {
104
            a = 320 ;
105
        } else if (b0 > b1) {
106
            a= a - 64;
107
            if (a < 0) a=0;
108
        }
109
    } else {
110
        a = a - 128;
111
        if (a < 0) a=0;
112
    }
113
    return a;
114
}
115

    
116
/* AC3 bit allocation. The algorithm is the one described in the AC3
117
   spec. */
118
void ac3_parametric_bit_allocation(AC3BitAllocParameters *s, uint8_t *bap,
119
                                   int8_t *exp, int start, int end,
120
                                   int snroffset, int fgain, int is_lfe,
121
                                   int deltbae,int deltnseg, 
122
                                   uint8_t *deltoffst, uint8_t *deltlen, uint8_t *deltba)
123
{
124
    int bin,i,j,k,end1,v,v1,bndstrt,bndend,lowcomp,begin;
125
    int fastleak,slowleak,address,tmp;
126
    int16_t psd[256]; /* scaled exponents */
127
    int16_t bndpsd[50]; /* interpolated exponents */
128
    int16_t excite[50]; /* excitation */
129
    int16_t mask[50];   /* masking value */
130

    
131
    /* exponent mapping to PSD */
132
    for(bin=start;bin<end;bin++) {
133
        psd[bin]=(3072 - (exp[bin] << 7));
134
    }
135

    
136
    /* PSD integration */
137
    j=start;
138
    k=masktab[start];
139
    do {
140
        v=psd[j];
141
        j++;
142
        end1=bndtab[k+1];
143
        if (end1 > end) end1=end;
144
        for(i=j;i<end1;i++) {
145
            int c,adr;
146
            /* logadd */
147
            v1=psd[j];
148
            c=v-v1;
149
            if (c >= 0) {
150
                adr=c >> 1;
151
                if (adr > 255) adr=255;
152
                v=v + latab[adr];
153
            } else {
154
                adr=(-c) >> 1;
155
                if (adr > 255) adr=255;
156
                v=v1 + latab[adr];
157
            }
158
            j++;
159
        }
160
        bndpsd[k]=v;
161
        k++;
162
    } while (end > bndtab[k]);
163

    
164
    /* excitation function */
165
    bndstrt = masktab[start];
166
    bndend = masktab[end-1] + 1;
167
    
168
    if (bndstrt == 0) {
169
        lowcomp = 0;
170
        lowcomp = calc_lowcomp1(lowcomp, bndpsd[0], bndpsd[1]) ;
171
        excite[0] = bndpsd[0] - fgain - lowcomp ;
172
        lowcomp = calc_lowcomp1(lowcomp, bndpsd[1], bndpsd[2]) ;
173
        excite[1] = bndpsd[1] - fgain - lowcomp ;
174
        begin = 7 ;
175
        for (bin = 2; bin < 7; bin++) {
176
            if (!(is_lfe && bin == 6))
177
                lowcomp = calc_lowcomp1(lowcomp, bndpsd[bin], bndpsd[bin+1]) ;
178
            fastleak = bndpsd[bin] - fgain ;
179
            slowleak = bndpsd[bin] - s->sgain ;
180
            excite[bin] = fastleak - lowcomp ;
181
            if (!(is_lfe && bin == 6)) {
182
                if (bndpsd[bin] <= bndpsd[bin+1]) {
183
                    begin = bin + 1 ;
184
                    break ;
185
                }
186
            }
187
        }
188
    
189
        end1=bndend;
190
        if (end1 > 22) end1=22;
191
    
192
        for (bin = begin; bin < end1; bin++) {
193
            if (!(is_lfe && bin == 6))
194
                lowcomp = calc_lowcomp(lowcomp, bndpsd[bin], bndpsd[bin+1], bin) ;
195
        
196
            fastleak -= s->fdecay ;
197
            v = bndpsd[bin] - fgain;
198
            if (fastleak < v) fastleak = v;
199
        
200
            slowleak -= s->sdecay ;
201
            v = bndpsd[bin] - s->sgain;
202
            if (slowleak < v) slowleak = v;
203
        
204
            v=fastleak - lowcomp;
205
            if (slowleak > v) v=slowleak;
206
        
207
            excite[bin] = v;
208
        }
209
        begin = 22;
210
    } else {
211
        /* coupling channel */
212
        begin = bndstrt;
213
        
214
        fastleak = (s->cplfleak << 8) + 768;
215
        slowleak = (s->cplsleak << 8) + 768;
216
    }
217

    
218
    for (bin = begin; bin < bndend; bin++) {
219
        fastleak -= s->fdecay ;
220
        v = bndpsd[bin] - fgain;
221
        if (fastleak < v) fastleak = v;
222
        slowleak -= s->sdecay ;
223
        v = bndpsd[bin] - s->sgain;
224
        if (slowleak < v) slowleak = v;
225

    
226
        v=fastleak;
227
        if (slowleak > v) v = slowleak;
228
        excite[bin] = v;
229
    }
230

    
231
    /* compute masking curve */
232

    
233
    for (bin = bndstrt; bin < bndend; bin++) {
234
        v1 = excite[bin];
235
        tmp = s->dbknee - bndpsd[bin];
236
        if (tmp > 0) {
237
            v1 += tmp >> 2;
238
        }
239
        v=hth[bin >> s->halfratecod][s->fscod];
240
        if (v1 > v) v=v1;
241
        mask[bin] = v;
242
    }
243

    
244
    /* delta bit allocation */
245

    
246
    if (deltbae == 0 || deltbae == 1) {
247
        int band, seg, delta;
248
        band = 0 ;
249
        for (seg = 0; seg < deltnseg; seg++) {
250
            band += deltoffst[seg] ;
251
            if (deltba[seg] >= 4) {
252
                delta = (deltba[seg] - 3) << 7;
253
            } else {
254
                delta = (deltba[seg] - 4) << 7;
255
            }
256
            for (k = 0; k < deltlen[seg]; k++) {
257
                mask[band] += delta ;
258
                band++ ;
259
            }
260
        }
261
    }
262

    
263
    /* compute bit allocation */
264
    
265
    i = start ;
266
    j = masktab[start] ;
267
    do {
268
        v=mask[j];
269
        v -= snroffset ;
270
        v -= s->floor ;
271
        if (v < 0) v = 0;
272
        v &= 0x1fe0 ;
273
        v += s->floor ;
274

    
275
        end1=bndtab[j] + bndsz[j];
276
        if (end1 > end) end1=end;
277

    
278
        for (k = i; k < end1; k++) {
279
            address = (psd[i] - v) >> 5 ;
280
            if (address < 0) address=0;
281
            else if (address > 63) address=63;
282
            bap[i] = baptab[address];
283
            i++;
284
        }
285
    } while (end > bndtab[j++]) ;
286
}
287

    
288
typedef struct IComplex {
289
    short re,im;
290
} IComplex;
291

    
292
static void fft_init(int ln)
293
{
294
    int i, j, m, n;
295
    float alpha;
296

    
297
    n = 1 << ln;
298

    
299
    for(i=0;i<(n/2);i++) {
300
        alpha = 2 * M_PI * (float)i / (float)n;
301
        costab[i] = fix15(cos(alpha));
302
        sintab[i] = fix15(sin(alpha));
303
    }
304

    
305
    for(i=0;i<n;i++) {
306
        m=0;
307
        for(j=0;j<ln;j++) {
308
            m |= ((i >> j) & 1) << (ln-j-1);
309
        }
310
        fft_rev[i]=m;
311
    }
312
}
313

    
314
/* butter fly op */
315
#define BF(pre, pim, qre, qim, pre1, pim1, qre1, qim1) \
316
{\
317
  int ax, ay, bx, by;\
318
  bx=pre1;\
319
  by=pim1;\
320
  ax=qre1;\
321
  ay=qim1;\
322
  pre = (bx + ax) >> 1;\
323
  pim = (by + ay) >> 1;\
324
  qre = (bx - ax) >> 1;\
325
  qim = (by - ay) >> 1;\
326
}
327

    
328
#define MUL16(a,b) ((a) * (b))
329

    
330
#define CMUL(pre, pim, are, aim, bre, bim) \
331
{\
332
   pre = (MUL16(are, bre) - MUL16(aim, bim)) >> 15;\
333
   pim = (MUL16(are, bim) + MUL16(bre, aim)) >> 15;\
334
}
335

    
336

    
337
/* do a 2^n point complex fft on 2^ln points. */
338
static void fft(IComplex *z, int ln)
339
{
340
    int        j, l, np, np2;
341
    int        nblocks, nloops;
342
    register IComplex *p,*q;
343
    int tmp_re, tmp_im;
344

    
345
    np = 1 << ln;
346

    
347
    /* reverse */
348
    for(j=0;j<np;j++) {
349
        int k;
350
        IComplex tmp;
351
        k = fft_rev[j];
352
        if (k < j) {
353
            tmp = z[k];
354
            z[k] = z[j];
355
            z[j] = tmp;
356
        }
357
    }
358

    
359
    /* pass 0 */
360

    
361
    p=&z[0];
362
    j=(np >> 1);
363
    do {
364
        BF(p[0].re, p[0].im, p[1].re, p[1].im, 
365
           p[0].re, p[0].im, p[1].re, p[1].im);
366
        p+=2;
367
    } while (--j != 0);
368

    
369
    /* pass 1 */
370

    
371
    p=&z[0];
372
    j=np >> 2;
373
    do {
374
        BF(p[0].re, p[0].im, p[2].re, p[2].im, 
375
           p[0].re, p[0].im, p[2].re, p[2].im);
376
        BF(p[1].re, p[1].im, p[3].re, p[3].im, 
377
           p[1].re, p[1].im, p[3].im, -p[3].re);
378
        p+=4;
379
    } while (--j != 0);
380

    
381
    /* pass 2 .. ln-1 */
382

    
383
    nblocks = np >> 3;
384
    nloops = 1 << 2;
385
    np2 = np >> 1;
386
    do {
387
        p = z;
388
        q = z + nloops;
389
        for (j = 0; j < nblocks; ++j) {
390

    
391
            BF(p->re, p->im, q->re, q->im,
392
               p->re, p->im, q->re, q->im);
393
            
394
            p++;
395
            q++;
396
            for(l = nblocks; l < np2; l += nblocks) {
397
                CMUL(tmp_re, tmp_im, costab[l], -sintab[l], q->re, q->im);
398
                BF(p->re, p->im, q->re, q->im,
399
                   p->re, p->im, tmp_re, tmp_im);
400
                p++;
401
                q++;
402
            }
403
            p += nloops;
404
            q += nloops;
405
        }
406
        nblocks = nblocks >> 1;
407
        nloops = nloops << 1;
408
    } while (nblocks != 0);
409
}
410

    
411
/* do a 512 point mdct */
412
static void mdct512(int32_t *out, int16_t *in)
413
{
414
    int i, re, im, re1, im1;
415
    int16_t rot[N]; 
416
    IComplex x[N/4];
417

    
418
    /* shift to simplify computations */
419
    for(i=0;i<N/4;i++)
420
        rot[i] = -in[i + 3*N/4];
421
    for(i=N/4;i<N;i++)
422
        rot[i] = in[i - N/4];
423
        
424
    /* pre rotation */
425
    for(i=0;i<N/4;i++) {
426
        re = ((int)rot[2*i] - (int)rot[N-1-2*i]) >> 1;
427
        im = -((int)rot[N/2+2*i] - (int)rot[N/2-1-2*i]) >> 1;
428
        CMUL(x[i].re, x[i].im, re, im, -xcos1[i], xsin1[i]);
429
    }
430

    
431
    fft(x, MDCT_NBITS - 2);
432
  
433
    /* post rotation */
434
    for(i=0;i<N/4;i++) {
435
        re = x[i].re;
436
        im = x[i].im;
437
        CMUL(re1, im1, re, im, xsin1[i], xcos1[i]);
438
        out[2*i] = im1;
439
        out[N/2-1-2*i] = re1;
440
    }
441
}
442

    
443
/* XXX: use another norm ? */
444
static int calc_exp_diff(uint8_t *exp1, uint8_t *exp2, int n)
445
{
446
    int sum, i;
447
    sum = 0;
448
    for(i=0;i<n;i++) {
449
        sum += abs(exp1[i] - exp2[i]);
450
    }
451
    return sum;
452
}
453

    
454
static void compute_exp_strategy(uint8_t exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS],
455
                                 uint8_t exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
456
                                 int ch, int is_lfe)
457
{
458
    int i, j;
459
    int exp_diff;
460
    
461
    /* estimate if the exponent variation & decide if they should be
462
       reused in the next frame */
463
    exp_strategy[0][ch] = EXP_NEW;
464
    for(i=1;i<NB_BLOCKS;i++) {
465
        exp_diff = calc_exp_diff(exp[i][ch], exp[i-1][ch], N/2);
466
#ifdef DEBUG            
467
        av_log(NULL, AV_LOG_DEBUG, "exp_diff=%d\n", exp_diff);
468
#endif
469
        if (exp_diff > EXP_DIFF_THRESHOLD)
470
            exp_strategy[i][ch] = EXP_NEW;
471
        else
472
            exp_strategy[i][ch] = EXP_REUSE;
473
    }
474
    if (is_lfe)
475
        return;
476

    
477
    /* now select the encoding strategy type : if exponents are often
478
       recoded, we use a coarse encoding */
479
    i = 0;
480
    while (i < NB_BLOCKS) {
481
        j = i + 1;
482
        while (j < NB_BLOCKS && exp_strategy[j][ch] == EXP_REUSE)
483
            j++;
484
        switch(j - i) {
485
        case 1:
486
            exp_strategy[i][ch] = EXP_D45;
487
            break;
488
        case 2:
489
        case 3:
490
            exp_strategy[i][ch] = EXP_D25;
491
            break;
492
        default:
493
            exp_strategy[i][ch] = EXP_D15;
494
            break;
495
        }
496
        i = j;
497
    }
498
}
499

    
500
/* set exp[i] to min(exp[i], exp1[i]) */
501
static void exponent_min(uint8_t exp[N/2], uint8_t exp1[N/2], int n)
502
{
503
    int i;
504

    
505
    for(i=0;i<n;i++) {
506
        if (exp1[i] < exp[i])
507
            exp[i] = exp1[i];
508
    }
509
}
510
                                 
511
/* update the exponents so that they are the ones the decoder will
512
   decode. Return the number of bits used to code the exponents */
513
static int encode_exp(uint8_t encoded_exp[N/2], 
514
                      uint8_t exp[N/2], 
515
                      int nb_exps,
516
                      int exp_strategy)
517
{
518
    int group_size, nb_groups, i, j, k, recurse, exp_min, delta;
519
    uint8_t exp1[N/2];
520

    
521
    switch(exp_strategy) {
522
    case EXP_D15:
523
        group_size = 1;
524
        break;
525
    case EXP_D25:
526
        group_size = 2;
527
        break;
528
    default:
529
    case EXP_D45:
530
        group_size = 4;
531
        break;
532
    }
533
    nb_groups = ((nb_exps + (group_size * 3) - 4) / (3 * group_size)) * 3;
534

    
535
    /* for each group, compute the minimum exponent */
536
    exp1[0] = exp[0]; /* DC exponent is handled separately */
537
    k = 1;
538
    for(i=1;i<=nb_groups;i++) {
539
        exp_min = exp[k];
540
        assert(exp_min >= 0 && exp_min <= 24);
541
        for(j=1;j<group_size;j++) {
542
            if (exp[k+j] < exp_min)
543
                exp_min = exp[k+j];
544
        }
545
        exp1[i] = exp_min;
546
        k += group_size;
547
    }
548

    
549
    /* constraint for DC exponent */
550
    if (exp1[0] > 15)
551
        exp1[0] = 15;
552

    
553
    /* Iterate until the delta constraints between each groups are
554
       satisfyed. I'm sure it is possible to find a better algorithm,
555
       but I am lazy */
556
    do {
557
        recurse = 0;
558
        for(i=1;i<=nb_groups;i++) {
559
            delta = exp1[i] - exp1[i-1];
560
            if (delta > 2) {
561
                /* if delta too big, we encode a smaller exponent */
562
                exp1[i] = exp1[i-1] + 2;
563
            } else if (delta < -2) {
564
                /* if delta is too small, we must decrease the previous
565
               exponent, which means we must recurse */
566
                recurse = 1;
567
                exp1[i-1] = exp1[i] + 2;
568
            }
569
        }
570
    } while (recurse);
571
    
572
    /* now we have the exponent values the decoder will see */
573
    encoded_exp[0] = exp1[0];
574
    k = 1;
575
    for(i=1;i<=nb_groups;i++) {
576
        for(j=0;j<group_size;j++) {
577
            encoded_exp[k+j] = exp1[i];
578
        }
579
        k += group_size;
580
    }
581
    
582
#if defined(DEBUG)
583
    av_log(NULL, AV_LOG_DEBUG, "exponents: strategy=%d\n", exp_strategy);
584
    for(i=0;i<=nb_groups * group_size;i++) {
585
        av_log(NULL, AV_LOG_DEBUG, "%d ", encoded_exp[i]);
586
    }
587
    av_log(NULL, AV_LOG_DEBUG, "\n");
588
#endif
589

    
590
    return 4 + (nb_groups / 3) * 7;
591
}
592

    
593
/* return the size in bits taken by the mantissa */
594
static int compute_mantissa_size(AC3EncodeContext *s, uint8_t *m, int nb_coefs)
595
{
596
    int bits, mant, i;
597

    
598
    bits = 0;
599
    for(i=0;i<nb_coefs;i++) {
600
        mant = m[i];
601
        switch(mant) {
602
        case 0:
603
            /* nothing */
604
            break;
605
        case 1:
606
            /* 3 mantissa in 5 bits */
607
            if (s->mant1_cnt == 0) 
608
                bits += 5;
609
            if (++s->mant1_cnt == 3)
610
                s->mant1_cnt = 0;
611
            break;
612
        case 2:
613
            /* 3 mantissa in 7 bits */
614
            if (s->mant2_cnt == 0) 
615
                bits += 7;
616
            if (++s->mant2_cnt == 3)
617
                s->mant2_cnt = 0;
618
            break;
619
        case 3:
620
            bits += 3;
621
            break;
622
        case 4:
623
            /* 2 mantissa in 7 bits */
624
            if (s->mant4_cnt == 0)
625
                bits += 7;
626
            if (++s->mant4_cnt == 2) 
627
                s->mant4_cnt = 0;
628
            break;
629
        case 14:
630
            bits += 14;
631
            break;
632
        case 15:
633
            bits += 16;
634
            break;
635
        default:
636
            bits += mant - 1;
637
            break;
638
        }
639
    }
640
    return bits;
641
}
642

    
643

    
644
static int bit_alloc(AC3EncodeContext *s,
645
                     uint8_t bap[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
646
                     uint8_t encoded_exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
647
                     uint8_t exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS],
648
                     int frame_bits, int csnroffst, int fsnroffst)
649
{
650
    int i, ch;
651

    
652
    /* compute size */
653
    for(i=0;i<NB_BLOCKS;i++) {
654
        s->mant1_cnt = 0;
655
        s->mant2_cnt = 0;
656
        s->mant4_cnt = 0;
657
        for(ch=0;ch<s->nb_all_channels;ch++) {
658
            ac3_parametric_bit_allocation(&s->bit_alloc, 
659
                                          bap[i][ch], (int8_t *)encoded_exp[i][ch], 
660
                                          0, s->nb_coefs[ch], 
661
                                          (((csnroffst-15) << 4) + 
662
                                           fsnroffst) << 2, 
663
                                          fgaintab[s->fgaincod[ch]],
664
                                          ch == s->lfe_channel,
665
                                          2, 0, NULL, NULL, NULL);
666
            frame_bits += compute_mantissa_size(s, bap[i][ch], 
667
                                                 s->nb_coefs[ch]);
668
        }
669
    }
670
#if 0
671
    printf("csnr=%d fsnr=%d frame_bits=%d diff=%d\n", 
672
           csnroffst, fsnroffst, frame_bits, 
673
           16 * s->frame_size - ((frame_bits + 7) & ~7));
674
#endif
675
    return 16 * s->frame_size - frame_bits;
676
}
677

    
678
#define SNR_INC1 4
679

    
680
static int compute_bit_allocation(AC3EncodeContext *s,
681
                                  uint8_t bap[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
682
                                  uint8_t encoded_exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
683
                                  uint8_t exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS],
684
                                  int frame_bits)
685
{
686
    int i, ch;
687
    int csnroffst, fsnroffst;
688
    uint8_t bap1[NB_BLOCKS][AC3_MAX_CHANNELS][N/2];
689
    static int frame_bits_inc[8] = { 0, 0, 2, 2, 2, 4, 2, 4 };
690

    
691
    /* init default parameters */
692
    s->sdecaycod = 2;
693
    s->fdecaycod = 1;
694
    s->sgaincod = 1;
695
    s->dbkneecod = 2;
696
    s->floorcod = 4;
697
    for(ch=0;ch<s->nb_all_channels;ch++) 
698
        s->fgaincod[ch] = 4;
699
    
700
    /* compute real values */
701
    s->bit_alloc.fscod = s->fscod;
702
    s->bit_alloc.halfratecod = s->halfratecod;
703
    s->bit_alloc.sdecay = sdecaytab[s->sdecaycod] >> s->halfratecod;
704
    s->bit_alloc.fdecay = fdecaytab[s->fdecaycod] >> s->halfratecod;
705
    s->bit_alloc.sgain = sgaintab[s->sgaincod];
706
    s->bit_alloc.dbknee = dbkneetab[s->dbkneecod];
707
    s->bit_alloc.floor = floortab[s->floorcod];
708
    
709
    /* header size */
710
    frame_bits += 65;
711
    // if (s->acmod == 2)
712
    //    frame_bits += 2;
713
    frame_bits += frame_bits_inc[s->acmod];
714

    
715
    /* audio blocks */
716
    for(i=0;i<NB_BLOCKS;i++) {
717
        frame_bits += s->nb_channels * 2 + 2; /* blksw * c, dithflag * c, dynrnge, cplstre */
718
        if (s->acmod == 2)
719
            frame_bits++; /* rematstr */
720
        frame_bits += 2 * s->nb_channels; /* chexpstr[2] * c */
721
        if (s->lfe)
722
            frame_bits++; /* lfeexpstr */
723
        for(ch=0;ch<s->nb_channels;ch++) {
724
            if (exp_strategy[i][ch] != EXP_REUSE)
725
                frame_bits += 6 + 2; /* chbwcod[6], gainrng[2] */
726
        }
727
        frame_bits++; /* baie */
728
        frame_bits++; /* snr */
729
        frame_bits += 2; /* delta / skip */
730
    }
731
    frame_bits++; /* cplinu for block 0 */
732
    /* bit alloc info */
733
    /* sdcycod[2], fdcycod[2], sgaincod[2], dbpbcod[2], floorcod[3] */
734
    /* csnroffset[6] */
735
    /* (fsnoffset[4] + fgaincod[4]) * c */
736
    frame_bits += 2*4 + 3 + 6 + s->nb_all_channels * (4 + 3);
737

    
738
    /* CRC */
739
    frame_bits += 16;
740

    
741
    /* now the big work begins : do the bit allocation. Modify the snr
742
       offset until we can pack everything in the requested frame size */
743

    
744
    csnroffst = s->csnroffst;
745
    while (csnroffst >= 0 && 
746
           bit_alloc(s, bap, encoded_exp, exp_strategy, frame_bits, csnroffst, 0) < 0)
747
        csnroffst -= SNR_INC1;
748
    if (csnroffst < 0) {
749
        av_log(NULL, AV_LOG_ERROR, "Yack, Error !!!\n");
750
        return -1;
751
    }
752
    while ((csnroffst + SNR_INC1) <= 63 && 
753
           bit_alloc(s, bap1, encoded_exp, exp_strategy, frame_bits, 
754
                     csnroffst + SNR_INC1, 0) >= 0) {
755
        csnroffst += SNR_INC1;
756
        memcpy(bap, bap1, sizeof(bap1));
757
    }
758
    while ((csnroffst + 1) <= 63 && 
759
           bit_alloc(s, bap1, encoded_exp, exp_strategy, frame_bits, csnroffst + 1, 0) >= 0) {
760
        csnroffst++;
761
        memcpy(bap, bap1, sizeof(bap1));
762
    }
763

    
764
    fsnroffst = 0;
765
    while ((fsnroffst + SNR_INC1) <= 15 && 
766
           bit_alloc(s, bap1, encoded_exp, exp_strategy, frame_bits, 
767
                     csnroffst, fsnroffst + SNR_INC1) >= 0) {
768
        fsnroffst += SNR_INC1;
769
        memcpy(bap, bap1, sizeof(bap1));
770
    }
771
    while ((fsnroffst + 1) <= 15 && 
772
           bit_alloc(s, bap1, encoded_exp, exp_strategy, frame_bits, 
773
                     csnroffst, fsnroffst + 1) >= 0) {
774
        fsnroffst++;
775
        memcpy(bap, bap1, sizeof(bap1));
776
    }
777
    
778
    s->csnroffst = csnroffst;
779
    for(ch=0;ch<s->nb_all_channels;ch++)
780
        s->fsnroffst[ch] = fsnroffst;
781
#if defined(DEBUG_BITALLOC)
782
    {
783
        int j;
784

    
785
        for(i=0;i<6;i++) {
786
            for(ch=0;ch<s->nb_all_channels;ch++) {
787
                printf("Block #%d Ch%d:\n", i, ch);
788
                printf("bap=");
789
                for(j=0;j<s->nb_coefs[ch];j++) {
790
                    printf("%d ",bap[i][ch][j]);
791
                }
792
                printf("\n");
793
            }
794
        }
795
    }
796
#endif
797
    return 0;
798
}
799

    
800
void ac3_common_init(void)
801
{
802
    int i, j, k, l, v;
803
    /* compute bndtab and masktab from bandsz */
804
    k = 0;
805
    l = 0;
806
    for(i=0;i<50;i++) {
807
        bndtab[i] = l;
808
        v = bndsz[i];
809
        for(j=0;j<v;j++) masktab[k++]=i;
810
        l += v;
811
    }
812
    bndtab[50] = 0;
813
}
814

    
815

    
816
static int AC3_encode_init(AVCodecContext *avctx)
817
{
818
    int freq = avctx->sample_rate;
819
    int bitrate = avctx->bit_rate;
820
    int channels = avctx->channels;
821
    AC3EncodeContext *s = avctx->priv_data;
822
    int i, j, ch;
823
    float alpha;
824
    static const uint8_t acmod_defs[6] = {
825
        0x01, /* C */
826
        0x02, /* L R */
827
        0x03, /* L C R */
828
        0x06, /* L R SL SR */
829
        0x07, /* L C R SL SR */
830
        0x07, /* L C R SL SR (+LFE) */
831
    };
832

    
833
    avctx->frame_size = AC3_FRAME_SIZE;
834
    
835
    /* number of channels */
836
    if (channels < 1 || channels > 6)
837
        return -1;
838
    s->acmod = acmod_defs[channels - 1];
839
    s->lfe = (channels == 6) ? 1 : 0;
840
    s->nb_all_channels = channels;
841
    s->nb_channels = channels > 5 ? 5 : channels;
842
    s->lfe_channel = s->lfe ? 5 : -1;
843

    
844
    /* frequency */
845
    for(i=0;i<3;i++) {
846
        for(j=0;j<3;j++) 
847
            if ((ac3_freqs[j] >> i) == freq)
848
                goto found;
849
    }
850
    return -1;
851
 found:    
852
    s->sample_rate = freq;
853
    s->halfratecod = i;
854
    s->fscod = j;
855
    s->bsid = 8 + s->halfratecod;
856
    s->bsmod = 0; /* complete main audio service */
857

    
858
    /* bitrate & frame size */
859
    bitrate /= 1000;
860
    for(i=0;i<19;i++) {
861
        if ((ac3_bitratetab[i] >> s->halfratecod) == bitrate)
862
            break;
863
    }
864
    if (i == 19)
865
        return -1;
866
    s->bit_rate = bitrate;
867
    s->frmsizecod = i << 1;
868
    s->frame_size_min = (bitrate * 1000 * AC3_FRAME_SIZE) / (freq * 16);
869
    /* for now we do not handle fractional sizes */
870
    s->frame_size = s->frame_size_min;
871
    
872
    /* bit allocation init */
873
    for(ch=0;ch<s->nb_channels;ch++) {
874
        /* bandwidth for each channel */
875
        /* XXX: should compute the bandwidth according to the frame
876
           size, so that we avoid anoying high freq artefacts */
877
        s->chbwcod[ch] = 50; /* sample bandwidth as mpeg audio layer 2 table 0 */
878
        s->nb_coefs[ch] = ((s->chbwcod[ch] + 12) * 3) + 37;
879
    }
880
    if (s->lfe) {
881
        s->nb_coefs[s->lfe_channel] = 7; /* fixed */
882
    }
883
    /* initial snr offset */
884
    s->csnroffst = 40;
885

    
886
    ac3_common_init();
887

    
888
    /* mdct init */
889
    fft_init(MDCT_NBITS - 2);
890
    for(i=0;i<N/4;i++) {
891
        alpha = 2 * M_PI * (i + 1.0 / 8.0) / (float)N;
892
        xcos1[i] = fix15(-cos(alpha));
893
        xsin1[i] = fix15(-sin(alpha));
894
    }
895

    
896
    ac3_crc_init();
897
    
898
    avctx->coded_frame= avcodec_alloc_frame();
899
    avctx->coded_frame->key_frame= 1;
900

    
901
    return 0;
902
}
903

    
904
/* output the AC3 frame header */
905
static void output_frame_header(AC3EncodeContext *s, unsigned char *frame)
906
{
907
    init_put_bits(&s->pb, frame, AC3_MAX_CODED_FRAME_SIZE);
908

    
909
    put_bits(&s->pb, 16, 0x0b77); /* frame header */
910
    put_bits(&s->pb, 16, 0); /* crc1: will be filled later */
911
    put_bits(&s->pb, 2, s->fscod);
912
    put_bits(&s->pb, 6, s->frmsizecod + (s->frame_size - s->frame_size_min));
913
    put_bits(&s->pb, 5, s->bsid);
914
    put_bits(&s->pb, 3, s->bsmod);
915
    put_bits(&s->pb, 3, s->acmod);
916
    if ((s->acmod & 0x01) && s->acmod != 0x01)
917
        put_bits(&s->pb, 2, 1); /* XXX -4.5 dB */
918
    if (s->acmod & 0x04)
919
        put_bits(&s->pb, 2, 1); /* XXX -6 dB */
920
    if (s->acmod == 0x02)
921
        put_bits(&s->pb, 2, 0); /* surround not indicated */
922
    put_bits(&s->pb, 1, s->lfe); /* LFE */
923
    put_bits(&s->pb, 5, 31); /* dialog norm: -31 db */
924
    put_bits(&s->pb, 1, 0); /* no compression control word */
925
    put_bits(&s->pb, 1, 0); /* no lang code */
926
    put_bits(&s->pb, 1, 0); /* no audio production info */
927
    put_bits(&s->pb, 1, 0); /* no copyright */
928
    put_bits(&s->pb, 1, 1); /* original bitstream */
929
    put_bits(&s->pb, 1, 0); /* no time code 1 */
930
    put_bits(&s->pb, 1, 0); /* no time code 2 */
931
    put_bits(&s->pb, 1, 0); /* no addtional bit stream info */
932
}
933

    
934
/* symetric quantization on 'levels' levels */
935
static inline int sym_quant(int c, int e, int levels)
936
{
937
    int v;
938

    
939
    if (c >= 0) {
940
        v = (levels * (c << e)) >> 24;
941
        v = (v + 1) >> 1;
942
        v = (levels >> 1) + v;
943
    } else {
944
        v = (levels * ((-c) << e)) >> 24;
945
        v = (v + 1) >> 1;
946
        v = (levels >> 1) - v;
947
    }
948
    assert (v >= 0 && v < levels);
949
    return v;
950
}
951

    
952
/* asymetric quantization on 2^qbits levels */
953
static inline int asym_quant(int c, int e, int qbits)
954
{
955
    int lshift, m, v;
956

    
957
    lshift = e + qbits - 24;
958
    if (lshift >= 0)
959
        v = c << lshift;
960
    else
961
        v = c >> (-lshift);
962
    /* rounding */
963
    v = (v + 1) >> 1;
964
    m = (1 << (qbits-1));
965
    if (v >= m)
966
        v = m - 1;
967
    assert(v >= -m);
968
    return v & ((1 << qbits)-1);
969
}
970

    
971
/* Output one audio block. There are NB_BLOCKS audio blocks in one AC3
972
   frame */
973
static void output_audio_block(AC3EncodeContext *s,
974
                               uint8_t exp_strategy[AC3_MAX_CHANNELS],
975
                               uint8_t encoded_exp[AC3_MAX_CHANNELS][N/2],
976
                               uint8_t bap[AC3_MAX_CHANNELS][N/2],
977
                               int32_t mdct_coefs[AC3_MAX_CHANNELS][N/2],
978
                               int8_t global_exp[AC3_MAX_CHANNELS],
979
                               int block_num)
980
{
981
    int ch, nb_groups, group_size, i, baie, rbnd;
982
    uint8_t *p;
983
    uint16_t qmant[AC3_MAX_CHANNELS][N/2];
984
    int exp0, exp1;
985
    int mant1_cnt, mant2_cnt, mant4_cnt;
986
    uint16_t *qmant1_ptr, *qmant2_ptr, *qmant4_ptr;
987
    int delta0, delta1, delta2;
988

    
989
    for(ch=0;ch<s->nb_channels;ch++) 
990
        put_bits(&s->pb, 1, 0); /* 512 point MDCT */
991
    for(ch=0;ch<s->nb_channels;ch++) 
992
        put_bits(&s->pb, 1, 1); /* no dither */
993
    put_bits(&s->pb, 1, 0); /* no dynamic range */
994
    if (block_num == 0) {
995
        /* for block 0, even if no coupling, we must say it. This is a
996
           waste of bit :-) */
997
        put_bits(&s->pb, 1, 1); /* coupling strategy present */
998
        put_bits(&s->pb, 1, 0); /* no coupling strategy */
999
    } else {
1000
        put_bits(&s->pb, 1, 0); /* no new coupling strategy */
1001
    }
1002

    
1003
    if (s->acmod == 2)
1004
      {
1005
        if(block_num==0)
1006
          {
1007
            /* first block must define rematrixing (rematstr)  */
1008
            put_bits(&s->pb, 1, 1); 
1009
            
1010
            /* dummy rematrixing rematflg(1:4)=0 */
1011
            for (rbnd=0;rbnd<4;rbnd++)
1012
              put_bits(&s->pb, 1, 0); 
1013
          }
1014
        else 
1015
          {
1016
            /* no matrixing (but should be used in the future) */
1017
            put_bits(&s->pb, 1, 0);
1018
          } 
1019
      }
1020

    
1021
#if defined(DEBUG) 
1022
    {
1023
      static int count = 0;
1024
      av_log(NULL, AV_LOG_DEBUG, "Block #%d (%d)\n", block_num, count++);
1025
    }
1026
#endif
1027
    /* exponent strategy */
1028
    for(ch=0;ch<s->nb_channels;ch++) {
1029
        put_bits(&s->pb, 2, exp_strategy[ch]);
1030
    }
1031
    
1032
    if (s->lfe) {
1033
        put_bits(&s->pb, 1, exp_strategy[s->lfe_channel]);
1034
    }
1035

    
1036
    for(ch=0;ch<s->nb_channels;ch++) {
1037
        if (exp_strategy[ch] != EXP_REUSE)
1038
            put_bits(&s->pb, 6, s->chbwcod[ch]);
1039
    }
1040
    
1041
    /* exponents */
1042
    for (ch = 0; ch < s->nb_all_channels; ch++) {
1043
        switch(exp_strategy[ch]) {
1044
        case EXP_REUSE:
1045
            continue;
1046
        case EXP_D15:
1047
            group_size = 1;
1048
            break;
1049
        case EXP_D25:
1050
            group_size = 2;
1051
            break;
1052
        default:
1053
        case EXP_D45:
1054
            group_size = 4;
1055
            break;
1056
        }
1057
        nb_groups = (s->nb_coefs[ch] + (group_size * 3) - 4) / (3 * group_size);
1058
        p = encoded_exp[ch];
1059

    
1060
        /* first exponent */
1061
        exp1 = *p++;
1062
        put_bits(&s->pb, 4, exp1);
1063

    
1064
        /* next ones are delta encoded */
1065
        for(i=0;i<nb_groups;i++) {
1066
            /* merge three delta in one code */
1067
            exp0 = exp1;
1068
            exp1 = p[0];
1069
            p += group_size;
1070
            delta0 = exp1 - exp0 + 2;
1071

    
1072
            exp0 = exp1;
1073
            exp1 = p[0];
1074
            p += group_size;
1075
            delta1 = exp1 - exp0 + 2;
1076

    
1077
            exp0 = exp1;
1078
            exp1 = p[0];
1079
            p += group_size;
1080
            delta2 = exp1 - exp0 + 2;
1081

    
1082
            put_bits(&s->pb, 7, ((delta0 * 5 + delta1) * 5) + delta2);
1083
        }
1084

    
1085
        if (ch != s->lfe_channel)
1086
            put_bits(&s->pb, 2, 0); /* no gain range info */
1087
    }
1088

    
1089
    /* bit allocation info */
1090
    baie = (block_num == 0);
1091
    put_bits(&s->pb, 1, baie);
1092
    if (baie) {
1093
        put_bits(&s->pb, 2, s->sdecaycod);
1094
        put_bits(&s->pb, 2, s->fdecaycod);
1095
        put_bits(&s->pb, 2, s->sgaincod);
1096
        put_bits(&s->pb, 2, s->dbkneecod);
1097
        put_bits(&s->pb, 3, s->floorcod);
1098
    }
1099

    
1100
    /* snr offset */
1101
    put_bits(&s->pb, 1, baie); /* always present with bai */
1102
    if (baie) {
1103
        put_bits(&s->pb, 6, s->csnroffst);
1104
        for(ch=0;ch<s->nb_all_channels;ch++) {
1105
            put_bits(&s->pb, 4, s->fsnroffst[ch]);
1106
            put_bits(&s->pb, 3, s->fgaincod[ch]);
1107
        }
1108
    }
1109
    
1110
    put_bits(&s->pb, 1, 0); /* no delta bit allocation */
1111
    put_bits(&s->pb, 1, 0); /* no data to skip */
1112

    
1113
    /* mantissa encoding : we use two passes to handle the grouping. A
1114
       one pass method may be faster, but it would necessitate to
1115
       modify the output stream. */
1116

    
1117
    /* first pass: quantize */
1118
    mant1_cnt = mant2_cnt = mant4_cnt = 0;
1119
    qmant1_ptr = qmant2_ptr = qmant4_ptr = NULL;
1120

    
1121
    for (ch = 0; ch < s->nb_all_channels; ch++) {
1122
        int b, c, e, v;
1123

    
1124
        for(i=0;i<s->nb_coefs[ch];i++) {
1125
            c = mdct_coefs[ch][i];
1126
            e = encoded_exp[ch][i] - global_exp[ch];
1127
            b = bap[ch][i];
1128
            switch(b) {
1129
            case 0:
1130
                v = 0;
1131
                break;
1132
            case 1:
1133
                v = sym_quant(c, e, 3);
1134
                switch(mant1_cnt) {
1135
                case 0:
1136
                    qmant1_ptr = &qmant[ch][i];
1137
                    v = 9 * v;
1138
                    mant1_cnt = 1;
1139
                    break;
1140
                case 1:
1141
                    *qmant1_ptr += 3 * v;
1142
                    mant1_cnt = 2;
1143
                    v = 128;
1144
                    break;
1145
                default:
1146
                    *qmant1_ptr += v;
1147
                    mant1_cnt = 0;
1148
                    v = 128;
1149
                    break;
1150
                }
1151
                break;
1152
            case 2:
1153
                v = sym_quant(c, e, 5);
1154
                switch(mant2_cnt) {
1155
                case 0:
1156
                    qmant2_ptr = &qmant[ch][i];
1157
                    v = 25 * v;
1158
                    mant2_cnt = 1;
1159
                    break;
1160
                case 1:
1161
                    *qmant2_ptr += 5 * v;
1162
                    mant2_cnt = 2;
1163
                    v = 128;
1164
                    break;
1165
                default:
1166
                    *qmant2_ptr += v;
1167
                    mant2_cnt = 0;
1168
                    v = 128;
1169
                    break;
1170
                }
1171
                break;
1172
            case 3:
1173
                v = sym_quant(c, e, 7);
1174
                break;
1175
            case 4:
1176
                v = sym_quant(c, e, 11);
1177
                switch(mant4_cnt) {
1178
                case 0:
1179
                    qmant4_ptr = &qmant[ch][i];
1180
                    v = 11 * v;
1181
                    mant4_cnt = 1;
1182
                    break;
1183
                default:
1184
                    *qmant4_ptr += v;
1185
                    mant4_cnt = 0;
1186
                    v = 128;
1187
                    break;
1188
                }
1189
                break;
1190
            case 5:
1191
                v = sym_quant(c, e, 15);
1192
                break;
1193
            case 14:
1194
                v = asym_quant(c, e, 14);
1195
                break;
1196
            case 15:
1197
                v = asym_quant(c, e, 16);
1198
                break;
1199
            default:
1200
                v = asym_quant(c, e, b - 1);
1201
                break;
1202
            }
1203
            qmant[ch][i] = v;
1204
        }
1205
    }
1206

    
1207
    /* second pass : output the values */
1208
    for (ch = 0; ch < s->nb_all_channels; ch++) {
1209
        int b, q;
1210
        
1211
        for(i=0;i<s->nb_coefs[ch];i++) {
1212
            q = qmant[ch][i];
1213
            b = bap[ch][i];
1214
            switch(b) {
1215
            case 0:
1216
                break;
1217
            case 1:
1218
                if (q != 128) 
1219
                    put_bits(&s->pb, 5, q);
1220
                break;
1221
            case 2:
1222
                if (q != 128) 
1223
                    put_bits(&s->pb, 7, q);
1224
                break;
1225
            case 3:
1226
                put_bits(&s->pb, 3, q);
1227
                break;
1228
            case 4:
1229
                if (q != 128)
1230
                    put_bits(&s->pb, 7, q);
1231
                break;
1232
            case 14:
1233
                put_bits(&s->pb, 14, q);
1234
                break;
1235
            case 15:
1236
                put_bits(&s->pb, 16, q);
1237
                break;
1238
            default:
1239
                put_bits(&s->pb, b - 1, q);
1240
                break;
1241
            }
1242
        }
1243
    }
1244
}
1245

    
1246
/* compute the ac3 crc */
1247

    
1248
#define CRC16_POLY ((1 << 0) | (1 << 2) | (1 << 15) | (1 << 16))
1249

    
1250
static void ac3_crc_init(void)
1251
{
1252
    unsigned int c, n, k;
1253

    
1254
    for(n=0;n<256;n++) {
1255
        c = n << 8;
1256
        for (k = 0; k < 8; k++) {
1257
            if (c & (1 << 15)) 
1258
                c = ((c << 1) & 0xffff) ^ (CRC16_POLY & 0xffff);
1259
            else
1260
                c = c << 1;
1261
        }
1262
        crc_table[n] = c;
1263
    }
1264
}
1265

    
1266
static unsigned int ac3_crc(uint8_t *data, int n, unsigned int crc)
1267
{
1268
    int i;
1269
    for(i=0;i<n;i++) {
1270
        crc = (crc_table[data[i] ^ (crc >> 8)] ^ (crc << 8)) & 0xffff;
1271
    }
1272
    return crc;
1273
}
1274

    
1275
static unsigned int mul_poly(unsigned int a, unsigned int b, unsigned int poly)
1276
{
1277
    unsigned int c;
1278

    
1279
    c = 0;
1280
    while (a) {
1281
        if (a & 1)
1282
            c ^= b;
1283
        a = a >> 1;
1284
        b = b << 1;
1285
        if (b & (1 << 16))
1286
            b ^= poly;
1287
    }
1288
    return c;
1289
}
1290

    
1291
static unsigned int pow_poly(unsigned int a, unsigned int n, unsigned int poly)
1292
{
1293
    unsigned int r;
1294
    r = 1;
1295
    while (n) {
1296
        if (n & 1)
1297
            r = mul_poly(r, a, poly);
1298
        a = mul_poly(a, a, poly);
1299
        n >>= 1;
1300
    }
1301
    return r;
1302
}
1303

    
1304

    
1305
/* compute log2(max(abs(tab[]))) */
1306
static int log2_tab(int16_t *tab, int n)
1307
{
1308
    int i, v;
1309

    
1310
    v = 0;
1311
    for(i=0;i<n;i++) {
1312
        v |= abs(tab[i]);
1313
    }
1314
    return av_log2(v);
1315
}
1316

    
1317
static void lshift_tab(int16_t *tab, int n, int lshift)
1318
{
1319
    int i;
1320

    
1321
    if (lshift > 0) {
1322
        for(i=0;i<n;i++) {
1323
            tab[i] <<= lshift;
1324
        }
1325
    } else if (lshift < 0) {
1326
        lshift = -lshift;
1327
        for(i=0;i<n;i++) {
1328
            tab[i] >>= lshift;
1329
        }
1330
    }
1331
}
1332

    
1333
/* fill the end of the frame and compute the two crcs */
1334
static int output_frame_end(AC3EncodeContext *s)
1335
{
1336
    int frame_size, frame_size_58, n, crc1, crc2, crc_inv;
1337
    uint8_t *frame;
1338

    
1339
    frame_size = s->frame_size; /* frame size in words */
1340
    /* align to 8 bits */
1341
    flush_put_bits(&s->pb);
1342
    /* add zero bytes to reach the frame size */
1343
    frame = s->pb.buf;
1344
    n = 2 * s->frame_size - (pbBufPtr(&s->pb) - frame) - 2;
1345
    assert(n >= 0);
1346
    if(n>0)
1347
      memset(pbBufPtr(&s->pb), 0, n);
1348
    
1349
    /* Now we must compute both crcs : this is not so easy for crc1
1350
       because it is at the beginning of the data... */
1351
    frame_size_58 = (frame_size >> 1) + (frame_size >> 3);
1352
    crc1 = ac3_crc(frame + 4, (2 * frame_size_58) - 4, 0);
1353
    /* XXX: could precompute crc_inv */
1354
    crc_inv = pow_poly((CRC16_POLY >> 1), (16 * frame_size_58) - 16, CRC16_POLY);
1355
    crc1 = mul_poly(crc_inv, crc1, CRC16_POLY);
1356
    frame[2] = crc1 >> 8;
1357
    frame[3] = crc1;
1358
    
1359
    crc2 = ac3_crc(frame + 2 * frame_size_58, (frame_size - frame_size_58) * 2 - 2, 0);
1360
    frame[2*frame_size - 2] = crc2 >> 8;
1361
    frame[2*frame_size - 1] = crc2;
1362

    
1363
    //    printf("n=%d frame_size=%d\n", n, frame_size);
1364
    return frame_size * 2;
1365
}
1366

    
1367
static int AC3_encode_frame(AVCodecContext *avctx,
1368
                            unsigned char *frame, int buf_size, void *data)
1369
{
1370
    AC3EncodeContext *s = avctx->priv_data;
1371
    short *samples = data;
1372
    int i, j, k, v, ch;
1373
    int16_t input_samples[N];
1374
    int32_t mdct_coef[NB_BLOCKS][AC3_MAX_CHANNELS][N/2];
1375
    uint8_t exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2];
1376
    uint8_t exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS];
1377
    uint8_t encoded_exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2];
1378
    uint8_t bap[NB_BLOCKS][AC3_MAX_CHANNELS][N/2];
1379
    int8_t exp_samples[NB_BLOCKS][AC3_MAX_CHANNELS];
1380
    int frame_bits;
1381

    
1382
    frame_bits = 0;
1383
    for(ch=0;ch<s->nb_all_channels;ch++) {
1384
        /* fixed mdct to the six sub blocks & exponent computation */
1385
        for(i=0;i<NB_BLOCKS;i++) {
1386
            int16_t *sptr;
1387
            int sinc;
1388

    
1389
            /* compute input samples */
1390
            memcpy(input_samples, s->last_samples[ch], N/2 * sizeof(int16_t));
1391
            sinc = s->nb_all_channels;
1392
            sptr = samples + (sinc * (N/2) * i) + ch;
1393
            for(j=0;j<N/2;j++) {
1394
                v = *sptr;
1395
                input_samples[j + N/2] = v;
1396
                s->last_samples[ch][j] = v; 
1397
                sptr += sinc;
1398
            }
1399

    
1400
            /* apply the MDCT window */
1401
            for(j=0;j<N/2;j++) {
1402
                input_samples[j] = MUL16(input_samples[j], 
1403
                                         ac3_window[j]) >> 15;
1404
                input_samples[N-j-1] = MUL16(input_samples[N-j-1], 
1405
                                             ac3_window[j]) >> 15;
1406
            }
1407
        
1408
            /* Normalize the samples to use the maximum available
1409
               precision */
1410
            v = 14 - log2_tab(input_samples, N);
1411
            if (v < 0)
1412
                v = 0;
1413
            exp_samples[i][ch] = v - 8;
1414
            lshift_tab(input_samples, N, v);
1415

    
1416
            /* do the MDCT */
1417
            mdct512(mdct_coef[i][ch], input_samples);
1418
            
1419
            /* compute "exponents". We take into account the
1420
               normalization there */
1421
            for(j=0;j<N/2;j++) {
1422
                int e;
1423
                v = abs(mdct_coef[i][ch][j]);
1424
                if (v == 0)
1425
                    e = 24;
1426
                else {
1427
                    e = 23 - av_log2(v) + exp_samples[i][ch];
1428
                    if (e >= 24) {
1429
                        e = 24;
1430
                        mdct_coef[i][ch][j] = 0;
1431
                    }
1432
                }
1433
                exp[i][ch][j] = e;
1434
            }
1435
        }
1436
        
1437
        compute_exp_strategy(exp_strategy, exp, ch, ch == s->lfe_channel);
1438

    
1439
        /* compute the exponents as the decoder will see them. The
1440
           EXP_REUSE case must be handled carefully : we select the
1441
           min of the exponents */
1442
        i = 0;
1443
        while (i < NB_BLOCKS) {
1444
            j = i + 1;
1445
            while (j < NB_BLOCKS && exp_strategy[j][ch] == EXP_REUSE) {
1446
                exponent_min(exp[i][ch], exp[j][ch], s->nb_coefs[ch]);
1447
                j++;
1448
            }
1449
            frame_bits += encode_exp(encoded_exp[i][ch],
1450
                                     exp[i][ch], s->nb_coefs[ch], 
1451
                                     exp_strategy[i][ch]);
1452
            /* copy encoded exponents for reuse case */
1453
            for(k=i+1;k<j;k++) {
1454
                memcpy(encoded_exp[k][ch], encoded_exp[i][ch], 
1455
                       s->nb_coefs[ch] * sizeof(uint8_t));
1456
            }
1457
            i = j;
1458
        }
1459
    }
1460

    
1461
    compute_bit_allocation(s, bap, encoded_exp, exp_strategy, frame_bits);
1462
    /* everything is known... let's output the frame */
1463
    output_frame_header(s, frame);
1464
        
1465
    for(i=0;i<NB_BLOCKS;i++) {
1466
        output_audio_block(s, exp_strategy[i], encoded_exp[i], 
1467
                           bap[i], mdct_coef[i], exp_samples[i], i);
1468
    }
1469
    return output_frame_end(s);
1470
}
1471

    
1472
static int AC3_encode_close(AVCodecContext *avctx)
1473
{
1474
    av_freep(&avctx->coded_frame);
1475
    return 0;
1476
}
1477

    
1478
#if 0
1479
/*************************************************************************/
1480
/* TEST */
1481

1482
#define FN (N/4)
1483

1484
void fft_test(void)
1485
{
1486
    IComplex in[FN], in1[FN];
1487
    int k, n, i;
1488
    float sum_re, sum_im, a;
1489

1490
    /* FFT test */
1491

1492
    for(i=0;i<FN;i++) {
1493
        in[i].re = random() % 65535 - 32767;
1494
        in[i].im = random() % 65535 - 32767;
1495
        in1[i] = in[i];
1496
    }
1497
    fft(in, 7);
1498

1499
    /* do it by hand */
1500
    for(k=0;k<FN;k++) {
1501
        sum_re = 0;
1502
        sum_im = 0;
1503
        for(n=0;n<FN;n++) {
1504
            a = -2 * M_PI * (n * k) / FN;
1505
            sum_re += in1[n].re * cos(a) - in1[n].im * sin(a);
1506
            sum_im += in1[n].re * sin(a) + in1[n].im * cos(a);
1507
        }
1508
        printf("%3d: %6d,%6d %6.0f,%6.0f\n", 
1509
               k, in[k].re, in[k].im, sum_re / FN, sum_im / FN); 
1510
    }
1511
}
1512

1513
void mdct_test(void)
1514
{
1515
    int16_t input[N];
1516
    int32_t output[N/2];
1517
    float input1[N];
1518
    float output1[N/2];
1519
    float s, a, err, e, emax;
1520
    int i, k, n;
1521

1522
    for(i=0;i<N;i++) {
1523
        input[i] = (random() % 65535 - 32767) * 9 / 10;
1524
        input1[i] = input[i];
1525
    }
1526

1527
    mdct512(output, input);
1528
    
1529
    /* do it by hand */
1530
    for(k=0;k<N/2;k++) {
1531
        s = 0;
1532
        for(n=0;n<N;n++) {
1533
            a = (2*M_PI*(2*n+1+N/2)*(2*k+1) / (4 * N));
1534
            s += input1[n] * cos(a);
1535
        }
1536
        output1[k] = -2 * s / N;
1537
    }
1538
    
1539
    err = 0;
1540
    emax = 0;
1541
    for(i=0;i<N/2;i++) {
1542
        printf("%3d: %7d %7.0f\n", i, output[i], output1[i]);
1543
        e = output[i] - output1[i];
1544
        if (e > emax)
1545
            emax = e;
1546
        err += e * e;
1547
    }
1548
    printf("err2=%f emax=%f\n", err / (N/2), emax);
1549
}
1550

1551
void test_ac3(void)
1552
{
1553
    AC3EncodeContext ctx;
1554
    unsigned char frame[AC3_MAX_CODED_FRAME_SIZE];
1555
    short samples[AC3_FRAME_SIZE];
1556
    int ret, i;
1557
    
1558
    AC3_encode_init(&ctx, 44100, 64000, 1);
1559

1560
    fft_test();
1561
    mdct_test();
1562

1563
    for(i=0;i<AC3_FRAME_SIZE;i++)
1564
        samples[i] = (int)(sin(2*M_PI*i*1000.0/44100) * 10000);
1565
    ret = AC3_encode_frame(&ctx, frame, samples);
1566
    printf("ret=%d\n", ret);
1567
}
1568
#endif
1569

    
1570
AVCodec ac3_encoder = {
1571
    "ac3",
1572
    CODEC_TYPE_AUDIO,
1573
    CODEC_ID_AC3,
1574
    sizeof(AC3EncodeContext),
1575
    AC3_encode_init,
1576
    AC3_encode_frame,
1577
    AC3_encode_close,
1578
    NULL,
1579
};