quicktime/ffmpeg/libavcodec/ac3enc.c

   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 #include "bitstream.h"
  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, exp_min;
 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     /* Decrease the delta between each groups to within 2
 554      * so that they can be differentially encoded */
 555     for (i=1;i<=nb_groups;i++)
 556         exp1[i] = FFMIN(exp1[i], exp1[i-1] + 2);
 557     for (i=nb_groups-1;i>=0;i--)
 558         exp1[i] = FFMIN(exp1[i], exp1[i+1] + 2);
 559
 560     /* now we have the exponent values the decoder will see */
 561     encoded_exp[0] = exp1[0];
 562     k = 1;
 563     for(i=1;i<=nb_groups;i++) {
 564         for(j=0;j<group_size;j++) {
 565             encoded_exp[k+j] = exp1[i];
 566         }
 567         k += group_size;
 568     }
 569
 570 #if defined(DEBUG)
 571     av_log(NULL, AV_LOG_DEBUG, "exponents: strategy=%d\n", exp_strategy);
 572     for(i=0;i<=nb_groups * group_size;i++) {
 573         av_log(NULL, AV_LOG_DEBUG, "%d ", encoded_exp[i]);
 574     }
 575     av_log(NULL, AV_LOG_DEBUG, "\n");
 576 #endif
 577
 578     return 4 + (nb_groups / 3) * 7;
 579 }
 580
 581 /* return the size in bits taken by the mantissa */
 582 static int compute_mantissa_size(AC3EncodeContext *s, uint8_t *m, int nb_coefs)
 583 {
 584     int bits, mant, i;
 585
 586     bits = 0;
 587     for(i=0;i<nb_coefs;i++) {
 588         mant = m[i];
 589         switch(mant) {
 590         case 0:
 591             /* nothing */
 592             break;
 593         case 1:
 594             /* 3 mantissa in 5 bits */
 595             if (s->mant1_cnt == 0)
 596                 bits += 5;
 597             if (++s->mant1_cnt == 3)
 598                 s->mant1_cnt = 0;
 599             break;
 600         case 2:
 601             /* 3 mantissa in 7 bits */
 602             if (s->mant2_cnt == 0)
 603                 bits += 7;
 604             if (++s->mant2_cnt == 3)
 605                 s->mant2_cnt = 0;
 606             break;
 607         case 3:
 608             bits += 3;
 609             break;
 610         case 4:
 611             /* 2 mantissa in 7 bits */
 612             if (s->mant4_cnt == 0)
 613                 bits += 7;
 614             if (++s->mant4_cnt == 2)
 615                 s->mant4_cnt = 0;
 616             break;
 617         case 14:
 618             bits += 14;
 619             break;
 620         case 15:
 621             bits += 16;
 622             break;
 623         default:
 624             bits += mant - 1;
 625             break;
 626         }
 627     }
 628     return bits;
 629 }
 630
 631
 632 static int bit_alloc(AC3EncodeContext *s,
 633                      uint8_t bap[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
 634                      uint8_t encoded_exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
 635                      uint8_t exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS],
 636                      int frame_bits, int csnroffst, int fsnroffst)
 637 {
 638     int i, ch;
 639
 640     /* compute size */
 641     for(i=0;i<NB_BLOCKS;i++) {
 642         s->mant1_cnt = 0;
 643         s->mant2_cnt = 0;
 644         s->mant4_cnt = 0;
 645         for(ch=0;ch<s->nb_all_channels;ch++) {
 646             ac3_parametric_bit_allocation(&s->bit_alloc,
 647                                           bap[i][ch], (int8_t *)encoded_exp[i][ch],
 648                                           0, s->nb_coefs[ch],
 649                                           (((csnroffst-15) << 4) +
 650                                            fsnroffst) << 2,
 651                                           fgaintab[s->fgaincod[ch]],
 652                                           ch == s->lfe_channel,
 653                                           2, 0, NULL, NULL, NULL);
 654             frame_bits += compute_mantissa_size(s, bap[i][ch],
 655                                                  s->nb_coefs[ch]);
 656         }
 657     }
 658 #if 0
 659     printf("csnr=%d fsnr=%d frame_bits=%d diff=%d\n",
 660            csnroffst, fsnroffst, frame_bits,
 661            16 * s->frame_size - ((frame_bits + 7) & ~7));
 662 #endif
 663     return 16 * s->frame_size - frame_bits;
 664 }
 665
 666 #define SNR_INC1 4
 667
 668 static int compute_bit_allocation(AC3EncodeContext *s,
 669                                   uint8_t bap[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
 670                                   uint8_t encoded_exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
 671                                   uint8_t exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS],
 672                                   int frame_bits)
 673 {
 674     int i, ch;
 675     int csnroffst, fsnroffst;
 676     uint8_t bap1[NB_BLOCKS][AC3_MAX_CHANNELS][N/2];
 677     static int frame_bits_inc[8] = { 0, 0, 2, 2, 2, 4, 2, 4 };
 678
 679     /* init default parameters */
 680     s->sdecaycod = 2;
 681     s->fdecaycod = 1;
 682     s->sgaincod = 1;
 683     s->dbkneecod = 2;
 684     s->floorcod = 4;
 685     for(ch=0;ch<s->nb_all_channels;ch++)
 686         s->fgaincod[ch] = 4;
 687
 688     /* compute real values */
 689     s->bit_alloc.fscod = s->fscod;
 690     s->bit_alloc.halfratecod = s->halfratecod;
 691     s->bit_alloc.sdecay = sdecaytab[s->sdecaycod] >> s->halfratecod;
 692     s->bit_alloc.fdecay = fdecaytab[s->fdecaycod] >> s->halfratecod;
 693     s->bit_alloc.sgain = sgaintab[s->sgaincod];
 694     s->bit_alloc.dbknee = dbkneetab[s->dbkneecod];
 695     s->bit_alloc.floor = floortab[s->floorcod];
 696
 697     /* header size */
 698     frame_bits += 65;
 699     // if (s->acmod == 2)
 700     //    frame_bits += 2;
 701     frame_bits += frame_bits_inc[s->acmod];
 702
 703     /* audio blocks */
 704     for(i=0;i<NB_BLOCKS;i++) {
 705         frame_bits += s->nb_channels * 2 + 2; /* blksw * c, dithflag * c, dynrnge, cplstre */
 706         if (s->acmod == 2) {
 707             frame_bits++; /* rematstr */
 708             if(i==0) frame_bits += 4;
 709         }
 710         frame_bits += 2 * s->nb_channels; /* chexpstr[2] * c */
 711         if (s->lfe)
 712             frame_bits++; /* lfeexpstr */
 713         for(ch=0;ch<s->nb_channels;ch++) {
 714             if (exp_strategy[i][ch] != EXP_REUSE)
 715                 frame_bits += 6 + 2; /* chbwcod[6], gainrng[2] */
 716         }
 717         frame_bits++; /* baie */
 718         frame_bits++; /* snr */
 719         frame_bits += 2; /* delta / skip */
 720     }
 721     frame_bits++; /* cplinu for block 0 */
 722     /* bit alloc info */
 723     /* sdcycod[2], fdcycod[2], sgaincod[2], dbpbcod[2], floorcod[3] */
 724     /* csnroffset[6] */
 725     /* (fsnoffset[4] + fgaincod[4]) * c */
 726     frame_bits += 2*4 + 3 + 6 + s->nb_all_channels * (4 + 3);
 727
 728     /* auxdatae, crcrsv */
 729     frame_bits += 2;
 730
 731     /* CRC */
 732     frame_bits += 16;
 733
 734     /* now the big work begins : do the bit allocation. Modify the snr
 735        offset until we can pack everything in the requested frame size */
 736
 737     csnroffst = s->csnroffst;
 738     while (csnroffst >= 0 &&
 739            bit_alloc(s, bap, encoded_exp, exp_strategy, frame_bits, csnroffst, 0) < 0)
 740         csnroffst -= SNR_INC1;
 741     if (csnroffst < 0) {
 742         av_log(NULL, AV_LOG_ERROR, "Yack, Error !!!\n");
 743         return -1;
 744     }
 745     while ((csnroffst + SNR_INC1) <= 63 &&
 746            bit_alloc(s, bap1, encoded_exp, exp_strategy, frame_bits,
 747                      csnroffst + SNR_INC1, 0) >= 0) {
 748         csnroffst += SNR_INC1;
 749         memcpy(bap, bap1, sizeof(bap1));
 750     }
 751     while ((csnroffst + 1) <= 63 &&
 752            bit_alloc(s, bap1, encoded_exp, exp_strategy, frame_bits, csnroffst + 1, 0) >= 0) {
 753         csnroffst++;
 754         memcpy(bap, bap1, sizeof(bap1));
 755     }
 756
 757     fsnroffst = 0;
 758     while ((fsnroffst + SNR_INC1) <= 15 &&
 759            bit_alloc(s, bap1, encoded_exp, exp_strategy, frame_bits,
 760                      csnroffst, fsnroffst + SNR_INC1) >= 0) {
 761         fsnroffst += SNR_INC1;
 762         memcpy(bap, bap1, sizeof(bap1));
 763     }
 764     while ((fsnroffst + 1) <= 15 &&
 765            bit_alloc(s, bap1, encoded_exp, exp_strategy, frame_bits,
 766                      csnroffst, fsnroffst + 1) >= 0) {
 767         fsnroffst++;
 768         memcpy(bap, bap1, sizeof(bap1));
 769     }
 770
 771     s->csnroffst = csnroffst;
 772     for(ch=0;ch<s->nb_all_channels;ch++)
 773         s->fsnroffst[ch] = fsnroffst;
 774 #if defined(DEBUG_BITALLOC)
 775     {
 776         int j;
 777
 778         for(i=0;i<6;i++) {
 779             for(ch=0;ch<s->nb_all_channels;ch++) {
 780                 printf("Block #%d Ch%d:\n", i, ch);
 781                 printf("bap=");
 782                 for(j=0;j<s->nb_coefs[ch];j++) {
 783                     printf("%d ",bap[i][ch][j]);
 784                 }
 785                 printf("\n");
 786             }
 787         }
 788     }
 789 #endif
 790     return 0;
 791 }
 792
 793 void ac3_common_init(void)
 794 {
 795     int i, j, k, l, v;
 796     /* compute bndtab and masktab from bandsz */
 797     k = 0;
 798     l = 0;
 799     for(i=0;i<50;i++) {
 800         bndtab[i] = l;
 801         v = bndsz[i];
 802         for(j=0;j<v;j++) masktab[k++]=i;
 803         l += v;
 804     }
 805     bndtab[50] = 0;
 806 }
 807
 808
 809 static int AC3_encode_init(AVCodecContext *avctx)
 810 {
 811     int freq = avctx->sample_rate;
 812     int bitrate = avctx->bit_rate;
 813     int channels = avctx->channels;
 814     AC3EncodeContext *s = avctx->priv_data;
 815     int i, j, ch;
 816     float alpha;
 817     static const uint8_t acmod_defs[6] = {
 818         0x01, /* C */
 819         0x02, /* L R */
 820         0x03, /* L C R */
 821         0x06, /* L R SL SR */
 822         0x07, /* L C R SL SR */
 823         0x07, /* L C R SL SR (+LFE) */
 824     };
 825
 826     avctx->frame_size = AC3_FRAME_SIZE;
 827
 828     /* number of channels */
 829     if (channels < 1 || channels > 6)
 830         return -1;
 831     s->acmod = acmod_defs[channels - 1];
 832     s->lfe = (channels == 6) ? 1 : 0;
 833     s->nb_all_channels = channels;
 834     s->nb_channels = channels > 5 ? 5 : channels;
 835     s->lfe_channel = s->lfe ? 5 : -1;
 836
 837     /* frequency */
 838     for(i=0;i<3;i++) {
 839         for(j=0;j<3;j++)
 840             if ((ac3_freqs[j] >> i) == freq)
 841                 goto found;
 842     }
 843     return -1;
 844  found:
 845     s->sample_rate = freq;
 846     s->halfratecod = i;
 847     s->fscod = j;
 848     s->bsid = 8 + s->halfratecod;
 849     s->bsmod = 0; /* complete main audio service */
 850
 851     /* bitrate & frame size */
 852     bitrate /= 1000;
 853     for(i=0;i<19;i++) {
 854         if ((ac3_bitratetab[i] >> s->halfratecod) == bitrate)
 855             break;
 856     }
 857     if (i == 19)
 858         return -1;
 859     s->bit_rate = bitrate;
 860     s->frmsizecod = i << 1;
 861     s->frame_size_min = (bitrate * 1000 * AC3_FRAME_SIZE) / (freq * 16);
 862     /* for now we do not handle fractional sizes */
 863     s->frame_size = s->frame_size_min;
 864
 865     /* bit allocation init */
 866     for(ch=0;ch<s->nb_channels;ch++) {
 867         /* bandwidth for each channel */
 868         /* XXX: should compute the bandwidth according to the frame
 869            size, so that we avoid anoying high freq artefacts */
 870         s->chbwcod[ch] = 50; /* sample bandwidth as mpeg audio layer 2 table 0 */
 871         s->nb_coefs[ch] = ((s->chbwcod[ch] + 12) * 3) + 37;
 872     }
 873     if (s->lfe) {
 874         s->nb_coefs[s->lfe_channel] = 7; /* fixed */
 875     }
 876     /* initial snr offset */
 877     s->csnroffst = 40;
 878
 879     ac3_common_init();
 880
 881     /* mdct init */
 882     fft_init(MDCT_NBITS - 2);
 883     for(i=0;i<N/4;i++) {
 884         alpha = 2 * M_PI * (i + 1.0 / 8.0) / (float)N;
 885         xcos1[i] = fix15(-cos(alpha));
 886         xsin1[i] = fix15(-sin(alpha));
 887     }
 888
 889     ac3_crc_init();
 890
 891     avctx->coded_frame= avcodec_alloc_frame();
 892     avctx->coded_frame->key_frame= 1;
 893
 894     return 0;
 895 }
 896
 897 /* output the AC3 frame header */
 898 static void output_frame_header(AC3EncodeContext *s, unsigned char *frame)
 899 {
 900     init_put_bits(&s->pb, frame, AC3_MAX_CODED_FRAME_SIZE);
 901
 902     put_bits(&s->pb, 16, 0x0b77); /* frame header */
 903     put_bits(&s->pb, 16, 0); /* crc1: will be filled later */
 904     put_bits(&s->pb, 2, s->fscod);
 905     put_bits(&s->pb, 6, s->frmsizecod + (s->frame_size - s->frame_size_min));
 906     put_bits(&s->pb, 5, s->bsid);
 907     put_bits(&s->pb, 3, s->bsmod);
 908     put_bits(&s->pb, 3, s->acmod);
 909     if ((s->acmod & 0x01) && s->acmod != 0x01)
 910         put_bits(&s->pb, 2, 1); /* XXX -4.5 dB */
 911     if (s->acmod & 0x04)
 912         put_bits(&s->pb, 2, 1); /* XXX -6 dB */
 913     if (s->acmod == 0x02)
 914         put_bits(&s->pb, 2, 0); /* surround not indicated */
 915     put_bits(&s->pb, 1, s->lfe); /* LFE */
 916     put_bits(&s->pb, 5, 31); /* dialog norm: -31 db */
 917     put_bits(&s->pb, 1, 0); /* no compression control word */
 918     put_bits(&s->pb, 1, 0); /* no lang code */
 919     put_bits(&s->pb, 1, 0); /* no audio production info */
 920     put_bits(&s->pb, 1, 0); /* no copyright */
 921     put_bits(&s->pb, 1, 1); /* original bitstream */
 922     put_bits(&s->pb, 1, 0); /* no time code 1 */
 923     put_bits(&s->pb, 1, 0); /* no time code 2 */
 924     put_bits(&s->pb, 1, 0); /* no addtional bit stream info */
 925 }
 926
 927 /* symetric quantization on 'levels' levels */
 928 static inline int sym_quant(int c, int e, int levels)
 929 {
 930     int v;
 931
 932     if (c >= 0) {
 933         v = (levels * (c << e)) >> 24;
 934         v = (v + 1) >> 1;
 935         v = (levels >> 1) + v;
 936     } else {
 937         v = (levels * ((-c) << e)) >> 24;
 938         v = (v + 1) >> 1;
 939         v = (levels >> 1) - v;
 940     }
 941     assert (v >= 0 && v < levels);
 942     return v;
 943 }
 944
 945 /* asymetric quantization on 2^qbits levels */
 946 static inline int asym_quant(int c, int e, int qbits)
 947 {
 948     int lshift, m, v;
 949
 950     lshift = e + qbits - 24;
 951     if (lshift >= 0)
 952         v = c << lshift;
 953     else
 954         v = c >> (-lshift);
 955     /* rounding */
 956     v = (v + 1) >> 1;
 957     m = (1 << (qbits-1));
 958     if (v >= m)
 959         v = m - 1;
 960     assert(v >= -m);
 961     return v & ((1 << qbits)-1);
 962 }
 963
 964 /* Output one audio block. There are NB_BLOCKS audio blocks in one AC3
 965    frame */
 966 static void output_audio_block(AC3EncodeContext *s,
 967                                uint8_t exp_strategy[AC3_MAX_CHANNELS],
 968                                uint8_t encoded_exp[AC3_MAX_CHANNELS][N/2],
 969                                uint8_t bap[AC3_MAX_CHANNELS][N/2],
 970                                int32_t mdct_coefs[AC3_MAX_CHANNELS][N/2],
 971                                int8_t global_exp[AC3_MAX_CHANNELS],
 972                                int block_num)
 973 {
 974     int ch, nb_groups, group_size, i, baie, rbnd;
 975     uint8_t *p;
 976     uint16_t qmant[AC3_MAX_CHANNELS][N/2];
 977     int exp0, exp1;
 978     int mant1_cnt, mant2_cnt, mant4_cnt;
 979     uint16_t *qmant1_ptr, *qmant2_ptr, *qmant4_ptr;
 980     int delta0, delta1, delta2;
 981
 982     for(ch=0;ch<s->nb_channels;ch++)
 983         put_bits(&s->pb, 1, 0); /* 512 point MDCT */
 984     for(ch=0;ch<s->nb_channels;ch++)
 985         put_bits(&s->pb, 1, 1); /* no dither */
 986     put_bits(&s->pb, 1, 0); /* no dynamic range */
 987     if (block_num == 0) {
 988         /* for block 0, even if no coupling, we must say it. This is a
 989            waste of bit :-) */
 990         put_bits(&s->pb, 1, 1); /* coupling strategy present */
 991         put_bits(&s->pb, 1, 0); /* no coupling strategy */
 992     } else {
 993         put_bits(&s->pb, 1, 0); /* no new coupling strategy */
 994     }
 995
 996     if (s->acmod == 2)
 997       {
 998         if(block_num==0)
 999           {
1000             /* first block must define rematrixing (rematstr)  */
1001             put_bits(&s->pb, 1, 1);
1002
1003             /* dummy rematrixing rematflg(1:4)=0 */
1004             for (rbnd=0;rbnd<4;rbnd++)
1005               put_bits(&s->pb, 1, 0);
1006           }
1007         else
1008           {
1009             /* no matrixing (but should be used in the future) */
1010             put_bits(&s->pb, 1, 0);
1011           }
1012       }
1013
1014 #if defined(DEBUG)
1015     {
1016       static int count = 0;
1017       av_log(NULL, AV_LOG_DEBUG, "Block #%d (%d)\n", block_num, count++);
1018     }
1019 #endif
1020     /* exponent strategy */
1021     for(ch=0;ch<s->nb_channels;ch++) {
1022         put_bits(&s->pb, 2, exp_strategy[ch]);
1023     }
1024
1025     if (s->lfe) {
1026         put_bits(&s->pb, 1, exp_strategy[s->lfe_channel]);
1027     }
1028
1029     for(ch=0;ch<s->nb_channels;ch++) {
1030         if (exp_strategy[ch] != EXP_REUSE)
1031             put_bits(&s->pb, 6, s->chbwcod[ch]);
1032     }
1033
1034     /* exponents */
1035     for (ch = 0; ch < s->nb_all_channels; ch++) {
1036         switch(exp_strategy[ch]) {
1037         case EXP_REUSE:
1038             continue;
1039         case EXP_D15:
1040             group_size = 1;
1041             break;
1042         case EXP_D25:
1043             group_size = 2;
1044             break;
1045         default:
1046         case EXP_D45:
1047             group_size = 4;
1048             break;
1049         }
1050         nb_groups = (s->nb_coefs[ch] + (group_size * 3) - 4) / (3 * group_size);
1051         p = encoded_exp[ch];
1052
1053         /* first exponent */
1054         exp1 = *p++;
1055         put_bits(&s->pb, 4, exp1);
1056
1057         /* next ones are delta encoded */
1058         for(i=0;i<nb_groups;i++) {
1059             /* merge three delta in one code */
1060             exp0 = exp1;
1061             exp1 = p[0];
1062             p += group_size;
1063             delta0 = exp1 - exp0 + 2;
1064
1065             exp0 = exp1;
1066             exp1 = p[0];
1067             p += group_size;
1068             delta1 = exp1 - exp0 + 2;
1069
1070             exp0 = exp1;
1071             exp1 = p[0];
1072             p += group_size;
1073             delta2 = exp1 - exp0 + 2;
1074
1075             put_bits(&s->pb, 7, ((delta0 * 5 + delta1) * 5) + delta2);
1076         }
1077
1078         if (ch != s->lfe_channel)
1079             put_bits(&s->pb, 2, 0); /* no gain range info */
1080     }
1081
1082     /* bit allocation info */
1083     baie = (block_num == 0);
1084     put_bits(&s->pb, 1, baie);
1085     if (baie) {
1086         put_bits(&s->pb, 2, s->sdecaycod);
1087         put_bits(&s->pb, 2, s->fdecaycod);
1088         put_bits(&s->pb, 2, s->sgaincod);
1089         put_bits(&s->pb, 2, s->dbkneecod);
1090         put_bits(&s->pb, 3, s->floorcod);
1091     }
1092
1093     /* snr offset */
1094     put_bits(&s->pb, 1, baie); /* always present with bai */
1095     if (baie) {
1096         put_bits(&s->pb, 6, s->csnroffst);
1097         for(ch=0;ch<s->nb_all_channels;ch++) {
1098             put_bits(&s->pb, 4, s->fsnroffst[ch]);
1099             put_bits(&s->pb, 3, s->fgaincod[ch]);
1100         }
1101     }
1102
1103     put_bits(&s->pb, 1, 0); /* no delta bit allocation */
1104     put_bits(&s->pb, 1, 0); /* no data to skip */
1105
1106     /* mantissa encoding : we use two passes to handle the grouping. A
1107        one pass method may be faster, but it would necessitate to
1108        modify the output stream. */
1109
1110     /* first pass: quantize */
1111     mant1_cnt = mant2_cnt = mant4_cnt = 0;
1112     qmant1_ptr = qmant2_ptr = qmant4_ptr = NULL;
1113
1114     for (ch = 0; ch < s->nb_all_channels; ch++) {
1115         int b, c, e, v;
1116
1117         for(i=0;i<s->nb_coefs[ch];i++) {
1118             c = mdct_coefs[ch][i];
1119             e = encoded_exp[ch][i] - global_exp[ch];
1120             b = bap[ch][i];
1121             switch(b) {
1122             case 0:
1123                 v = 0;
1124                 break;
1125             case 1:
1126                 v = sym_quant(c, e, 3);
1127                 switch(mant1_cnt) {
1128                 case 0:
1129                     qmant1_ptr = &qmant[ch][i];
1130                     v = 9 * v;
1131                     mant1_cnt = 1;
1132                     break;
1133                 case 1:
1134                     *qmant1_ptr += 3 * v;
1135                     mant1_cnt = 2;
1136                     v = 128;
1137                     break;
1138                 default:
1139                     *qmant1_ptr += v;
1140                     mant1_cnt = 0;
1141                     v = 128;
1142                     break;
1143                 }
1144                 break;
1145             case 2:
1146                 v = sym_quant(c, e, 5);
1147                 switch(mant2_cnt) {
1148                 case 0:
1149                     qmant2_ptr = &qmant[ch][i];
1150                     v = 25 * v;
1151                     mant2_cnt = 1;
1152                     break;
1153                 case 1:
1154                     *qmant2_ptr += 5 * v;
1155                     mant2_cnt = 2;
1156                     v = 128;
1157                     break;
1158                 default:
1159                     *qmant2_ptr += v;
1160                     mant2_cnt = 0;
1161                     v = 128;
1162                     break;
1163                 }
1164                 break;
1165             case 3:
1166                 v = sym_quant(c, e, 7);
1167                 break;
1168             case 4:
1169                 v = sym_quant(c, e, 11);
1170                 switch(mant4_cnt) {
1171                 case 0:
1172                     qmant4_ptr = &qmant[ch][i];
1173                     v = 11 * v;
1174                     mant4_cnt = 1;
1175                     break;
1176                 default:
1177                     *qmant4_ptr += v;
1178                     mant4_cnt = 0;
1179                     v = 128;
1180                     break;
1181                 }
1182                 break;
1183             case 5:
1184                 v = sym_quant(c, e, 15);
1185                 break;
1186             case 14:
1187                 v = asym_quant(c, e, 14);
1188                 break;
1189             case 15:
1190                 v = asym_quant(c, e, 16);
1191                 break;
1192             default:
1193                 v = asym_quant(c, e, b - 1);
1194                 break;
1195             }
1196             qmant[ch][i] = v;
1197         }
1198     }
1199
1200     /* second pass : output the values */
1201     for (ch = 0; ch < s->nb_all_channels; ch++) {
1202         int b, q;
1203
1204         for(i=0;i<s->nb_coefs[ch];i++) {
1205             q = qmant[ch][i];
1206             b = bap[ch][i];
1207             switch(b) {
1208             case 0:
1209                 break;
1210             case 1:
1211                 if (q != 128)
1212                     put_bits(&s->pb, 5, q);
1213                 break;
1214             case 2:
1215                 if (q != 128)
1216                     put_bits(&s->pb, 7, q);
1217                 break;
1218             case 3:
1219                 put_bits(&s->pb, 3, q);
1220                 break;
1221             case 4:
1222                 if (q != 128)
1223                     put_bits(&s->pb, 7, q);
1224                 break;
1225             case 14:
1226                 put_bits(&s->pb, 14, q);
1227                 break;
1228             case 15:
1229                 put_bits(&s->pb, 16, q);
1230                 break;
1231             default:
1232                 put_bits(&s->pb, b - 1, q);
1233                 break;
1234             }
1235         }
1236     }
1237 }
1238
1239 /* compute the ac3 crc */
1240
1241 #define CRC16_POLY ((1 << 0) | (1 << 2) | (1 << 15) | (1 << 16))
1242
1243 static void ac3_crc_init(void)
1244 {
1245     unsigned int c, n, k;
1246
1247     for(n=0;n<256;n++) {
1248         c = n << 8;
1249         for (k = 0; k < 8; k++) {
1250             if (c & (1 << 15))
1251                 c = ((c << 1) & 0xffff) ^ (CRC16_POLY & 0xffff);
1252             else
1253                 c = c << 1;
1254         }
1255         crc_table[n] = c;
1256     }
1257 }
1258
1259 static unsigned int ac3_crc(uint8_t *data, int n, unsigned int crc)
1260 {
1261     int i;
1262     for(i=0;i<n;i++) {
1263         crc = (crc_table[data[i] ^ (crc >> 8)] ^ (crc << 8)) & 0xffff;
1264     }
1265     return crc;
1266 }
1267
1268 static unsigned int mul_poly(unsigned int a, unsigned int b, unsigned int poly)
1269 {
1270     unsigned int c;
1271
1272     c = 0;
1273     while (a) {
1274         if (a & 1)
1275             c ^= b;
1276         a = a >> 1;
1277         b = b << 1;
1278         if (b & (1 << 16))
1279             b ^= poly;
1280     }
1281     return c;
1282 }
1283
1284 static unsigned int pow_poly(unsigned int a, unsigned int n, unsigned int poly)
1285 {
1286     unsigned int r;
1287     r = 1;
1288     while (n) {
1289         if (n & 1)
1290             r = mul_poly(r, a, poly);
1291         a = mul_poly(a, a, poly);
1292         n >>= 1;
1293     }
1294     return r;
1295 }
1296
1297
1298 /* compute log2(max(abs(tab[]))) */
1299 static int log2_tab(int16_t *tab, int n)
1300 {
1301     int i, v;
1302
1303     v = 0;
1304     for(i=0;i<n;i++) {
1305         v |= abs(tab[i]);
1306     }
1307     return av_log2(v);
1308 }
1309
1310 static void lshift_tab(int16_t *tab, int n, int lshift)
1311 {
1312     int i;
1313
1314     if (lshift > 0) {
1315         for(i=0;i<n;i++) {
1316             tab[i] <<= lshift;
1317         }
1318     } else if (lshift < 0) {
1319         lshift = -lshift;
1320         for(i=0;i<n;i++) {
1321             tab[i] >>= lshift;
1322         }
1323     }
1324 }
1325
1326 /* fill the end of the frame and compute the two crcs */
1327 static int output_frame_end(AC3EncodeContext *s)
1328 {
1329     int frame_size, frame_size_58, n, crc1, crc2, crc_inv;
1330     uint8_t *frame;
1331
1332     frame_size = s->frame_size; /* frame size in words */
1333     /* align to 8 bits */
1334     flush_put_bits(&s->pb);
1335     /* add zero bytes to reach the frame size */
1336     frame = s->pb.buf;
1337     n = 2 * s->frame_size - (pbBufPtr(&s->pb) - frame) - 2;
1338     assert(n >= 0);
1339     if(n>0)
1340       memset(pbBufPtr(&s->pb), 0, n);
1341
1342     /* Now we must compute both crcs : this is not so easy for crc1
1343        because it is at the beginning of the data... */
1344     frame_size_58 = (frame_size >> 1) + (frame_size >> 3);
1345     crc1 = ac3_crc(frame + 4, (2 * frame_size_58) - 4, 0);
1346     /* XXX: could precompute crc_inv */
1347     crc_inv = pow_poly((CRC16_POLY >> 1), (16 * frame_size_58) - 16, CRC16_POLY);
1348     crc1 = mul_poly(crc_inv, crc1, CRC16_POLY);
1349     frame[2] = crc1 >> 8;
1350     frame[3] = crc1;
1351
1352     crc2 = ac3_crc(frame + 2 * frame_size_58, (frame_size - frame_size_58) * 2 - 2, 0);
1353     frame[2*frame_size - 2] = crc2 >> 8;
1354     frame[2*frame_size - 1] = crc2;
1355
1356     //    printf("n=%d frame_size=%d\n", n, frame_size);
1357     return frame_size * 2;
1358 }
1359
1360 static int AC3_encode_frame(AVCodecContext *avctx,
1361                             unsigned char *frame, int buf_size, void *data)
1362 {
1363     AC3EncodeContext *s = avctx->priv_data;
1364     int16_t *samples = data;
1365     int i, j, k, v, ch;
1366     int16_t input_samples[N];
1367     int32_t mdct_coef[NB_BLOCKS][AC3_MAX_CHANNELS][N/2];
1368     uint8_t exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2];
1369     uint8_t exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS];
1370     uint8_t encoded_exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2];
1371     uint8_t bap[NB_BLOCKS][AC3_MAX_CHANNELS][N/2];
1372     int8_t exp_samples[NB_BLOCKS][AC3_MAX_CHANNELS];
1373     int frame_bits;
1374
1375     frame_bits = 0;
1376     for(ch=0;ch<s->nb_all_channels;ch++) {
1377         /* fixed mdct to the six sub blocks & exponent computation */
1378         for(i=0;i<NB_BLOCKS;i++) {
1379             int16_t *sptr;
1380             int sinc;
1381
1382             /* compute input samples */
1383             memcpy(input_samples, s->last_samples[ch], N/2 * sizeof(int16_t));
1384             sinc = s->nb_all_channels;
1385             sptr = samples + (sinc * (N/2) * i) + ch;
1386             for(j=0;j<N/2;j++) {
1387                 v = *sptr;
1388                 input_samples[j + N/2] = v;
1389                 s->last_samples[ch][j] = v;
1390                 sptr += sinc;
1391             }
1392
1393             /* apply the MDCT window */
1394             for(j=0;j<N/2;j++) {
1395                 input_samples[j] = MUL16(input_samples[j],
1396                                          ac3_window[j]) >> 15;
1397                 input_samples[N-j-1] = MUL16(input_samples[N-j-1],
1398                                              ac3_window[j]) >> 15;
1399             }
1400
1401             /* Normalize the samples to use the maximum available
1402                precision */
1403             v = 14 - log2_tab(input_samples, N);
1404             if (v < 0)
1405                 v = 0;
1406             exp_samples[i][ch] = v - 8;
1407             lshift_tab(input_samples, N, v);
1408
1409             /* do the MDCT */
1410             mdct512(mdct_coef[i][ch], input_samples);
1411
1412             /* compute "exponents". We take into account the
1413                normalization there */
1414             for(j=0;j<N/2;j++) {
1415                 int e;
1416                 v = abs(mdct_coef[i][ch][j]);
1417                 if (v == 0)
1418                     e = 24;
1419                 else {
1420                     e = 23 - av_log2(v) + exp_samples[i][ch];
1421                     if (e >= 24) {
1422                         e = 24;
1423                         mdct_coef[i][ch][j] = 0;
1424                     }
1425                 }
1426                 exp[i][ch][j] = e;
1427             }
1428         }
1429
1430         compute_exp_strategy(exp_strategy, exp, ch, ch == s->lfe_channel);
1431
1432         /* compute the exponents as the decoder will see them. The
1433            EXP_REUSE case must be handled carefully : we select the
1434            min of the exponents */
1435         i = 0;
1436         while (i < NB_BLOCKS) {
1437             j = i + 1;
1438             while (j < NB_BLOCKS && exp_strategy[j][ch] == EXP_REUSE) {
1439                 exponent_min(exp[i][ch], exp[j][ch], s->nb_coefs[ch]);
1440                 j++;
1441             }
1442             frame_bits += encode_exp(encoded_exp[i][ch],
1443                                      exp[i][ch], s->nb_coefs[ch],
1444                                      exp_strategy[i][ch]);
1445             /* copy encoded exponents for reuse case */
1446             for(k=i+1;k<j;k++) {
1447                 memcpy(encoded_exp[k][ch], encoded_exp[i][ch],
1448                        s->nb_coefs[ch] * sizeof(uint8_t));
1449             }
1450             i = j;
1451         }
1452     }
1453
1454     compute_bit_allocation(s, bap, encoded_exp, exp_strategy, frame_bits);
1455     /* everything is known... let's output the frame */
1456     output_frame_header(s, frame);
1457
1458     for(i=0;i<NB_BLOCKS;i++) {
1459         output_audio_block(s, exp_strategy[i], encoded_exp[i],
1460                            bap[i], mdct_coef[i], exp_samples[i], i);
1461     }
1462     return output_frame_end(s);
1463 }
1464
1465 static int AC3_encode_close(AVCodecContext *avctx)
1466 {
1467     av_freep(&avctx->coded_frame);
1468     return 0;
1469 }
1470
1471 #if 0
1472 /*************************************************************************/
1473 /* TEST */
1474
1475 #define FN (N/4)
1476
1477 void fft_test(void)
1478 {
1479     IComplex in[FN], in1[FN];
1480     int k, n, i;
1481     float sum_re, sum_im, a;
1482
1483     /* FFT test */
1484
1485     for(i=0;i<FN;i++) {
1486         in[i].re = random() % 65535 - 32767;
1487         in[i].im = random() % 65535 - 32767;
1488         in1[i] = in[i];
1489     }
1490     fft(in, 7);
1491
1492     /* do it by hand */
1493     for(k=0;k<FN;k++) {
1494         sum_re = 0;
1495         sum_im = 0;
1496         for(n=0;n<FN;n++) {
1497             a = -2 * M_PI * (n * k) / FN;
1498             sum_re += in1[n].re * cos(a) - in1[n].im * sin(a);
1499             sum_im += in1[n].re * sin(a) + in1[n].im * cos(a);
1500         }
1501         printf("%3d: %6d,%6d %6.0f,%6.0f\n",
1502                k, in[k].re, in[k].im, sum_re / FN, sum_im / FN);
1503     }
1504 }
1505
1506 void mdct_test(void)
1507 {
1508     int16_t input[N];
1509     int32_t output[N/2];
1510     float input1[N];
1511     float output1[N/2];
1512     float s, a, err, e, emax;
1513     int i, k, n;
1514
1515     for(i=0;i<N;i++) {
1516         input[i] = (random() % 65535 - 32767) * 9 / 10;
1517         input1[i] = input[i];
1518     }
1519
1520     mdct512(output, input);
1521
1522     /* do it by hand */
1523     for(k=0;k<N/2;k++) {
1524         s = 0;
1525         for(n=0;n<N;n++) {
1526             a = (2*M_PI*(2*n+1+N/2)*(2*k+1) / (4 * N));
1527             s += input1[n] * cos(a);
1528         }
1529         output1[k] = -2 * s / N;
1530     }
1531
1532     err = 0;
1533     emax = 0;
1534     for(i=0;i<N/2;i++) {
1535         printf("%3d: %7d %7.0f\n", i, output[i], output1[i]);
1536         e = output[i] - output1[i];
1537         if (e > emax)
1538             emax = e;
1539         err += e * e;
1540     }
1541     printf("err2=%f emax=%f\n", err / (N/2), emax);
1542 }
1543
1544 void test_ac3(void)
1545 {
1546     AC3EncodeContext ctx;
1547     unsigned char frame[AC3_MAX_CODED_FRAME_SIZE];
1548     short samples[AC3_FRAME_SIZE];
1549     int ret, i;
1550
1551     AC3_encode_init(&ctx, 44100, 64000, 1);
1552
1553     fft_test();
1554     mdct_test();
1555
1556     for(i=0;i<AC3_FRAME_SIZE;i++)
1557         samples[i] = (int)(sin(2*M_PI*i*1000.0/44100) * 10000);
1558     ret = AC3_encode_frame(&ctx, frame, samples);
1559     printf("ret=%d\n", ret);
1560 }
1561 #endif
1562
1563 AVCodec ac3_encoder = {
1564     "ac3",
1565     CODEC_TYPE_AUDIO,
1566     CODEC_ID_AC3,
1567     sizeof(AC3EncodeContext),
1568     AC3_encode_init,
1569     AC3_encode_frame,
1570     AC3_encode_close,
1571     NULL,
1572 };