Mercurial > libavcodec.hg
annotate ac3enc.c @ 305:1de85b419387 libavcodec
emms was missing, found by juanjo but he didnt commit it?!
author | michaelni |
---|---|
date | Thu, 04 Apr 2002 19:50:46 +0000 |
parents | 5fc0c3af3fe4 |
children | 289eb941b8ba |
rev | line source |
---|---|
0 | 1 /* |
2 * The simplest AC3 encoder | |
3 * Copyright (c) 2000 Gerard Lantau. | |
4 * | |
5 * This program is free software; you can redistribute it and/or modify | |
6 * it under the terms of the GNU General Public License as published by | |
7 * the Free Software Foundation; either version 2 of the License, or | |
8 * (at your option) any later version. | |
9 * | |
10 * This program 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 | |
13 * GNU General Public License for more details. | |
14 * | |
15 * You should have received a copy of the GNU General Public License | |
16 * along with this program; if not, write to the Free Software | |
17 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
18 */ | |
64 | 19 //#define DEBUG |
20 //#define DEBUG_BITALLOC | |
21 #include "avcodec.h" | |
0 | 22 #include <math.h> |
23 | |
24 #include "ac3enc.h" | |
25 #include "ac3tab.h" | |
26 | |
27 | |
28 #define MDCT_NBITS 9 | |
29 #define N (1 << MDCT_NBITS) | |
30 #define NB_BLOCKS 6 /* number of PCM blocks inside an AC3 frame */ | |
31 | |
32 /* new exponents are sent if their Norm 1 exceed this number */ | |
33 #define EXP_DIFF_THRESHOLD 1000 | |
34 | |
35 /* exponent encoding strategy */ | |
36 #define EXP_REUSE 0 | |
37 #define EXP_NEW 1 | |
38 | |
39 #define EXP_D15 1 | |
40 #define EXP_D25 2 | |
41 #define EXP_D45 3 | |
42 | |
43 static void fft_init(int ln); | |
44 static void ac3_crc_init(void); | |
45 | |
46 static inline INT16 fix15(float a) | |
47 { | |
48 int v; | |
49 v = (int)(a * (float)(1 << 15)); | |
50 if (v < -32767) | |
51 v = -32767; | |
52 else if (v > 32767) | |
53 v = 32767; | |
54 return v; | |
55 } | |
56 | |
57 static inline int calc_lowcomp1(int a, int b0, int b1) | |
58 { | |
59 if ((b0 + 256) == b1) { | |
60 a = 384 ; | |
61 } else if (b0 > b1) { | |
62 a = a - 64; | |
63 if (a < 0) a=0; | |
64 } | |
65 return a; | |
66 } | |
67 | |
68 static inline int calc_lowcomp(int a, int b0, int b1, int bin) | |
69 { | |
70 if (bin < 7) { | |
71 if ((b0 + 256) == b1) { | |
72 a = 384 ; | |
73 } else if (b0 > b1) { | |
74 a = a - 64; | |
75 if (a < 0) a=0; | |
76 } | |
77 } else if (bin < 20) { | |
78 if ((b0 + 256) == b1) { | |
79 a = 320 ; | |
80 } else if (b0 > b1) { | |
81 a= a - 64; | |
82 if (a < 0) a=0; | |
83 } | |
84 } else { | |
85 a = a - 128; | |
86 if (a < 0) a=0; | |
87 } | |
88 return a; | |
89 } | |
90 | |
91 /* AC3 bit allocation. The algorithm is the one described in the AC3 | |
92 spec with some optimizations because of our simplified encoding | |
93 assumptions. */ | |
94 void parametric_bit_allocation(AC3EncodeContext *s, UINT8 *bap, | |
95 INT8 *exp, int start, int end, | |
96 int snroffset, int fgain) | |
97 { | |
98 int bin,i,j,k,end1,v,v1,bndstrt,bndend,lowcomp,begin; | |
99 int fastleak,slowleak,address,tmp; | |
100 INT16 psd[256]; /* scaled exponents */ | |
101 INT16 bndpsd[50]; /* interpolated exponents */ | |
102 INT16 excite[50]; /* excitation */ | |
103 INT16 mask[50]; /* masking value */ | |
104 | |
105 /* exponent mapping to PSD */ | |
106 for(bin=start;bin<end;bin++) { | |
107 psd[bin]=(3072 - (exp[bin] << 7)); | |
108 } | |
109 | |
110 /* PSD integration */ | |
111 j=start; | |
112 k=masktab[start]; | |
113 do { | |
114 v=psd[j]; | |
115 j++; | |
116 end1=bndtab[k+1]; | |
117 if (end1 > end) end1=end; | |
118 for(i=j;i<end1;i++) { | |
119 int c,adr; | |
120 /* logadd */ | |
121 v1=psd[j]; | |
122 c=v-v1; | |
123 if (c >= 0) { | |
124 adr=c >> 1; | |
125 if (adr > 255) adr=255; | |
126 v=v + latab[adr]; | |
127 } else { | |
128 adr=(-c) >> 1; | |
129 if (adr > 255) adr=255; | |
130 v=v1 + latab[adr]; | |
131 } | |
132 j++; | |
133 } | |
134 bndpsd[k]=v; | |
135 k++; | |
136 } while (end > bndtab[k]); | |
137 | |
138 /* excitation function */ | |
139 bndstrt = masktab[start]; | |
140 bndend = masktab[end-1] + 1; | |
141 | |
142 lowcomp = 0; | |
143 lowcomp = calc_lowcomp1(lowcomp, bndpsd[0], bndpsd[1]) ; | |
144 excite[0] = bndpsd[0] - fgain - lowcomp ; | |
145 lowcomp = calc_lowcomp1(lowcomp, bndpsd[1], bndpsd[2]) ; | |
146 excite[1] = bndpsd[1] - fgain - lowcomp ; | |
147 begin = 7 ; | |
148 for (bin = 2; bin < 7; bin++) { | |
149 lowcomp = calc_lowcomp1(lowcomp, bndpsd[bin], bndpsd[bin+1]) ; | |
150 fastleak = bndpsd[bin] - fgain ; | |
151 slowleak = bndpsd[bin] - s->sgain ; | |
152 excite[bin] = fastleak - lowcomp ; | |
153 if (bndpsd[bin] <= bndpsd[bin+1]) { | |
154 begin = bin + 1 ; | |
155 break ; | |
156 } | |
157 } | |
158 | |
159 end1=bndend; | |
160 if (end1 > 22) end1=22; | |
161 | |
162 for (bin = begin; bin < end1; bin++) { | |
163 lowcomp = calc_lowcomp(lowcomp, bndpsd[bin], bndpsd[bin+1], bin) ; | |
164 | |
165 fastleak -= s->fdecay ; | |
166 v = bndpsd[bin] - fgain; | |
167 if (fastleak < v) fastleak = v; | |
168 | |
169 slowleak -= s->sdecay ; | |
170 v = bndpsd[bin] - s->sgain; | |
171 if (slowleak < v) slowleak = v; | |
172 | |
173 v=fastleak - lowcomp; | |
174 if (slowleak > v) v=slowleak; | |
175 | |
176 excite[bin] = v; | |
177 } | |
178 | |
179 for (bin = 22; bin < bndend; bin++) { | |
180 fastleak -= s->fdecay ; | |
181 v = bndpsd[bin] - fgain; | |
182 if (fastleak < v) fastleak = v; | |
183 slowleak -= s->sdecay ; | |
184 v = bndpsd[bin] - s->sgain; | |
185 if (slowleak < v) slowleak = v; | |
186 | |
187 v=fastleak; | |
188 if (slowleak > v) v = slowleak; | |
189 excite[bin] = v; | |
190 } | |
191 | |
192 /* compute masking curve */ | |
193 | |
194 for (bin = bndstrt; bin < bndend; bin++) { | |
195 v1 = excite[bin]; | |
196 tmp = s->dbknee - bndpsd[bin]; | |
197 if (tmp > 0) { | |
198 v1 += tmp >> 2; | |
199 } | |
200 v=hth[bin >> s->halfratecod][s->fscod]; | |
201 if (v1 > v) v=v1; | |
202 mask[bin] = v; | |
203 } | |
204 | |
205 /* compute bit allocation */ | |
206 | |
207 i = start ; | |
208 j = masktab[start] ; | |
209 do { | |
210 v=mask[j]; | |
211 v -= snroffset ; | |
212 v -= s->floor ; | |
213 if (v < 0) v = 0; | |
214 v &= 0x1fe0 ; | |
215 v += s->floor ; | |
216 | |
217 end1=bndtab[j] + bndsz[j]; | |
218 if (end1 > end) end1=end; | |
219 | |
220 for (k = i; k < end1; k++) { | |
221 address = (psd[i] - v) >> 5 ; | |
222 if (address < 0) address=0; | |
223 else if (address > 63) address=63; | |
224 bap[i] = baptab[address]; | |
225 i++; | |
226 } | |
227 } while (end > bndtab[j++]) ; | |
228 } | |
229 | |
230 typedef struct IComplex { | |
231 short re,im; | |
232 } IComplex; | |
233 | |
234 static void fft_init(int ln) | |
235 { | |
236 int i, j, m, n; | |
237 float alpha; | |
238 | |
239 n = 1 << ln; | |
240 | |
241 for(i=0;i<(n/2);i++) { | |
242 alpha = 2 * M_PI * (float)i / (float)n; | |
243 costab[i] = fix15(cos(alpha)); | |
244 sintab[i] = fix15(sin(alpha)); | |
245 } | |
246 | |
247 for(i=0;i<n;i++) { | |
248 m=0; | |
249 for(j=0;j<ln;j++) { | |
250 m |= ((i >> j) & 1) << (ln-j-1); | |
251 } | |
252 fft_rev[i]=m; | |
253 } | |
254 } | |
255 | |
256 /* butter fly op */ | |
257 #define BF(pre, pim, qre, qim, pre1, pim1, qre1, qim1) \ | |
258 {\ | |
259 int ax, ay, bx, by;\ | |
260 bx=pre1;\ | |
261 by=pim1;\ | |
262 ax=qre1;\ | |
263 ay=qim1;\ | |
264 pre = (bx + ax) >> 1;\ | |
265 pim = (by + ay) >> 1;\ | |
266 qre = (bx - ax) >> 1;\ | |
267 qim = (by - ay) >> 1;\ | |
268 } | |
269 | |
270 #define MUL16(a,b) ((a) * (b)) | |
271 | |
272 #define CMUL(pre, pim, are, aim, bre, bim) \ | |
273 {\ | |
274 pre = (MUL16(are, bre) - MUL16(aim, bim)) >> 15;\ | |
275 pim = (MUL16(are, bim) + MUL16(bre, aim)) >> 15;\ | |
276 } | |
277 | |
278 | |
279 /* do a 2^n point complex fft on 2^ln points. */ | |
280 static void fft(IComplex *z, int ln) | |
281 { | |
282 int j, l, np, np2; | |
283 int nblocks, nloops; | |
284 register IComplex *p,*q; | |
285 int tmp_re, tmp_im; | |
286 | |
287 np = 1 << ln; | |
288 | |
289 /* reverse */ | |
290 for(j=0;j<np;j++) { | |
291 int k; | |
292 IComplex tmp; | |
293 k = fft_rev[j]; | |
294 if (k < j) { | |
295 tmp = z[k]; | |
296 z[k] = z[j]; | |
297 z[j] = tmp; | |
298 } | |
299 } | |
300 | |
301 /* pass 0 */ | |
302 | |
303 p=&z[0]; | |
304 j=(np >> 1); | |
305 do { | |
306 BF(p[0].re, p[0].im, p[1].re, p[1].im, | |
307 p[0].re, p[0].im, p[1].re, p[1].im); | |
308 p+=2; | |
309 } while (--j != 0); | |
310 | |
311 /* pass 1 */ | |
312 | |
313 p=&z[0]; | |
314 j=np >> 2; | |
315 do { | |
316 BF(p[0].re, p[0].im, p[2].re, p[2].im, | |
317 p[0].re, p[0].im, p[2].re, p[2].im); | |
318 BF(p[1].re, p[1].im, p[3].re, p[3].im, | |
319 p[1].re, p[1].im, p[3].im, -p[3].re); | |
320 p+=4; | |
321 } while (--j != 0); | |
322 | |
323 /* pass 2 .. ln-1 */ | |
324 | |
325 nblocks = np >> 3; | |
326 nloops = 1 << 2; | |
327 np2 = np >> 1; | |
328 do { | |
329 p = z; | |
330 q = z + nloops; | |
331 for (j = 0; j < nblocks; ++j) { | |
332 | |
333 BF(p->re, p->im, q->re, q->im, | |
334 p->re, p->im, q->re, q->im); | |
335 | |
336 p++; | |
337 q++; | |
338 for(l = nblocks; l < np2; l += nblocks) { | |
339 CMUL(tmp_re, tmp_im, costab[l], -sintab[l], q->re, q->im); | |
340 BF(p->re, p->im, q->re, q->im, | |
341 p->re, p->im, tmp_re, tmp_im); | |
342 p++; | |
343 q++; | |
344 } | |
345 p += nloops; | |
346 q += nloops; | |
347 } | |
348 nblocks = nblocks >> 1; | |
349 nloops = nloops << 1; | |
350 } while (nblocks != 0); | |
351 } | |
352 | |
353 /* do a 512 point mdct */ | |
354 static void mdct512(INT32 *out, INT16 *in) | |
355 { | |
356 int i, re, im, re1, im1; | |
357 INT16 rot[N]; | |
358 IComplex x[N/4]; | |
359 | |
360 /* shift to simplify computations */ | |
361 for(i=0;i<N/4;i++) | |
362 rot[i] = -in[i + 3*N/4]; | |
363 for(i=N/4;i<N;i++) | |
364 rot[i] = in[i - N/4]; | |
365 | |
366 /* pre rotation */ | |
367 for(i=0;i<N/4;i++) { | |
368 re = ((int)rot[2*i] - (int)rot[N-1-2*i]) >> 1; | |
369 im = -((int)rot[N/2+2*i] - (int)rot[N/2-1-2*i]) >> 1; | |
370 CMUL(x[i].re, x[i].im, re, im, -xcos1[i], xsin1[i]); | |
371 } | |
372 | |
373 fft(x, MDCT_NBITS - 2); | |
374 | |
375 /* post rotation */ | |
376 for(i=0;i<N/4;i++) { | |
377 re = x[i].re; | |
378 im = x[i].im; | |
379 CMUL(re1, im1, re, im, xsin1[i], xcos1[i]); | |
380 out[2*i] = im1; | |
381 out[N/2-1-2*i] = re1; | |
382 } | |
383 } | |
384 | |
385 /* XXX: use another norm ? */ | |
386 static int calc_exp_diff(UINT8 *exp1, UINT8 *exp2, int n) | |
387 { | |
388 int sum, i; | |
389 sum = 0; | |
390 for(i=0;i<n;i++) { | |
391 sum += abs(exp1[i] - exp2[i]); | |
392 } | |
393 return sum; | |
394 } | |
395 | |
396 static void compute_exp_strategy(UINT8 exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS], | |
397 UINT8 exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2], | |
398 int ch) | |
399 { | |
400 int i, j; | |
401 int exp_diff; | |
402 | |
403 /* estimate if the exponent variation & decide if they should be | |
404 reused in the next frame */ | |
405 exp_strategy[0][ch] = EXP_NEW; | |
406 for(i=1;i<NB_BLOCKS;i++) { | |
407 exp_diff = calc_exp_diff(exp[i][ch], exp[i-1][ch], N/2); | |
408 #ifdef DEBUG | |
409 printf("exp_diff=%d\n", exp_diff); | |
410 #endif | |
411 if (exp_diff > EXP_DIFF_THRESHOLD) | |
412 exp_strategy[i][ch] = EXP_NEW; | |
413 else | |
414 exp_strategy[i][ch] = EXP_REUSE; | |
415 } | |
416 /* now select the encoding strategy type : if exponents are often | |
417 recoded, we use a coarse encoding */ | |
418 i = 0; | |
419 while (i < NB_BLOCKS) { | |
420 j = i + 1; | |
421 while (j < NB_BLOCKS && exp_strategy[j][ch] == EXP_REUSE) | |
422 j++; | |
423 switch(j - i) { | |
424 case 1: | |
425 exp_strategy[i][ch] = EXP_D45; | |
426 break; | |
427 case 2: | |
428 case 3: | |
429 exp_strategy[i][ch] = EXP_D25; | |
430 break; | |
431 default: | |
432 exp_strategy[i][ch] = EXP_D15; | |
433 break; | |
434 } | |
435 i = j; | |
436 } | |
437 } | |
438 | |
439 /* set exp[i] to min(exp[i], exp1[i]) */ | |
440 static void exponent_min(UINT8 exp[N/2], UINT8 exp1[N/2], int n) | |
441 { | |
442 int i; | |
443 | |
444 for(i=0;i<n;i++) { | |
445 if (exp1[i] < exp[i]) | |
446 exp[i] = exp1[i]; | |
447 } | |
448 } | |
449 | |
450 /* update the exponents so that they are the ones the decoder will | |
451 decode. Return the number of bits used to code the exponents */ | |
452 static int encode_exp(UINT8 encoded_exp[N/2], | |
453 UINT8 exp[N/2], | |
454 int nb_exps, | |
455 int exp_strategy) | |
456 { | |
457 int group_size, nb_groups, i, j, k, recurse, exp_min, delta; | |
458 UINT8 exp1[N/2]; | |
459 | |
460 switch(exp_strategy) { | |
461 case EXP_D15: | |
462 group_size = 1; | |
463 break; | |
464 case EXP_D25: | |
465 group_size = 2; | |
466 break; | |
467 default: | |
468 case EXP_D45: | |
469 group_size = 4; | |
470 break; | |
471 } | |
472 nb_groups = ((nb_exps + (group_size * 3) - 4) / (3 * group_size)) * 3; | |
473 | |
474 /* for each group, compute the minimum exponent */ | |
475 exp1[0] = exp[0]; /* DC exponent is handled separately */ | |
476 k = 1; | |
477 for(i=1;i<=nb_groups;i++) { | |
478 exp_min = exp[k]; | |
479 assert(exp_min >= 0 && exp_min <= 24); | |
480 for(j=1;j<group_size;j++) { | |
481 if (exp[k+j] < exp_min) | |
482 exp_min = exp[k+j]; | |
483 } | |
484 exp1[i] = exp_min; | |
485 k += group_size; | |
486 } | |
487 | |
488 /* constraint for DC exponent */ | |
489 if (exp1[0] > 15) | |
490 exp1[0] = 15; | |
491 | |
492 /* Iterate until the delta constraints between each groups are | |
493 satisfyed. I'm sure it is possible to find a better algorithm, | |
494 but I am lazy */ | |
495 do { | |
496 recurse = 0; | |
497 for(i=1;i<=nb_groups;i++) { | |
498 delta = exp1[i] - exp1[i-1]; | |
499 if (delta > 2) { | |
500 /* if delta too big, we encode a smaller exponent */ | |
501 exp1[i] = exp1[i-1] + 2; | |
502 } else if (delta < -2) { | |
503 /* if delta is too small, we must decrease the previous | |
504 exponent, which means we must recurse */ | |
505 recurse = 1; | |
506 exp1[i-1] = exp1[i] + 2; | |
507 } | |
508 } | |
509 } while (recurse); | |
510 | |
511 /* now we have the exponent values the decoder will see */ | |
512 encoded_exp[0] = exp1[0]; | |
513 k = 1; | |
514 for(i=1;i<=nb_groups;i++) { | |
515 for(j=0;j<group_size;j++) { | |
516 encoded_exp[k+j] = exp1[i]; | |
517 } | |
518 k += group_size; | |
519 } | |
520 | |
521 #if defined(DEBUG) | |
522 printf("exponents: strategy=%d\n", exp_strategy); | |
523 for(i=0;i<=nb_groups * group_size;i++) { | |
524 printf("%d ", encoded_exp[i]); | |
525 } | |
526 printf("\n"); | |
527 #endif | |
528 | |
529 return 4 + (nb_groups / 3) * 7; | |
530 } | |
531 | |
532 /* return the size in bits taken by the mantissa */ | |
533 int compute_mantissa_size(AC3EncodeContext *s, UINT8 *m, int nb_coefs) | |
534 { | |
535 int bits, mant, i; | |
536 | |
537 bits = 0; | |
538 for(i=0;i<nb_coefs;i++) { | |
539 mant = m[i]; | |
540 switch(mant) { | |
541 case 0: | |
542 /* nothing */ | |
543 break; | |
544 case 1: | |
545 /* 3 mantissa in 5 bits */ | |
546 if (s->mant1_cnt == 0) | |
547 bits += 5; | |
548 if (++s->mant1_cnt == 3) | |
549 s->mant1_cnt = 0; | |
550 break; | |
551 case 2: | |
552 /* 3 mantissa in 7 bits */ | |
553 if (s->mant2_cnt == 0) | |
554 bits += 7; | |
555 if (++s->mant2_cnt == 3) | |
556 s->mant2_cnt = 0; | |
557 break; | |
558 case 3: | |
559 bits += 3; | |
560 break; | |
561 case 4: | |
562 /* 2 mantissa in 7 bits */ | |
563 if (s->mant4_cnt == 0) | |
564 bits += 7; | |
565 if (++s->mant4_cnt == 2) | |
566 s->mant4_cnt = 0; | |
567 break; | |
568 case 14: | |
569 bits += 14; | |
570 break; | |
571 case 15: | |
572 bits += 16; | |
573 break; | |
574 default: | |
575 bits += mant - 1; | |
576 break; | |
577 } | |
578 } | |
579 return bits; | |
580 } | |
581 | |
582 | |
583 static int bit_alloc(AC3EncodeContext *s, | |
584 UINT8 bap[NB_BLOCKS][AC3_MAX_CHANNELS][N/2], | |
585 UINT8 encoded_exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2], | |
586 UINT8 exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS], | |
587 int frame_bits, int csnroffst, int fsnroffst) | |
588 { | |
589 int i, ch; | |
590 | |
591 /* compute size */ | |
592 for(i=0;i<NB_BLOCKS;i++) { | |
593 s->mant1_cnt = 0; | |
594 s->mant2_cnt = 0; | |
595 s->mant4_cnt = 0; | |
596 for(ch=0;ch<s->nb_channels;ch++) { | |
25 | 597 parametric_bit_allocation(s, bap[i][ch], (INT8 *)encoded_exp[i][ch], |
0 | 598 0, s->nb_coefs[ch], |
599 (((csnroffst-15) << 4) + | |
600 fsnroffst) << 2, | |
601 fgaintab[s->fgaincod[ch]]); | |
602 frame_bits += compute_mantissa_size(s, bap[i][ch], | |
603 s->nb_coefs[ch]); | |
604 } | |
605 } | |
606 #if 0 | |
607 printf("csnr=%d fsnr=%d frame_bits=%d diff=%d\n", | |
608 csnroffst, fsnroffst, frame_bits, | |
609 16 * s->frame_size - ((frame_bits + 7) & ~7)); | |
610 #endif | |
611 return 16 * s->frame_size - frame_bits; | |
612 } | |
613 | |
614 #define SNR_INC1 4 | |
615 | |
616 static int compute_bit_allocation(AC3EncodeContext *s, | |
617 UINT8 bap[NB_BLOCKS][AC3_MAX_CHANNELS][N/2], | |
618 UINT8 encoded_exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2], | |
619 UINT8 exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS], | |
620 int frame_bits) | |
621 { | |
622 int i, ch; | |
623 int csnroffst, fsnroffst; | |
624 UINT8 bap1[NB_BLOCKS][AC3_MAX_CHANNELS][N/2]; | |
625 | |
626 /* init default parameters */ | |
627 s->sdecaycod = 2; | |
628 s->fdecaycod = 1; | |
629 s->sgaincod = 1; | |
630 s->dbkneecod = 2; | |
631 s->floorcod = 4; | |
632 for(ch=0;ch<s->nb_channels;ch++) | |
633 s->fgaincod[ch] = 4; | |
634 | |
635 /* compute real values */ | |
636 s->sdecay = sdecaytab[s->sdecaycod] >> s->halfratecod; | |
637 s->fdecay = fdecaytab[s->fdecaycod] >> s->halfratecod; | |
638 s->sgain = sgaintab[s->sgaincod]; | |
639 s->dbknee = dbkneetab[s->dbkneecod]; | |
640 s->floor = floortab[s->floorcod]; | |
641 | |
642 /* header size */ | |
643 frame_bits += 65; | |
644 if (s->acmod == 2) | |
645 frame_bits += 2; | |
646 | |
647 /* audio blocks */ | |
648 for(i=0;i<NB_BLOCKS;i++) { | |
649 frame_bits += s->nb_channels * 2 + 2; | |
650 if (s->acmod == 2) | |
651 frame_bits++; | |
652 frame_bits += 2 * s->nb_channels; | |
653 for(ch=0;ch<s->nb_channels;ch++) { | |
654 if (exp_strategy[i][ch] != EXP_REUSE) | |
655 frame_bits += 6 + 2; | |
656 } | |
657 frame_bits++; /* baie */ | |
658 frame_bits++; /* snr */ | |
659 frame_bits += 2; /* delta / skip */ | |
660 } | |
661 frame_bits++; /* cplinu for block 0 */ | |
662 /* bit alloc info */ | |
663 frame_bits += 2*4 + 3 + 6 + s->nb_channels * (4 + 3); | |
664 | |
665 /* CRC */ | |
666 frame_bits += 16; | |
667 | |
668 /* now the big work begins : do the bit allocation. Modify the snr | |
669 offset until we can pack everything in the requested frame size */ | |
670 | |
671 csnroffst = s->csnroffst; | |
672 while (csnroffst >= 0 && | |
673 bit_alloc(s, bap, encoded_exp, exp_strategy, frame_bits, csnroffst, 0) < 0) | |
674 csnroffst -= SNR_INC1; | |
675 if (csnroffst < 0) { | |
676 fprintf(stderr, "Error !!!\n"); | |
677 return -1; | |
678 } | |
679 while ((csnroffst + SNR_INC1) <= 63 && | |
680 bit_alloc(s, bap1, encoded_exp, exp_strategy, frame_bits, | |
681 csnroffst + SNR_INC1, 0) >= 0) { | |
682 csnroffst += SNR_INC1; | |
683 memcpy(bap, bap1, sizeof(bap1)); | |
684 } | |
685 while ((csnroffst + 1) <= 63 && | |
686 bit_alloc(s, bap1, encoded_exp, exp_strategy, frame_bits, csnroffst + 1, 0) >= 0) { | |
687 csnroffst++; | |
688 memcpy(bap, bap1, sizeof(bap1)); | |
689 } | |
690 | |
691 fsnroffst = 0; | |
692 while ((fsnroffst + SNR_INC1) <= 15 && | |
693 bit_alloc(s, bap1, encoded_exp, exp_strategy, frame_bits, | |
694 csnroffst, fsnroffst + SNR_INC1) >= 0) { | |
695 fsnroffst += SNR_INC1; | |
696 memcpy(bap, bap1, sizeof(bap1)); | |
697 } | |
698 while ((fsnroffst + 1) <= 15 && | |
699 bit_alloc(s, bap1, encoded_exp, exp_strategy, frame_bits, | |
700 csnroffst, fsnroffst + 1) >= 0) { | |
701 fsnroffst++; | |
702 memcpy(bap, bap1, sizeof(bap1)); | |
703 } | |
704 | |
705 s->csnroffst = csnroffst; | |
706 for(ch=0;ch<s->nb_channels;ch++) | |
707 s->fsnroffst[ch] = fsnroffst; | |
708 #if defined(DEBUG_BITALLOC) | |
709 { | |
710 int j; | |
711 | |
712 for(i=0;i<6;i++) { | |
713 for(ch=0;ch<s->nb_channels;ch++) { | |
714 printf("Block #%d Ch%d:\n", i, ch); | |
715 printf("bap="); | |
716 for(j=0;j<s->nb_coefs[ch];j++) { | |
717 printf("%d ",bap[i][ch][j]); | |
718 } | |
719 printf("\n"); | |
720 } | |
721 } | |
722 } | |
723 #endif | |
724 return 0; | |
725 } | |
726 | |
727 static int AC3_encode_init(AVCodecContext *avctx) | |
728 { | |
729 int freq = avctx->sample_rate; | |
730 int bitrate = avctx->bit_rate; | |
731 int channels = avctx->channels; | |
732 AC3EncodeContext *s = avctx->priv_data; | |
733 int i, j, k, l, ch, v; | |
734 float alpha; | |
735 static unsigned short freqs[3] = { 48000, 44100, 32000 }; | |
736 | |
737 avctx->frame_size = AC3_FRAME_SIZE; | |
738 avctx->key_frame = 1; /* always key frame */ | |
739 | |
740 /* number of channels */ | |
741 if (channels == 1) | |
742 s->acmod = 1; | |
743 else if (channels == 2) | |
744 s->acmod = 2; | |
745 else | |
746 return -1; | |
747 s->nb_channels = channels; | |
748 | |
749 /* frequency */ | |
750 for(i=0;i<3;i++) { | |
751 for(j=0;j<3;j++) | |
752 if ((freqs[j] >> i) == freq) | |
753 goto found; | |
754 } | |
755 return -1; | |
756 found: | |
757 s->sample_rate = freq; | |
758 s->halfratecod = i; | |
759 s->fscod = j; | |
760 s->bsid = 8 + s->halfratecod; | |
761 s->bsmod = 0; /* complete main audio service */ | |
762 | |
763 /* bitrate & frame size */ | |
764 bitrate /= 1000; | |
765 for(i=0;i<19;i++) { | |
766 if ((bitratetab[i] >> s->halfratecod) == bitrate) | |
767 break; | |
768 } | |
769 if (i == 19) | |
770 return -1; | |
771 s->bit_rate = bitrate; | |
772 s->frmsizecod = i << 1; | |
773 s->frame_size_min = (bitrate * 1000 * AC3_FRAME_SIZE) / (freq * 16); | |
774 /* for now we do not handle fractional sizes */ | |
775 s->frame_size = s->frame_size_min; | |
776 | |
777 /* bit allocation init */ | |
778 for(ch=0;ch<s->nb_channels;ch++) { | |
779 /* bandwidth for each channel */ | |
780 /* XXX: should compute the bandwidth according to the frame | |
781 size, so that we avoid anoying high freq artefacts */ | |
782 s->chbwcod[ch] = 50; /* sample bandwidth as mpeg audio layer 2 table 0 */ | |
783 s->nb_coefs[ch] = ((s->chbwcod[ch] + 12) * 3) + 37; | |
784 } | |
785 /* initial snr offset */ | |
786 s->csnroffst = 40; | |
787 | |
788 /* compute bndtab and masktab from bandsz */ | |
789 k = 0; | |
790 l = 0; | |
791 for(i=0;i<50;i++) { | |
792 bndtab[i] = l; | |
793 v = bndsz[i]; | |
794 for(j=0;j<v;j++) masktab[k++]=i; | |
795 l += v; | |
796 } | |
797 bndtab[50] = 0; | |
798 | |
799 /* mdct init */ | |
800 fft_init(MDCT_NBITS - 2); | |
801 for(i=0;i<N/4;i++) { | |
802 alpha = 2 * M_PI * (i + 1.0 / 8.0) / (float)N; | |
803 xcos1[i] = fix15(-cos(alpha)); | |
804 xsin1[i] = fix15(-sin(alpha)); | |
805 } | |
806 | |
807 ac3_crc_init(); | |
808 | |
809 return 0; | |
810 } | |
811 | |
812 /* output the AC3 frame header */ | |
813 static void output_frame_header(AC3EncodeContext *s, unsigned char *frame) | |
814 { | |
815 init_put_bits(&s->pb, frame, AC3_MAX_CODED_FRAME_SIZE, NULL, NULL); | |
816 | |
817 put_bits(&s->pb, 16, 0x0b77); /* frame header */ | |
818 put_bits(&s->pb, 16, 0); /* crc1: will be filled later */ | |
819 put_bits(&s->pb, 2, s->fscod); | |
820 put_bits(&s->pb, 6, s->frmsizecod + (s->frame_size - s->frame_size_min)); | |
821 put_bits(&s->pb, 5, s->bsid); | |
822 put_bits(&s->pb, 3, s->bsmod); | |
823 put_bits(&s->pb, 3, s->acmod); | |
824 if (s->acmod == 2) { | |
825 put_bits(&s->pb, 2, 0); /* surround not indicated */ | |
826 } | |
827 put_bits(&s->pb, 1, 0); /* no LFE */ | |
828 put_bits(&s->pb, 5, 31); /* dialog norm: -31 db */ | |
829 put_bits(&s->pb, 1, 0); /* no compression control word */ | |
830 put_bits(&s->pb, 1, 0); /* no lang code */ | |
831 put_bits(&s->pb, 1, 0); /* no audio production info */ | |
832 put_bits(&s->pb, 1, 0); /* no copyright */ | |
833 put_bits(&s->pb, 1, 1); /* original bitstream */ | |
834 put_bits(&s->pb, 1, 0); /* no time code 1 */ | |
835 put_bits(&s->pb, 1, 0); /* no time code 2 */ | |
836 put_bits(&s->pb, 1, 0); /* no addtional bit stream info */ | |
837 } | |
838 | |
839 /* symetric quantization on 'levels' levels */ | |
840 static inline int sym_quant(int c, int e, int levels) | |
841 { | |
842 int v; | |
843 | |
844 if (c >= 0) { | |
87 | 845 v = (levels * (c << e)) >> 24; |
846 v = (v + 1) >> 1; | |
0 | 847 v = (levels >> 1) + v; |
848 } else { | |
87 | 849 v = (levels * ((-c) << e)) >> 24; |
850 v = (v + 1) >> 1; | |
0 | 851 v = (levels >> 1) - v; |
852 } | |
853 assert (v >= 0 && v < levels); | |
854 return v; | |
855 } | |
856 | |
857 /* asymetric quantization on 2^qbits levels */ | |
858 static inline int asym_quant(int c, int e, int qbits) | |
859 { | |
860 int lshift, m, v; | |
861 | |
862 lshift = e + qbits - 24; | |
863 if (lshift >= 0) | |
864 v = c << lshift; | |
865 else | |
866 v = c >> (-lshift); | |
867 /* rounding */ | |
868 v = (v + 1) >> 1; | |
869 m = (1 << (qbits-1)); | |
870 if (v >= m) | |
871 v = m - 1; | |
872 assert(v >= -m); | |
873 return v & ((1 << qbits)-1); | |
874 } | |
875 | |
876 /* Output one audio block. There are NB_BLOCKS audio blocks in one AC3 | |
877 frame */ | |
878 static void output_audio_block(AC3EncodeContext *s, | |
879 UINT8 exp_strategy[AC3_MAX_CHANNELS], | |
880 UINT8 encoded_exp[AC3_MAX_CHANNELS][N/2], | |
881 UINT8 bap[AC3_MAX_CHANNELS][N/2], | |
882 INT32 mdct_coefs[AC3_MAX_CHANNELS][N/2], | |
883 INT8 global_exp[AC3_MAX_CHANNELS], | |
884 int block_num) | |
885 { | |
886 int ch, nb_groups, group_size, i, baie; | |
887 UINT8 *p; | |
888 UINT16 qmant[AC3_MAX_CHANNELS][N/2]; | |
889 int exp0, exp1; | |
890 int mant1_cnt, mant2_cnt, mant4_cnt; | |
891 UINT16 *qmant1_ptr, *qmant2_ptr, *qmant4_ptr; | |
892 int delta0, delta1, delta2; | |
893 | |
894 for(ch=0;ch<s->nb_channels;ch++) | |
895 put_bits(&s->pb, 1, 0); /* 512 point MDCT */ | |
896 for(ch=0;ch<s->nb_channels;ch++) | |
897 put_bits(&s->pb, 1, 1); /* no dither */ | |
898 put_bits(&s->pb, 1, 0); /* no dynamic range */ | |
899 if (block_num == 0) { | |
900 /* for block 0, even if no coupling, we must say it. This is a | |
901 waste of bit :-) */ | |
902 put_bits(&s->pb, 1, 1); /* coupling strategy present */ | |
903 put_bits(&s->pb, 1, 0); /* no coupling strategy */ | |
904 } else { | |
905 put_bits(&s->pb, 1, 0); /* no new coupling strategy */ | |
906 } | |
907 | |
908 if (s->acmod == 2) { | |
909 put_bits(&s->pb, 1, 0); /* no matrixing (but should be used in the future) */ | |
910 } | |
911 | |
912 #if defined(DEBUG) | |
913 { | |
914 static int count = 0; | |
915 printf("Block #%d (%d)\n", block_num, count++); | |
916 } | |
917 #endif | |
918 /* exponent strategy */ | |
919 for(ch=0;ch<s->nb_channels;ch++) { | |
920 put_bits(&s->pb, 2, exp_strategy[ch]); | |
921 } | |
922 | |
923 for(ch=0;ch<s->nb_channels;ch++) { | |
924 if (exp_strategy[ch] != EXP_REUSE) | |
925 put_bits(&s->pb, 6, s->chbwcod[ch]); | |
926 } | |
927 | |
928 /* exponents */ | |
929 for (ch = 0; ch < s->nb_channels; ch++) { | |
930 switch(exp_strategy[ch]) { | |
931 case EXP_REUSE: | |
932 continue; | |
933 case EXP_D15: | |
934 group_size = 1; | |
935 break; | |
936 case EXP_D25: | |
937 group_size = 2; | |
938 break; | |
939 default: | |
940 case EXP_D45: | |
941 group_size = 4; | |
942 break; | |
943 } | |
944 nb_groups = (s->nb_coefs[ch] + (group_size * 3) - 4) / (3 * group_size); | |
945 p = encoded_exp[ch]; | |
946 | |
947 /* first exponent */ | |
948 exp1 = *p++; | |
949 put_bits(&s->pb, 4, exp1); | |
950 | |
951 /* next ones are delta encoded */ | |
952 for(i=0;i<nb_groups;i++) { | |
953 /* merge three delta in one code */ | |
954 exp0 = exp1; | |
955 exp1 = p[0]; | |
956 p += group_size; | |
957 delta0 = exp1 - exp0 + 2; | |
958 | |
959 exp0 = exp1; | |
960 exp1 = p[0]; | |
961 p += group_size; | |
962 delta1 = exp1 - exp0 + 2; | |
963 | |
964 exp0 = exp1; | |
965 exp1 = p[0]; | |
966 p += group_size; | |
967 delta2 = exp1 - exp0 + 2; | |
968 | |
969 put_bits(&s->pb, 7, ((delta0 * 5 + delta1) * 5) + delta2); | |
970 } | |
971 | |
972 put_bits(&s->pb, 2, 0); /* no gain range info */ | |
973 } | |
974 | |
975 /* bit allocation info */ | |
976 baie = (block_num == 0); | |
977 put_bits(&s->pb, 1, baie); | |
978 if (baie) { | |
979 put_bits(&s->pb, 2, s->sdecaycod); | |
980 put_bits(&s->pb, 2, s->fdecaycod); | |
981 put_bits(&s->pb, 2, s->sgaincod); | |
982 put_bits(&s->pb, 2, s->dbkneecod); | |
983 put_bits(&s->pb, 3, s->floorcod); | |
984 } | |
985 | |
986 /* snr offset */ | |
987 put_bits(&s->pb, 1, baie); /* always present with bai */ | |
988 if (baie) { | |
989 put_bits(&s->pb, 6, s->csnroffst); | |
990 for(ch=0;ch<s->nb_channels;ch++) { | |
991 put_bits(&s->pb, 4, s->fsnroffst[ch]); | |
992 put_bits(&s->pb, 3, s->fgaincod[ch]); | |
993 } | |
994 } | |
995 | |
996 put_bits(&s->pb, 1, 0); /* no delta bit allocation */ | |
997 put_bits(&s->pb, 1, 0); /* no data to skip */ | |
998 | |
999 /* mantissa encoding : we use two passes to handle the grouping. A | |
1000 one pass method may be faster, but it would necessitate to | |
1001 modify the output stream. */ | |
1002 | |
1003 /* first pass: quantize */ | |
1004 mant1_cnt = mant2_cnt = mant4_cnt = 0; | |
1005 qmant1_ptr = qmant2_ptr = qmant4_ptr = NULL; | |
1006 | |
1007 for (ch = 0; ch < s->nb_channels; ch++) { | |
1008 int b, c, e, v; | |
1009 | |
1010 for(i=0;i<s->nb_coefs[ch];i++) { | |
1011 c = mdct_coefs[ch][i]; | |
1012 e = encoded_exp[ch][i] - global_exp[ch]; | |
1013 b = bap[ch][i]; | |
1014 switch(b) { | |
1015 case 0: | |
1016 v = 0; | |
1017 break; | |
1018 case 1: | |
1019 v = sym_quant(c, e, 3); | |
1020 switch(mant1_cnt) { | |
1021 case 0: | |
1022 qmant1_ptr = &qmant[ch][i]; | |
1023 v = 9 * v; | |
1024 mant1_cnt = 1; | |
1025 break; | |
1026 case 1: | |
1027 *qmant1_ptr += 3 * v; | |
1028 mant1_cnt = 2; | |
1029 v = 128; | |
1030 break; | |
1031 default: | |
1032 *qmant1_ptr += v; | |
1033 mant1_cnt = 0; | |
1034 v = 128; | |
1035 break; | |
1036 } | |
1037 break; | |
1038 case 2: | |
1039 v = sym_quant(c, e, 5); | |
1040 switch(mant2_cnt) { | |
1041 case 0: | |
1042 qmant2_ptr = &qmant[ch][i]; | |
1043 v = 25 * v; | |
1044 mant2_cnt = 1; | |
1045 break; | |
1046 case 1: | |
1047 *qmant2_ptr += 5 * v; | |
1048 mant2_cnt = 2; | |
1049 v = 128; | |
1050 break; | |
1051 default: | |
1052 *qmant2_ptr += v; | |
1053 mant2_cnt = 0; | |
1054 v = 128; | |
1055 break; | |
1056 } | |
1057 break; | |
1058 case 3: | |
1059 v = sym_quant(c, e, 7); | |
1060 break; | |
1061 case 4: | |
1062 v = sym_quant(c, e, 11); | |
1063 switch(mant4_cnt) { | |
1064 case 0: | |
1065 qmant4_ptr = &qmant[ch][i]; | |
1066 v = 11 * v; | |
1067 mant4_cnt = 1; | |
1068 break; | |
1069 default: | |
1070 *qmant4_ptr += v; | |
1071 mant4_cnt = 0; | |
1072 v = 128; | |
1073 break; | |
1074 } | |
1075 break; | |
1076 case 5: | |
1077 v = sym_quant(c, e, 15); | |
1078 break; | |
1079 case 14: | |
1080 v = asym_quant(c, e, 14); | |
1081 break; | |
1082 case 15: | |
1083 v = asym_quant(c, e, 16); | |
1084 break; | |
1085 default: | |
1086 v = asym_quant(c, e, b - 1); | |
1087 break; | |
1088 } | |
1089 qmant[ch][i] = v; | |
1090 } | |
1091 } | |
1092 | |
1093 /* second pass : output the values */ | |
1094 for (ch = 0; ch < s->nb_channels; ch++) { | |
1095 int b, q; | |
1096 | |
1097 for(i=0;i<s->nb_coefs[ch];i++) { | |
1098 q = qmant[ch][i]; | |
1099 b = bap[ch][i]; | |
1100 switch(b) { | |
1101 case 0: | |
1102 break; | |
1103 case 1: | |
1104 if (q != 128) | |
1105 put_bits(&s->pb, 5, q); | |
1106 break; | |
1107 case 2: | |
1108 if (q != 128) | |
1109 put_bits(&s->pb, 7, q); | |
1110 break; | |
1111 case 3: | |
1112 put_bits(&s->pb, 3, q); | |
1113 break; | |
1114 case 4: | |
1115 if (q != 128) | |
1116 put_bits(&s->pb, 7, q); | |
1117 break; | |
1118 case 14: | |
1119 put_bits(&s->pb, 14, q); | |
1120 break; | |
1121 case 15: | |
1122 put_bits(&s->pb, 16, q); | |
1123 break; | |
1124 default: | |
1125 put_bits(&s->pb, b - 1, q); | |
1126 break; | |
1127 } | |
1128 } | |
1129 } | |
1130 } | |
1131 | |
1132 /* compute the ac3 crc */ | |
1133 | |
1134 #define CRC16_POLY ((1 << 0) | (1 << 2) | (1 << 15) | (1 << 16)) | |
1135 | |
1136 static void ac3_crc_init(void) | |
1137 { | |
1138 unsigned int c, n, k; | |
1139 | |
1140 for(n=0;n<256;n++) { | |
1141 c = n << 8; | |
1142 for (k = 0; k < 8; k++) { | |
1143 if (c & (1 << 15)) | |
1144 c = ((c << 1) & 0xffff) ^ (CRC16_POLY & 0xffff); | |
1145 else | |
1146 c = c << 1; | |
1147 } | |
1148 crc_table[n] = c; | |
1149 } | |
1150 } | |
1151 | |
1152 static unsigned int ac3_crc(UINT8 *data, int n, unsigned int crc) | |
1153 { | |
1154 int i; | |
1155 for(i=0;i<n;i++) { | |
1156 crc = (crc_table[data[i] ^ (crc >> 8)] ^ (crc << 8)) & 0xffff; | |
1157 } | |
1158 return crc; | |
1159 } | |
1160 | |
1161 static unsigned int mul_poly(unsigned int a, unsigned int b, unsigned int poly) | |
1162 { | |
1163 unsigned int c; | |
1164 | |
1165 c = 0; | |
1166 while (a) { | |
1167 if (a & 1) | |
1168 c ^= b; | |
1169 a = a >> 1; | |
1170 b = b << 1; | |
1171 if (b & (1 << 16)) | |
1172 b ^= poly; | |
1173 } | |
1174 return c; | |
1175 } | |
1176 | |
1177 static unsigned int pow_poly(unsigned int a, unsigned int n, unsigned int poly) | |
1178 { | |
1179 unsigned int r; | |
1180 r = 1; | |
1181 while (n) { | |
1182 if (n & 1) | |
1183 r = mul_poly(r, a, poly); | |
1184 a = mul_poly(a, a, poly); | |
1185 n >>= 1; | |
1186 } | |
1187 return r; | |
1188 } | |
1189 | |
1190 | |
1191 /* compute log2(max(abs(tab[]))) */ | |
1192 static int log2_tab(INT16 *tab, int n) | |
1193 { | |
1194 int i, v; | |
1195 | |
1196 v = 0; | |
1197 for(i=0;i<n;i++) { | |
1198 v |= abs(tab[i]); | |
1199 } | |
65 | 1200 return av_log2(v); |
0 | 1201 } |
1202 | |
1203 static void lshift_tab(INT16 *tab, int n, int lshift) | |
1204 { | |
1205 int i; | |
1206 | |
1207 if (lshift > 0) { | |
1208 for(i=0;i<n;i++) { | |
1209 tab[i] <<= lshift; | |
1210 } | |
1211 } else if (lshift < 0) { | |
1212 lshift = -lshift; | |
1213 for(i=0;i<n;i++) { | |
1214 tab[i] >>= lshift; | |
1215 } | |
1216 } | |
1217 } | |
1218 | |
1219 /* fill the end of the frame and compute the two crcs */ | |
1220 static int output_frame_end(AC3EncodeContext *s) | |
1221 { | |
1222 int frame_size, frame_size_58, n, crc1, crc2, crc_inv; | |
1223 UINT8 *frame; | |
1224 | |
1225 frame_size = s->frame_size; /* frame size in words */ | |
1226 /* align to 8 bits */ | |
1227 flush_put_bits(&s->pb); | |
1228 /* add zero bytes to reach the frame size */ | |
1229 frame = s->pb.buf; | |
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
87
diff
changeset
|
1230 n = 2 * s->frame_size - (pbBufPtr(&s->pb) - frame) - 2; |
0 | 1231 assert(n >= 0); |
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
87
diff
changeset
|
1232 memset(pbBufPtr(&s->pb), 0, n); |
0 | 1233 |
1234 /* Now we must compute both crcs : this is not so easy for crc1 | |
1235 because it is at the beginning of the data... */ | |
1236 frame_size_58 = (frame_size >> 1) + (frame_size >> 3); | |
1237 crc1 = ac3_crc(frame + 4, (2 * frame_size_58) - 4, 0); | |
1238 /* XXX: could precompute crc_inv */ | |
1239 crc_inv = pow_poly((CRC16_POLY >> 1), (16 * frame_size_58) - 16, CRC16_POLY); | |
1240 crc1 = mul_poly(crc_inv, crc1, CRC16_POLY); | |
1241 frame[2] = crc1 >> 8; | |
1242 frame[3] = crc1; | |
1243 | |
1244 crc2 = ac3_crc(frame + 2 * frame_size_58, (frame_size - frame_size_58) * 2 - 2, 0); | |
1245 frame[2*frame_size - 2] = crc2 >> 8; | |
1246 frame[2*frame_size - 1] = crc2; | |
1247 | |
1248 // printf("n=%d frame_size=%d\n", n, frame_size); | |
1249 return frame_size * 2; | |
1250 } | |
1251 | |
1252 int AC3_encode_frame(AVCodecContext *avctx, | |
1253 unsigned char *frame, int buf_size, void *data) | |
1254 { | |
1255 AC3EncodeContext *s = avctx->priv_data; | |
1256 short *samples = data; | |
1257 int i, j, k, v, ch; | |
1258 INT16 input_samples[N]; | |
1259 INT32 mdct_coef[NB_BLOCKS][AC3_MAX_CHANNELS][N/2]; | |
1260 UINT8 exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2]; | |
1261 UINT8 exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS]; | |
1262 UINT8 encoded_exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2]; | |
1263 UINT8 bap[NB_BLOCKS][AC3_MAX_CHANNELS][N/2]; | |
1264 INT8 exp_samples[NB_BLOCKS][AC3_MAX_CHANNELS]; | |
1265 int frame_bits; | |
1266 | |
1267 frame_bits = 0; | |
1268 for(ch=0;ch<s->nb_channels;ch++) { | |
1269 /* fixed mdct to the six sub blocks & exponent computation */ | |
1270 for(i=0;i<NB_BLOCKS;i++) { | |
1271 INT16 *sptr; | |
1272 int sinc; | |
1273 | |
1274 /* compute input samples */ | |
1275 memcpy(input_samples, s->last_samples[ch], N/2 * sizeof(INT16)); | |
1276 sinc = s->nb_channels; | |
1277 sptr = samples + (sinc * (N/2) * i) + ch; | |
1278 for(j=0;j<N/2;j++) { | |
1279 v = *sptr; | |
1280 input_samples[j + N/2] = v; | |
1281 s->last_samples[ch][j] = v; | |
1282 sptr += sinc; | |
1283 } | |
1284 | |
1285 /* apply the MDCT window */ | |
1286 for(j=0;j<N/2;j++) { | |
1287 input_samples[j] = MUL16(input_samples[j], | |
1288 ac3_window[j]) >> 15; | |
1289 input_samples[N-j-1] = MUL16(input_samples[N-j-1], | |
1290 ac3_window[j]) >> 15; | |
1291 } | |
1292 | |
1293 /* Normalize the samples to use the maximum available | |
1294 precision */ | |
1295 v = 14 - log2_tab(input_samples, N); | |
1296 if (v < 0) | |
1297 v = 0; | |
1298 exp_samples[i][ch] = v - 8; | |
1299 lshift_tab(input_samples, N, v); | |
1300 | |
1301 /* do the MDCT */ | |
1302 mdct512(mdct_coef[i][ch], input_samples); | |
1303 | |
1304 /* compute "exponents". We take into account the | |
1305 normalization there */ | |
1306 for(j=0;j<N/2;j++) { | |
1307 int e; | |
1308 v = abs(mdct_coef[i][ch][j]); | |
1309 if (v == 0) | |
1310 e = 24; | |
1311 else { | |
65 | 1312 e = 23 - av_log2(v) + exp_samples[i][ch]; |
0 | 1313 if (e >= 24) { |
1314 e = 24; | |
1315 mdct_coef[i][ch][j] = 0; | |
1316 } | |
1317 } | |
1318 exp[i][ch][j] = e; | |
1319 } | |
1320 } | |
1321 | |
1322 compute_exp_strategy(exp_strategy, exp, ch); | |
1323 | |
1324 /* compute the exponents as the decoder will see them. The | |
1325 EXP_REUSE case must be handled carefully : we select the | |
1326 min of the exponents */ | |
1327 i = 0; | |
1328 while (i < NB_BLOCKS) { | |
1329 j = i + 1; | |
1330 while (j < NB_BLOCKS && exp_strategy[j][ch] == EXP_REUSE) { | |
1331 exponent_min(exp[i][ch], exp[j][ch], s->nb_coefs[ch]); | |
1332 j++; | |
1333 } | |
1334 frame_bits += encode_exp(encoded_exp[i][ch], | |
1335 exp[i][ch], s->nb_coefs[ch], | |
1336 exp_strategy[i][ch]); | |
1337 /* copy encoded exponents for reuse case */ | |
1338 for(k=i+1;k<j;k++) { | |
1339 memcpy(encoded_exp[k][ch], encoded_exp[i][ch], | |
1340 s->nb_coefs[ch] * sizeof(UINT8)); | |
1341 } | |
1342 i = j; | |
1343 } | |
1344 } | |
1345 | |
1346 compute_bit_allocation(s, bap, encoded_exp, exp_strategy, frame_bits); | |
1347 /* everything is known... let's output the frame */ | |
1348 output_frame_header(s, frame); | |
1349 | |
1350 for(i=0;i<NB_BLOCKS;i++) { | |
1351 output_audio_block(s, exp_strategy[i], encoded_exp[i], | |
1352 bap[i], mdct_coef[i], exp_samples[i], i); | |
1353 } | |
1354 return output_frame_end(s); | |
1355 } | |
1356 | |
1357 #if 0 | |
1358 /*************************************************************************/ | |
1359 /* TEST */ | |
1360 | |
1361 #define FN (N/4) | |
1362 | |
1363 void fft_test(void) | |
1364 { | |
1365 IComplex in[FN], in1[FN]; | |
1366 int k, n, i; | |
1367 float sum_re, sum_im, a; | |
1368 | |
1369 /* FFT test */ | |
1370 | |
1371 for(i=0;i<FN;i++) { | |
1372 in[i].re = random() % 65535 - 32767; | |
1373 in[i].im = random() % 65535 - 32767; | |
1374 in1[i] = in[i]; | |
1375 } | |
1376 fft(in, 7); | |
1377 | |
1378 /* do it by hand */ | |
1379 for(k=0;k<FN;k++) { | |
1380 sum_re = 0; | |
1381 sum_im = 0; | |
1382 for(n=0;n<FN;n++) { | |
1383 a = -2 * M_PI * (n * k) / FN; | |
1384 sum_re += in1[n].re * cos(a) - in1[n].im * sin(a); | |
1385 sum_im += in1[n].re * sin(a) + in1[n].im * cos(a); | |
1386 } | |
1387 printf("%3d: %6d,%6d %6.0f,%6.0f\n", | |
1388 k, in[k].re, in[k].im, sum_re / FN, sum_im / FN); | |
1389 } | |
1390 } | |
1391 | |
1392 void mdct_test(void) | |
1393 { | |
1394 INT16 input[N]; | |
1395 INT32 output[N/2]; | |
1396 float input1[N]; | |
1397 float output1[N/2]; | |
1398 float s, a, err, e, emax; | |
1399 int i, k, n; | |
1400 | |
1401 for(i=0;i<N;i++) { | |
1402 input[i] = (random() % 65535 - 32767) * 9 / 10; | |
1403 input1[i] = input[i]; | |
1404 } | |
1405 | |
1406 mdct512(output, input); | |
1407 | |
1408 /* do it by hand */ | |
1409 for(k=0;k<N/2;k++) { | |
1410 s = 0; | |
1411 for(n=0;n<N;n++) { | |
1412 a = (2*M_PI*(2*n+1+N/2)*(2*k+1) / (4 * N)); | |
1413 s += input1[n] * cos(a); | |
1414 } | |
1415 output1[k] = -2 * s / N; | |
1416 } | |
1417 | |
1418 err = 0; | |
1419 emax = 0; | |
1420 for(i=0;i<N/2;i++) { | |
1421 printf("%3d: %7d %7.0f\n", i, output[i], output1[i]); | |
1422 e = output[i] - output1[i]; | |
1423 if (e > emax) | |
1424 emax = e; | |
1425 err += e * e; | |
1426 } | |
1427 printf("err2=%f emax=%f\n", err / (N/2), emax); | |
1428 } | |
1429 | |
1430 void test_ac3(void) | |
1431 { | |
1432 AC3EncodeContext ctx; | |
1433 unsigned char frame[AC3_MAX_CODED_FRAME_SIZE]; | |
1434 short samples[AC3_FRAME_SIZE]; | |
1435 int ret, i; | |
1436 | |
1437 AC3_encode_init(&ctx, 44100, 64000, 1); | |
1438 | |
1439 fft_test(); | |
1440 mdct_test(); | |
1441 | |
1442 for(i=0;i<AC3_FRAME_SIZE;i++) | |
1443 samples[i] = (int)(sin(2*M_PI*i*1000.0/44100) * 10000); | |
1444 ret = AC3_encode_frame(&ctx, frame, samples); | |
1445 printf("ret=%d\n", ret); | |
1446 } | |
1447 #endif | |
1448 | |
1449 AVCodec ac3_encoder = { | |
1450 "ac3", | |
1451 CODEC_TYPE_AUDIO, | |
1452 CODEC_ID_AC3, | |
1453 sizeof(AC3EncodeContext), | |
1454 AC3_encode_init, | |
1455 AC3_encode_frame, | |
1456 NULL, | |
1457 }; |