Mercurial > libavcodec.hg
annotate ac3enc.c @ 372:19b6a1fa6f6d libavcodec
* Every frame is a key_frame
author | philipjsg |
---|---|
date | Thu, 09 May 2002 01:22:51 +0000 |
parents | 289eb941b8ba |
children | fce0a2520551 |
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, | |
314 | 96 int snroffset, int fgain, int is_lfe) |
0 | 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++) { | |
314 | 149 if (!(is_lfe && bin == 6)) |
150 lowcomp = calc_lowcomp1(lowcomp, bndpsd[bin], bndpsd[bin+1]) ; | |
0 | 151 fastleak = bndpsd[bin] - fgain ; |
152 slowleak = bndpsd[bin] - s->sgain ; | |
153 excite[bin] = fastleak - lowcomp ; | |
314 | 154 if (!(is_lfe && bin == 6)) { |
155 if (bndpsd[bin] <= bndpsd[bin+1]) { | |
156 begin = bin + 1 ; | |
157 break ; | |
158 } | |
159 } | |
0 | 160 } |
161 | |
162 end1=bndend; | |
163 if (end1 > 22) end1=22; | |
164 | |
165 for (bin = begin; bin < end1; bin++) { | |
314 | 166 if (!(is_lfe && bin == 6)) |
167 lowcomp = calc_lowcomp(lowcomp, bndpsd[bin], bndpsd[bin+1], bin) ; | |
0 | 168 |
169 fastleak -= s->fdecay ; | |
170 v = bndpsd[bin] - fgain; | |
171 if (fastleak < v) fastleak = v; | |
172 | |
173 slowleak -= s->sdecay ; | |
174 v = bndpsd[bin] - s->sgain; | |
175 if (slowleak < v) slowleak = v; | |
176 | |
177 v=fastleak - lowcomp; | |
178 if (slowleak > v) v=slowleak; | |
179 | |
180 excite[bin] = v; | |
181 } | |
182 | |
183 for (bin = 22; bin < bndend; bin++) { | |
184 fastleak -= s->fdecay ; | |
185 v = bndpsd[bin] - fgain; | |
186 if (fastleak < v) fastleak = v; | |
187 slowleak -= s->sdecay ; | |
188 v = bndpsd[bin] - s->sgain; | |
189 if (slowleak < v) slowleak = v; | |
190 | |
191 v=fastleak; | |
192 if (slowleak > v) v = slowleak; | |
193 excite[bin] = v; | |
194 } | |
195 | |
196 /* compute masking curve */ | |
197 | |
198 for (bin = bndstrt; bin < bndend; bin++) { | |
199 v1 = excite[bin]; | |
200 tmp = s->dbknee - bndpsd[bin]; | |
201 if (tmp > 0) { | |
202 v1 += tmp >> 2; | |
203 } | |
204 v=hth[bin >> s->halfratecod][s->fscod]; | |
205 if (v1 > v) v=v1; | |
206 mask[bin] = v; | |
207 } | |
208 | |
209 /* compute bit allocation */ | |
210 | |
211 i = start ; | |
212 j = masktab[start] ; | |
213 do { | |
214 v=mask[j]; | |
215 v -= snroffset ; | |
216 v -= s->floor ; | |
217 if (v < 0) v = 0; | |
218 v &= 0x1fe0 ; | |
219 v += s->floor ; | |
220 | |
221 end1=bndtab[j] + bndsz[j]; | |
222 if (end1 > end) end1=end; | |
223 | |
224 for (k = i; k < end1; k++) { | |
225 address = (psd[i] - v) >> 5 ; | |
226 if (address < 0) address=0; | |
227 else if (address > 63) address=63; | |
228 bap[i] = baptab[address]; | |
229 i++; | |
230 } | |
231 } while (end > bndtab[j++]) ; | |
232 } | |
233 | |
234 typedef struct IComplex { | |
235 short re,im; | |
236 } IComplex; | |
237 | |
238 static void fft_init(int ln) | |
239 { | |
240 int i, j, m, n; | |
241 float alpha; | |
242 | |
243 n = 1 << ln; | |
244 | |
245 for(i=0;i<(n/2);i++) { | |
246 alpha = 2 * M_PI * (float)i / (float)n; | |
247 costab[i] = fix15(cos(alpha)); | |
248 sintab[i] = fix15(sin(alpha)); | |
249 } | |
250 | |
251 for(i=0;i<n;i++) { | |
252 m=0; | |
253 for(j=0;j<ln;j++) { | |
254 m |= ((i >> j) & 1) << (ln-j-1); | |
255 } | |
256 fft_rev[i]=m; | |
257 } | |
258 } | |
259 | |
260 /* butter fly op */ | |
261 #define BF(pre, pim, qre, qim, pre1, pim1, qre1, qim1) \ | |
262 {\ | |
263 int ax, ay, bx, by;\ | |
264 bx=pre1;\ | |
265 by=pim1;\ | |
266 ax=qre1;\ | |
267 ay=qim1;\ | |
268 pre = (bx + ax) >> 1;\ | |
269 pim = (by + ay) >> 1;\ | |
270 qre = (bx - ax) >> 1;\ | |
271 qim = (by - ay) >> 1;\ | |
272 } | |
273 | |
274 #define MUL16(a,b) ((a) * (b)) | |
275 | |
276 #define CMUL(pre, pim, are, aim, bre, bim) \ | |
277 {\ | |
278 pre = (MUL16(are, bre) - MUL16(aim, bim)) >> 15;\ | |
279 pim = (MUL16(are, bim) + MUL16(bre, aim)) >> 15;\ | |
280 } | |
281 | |
282 | |
283 /* do a 2^n point complex fft on 2^ln points. */ | |
284 static void fft(IComplex *z, int ln) | |
285 { | |
286 int j, l, np, np2; | |
287 int nblocks, nloops; | |
288 register IComplex *p,*q; | |
289 int tmp_re, tmp_im; | |
290 | |
291 np = 1 << ln; | |
292 | |
293 /* reverse */ | |
294 for(j=0;j<np;j++) { | |
295 int k; | |
296 IComplex tmp; | |
297 k = fft_rev[j]; | |
298 if (k < j) { | |
299 tmp = z[k]; | |
300 z[k] = z[j]; | |
301 z[j] = tmp; | |
302 } | |
303 } | |
304 | |
305 /* pass 0 */ | |
306 | |
307 p=&z[0]; | |
308 j=(np >> 1); | |
309 do { | |
310 BF(p[0].re, p[0].im, p[1].re, p[1].im, | |
311 p[0].re, p[0].im, p[1].re, p[1].im); | |
312 p+=2; | |
313 } while (--j != 0); | |
314 | |
315 /* pass 1 */ | |
316 | |
317 p=&z[0]; | |
318 j=np >> 2; | |
319 do { | |
320 BF(p[0].re, p[0].im, p[2].re, p[2].im, | |
321 p[0].re, p[0].im, p[2].re, p[2].im); | |
322 BF(p[1].re, p[1].im, p[3].re, p[3].im, | |
323 p[1].re, p[1].im, p[3].im, -p[3].re); | |
324 p+=4; | |
325 } while (--j != 0); | |
326 | |
327 /* pass 2 .. ln-1 */ | |
328 | |
329 nblocks = np >> 3; | |
330 nloops = 1 << 2; | |
331 np2 = np >> 1; | |
332 do { | |
333 p = z; | |
334 q = z + nloops; | |
335 for (j = 0; j < nblocks; ++j) { | |
336 | |
337 BF(p->re, p->im, q->re, q->im, | |
338 p->re, p->im, q->re, q->im); | |
339 | |
340 p++; | |
341 q++; | |
342 for(l = nblocks; l < np2; l += nblocks) { | |
343 CMUL(tmp_re, tmp_im, costab[l], -sintab[l], q->re, q->im); | |
344 BF(p->re, p->im, q->re, q->im, | |
345 p->re, p->im, tmp_re, tmp_im); | |
346 p++; | |
347 q++; | |
348 } | |
349 p += nloops; | |
350 q += nloops; | |
351 } | |
352 nblocks = nblocks >> 1; | |
353 nloops = nloops << 1; | |
354 } while (nblocks != 0); | |
355 } | |
356 | |
357 /* do a 512 point mdct */ | |
358 static void mdct512(INT32 *out, INT16 *in) | |
359 { | |
360 int i, re, im, re1, im1; | |
361 INT16 rot[N]; | |
362 IComplex x[N/4]; | |
363 | |
364 /* shift to simplify computations */ | |
365 for(i=0;i<N/4;i++) | |
366 rot[i] = -in[i + 3*N/4]; | |
367 for(i=N/4;i<N;i++) | |
368 rot[i] = in[i - N/4]; | |
369 | |
370 /* pre rotation */ | |
371 for(i=0;i<N/4;i++) { | |
372 re = ((int)rot[2*i] - (int)rot[N-1-2*i]) >> 1; | |
373 im = -((int)rot[N/2+2*i] - (int)rot[N/2-1-2*i]) >> 1; | |
374 CMUL(x[i].re, x[i].im, re, im, -xcos1[i], xsin1[i]); | |
375 } | |
376 | |
377 fft(x, MDCT_NBITS - 2); | |
378 | |
379 /* post rotation */ | |
380 for(i=0;i<N/4;i++) { | |
381 re = x[i].re; | |
382 im = x[i].im; | |
383 CMUL(re1, im1, re, im, xsin1[i], xcos1[i]); | |
384 out[2*i] = im1; | |
385 out[N/2-1-2*i] = re1; | |
386 } | |
387 } | |
388 | |
389 /* XXX: use another norm ? */ | |
390 static int calc_exp_diff(UINT8 *exp1, UINT8 *exp2, int n) | |
391 { | |
392 int sum, i; | |
393 sum = 0; | |
394 for(i=0;i<n;i++) { | |
395 sum += abs(exp1[i] - exp2[i]); | |
396 } | |
397 return sum; | |
398 } | |
399 | |
400 static void compute_exp_strategy(UINT8 exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS], | |
401 UINT8 exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2], | |
314 | 402 int ch, int is_lfe) |
0 | 403 { |
404 int i, j; | |
405 int exp_diff; | |
406 | |
407 /* estimate if the exponent variation & decide if they should be | |
408 reused in the next frame */ | |
409 exp_strategy[0][ch] = EXP_NEW; | |
410 for(i=1;i<NB_BLOCKS;i++) { | |
411 exp_diff = calc_exp_diff(exp[i][ch], exp[i-1][ch], N/2); | |
412 #ifdef DEBUG | |
413 printf("exp_diff=%d\n", exp_diff); | |
414 #endif | |
415 if (exp_diff > EXP_DIFF_THRESHOLD) | |
416 exp_strategy[i][ch] = EXP_NEW; | |
417 else | |
418 exp_strategy[i][ch] = EXP_REUSE; | |
419 } | |
314 | 420 if (is_lfe) |
421 return; | |
422 | |
0 | 423 /* now select the encoding strategy type : if exponents are often |
424 recoded, we use a coarse encoding */ | |
425 i = 0; | |
426 while (i < NB_BLOCKS) { | |
427 j = i + 1; | |
428 while (j < NB_BLOCKS && exp_strategy[j][ch] == EXP_REUSE) | |
429 j++; | |
430 switch(j - i) { | |
431 case 1: | |
432 exp_strategy[i][ch] = EXP_D45; | |
433 break; | |
434 case 2: | |
435 case 3: | |
436 exp_strategy[i][ch] = EXP_D25; | |
437 break; | |
438 default: | |
439 exp_strategy[i][ch] = EXP_D15; | |
440 break; | |
441 } | |
314 | 442 i = j; |
0 | 443 } |
444 } | |
445 | |
446 /* set exp[i] to min(exp[i], exp1[i]) */ | |
447 static void exponent_min(UINT8 exp[N/2], UINT8 exp1[N/2], int n) | |
448 { | |
449 int i; | |
450 | |
451 for(i=0;i<n;i++) { | |
452 if (exp1[i] < exp[i]) | |
453 exp[i] = exp1[i]; | |
454 } | |
455 } | |
456 | |
457 /* update the exponents so that they are the ones the decoder will | |
458 decode. Return the number of bits used to code the exponents */ | |
459 static int encode_exp(UINT8 encoded_exp[N/2], | |
460 UINT8 exp[N/2], | |
461 int nb_exps, | |
462 int exp_strategy) | |
463 { | |
464 int group_size, nb_groups, i, j, k, recurse, exp_min, delta; | |
465 UINT8 exp1[N/2]; | |
466 | |
467 switch(exp_strategy) { | |
468 case EXP_D15: | |
469 group_size = 1; | |
470 break; | |
471 case EXP_D25: | |
472 group_size = 2; | |
473 break; | |
474 default: | |
475 case EXP_D45: | |
476 group_size = 4; | |
477 break; | |
478 } | |
479 nb_groups = ((nb_exps + (group_size * 3) - 4) / (3 * group_size)) * 3; | |
480 | |
481 /* for each group, compute the minimum exponent */ | |
482 exp1[0] = exp[0]; /* DC exponent is handled separately */ | |
483 k = 1; | |
484 for(i=1;i<=nb_groups;i++) { | |
485 exp_min = exp[k]; | |
486 assert(exp_min >= 0 && exp_min <= 24); | |
487 for(j=1;j<group_size;j++) { | |
488 if (exp[k+j] < exp_min) | |
489 exp_min = exp[k+j]; | |
490 } | |
491 exp1[i] = exp_min; | |
492 k += group_size; | |
493 } | |
494 | |
495 /* constraint for DC exponent */ | |
496 if (exp1[0] > 15) | |
497 exp1[0] = 15; | |
498 | |
499 /* Iterate until the delta constraints between each groups are | |
500 satisfyed. I'm sure it is possible to find a better algorithm, | |
501 but I am lazy */ | |
502 do { | |
503 recurse = 0; | |
504 for(i=1;i<=nb_groups;i++) { | |
505 delta = exp1[i] - exp1[i-1]; | |
506 if (delta > 2) { | |
507 /* if delta too big, we encode a smaller exponent */ | |
508 exp1[i] = exp1[i-1] + 2; | |
509 } else if (delta < -2) { | |
510 /* if delta is too small, we must decrease the previous | |
511 exponent, which means we must recurse */ | |
512 recurse = 1; | |
513 exp1[i-1] = exp1[i] + 2; | |
514 } | |
515 } | |
516 } while (recurse); | |
517 | |
518 /* now we have the exponent values the decoder will see */ | |
519 encoded_exp[0] = exp1[0]; | |
520 k = 1; | |
521 for(i=1;i<=nb_groups;i++) { | |
522 for(j=0;j<group_size;j++) { | |
523 encoded_exp[k+j] = exp1[i]; | |
524 } | |
525 k += group_size; | |
526 } | |
527 | |
528 #if defined(DEBUG) | |
529 printf("exponents: strategy=%d\n", exp_strategy); | |
530 for(i=0;i<=nb_groups * group_size;i++) { | |
531 printf("%d ", encoded_exp[i]); | |
532 } | |
533 printf("\n"); | |
534 #endif | |
535 | |
536 return 4 + (nb_groups / 3) * 7; | |
537 } | |
538 | |
539 /* return the size in bits taken by the mantissa */ | |
540 int compute_mantissa_size(AC3EncodeContext *s, UINT8 *m, int nb_coefs) | |
541 { | |
542 int bits, mant, i; | |
543 | |
544 bits = 0; | |
545 for(i=0;i<nb_coefs;i++) { | |
546 mant = m[i]; | |
547 switch(mant) { | |
548 case 0: | |
549 /* nothing */ | |
550 break; | |
551 case 1: | |
552 /* 3 mantissa in 5 bits */ | |
553 if (s->mant1_cnt == 0) | |
554 bits += 5; | |
555 if (++s->mant1_cnt == 3) | |
556 s->mant1_cnt = 0; | |
557 break; | |
558 case 2: | |
559 /* 3 mantissa in 7 bits */ | |
560 if (s->mant2_cnt == 0) | |
561 bits += 7; | |
562 if (++s->mant2_cnt == 3) | |
563 s->mant2_cnt = 0; | |
564 break; | |
565 case 3: | |
566 bits += 3; | |
567 break; | |
568 case 4: | |
569 /* 2 mantissa in 7 bits */ | |
570 if (s->mant4_cnt == 0) | |
571 bits += 7; | |
572 if (++s->mant4_cnt == 2) | |
573 s->mant4_cnt = 0; | |
574 break; | |
575 case 14: | |
576 bits += 14; | |
577 break; | |
578 case 15: | |
579 bits += 16; | |
580 break; | |
581 default: | |
582 bits += mant - 1; | |
583 break; | |
584 } | |
585 } | |
586 return bits; | |
587 } | |
588 | |
589 | |
590 static int bit_alloc(AC3EncodeContext *s, | |
591 UINT8 bap[NB_BLOCKS][AC3_MAX_CHANNELS][N/2], | |
592 UINT8 encoded_exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2], | |
593 UINT8 exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS], | |
594 int frame_bits, int csnroffst, int fsnroffst) | |
595 { | |
596 int i, ch; | |
597 | |
598 /* compute size */ | |
599 for(i=0;i<NB_BLOCKS;i++) { | |
600 s->mant1_cnt = 0; | |
601 s->mant2_cnt = 0; | |
602 s->mant4_cnt = 0; | |
314 | 603 for(ch=0;ch<s->nb_all_channels;ch++) { |
25 | 604 parametric_bit_allocation(s, bap[i][ch], (INT8 *)encoded_exp[i][ch], |
0 | 605 0, s->nb_coefs[ch], |
606 (((csnroffst-15) << 4) + | |
607 fsnroffst) << 2, | |
314 | 608 fgaintab[s->fgaincod[ch]], |
609 ch == s->lfe_channel); | |
0 | 610 frame_bits += compute_mantissa_size(s, bap[i][ch], |
611 s->nb_coefs[ch]); | |
612 } | |
613 } | |
614 #if 0 | |
615 printf("csnr=%d fsnr=%d frame_bits=%d diff=%d\n", | |
616 csnroffst, fsnroffst, frame_bits, | |
617 16 * s->frame_size - ((frame_bits + 7) & ~7)); | |
618 #endif | |
619 return 16 * s->frame_size - frame_bits; | |
620 } | |
621 | |
622 #define SNR_INC1 4 | |
623 | |
624 static int compute_bit_allocation(AC3EncodeContext *s, | |
625 UINT8 bap[NB_BLOCKS][AC3_MAX_CHANNELS][N/2], | |
626 UINT8 encoded_exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2], | |
627 UINT8 exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS], | |
628 int frame_bits) | |
629 { | |
630 int i, ch; | |
631 int csnroffst, fsnroffst; | |
632 UINT8 bap1[NB_BLOCKS][AC3_MAX_CHANNELS][N/2]; | |
314 | 633 static int frame_bits_inc[8] = { 0, 0, 2, 2, 2, 4, 2, 4 }; |
0 | 634 |
635 /* init default parameters */ | |
636 s->sdecaycod = 2; | |
637 s->fdecaycod = 1; | |
638 s->sgaincod = 1; | |
639 s->dbkneecod = 2; | |
640 s->floorcod = 4; | |
314 | 641 for(ch=0;ch<s->nb_all_channels;ch++) |
0 | 642 s->fgaincod[ch] = 4; |
643 | |
644 /* compute real values */ | |
645 s->sdecay = sdecaytab[s->sdecaycod] >> s->halfratecod; | |
646 s->fdecay = fdecaytab[s->fdecaycod] >> s->halfratecod; | |
647 s->sgain = sgaintab[s->sgaincod]; | |
648 s->dbknee = dbkneetab[s->dbkneecod]; | |
649 s->floor = floortab[s->floorcod]; | |
650 | |
651 /* header size */ | |
652 frame_bits += 65; | |
314 | 653 // if (s->acmod == 2) |
654 // frame_bits += 2; | |
655 frame_bits += frame_bits_inc[s->acmod]; | |
0 | 656 |
657 /* audio blocks */ | |
658 for(i=0;i<NB_BLOCKS;i++) { | |
314 | 659 frame_bits += s->nb_channels * 2 + 2; /* blksw * c, dithflag * c, dynrnge, cplstre */ |
0 | 660 if (s->acmod == 2) |
314 | 661 frame_bits++; /* rematstr */ |
662 frame_bits += 2 * s->nb_channels; /* chexpstr[2] * c */ | |
663 if (s->lfe) | |
664 frame_bits++; /* lfeexpstr */ | |
0 | 665 for(ch=0;ch<s->nb_channels;ch++) { |
666 if (exp_strategy[i][ch] != EXP_REUSE) | |
314 | 667 frame_bits += 6 + 2; /* chbwcod[6], gainrng[2] */ |
0 | 668 } |
669 frame_bits++; /* baie */ | |
670 frame_bits++; /* snr */ | |
671 frame_bits += 2; /* delta / skip */ | |
672 } | |
673 frame_bits++; /* cplinu for block 0 */ | |
674 /* bit alloc info */ | |
314 | 675 /* sdcycod[2], fdcycod[2], sgaincod[2], dbpbcod[2], floorcod[3] */ |
676 /* csnroffset[6] */ | |
677 /* (fsnoffset[4] + fgaincod[4]) * c */ | |
678 frame_bits += 2*4 + 3 + 6 + s->nb_all_channels * (4 + 3); | |
0 | 679 |
680 /* CRC */ | |
681 frame_bits += 16; | |
682 | |
683 /* now the big work begins : do the bit allocation. Modify the snr | |
684 offset until we can pack everything in the requested frame size */ | |
685 | |
686 csnroffst = s->csnroffst; | |
687 while (csnroffst >= 0 && | |
314 | 688 bit_alloc(s, bap, encoded_exp, exp_strategy, frame_bits, csnroffst, 0) < 0) |
689 csnroffst -= SNR_INC1; | |
0 | 690 if (csnroffst < 0) { |
314 | 691 fprintf(stderr, "Yack, Error !!!\n"); |
692 return -1; | |
0 | 693 } |
694 while ((csnroffst + SNR_INC1) <= 63 && | |
695 bit_alloc(s, bap1, encoded_exp, exp_strategy, frame_bits, | |
696 csnroffst + SNR_INC1, 0) >= 0) { | |
697 csnroffst += SNR_INC1; | |
698 memcpy(bap, bap1, sizeof(bap1)); | |
699 } | |
700 while ((csnroffst + 1) <= 63 && | |
701 bit_alloc(s, bap1, encoded_exp, exp_strategy, frame_bits, csnroffst + 1, 0) >= 0) { | |
702 csnroffst++; | |
703 memcpy(bap, bap1, sizeof(bap1)); | |
704 } | |
705 | |
706 fsnroffst = 0; | |
707 while ((fsnroffst + SNR_INC1) <= 15 && | |
708 bit_alloc(s, bap1, encoded_exp, exp_strategy, frame_bits, | |
709 csnroffst, fsnroffst + SNR_INC1) >= 0) { | |
710 fsnroffst += SNR_INC1; | |
711 memcpy(bap, bap1, sizeof(bap1)); | |
712 } | |
713 while ((fsnroffst + 1) <= 15 && | |
714 bit_alloc(s, bap1, encoded_exp, exp_strategy, frame_bits, | |
715 csnroffst, fsnroffst + 1) >= 0) { | |
716 fsnroffst++; | |
717 memcpy(bap, bap1, sizeof(bap1)); | |
718 } | |
719 | |
720 s->csnroffst = csnroffst; | |
314 | 721 for(ch=0;ch<s->nb_all_channels;ch++) |
0 | 722 s->fsnroffst[ch] = fsnroffst; |
723 #if defined(DEBUG_BITALLOC) | |
724 { | |
725 int j; | |
726 | |
727 for(i=0;i<6;i++) { | |
314 | 728 for(ch=0;ch<s->nb_all_channels;ch++) { |
0 | 729 printf("Block #%d Ch%d:\n", i, ch); |
730 printf("bap="); | |
731 for(j=0;j<s->nb_coefs[ch];j++) { | |
732 printf("%d ",bap[i][ch][j]); | |
733 } | |
734 printf("\n"); | |
735 } | |
736 } | |
737 } | |
738 #endif | |
739 return 0; | |
740 } | |
741 | |
742 static int AC3_encode_init(AVCodecContext *avctx) | |
743 { | |
744 int freq = avctx->sample_rate; | |
745 int bitrate = avctx->bit_rate; | |
746 int channels = avctx->channels; | |
747 AC3EncodeContext *s = avctx->priv_data; | |
748 int i, j, k, l, ch, v; | |
749 float alpha; | |
750 static unsigned short freqs[3] = { 48000, 44100, 32000 }; | |
314 | 751 static int acmod_defs[6] = { |
752 0x01, /* C */ | |
753 0x02, /* L R */ | |
754 0x03, /* L C R */ | |
755 0x06, /* L R SL SR */ | |
756 0x07, /* L C R SL SR */ | |
757 0x07, /* L C R SL SR (+LFE) */ | |
758 }; | |
0 | 759 |
760 avctx->frame_size = AC3_FRAME_SIZE; | |
761 avctx->key_frame = 1; /* always key frame */ | |
762 | |
763 /* number of channels */ | |
314 | 764 if (channels < 1 || channels > 6) |
765 return -1; | |
766 s->acmod = acmod_defs[channels - 1]; | |
767 s->lfe = (channels == 6) ? 1 : 0; | |
768 s->nb_all_channels = channels; | |
769 s->nb_channels = channels > 5 ? 5 : channels; | |
770 s->lfe_channel = s->lfe ? 5 : -1; | |
0 | 771 |
772 /* frequency */ | |
773 for(i=0;i<3;i++) { | |
774 for(j=0;j<3;j++) | |
775 if ((freqs[j] >> i) == freq) | |
776 goto found; | |
777 } | |
778 return -1; | |
779 found: | |
780 s->sample_rate = freq; | |
781 s->halfratecod = i; | |
782 s->fscod = j; | |
783 s->bsid = 8 + s->halfratecod; | |
784 s->bsmod = 0; /* complete main audio service */ | |
785 | |
786 /* bitrate & frame size */ | |
787 bitrate /= 1000; | |
788 for(i=0;i<19;i++) { | |
789 if ((bitratetab[i] >> s->halfratecod) == bitrate) | |
790 break; | |
791 } | |
792 if (i == 19) | |
793 return -1; | |
794 s->bit_rate = bitrate; | |
795 s->frmsizecod = i << 1; | |
796 s->frame_size_min = (bitrate * 1000 * AC3_FRAME_SIZE) / (freq * 16); | |
797 /* for now we do not handle fractional sizes */ | |
798 s->frame_size = s->frame_size_min; | |
799 | |
800 /* bit allocation init */ | |
801 for(ch=0;ch<s->nb_channels;ch++) { | |
802 /* bandwidth for each channel */ | |
803 /* XXX: should compute the bandwidth according to the frame | |
804 size, so that we avoid anoying high freq artefacts */ | |
805 s->chbwcod[ch] = 50; /* sample bandwidth as mpeg audio layer 2 table 0 */ | |
806 s->nb_coefs[ch] = ((s->chbwcod[ch] + 12) * 3) + 37; | |
807 } | |
314 | 808 if (s->lfe) { |
809 s->nb_coefs[s->lfe_channel] = 7; /* fixed */ | |
810 } | |
0 | 811 /* initial snr offset */ |
812 s->csnroffst = 40; | |
813 | |
814 /* compute bndtab and masktab from bandsz */ | |
815 k = 0; | |
816 l = 0; | |
817 for(i=0;i<50;i++) { | |
818 bndtab[i] = l; | |
819 v = bndsz[i]; | |
820 for(j=0;j<v;j++) masktab[k++]=i; | |
821 l += v; | |
822 } | |
823 bndtab[50] = 0; | |
824 | |
825 /* mdct init */ | |
826 fft_init(MDCT_NBITS - 2); | |
827 for(i=0;i<N/4;i++) { | |
828 alpha = 2 * M_PI * (i + 1.0 / 8.0) / (float)N; | |
829 xcos1[i] = fix15(-cos(alpha)); | |
830 xsin1[i] = fix15(-sin(alpha)); | |
831 } | |
832 | |
833 ac3_crc_init(); | |
834 | |
835 return 0; | |
836 } | |
837 | |
838 /* output the AC3 frame header */ | |
839 static void output_frame_header(AC3EncodeContext *s, unsigned char *frame) | |
840 { | |
841 init_put_bits(&s->pb, frame, AC3_MAX_CODED_FRAME_SIZE, NULL, NULL); | |
842 | |
843 put_bits(&s->pb, 16, 0x0b77); /* frame header */ | |
844 put_bits(&s->pb, 16, 0); /* crc1: will be filled later */ | |
845 put_bits(&s->pb, 2, s->fscod); | |
846 put_bits(&s->pb, 6, s->frmsizecod + (s->frame_size - s->frame_size_min)); | |
847 put_bits(&s->pb, 5, s->bsid); | |
848 put_bits(&s->pb, 3, s->bsmod); | |
849 put_bits(&s->pb, 3, s->acmod); | |
314 | 850 if ((s->acmod & 0x01) && s->acmod != 0x01) |
851 put_bits(&s->pb, 2, 1); /* XXX -4.5 dB */ | |
852 if (s->acmod & 0x04) | |
853 put_bits(&s->pb, 2, 1); /* XXX -6 dB */ | |
854 if (s->acmod == 0x02) | |
0 | 855 put_bits(&s->pb, 2, 0); /* surround not indicated */ |
314 | 856 put_bits(&s->pb, 1, s->lfe); /* LFE */ |
0 | 857 put_bits(&s->pb, 5, 31); /* dialog norm: -31 db */ |
858 put_bits(&s->pb, 1, 0); /* no compression control word */ | |
859 put_bits(&s->pb, 1, 0); /* no lang code */ | |
860 put_bits(&s->pb, 1, 0); /* no audio production info */ | |
861 put_bits(&s->pb, 1, 0); /* no copyright */ | |
862 put_bits(&s->pb, 1, 1); /* original bitstream */ | |
863 put_bits(&s->pb, 1, 0); /* no time code 1 */ | |
864 put_bits(&s->pb, 1, 0); /* no time code 2 */ | |
865 put_bits(&s->pb, 1, 0); /* no addtional bit stream info */ | |
866 } | |
867 | |
868 /* symetric quantization on 'levels' levels */ | |
869 static inline int sym_quant(int c, int e, int levels) | |
870 { | |
871 int v; | |
872 | |
873 if (c >= 0) { | |
87 | 874 v = (levels * (c << e)) >> 24; |
875 v = (v + 1) >> 1; | |
0 | 876 v = (levels >> 1) + v; |
877 } else { | |
87 | 878 v = (levels * ((-c) << e)) >> 24; |
879 v = (v + 1) >> 1; | |
0 | 880 v = (levels >> 1) - v; |
881 } | |
882 assert (v >= 0 && v < levels); | |
883 return v; | |
884 } | |
885 | |
886 /* asymetric quantization on 2^qbits levels */ | |
887 static inline int asym_quant(int c, int e, int qbits) | |
888 { | |
889 int lshift, m, v; | |
890 | |
891 lshift = e + qbits - 24; | |
892 if (lshift >= 0) | |
893 v = c << lshift; | |
894 else | |
895 v = c >> (-lshift); | |
896 /* rounding */ | |
897 v = (v + 1) >> 1; | |
898 m = (1 << (qbits-1)); | |
899 if (v >= m) | |
900 v = m - 1; | |
901 assert(v >= -m); | |
902 return v & ((1 << qbits)-1); | |
903 } | |
904 | |
905 /* Output one audio block. There are NB_BLOCKS audio blocks in one AC3 | |
906 frame */ | |
907 static void output_audio_block(AC3EncodeContext *s, | |
908 UINT8 exp_strategy[AC3_MAX_CHANNELS], | |
909 UINT8 encoded_exp[AC3_MAX_CHANNELS][N/2], | |
910 UINT8 bap[AC3_MAX_CHANNELS][N/2], | |
911 INT32 mdct_coefs[AC3_MAX_CHANNELS][N/2], | |
912 INT8 global_exp[AC3_MAX_CHANNELS], | |
913 int block_num) | |
914 { | |
915 int ch, nb_groups, group_size, i, baie; | |
916 UINT8 *p; | |
917 UINT16 qmant[AC3_MAX_CHANNELS][N/2]; | |
918 int exp0, exp1; | |
919 int mant1_cnt, mant2_cnt, mant4_cnt; | |
920 UINT16 *qmant1_ptr, *qmant2_ptr, *qmant4_ptr; | |
921 int delta0, delta1, delta2; | |
922 | |
923 for(ch=0;ch<s->nb_channels;ch++) | |
924 put_bits(&s->pb, 1, 0); /* 512 point MDCT */ | |
925 for(ch=0;ch<s->nb_channels;ch++) | |
926 put_bits(&s->pb, 1, 1); /* no dither */ | |
927 put_bits(&s->pb, 1, 0); /* no dynamic range */ | |
928 if (block_num == 0) { | |
929 /* for block 0, even if no coupling, we must say it. This is a | |
930 waste of bit :-) */ | |
931 put_bits(&s->pb, 1, 1); /* coupling strategy present */ | |
932 put_bits(&s->pb, 1, 0); /* no coupling strategy */ | |
933 } else { | |
934 put_bits(&s->pb, 1, 0); /* no new coupling strategy */ | |
935 } | |
936 | |
937 if (s->acmod == 2) { | |
938 put_bits(&s->pb, 1, 0); /* no matrixing (but should be used in the future) */ | |
939 } | |
940 | |
941 #if defined(DEBUG) | |
942 { | |
943 static int count = 0; | |
944 printf("Block #%d (%d)\n", block_num, count++); | |
945 } | |
946 #endif | |
947 /* exponent strategy */ | |
948 for(ch=0;ch<s->nb_channels;ch++) { | |
949 put_bits(&s->pb, 2, exp_strategy[ch]); | |
950 } | |
951 | |
314 | 952 if (s->lfe) { |
953 put_bits(&s->pb, 1, exp_strategy[s->lfe_channel]); | |
954 } | |
955 | |
0 | 956 for(ch=0;ch<s->nb_channels;ch++) { |
957 if (exp_strategy[ch] != EXP_REUSE) | |
958 put_bits(&s->pb, 6, s->chbwcod[ch]); | |
959 } | |
960 | |
961 /* exponents */ | |
314 | 962 for (ch = 0; ch < s->nb_all_channels; ch++) { |
0 | 963 switch(exp_strategy[ch]) { |
964 case EXP_REUSE: | |
965 continue; | |
966 case EXP_D15: | |
967 group_size = 1; | |
968 break; | |
969 case EXP_D25: | |
970 group_size = 2; | |
971 break; | |
972 default: | |
973 case EXP_D45: | |
974 group_size = 4; | |
975 break; | |
976 } | |
314 | 977 nb_groups = (s->nb_coefs[ch] + (group_size * 3) - 4) / (3 * group_size); |
0 | 978 p = encoded_exp[ch]; |
979 | |
980 /* first exponent */ | |
981 exp1 = *p++; | |
982 put_bits(&s->pb, 4, exp1); | |
983 | |
984 /* next ones are delta encoded */ | |
985 for(i=0;i<nb_groups;i++) { | |
986 /* merge three delta in one code */ | |
987 exp0 = exp1; | |
988 exp1 = p[0]; | |
989 p += group_size; | |
990 delta0 = exp1 - exp0 + 2; | |
991 | |
992 exp0 = exp1; | |
993 exp1 = p[0]; | |
994 p += group_size; | |
995 delta1 = exp1 - exp0 + 2; | |
996 | |
997 exp0 = exp1; | |
998 exp1 = p[0]; | |
999 p += group_size; | |
1000 delta2 = exp1 - exp0 + 2; | |
1001 | |
1002 put_bits(&s->pb, 7, ((delta0 * 5 + delta1) * 5) + delta2); | |
1003 } | |
1004 | |
314 | 1005 if (ch != s->lfe_channel) |
1006 put_bits(&s->pb, 2, 0); /* no gain range info */ | |
0 | 1007 } |
1008 | |
1009 /* bit allocation info */ | |
1010 baie = (block_num == 0); | |
1011 put_bits(&s->pb, 1, baie); | |
1012 if (baie) { | |
1013 put_bits(&s->pb, 2, s->sdecaycod); | |
1014 put_bits(&s->pb, 2, s->fdecaycod); | |
1015 put_bits(&s->pb, 2, s->sgaincod); | |
1016 put_bits(&s->pb, 2, s->dbkneecod); | |
1017 put_bits(&s->pb, 3, s->floorcod); | |
1018 } | |
1019 | |
1020 /* snr offset */ | |
1021 put_bits(&s->pb, 1, baie); /* always present with bai */ | |
1022 if (baie) { | |
1023 put_bits(&s->pb, 6, s->csnroffst); | |
314 | 1024 for(ch=0;ch<s->nb_all_channels;ch++) { |
0 | 1025 put_bits(&s->pb, 4, s->fsnroffst[ch]); |
1026 put_bits(&s->pb, 3, s->fgaincod[ch]); | |
1027 } | |
1028 } | |
1029 | |
1030 put_bits(&s->pb, 1, 0); /* no delta bit allocation */ | |
1031 put_bits(&s->pb, 1, 0); /* no data to skip */ | |
1032 | |
1033 /* mantissa encoding : we use two passes to handle the grouping. A | |
1034 one pass method may be faster, but it would necessitate to | |
1035 modify the output stream. */ | |
1036 | |
1037 /* first pass: quantize */ | |
1038 mant1_cnt = mant2_cnt = mant4_cnt = 0; | |
1039 qmant1_ptr = qmant2_ptr = qmant4_ptr = NULL; | |
1040 | |
314 | 1041 for (ch = 0; ch < s->nb_all_channels; ch++) { |
0 | 1042 int b, c, e, v; |
1043 | |
1044 for(i=0;i<s->nb_coefs[ch];i++) { | |
1045 c = mdct_coefs[ch][i]; | |
1046 e = encoded_exp[ch][i] - global_exp[ch]; | |
1047 b = bap[ch][i]; | |
1048 switch(b) { | |
1049 case 0: | |
1050 v = 0; | |
1051 break; | |
1052 case 1: | |
1053 v = sym_quant(c, e, 3); | |
1054 switch(mant1_cnt) { | |
1055 case 0: | |
1056 qmant1_ptr = &qmant[ch][i]; | |
1057 v = 9 * v; | |
1058 mant1_cnt = 1; | |
1059 break; | |
1060 case 1: | |
1061 *qmant1_ptr += 3 * v; | |
1062 mant1_cnt = 2; | |
1063 v = 128; | |
1064 break; | |
1065 default: | |
1066 *qmant1_ptr += v; | |
1067 mant1_cnt = 0; | |
1068 v = 128; | |
1069 break; | |
1070 } | |
1071 break; | |
1072 case 2: | |
1073 v = sym_quant(c, e, 5); | |
1074 switch(mant2_cnt) { | |
1075 case 0: | |
1076 qmant2_ptr = &qmant[ch][i]; | |
1077 v = 25 * v; | |
1078 mant2_cnt = 1; | |
1079 break; | |
1080 case 1: | |
1081 *qmant2_ptr += 5 * v; | |
1082 mant2_cnt = 2; | |
1083 v = 128; | |
1084 break; | |
1085 default: | |
1086 *qmant2_ptr += v; | |
1087 mant2_cnt = 0; | |
1088 v = 128; | |
1089 break; | |
1090 } | |
1091 break; | |
1092 case 3: | |
1093 v = sym_quant(c, e, 7); | |
1094 break; | |
1095 case 4: | |
1096 v = sym_quant(c, e, 11); | |
1097 switch(mant4_cnt) { | |
1098 case 0: | |
1099 qmant4_ptr = &qmant[ch][i]; | |
1100 v = 11 * v; | |
1101 mant4_cnt = 1; | |
1102 break; | |
1103 default: | |
1104 *qmant4_ptr += v; | |
1105 mant4_cnt = 0; | |
1106 v = 128; | |
1107 break; | |
1108 } | |
1109 break; | |
1110 case 5: | |
1111 v = sym_quant(c, e, 15); | |
1112 break; | |
1113 case 14: | |
1114 v = asym_quant(c, e, 14); | |
1115 break; | |
1116 case 15: | |
1117 v = asym_quant(c, e, 16); | |
1118 break; | |
1119 default: | |
1120 v = asym_quant(c, e, b - 1); | |
1121 break; | |
1122 } | |
1123 qmant[ch][i] = v; | |
1124 } | |
1125 } | |
1126 | |
1127 /* second pass : output the values */ | |
314 | 1128 for (ch = 0; ch < s->nb_all_channels; ch++) { |
0 | 1129 int b, q; |
1130 | |
1131 for(i=0;i<s->nb_coefs[ch];i++) { | |
1132 q = qmant[ch][i]; | |
1133 b = bap[ch][i]; | |
1134 switch(b) { | |
1135 case 0: | |
1136 break; | |
1137 case 1: | |
1138 if (q != 128) | |
1139 put_bits(&s->pb, 5, q); | |
1140 break; | |
1141 case 2: | |
1142 if (q != 128) | |
1143 put_bits(&s->pb, 7, q); | |
1144 break; | |
1145 case 3: | |
1146 put_bits(&s->pb, 3, q); | |
1147 break; | |
1148 case 4: | |
1149 if (q != 128) | |
1150 put_bits(&s->pb, 7, q); | |
1151 break; | |
1152 case 14: | |
1153 put_bits(&s->pb, 14, q); | |
1154 break; | |
1155 case 15: | |
1156 put_bits(&s->pb, 16, q); | |
1157 break; | |
1158 default: | |
1159 put_bits(&s->pb, b - 1, q); | |
1160 break; | |
1161 } | |
1162 } | |
1163 } | |
1164 } | |
1165 | |
1166 /* compute the ac3 crc */ | |
1167 | |
1168 #define CRC16_POLY ((1 << 0) | (1 << 2) | (1 << 15) | (1 << 16)) | |
1169 | |
1170 static void ac3_crc_init(void) | |
1171 { | |
1172 unsigned int c, n, k; | |
1173 | |
1174 for(n=0;n<256;n++) { | |
1175 c = n << 8; | |
1176 for (k = 0; k < 8; k++) { | |
1177 if (c & (1 << 15)) | |
1178 c = ((c << 1) & 0xffff) ^ (CRC16_POLY & 0xffff); | |
1179 else | |
1180 c = c << 1; | |
1181 } | |
1182 crc_table[n] = c; | |
1183 } | |
1184 } | |
1185 | |
1186 static unsigned int ac3_crc(UINT8 *data, int n, unsigned int crc) | |
1187 { | |
1188 int i; | |
1189 for(i=0;i<n;i++) { | |
1190 crc = (crc_table[data[i] ^ (crc >> 8)] ^ (crc << 8)) & 0xffff; | |
1191 } | |
1192 return crc; | |
1193 } | |
1194 | |
1195 static unsigned int mul_poly(unsigned int a, unsigned int b, unsigned int poly) | |
1196 { | |
1197 unsigned int c; | |
1198 | |
1199 c = 0; | |
1200 while (a) { | |
1201 if (a & 1) | |
1202 c ^= b; | |
1203 a = a >> 1; | |
1204 b = b << 1; | |
1205 if (b & (1 << 16)) | |
1206 b ^= poly; | |
1207 } | |
1208 return c; | |
1209 } | |
1210 | |
1211 static unsigned int pow_poly(unsigned int a, unsigned int n, unsigned int poly) | |
1212 { | |
1213 unsigned int r; | |
1214 r = 1; | |
1215 while (n) { | |
1216 if (n & 1) | |
1217 r = mul_poly(r, a, poly); | |
1218 a = mul_poly(a, a, poly); | |
1219 n >>= 1; | |
1220 } | |
1221 return r; | |
1222 } | |
1223 | |
1224 | |
1225 /* compute log2(max(abs(tab[]))) */ | |
1226 static int log2_tab(INT16 *tab, int n) | |
1227 { | |
1228 int i, v; | |
1229 | |
1230 v = 0; | |
1231 for(i=0;i<n;i++) { | |
1232 v |= abs(tab[i]); | |
1233 } | |
65 | 1234 return av_log2(v); |
0 | 1235 } |
1236 | |
1237 static void lshift_tab(INT16 *tab, int n, int lshift) | |
1238 { | |
1239 int i; | |
1240 | |
1241 if (lshift > 0) { | |
1242 for(i=0;i<n;i++) { | |
1243 tab[i] <<= lshift; | |
1244 } | |
1245 } else if (lshift < 0) { | |
1246 lshift = -lshift; | |
1247 for(i=0;i<n;i++) { | |
1248 tab[i] >>= lshift; | |
1249 } | |
1250 } | |
1251 } | |
1252 | |
1253 /* fill the end of the frame and compute the two crcs */ | |
1254 static int output_frame_end(AC3EncodeContext *s) | |
1255 { | |
1256 int frame_size, frame_size_58, n, crc1, crc2, crc_inv; | |
1257 UINT8 *frame; | |
1258 | |
1259 frame_size = s->frame_size; /* frame size in words */ | |
1260 /* align to 8 bits */ | |
1261 flush_put_bits(&s->pb); | |
1262 /* add zero bytes to reach the frame size */ | |
1263 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
|
1264 n = 2 * s->frame_size - (pbBufPtr(&s->pb) - frame) - 2; |
0 | 1265 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
|
1266 memset(pbBufPtr(&s->pb), 0, n); |
0 | 1267 |
1268 /* Now we must compute both crcs : this is not so easy for crc1 | |
1269 because it is at the beginning of the data... */ | |
1270 frame_size_58 = (frame_size >> 1) + (frame_size >> 3); | |
1271 crc1 = ac3_crc(frame + 4, (2 * frame_size_58) - 4, 0); | |
1272 /* XXX: could precompute crc_inv */ | |
1273 crc_inv = pow_poly((CRC16_POLY >> 1), (16 * frame_size_58) - 16, CRC16_POLY); | |
1274 crc1 = mul_poly(crc_inv, crc1, CRC16_POLY); | |
1275 frame[2] = crc1 >> 8; | |
1276 frame[3] = crc1; | |
1277 | |
1278 crc2 = ac3_crc(frame + 2 * frame_size_58, (frame_size - frame_size_58) * 2 - 2, 0); | |
1279 frame[2*frame_size - 2] = crc2 >> 8; | |
1280 frame[2*frame_size - 1] = crc2; | |
1281 | |
1282 // printf("n=%d frame_size=%d\n", n, frame_size); | |
1283 return frame_size * 2; | |
1284 } | |
1285 | |
1286 int AC3_encode_frame(AVCodecContext *avctx, | |
1287 unsigned char *frame, int buf_size, void *data) | |
1288 { | |
1289 AC3EncodeContext *s = avctx->priv_data; | |
1290 short *samples = data; | |
1291 int i, j, k, v, ch; | |
1292 INT16 input_samples[N]; | |
1293 INT32 mdct_coef[NB_BLOCKS][AC3_MAX_CHANNELS][N/2]; | |
1294 UINT8 exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2]; | |
1295 UINT8 exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS]; | |
1296 UINT8 encoded_exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2]; | |
1297 UINT8 bap[NB_BLOCKS][AC3_MAX_CHANNELS][N/2]; | |
1298 INT8 exp_samples[NB_BLOCKS][AC3_MAX_CHANNELS]; | |
1299 int frame_bits; | |
1300 | |
1301 frame_bits = 0; | |
314 | 1302 for(ch=0;ch<s->nb_all_channels;ch++) { |
0 | 1303 /* fixed mdct to the six sub blocks & exponent computation */ |
1304 for(i=0;i<NB_BLOCKS;i++) { | |
1305 INT16 *sptr; | |
1306 int sinc; | |
1307 | |
1308 /* compute input samples */ | |
1309 memcpy(input_samples, s->last_samples[ch], N/2 * sizeof(INT16)); | |
314 | 1310 sinc = s->nb_all_channels; |
0 | 1311 sptr = samples + (sinc * (N/2) * i) + ch; |
1312 for(j=0;j<N/2;j++) { | |
1313 v = *sptr; | |
1314 input_samples[j + N/2] = v; | |
1315 s->last_samples[ch][j] = v; | |
1316 sptr += sinc; | |
1317 } | |
1318 | |
1319 /* apply the MDCT window */ | |
1320 for(j=0;j<N/2;j++) { | |
1321 input_samples[j] = MUL16(input_samples[j], | |
1322 ac3_window[j]) >> 15; | |
1323 input_samples[N-j-1] = MUL16(input_samples[N-j-1], | |
1324 ac3_window[j]) >> 15; | |
1325 } | |
1326 | |
1327 /* Normalize the samples to use the maximum available | |
1328 precision */ | |
1329 v = 14 - log2_tab(input_samples, N); | |
1330 if (v < 0) | |
1331 v = 0; | |
1332 exp_samples[i][ch] = v - 8; | |
1333 lshift_tab(input_samples, N, v); | |
1334 | |
1335 /* do the MDCT */ | |
1336 mdct512(mdct_coef[i][ch], input_samples); | |
1337 | |
1338 /* compute "exponents". We take into account the | |
1339 normalization there */ | |
1340 for(j=0;j<N/2;j++) { | |
1341 int e; | |
1342 v = abs(mdct_coef[i][ch][j]); | |
1343 if (v == 0) | |
1344 e = 24; | |
1345 else { | |
65 | 1346 e = 23 - av_log2(v) + exp_samples[i][ch]; |
0 | 1347 if (e >= 24) { |
1348 e = 24; | |
1349 mdct_coef[i][ch][j] = 0; | |
1350 } | |
1351 } | |
1352 exp[i][ch][j] = e; | |
1353 } | |
1354 } | |
1355 | |
314 | 1356 compute_exp_strategy(exp_strategy, exp, ch, ch == s->lfe_channel); |
0 | 1357 |
1358 /* compute the exponents as the decoder will see them. The | |
1359 EXP_REUSE case must be handled carefully : we select the | |
1360 min of the exponents */ | |
1361 i = 0; | |
1362 while (i < NB_BLOCKS) { | |
1363 j = i + 1; | |
1364 while (j < NB_BLOCKS && exp_strategy[j][ch] == EXP_REUSE) { | |
1365 exponent_min(exp[i][ch], exp[j][ch], s->nb_coefs[ch]); | |
1366 j++; | |
1367 } | |
1368 frame_bits += encode_exp(encoded_exp[i][ch], | |
1369 exp[i][ch], s->nb_coefs[ch], | |
1370 exp_strategy[i][ch]); | |
1371 /* copy encoded exponents for reuse case */ | |
1372 for(k=i+1;k<j;k++) { | |
1373 memcpy(encoded_exp[k][ch], encoded_exp[i][ch], | |
1374 s->nb_coefs[ch] * sizeof(UINT8)); | |
1375 } | |
1376 i = j; | |
1377 } | |
1378 } | |
1379 | |
1380 compute_bit_allocation(s, bap, encoded_exp, exp_strategy, frame_bits); | |
1381 /* everything is known... let's output the frame */ | |
1382 output_frame_header(s, frame); | |
1383 | |
1384 for(i=0;i<NB_BLOCKS;i++) { | |
1385 output_audio_block(s, exp_strategy[i], encoded_exp[i], | |
1386 bap[i], mdct_coef[i], exp_samples[i], i); | |
1387 } | |
1388 return output_frame_end(s); | |
1389 } | |
1390 | |
1391 #if 0 | |
1392 /*************************************************************************/ | |
1393 /* TEST */ | |
1394 | |
1395 #define FN (N/4) | |
1396 | |
1397 void fft_test(void) | |
1398 { | |
1399 IComplex in[FN], in1[FN]; | |
1400 int k, n, i; | |
1401 float sum_re, sum_im, a; | |
1402 | |
1403 /* FFT test */ | |
1404 | |
1405 for(i=0;i<FN;i++) { | |
1406 in[i].re = random() % 65535 - 32767; | |
1407 in[i].im = random() % 65535 - 32767; | |
1408 in1[i] = in[i]; | |
1409 } | |
1410 fft(in, 7); | |
1411 | |
1412 /* do it by hand */ | |
1413 for(k=0;k<FN;k++) { | |
1414 sum_re = 0; | |
1415 sum_im = 0; | |
1416 for(n=0;n<FN;n++) { | |
1417 a = -2 * M_PI * (n * k) / FN; | |
1418 sum_re += in1[n].re * cos(a) - in1[n].im * sin(a); | |
1419 sum_im += in1[n].re * sin(a) + in1[n].im * cos(a); | |
1420 } | |
1421 printf("%3d: %6d,%6d %6.0f,%6.0f\n", | |
1422 k, in[k].re, in[k].im, sum_re / FN, sum_im / FN); | |
1423 } | |
1424 } | |
1425 | |
1426 void mdct_test(void) | |
1427 { | |
1428 INT16 input[N]; | |
1429 INT32 output[N/2]; | |
1430 float input1[N]; | |
1431 float output1[N/2]; | |
1432 float s, a, err, e, emax; | |
1433 int i, k, n; | |
1434 | |
1435 for(i=0;i<N;i++) { | |
1436 input[i] = (random() % 65535 - 32767) * 9 / 10; | |
1437 input1[i] = input[i]; | |
1438 } | |
1439 | |
1440 mdct512(output, input); | |
1441 | |
1442 /* do it by hand */ | |
1443 for(k=0;k<N/2;k++) { | |
1444 s = 0; | |
1445 for(n=0;n<N;n++) { | |
1446 a = (2*M_PI*(2*n+1+N/2)*(2*k+1) / (4 * N)); | |
1447 s += input1[n] * cos(a); | |
1448 } | |
1449 output1[k] = -2 * s / N; | |
1450 } | |
1451 | |
1452 err = 0; | |
1453 emax = 0; | |
1454 for(i=0;i<N/2;i++) { | |
1455 printf("%3d: %7d %7.0f\n", i, output[i], output1[i]); | |
1456 e = output[i] - output1[i]; | |
1457 if (e > emax) | |
1458 emax = e; | |
1459 err += e * e; | |
1460 } | |
1461 printf("err2=%f emax=%f\n", err / (N/2), emax); | |
1462 } | |
1463 | |
1464 void test_ac3(void) | |
1465 { | |
1466 AC3EncodeContext ctx; | |
1467 unsigned char frame[AC3_MAX_CODED_FRAME_SIZE]; | |
1468 short samples[AC3_FRAME_SIZE]; | |
1469 int ret, i; | |
1470 | |
1471 AC3_encode_init(&ctx, 44100, 64000, 1); | |
1472 | |
1473 fft_test(); | |
1474 mdct_test(); | |
1475 | |
1476 for(i=0;i<AC3_FRAME_SIZE;i++) | |
1477 samples[i] = (int)(sin(2*M_PI*i*1000.0/44100) * 10000); | |
1478 ret = AC3_encode_frame(&ctx, frame, samples); | |
1479 printf("ret=%d\n", ret); | |
1480 } | |
1481 #endif | |
1482 | |
1483 AVCodec ac3_encoder = { | |
1484 "ac3", | |
1485 CODEC_TYPE_AUDIO, | |
1486 CODEC_ID_AC3, | |
1487 sizeof(AC3EncodeContext), | |
1488 AC3_encode_init, | |
1489 AC3_encode_frame, | |
1490 NULL, | |
1491 }; |