Mercurial > libavcodec.hg
annotate ac3enc.c @ 3637:1c32151e7a61 libavcodec
make mpa_decode_header() 10 times faster
author | michael |
---|---|
date | Sat, 26 Aug 2006 18:02:52 +0000 |
parents | 894ef0dba687 |
children | 34d76180e5d0 |
rev | line source |
---|---|
0 | 1 /* |
2 * The simplest AC3 encoder | |
429 | 3 * Copyright (c) 2000 Fabrice Bellard. |
0 | 4 * |
429 | 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. | |
0 | 9 * |
429 | 10 * This library is distributed in the hope that it will be useful, |
0 | 11 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
429 | 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
13 * Lesser General Public License for more details. | |
0 | 14 * |
429 | 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 | |
3036
0b546eab515d
Update licensing information: The FSF changed postal address.
diego
parents:
2979
diff
changeset
|
17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
0 | 18 */ |
1106 | 19 |
20 /** | |
21 * @file ac3enc.c | |
22 * The simplest AC3 encoder. | |
23 */ | |
64 | 24 //#define DEBUG |
25 //#define DEBUG_BITALLOC | |
26 #include "avcodec.h" | |
2398
582e635cfa08
common.c -> bitstream.c (and the single non bitstream func -> utils.c)
michael
parents:
2367
diff
changeset
|
27 #include "bitstream.h" |
3170 | 28 #include "crc.h" |
782 | 29 #include "ac3.h" |
30 | |
31 typedef struct AC3EncodeContext { | |
32 PutBitContext pb; | |
33 int nb_channels; | |
34 int nb_all_channels; | |
35 int lfe_channel; | |
36 int bit_rate; | |
1057 | 37 unsigned int sample_rate; |
38 unsigned int bsid; | |
39 unsigned int frame_size_min; /* minimum frame size in case rounding is necessary */ | |
40 unsigned int frame_size; /* current frame size in words */ | |
3246 | 41 unsigned int bits_written; |
42 unsigned int samples_written; | |
782 | 43 int halfratecod; |
1057 | 44 unsigned int frmsizecod; |
45 unsigned int fscod; /* frequency */ | |
46 unsigned int acmod; | |
782 | 47 int lfe; |
1057 | 48 unsigned int bsmod; |
782 | 49 short last_samples[AC3_MAX_CHANNELS][256]; |
1057 | 50 unsigned int chbwcod[AC3_MAX_CHANNELS]; |
782 | 51 int nb_coefs[AC3_MAX_CHANNELS]; |
2967 | 52 |
782 | 53 /* bitrate allocation control */ |
2967 | 54 int sgaincod, sdecaycod, fdecaycod, dbkneecod, floorcod; |
782 | 55 AC3BitAllocParameters bit_alloc; |
56 int csnroffst; | |
57 int fgaincod[AC3_MAX_CHANNELS]; | |
58 int fsnroffst[AC3_MAX_CHANNELS]; | |
59 /* mantissa encoding */ | |
60 int mant1_cnt, mant2_cnt, mant4_cnt; | |
61 } AC3EncodeContext; | |
62 | |
0 | 63 #include "ac3tab.h" |
64 | |
65 #define MDCT_NBITS 9 | |
66 #define N (1 << MDCT_NBITS) | |
67 | |
68 /* new exponents are sent if their Norm 1 exceed this number */ | |
69 #define EXP_DIFF_THRESHOLD 1000 | |
70 | |
71 static void fft_init(int ln); | |
72 | |
1064 | 73 static inline int16_t fix15(float a) |
0 | 74 { |
75 int v; | |
76 v = (int)(a * (float)(1 << 15)); | |
77 if (v < -32767) | |
78 v = -32767; | |
2967 | 79 else if (v > 32767) |
0 | 80 v = 32767; |
81 return v; | |
82 } | |
83 | |
84 static inline int calc_lowcomp1(int a, int b0, int b1) | |
85 { | |
86 if ((b0 + 256) == b1) { | |
87 a = 384 ; | |
2967 | 88 } else if (b0 > b1) { |
0 | 89 a = a - 64; |
90 if (a < 0) a=0; | |
91 } | |
92 return a; | |
93 } | |
94 | |
95 static inline int calc_lowcomp(int a, int b0, int b1, int bin) | |
96 { | |
97 if (bin < 7) { | |
98 if ((b0 + 256) == b1) { | |
99 a = 384 ; | |
2967 | 100 } else if (b0 > b1) { |
0 | 101 a = a - 64; |
102 if (a < 0) a=0; | |
103 } | |
104 } else if (bin < 20) { | |
105 if ((b0 + 256) == b1) { | |
106 a = 320 ; | |
107 } else if (b0 > b1) { | |
108 a= a - 64; | |
109 if (a < 0) a=0; | |
110 } | |
111 } else { | |
112 a = a - 128; | |
113 if (a < 0) a=0; | |
114 } | |
115 return a; | |
116 } | |
117 | |
118 /* AC3 bit allocation. The algorithm is the one described in the AC3 | |
782 | 119 spec. */ |
1064 | 120 void ac3_parametric_bit_allocation(AC3BitAllocParameters *s, uint8_t *bap, |
121 int8_t *exp, int start, int end, | |
782 | 122 int snroffset, int fgain, int is_lfe, |
2967 | 123 int deltbae,int deltnseg, |
1064 | 124 uint8_t *deltoffst, uint8_t *deltlen, uint8_t *deltba) |
0 | 125 { |
126 int bin,i,j,k,end1,v,v1,bndstrt,bndend,lowcomp,begin; | |
127 int fastleak,slowleak,address,tmp; | |
1064 | 128 int16_t psd[256]; /* scaled exponents */ |
129 int16_t bndpsd[50]; /* interpolated exponents */ | |
130 int16_t excite[50]; /* excitation */ | |
131 int16_t mask[50]; /* masking value */ | |
0 | 132 |
133 /* exponent mapping to PSD */ | |
134 for(bin=start;bin<end;bin++) { | |
135 psd[bin]=(3072 - (exp[bin] << 7)); | |
136 } | |
137 | |
138 /* PSD integration */ | |
139 j=start; | |
140 k=masktab[start]; | |
141 do { | |
142 v=psd[j]; | |
143 j++; | |
144 end1=bndtab[k+1]; | |
145 if (end1 > end) end1=end; | |
146 for(i=j;i<end1;i++) { | |
147 int c,adr; | |
148 /* logadd */ | |
149 v1=psd[j]; | |
150 c=v-v1; | |
151 if (c >= 0) { | |
152 adr=c >> 1; | |
153 if (adr > 255) adr=255; | |
154 v=v + latab[adr]; | |
155 } else { | |
156 adr=(-c) >> 1; | |
157 if (adr > 255) adr=255; | |
158 v=v1 + latab[adr]; | |
159 } | |
160 j++; | |
161 } | |
162 bndpsd[k]=v; | |
163 k++; | |
164 } while (end > bndtab[k]); | |
165 | |
166 /* excitation function */ | |
167 bndstrt = masktab[start]; | |
168 bndend = masktab[end-1] + 1; | |
2967 | 169 |
782 | 170 if (bndstrt == 0) { |
171 lowcomp = 0; | |
172 lowcomp = calc_lowcomp1(lowcomp, bndpsd[0], bndpsd[1]) ; | |
173 excite[0] = bndpsd[0] - fgain - lowcomp ; | |
174 lowcomp = calc_lowcomp1(lowcomp, bndpsd[1], bndpsd[2]) ; | |
175 excite[1] = bndpsd[1] - fgain - lowcomp ; | |
176 begin = 7 ; | |
177 for (bin = 2; bin < 7; bin++) { | |
178 if (!(is_lfe && bin == 6)) | |
179 lowcomp = calc_lowcomp1(lowcomp, bndpsd[bin], bndpsd[bin+1]) ; | |
180 fastleak = bndpsd[bin] - fgain ; | |
181 slowleak = bndpsd[bin] - s->sgain ; | |
182 excite[bin] = fastleak - lowcomp ; | |
183 if (!(is_lfe && bin == 6)) { | |
184 if (bndpsd[bin] <= bndpsd[bin+1]) { | |
185 begin = bin + 1 ; | |
186 break ; | |
187 } | |
188 } | |
189 } | |
2967 | 190 |
782 | 191 end1=bndend; |
192 if (end1 > 22) end1=22; | |
2967 | 193 |
782 | 194 for (bin = begin; bin < end1; bin++) { |
195 if (!(is_lfe && bin == 6)) | |
196 lowcomp = calc_lowcomp(lowcomp, bndpsd[bin], bndpsd[bin+1], bin) ; | |
2967 | 197 |
782 | 198 fastleak -= s->fdecay ; |
199 v = bndpsd[bin] - fgain; | |
200 if (fastleak < v) fastleak = v; | |
2967 | 201 |
782 | 202 slowleak -= s->sdecay ; |
203 v = bndpsd[bin] - s->sgain; | |
204 if (slowleak < v) slowleak = v; | |
2967 | 205 |
782 | 206 v=fastleak - lowcomp; |
207 if (slowleak > v) v=slowleak; | |
2967 | 208 |
782 | 209 excite[bin] = v; |
210 } | |
211 begin = 22; | |
212 } else { | |
213 /* coupling channel */ | |
214 begin = bndstrt; | |
2967 | 215 |
782 | 216 fastleak = (s->cplfleak << 8) + 768; |
217 slowleak = (s->cplsleak << 8) + 768; | |
0 | 218 } |
219 | |
782 | 220 for (bin = begin; bin < bndend; bin++) { |
0 | 221 fastleak -= s->fdecay ; |
222 v = bndpsd[bin] - fgain; | |
223 if (fastleak < v) fastleak = v; | |
224 slowleak -= s->sdecay ; | |
225 v = bndpsd[bin] - s->sgain; | |
226 if (slowleak < v) slowleak = v; | |
227 | |
228 v=fastleak; | |
229 if (slowleak > v) v = slowleak; | |
230 excite[bin] = v; | |
231 } | |
232 | |
233 /* compute masking curve */ | |
234 | |
235 for (bin = bndstrt; bin < bndend; bin++) { | |
236 v1 = excite[bin]; | |
237 tmp = s->dbknee - bndpsd[bin]; | |
238 if (tmp > 0) { | |
239 v1 += tmp >> 2; | |
240 } | |
241 v=hth[bin >> s->halfratecod][s->fscod]; | |
242 if (v1 > v) v=v1; | |
243 mask[bin] = v; | |
244 } | |
245 | |
782 | 246 /* delta bit allocation */ |
247 | |
248 if (deltbae == 0 || deltbae == 1) { | |
249 int band, seg, delta; | |
250 band = 0 ; | |
251 for (seg = 0; seg < deltnseg; seg++) { | |
252 band += deltoffst[seg] ; | |
253 if (deltba[seg] >= 4) { | |
254 delta = (deltba[seg] - 3) << 7; | |
255 } else { | |
256 delta = (deltba[seg] - 4) << 7; | |
257 } | |
258 for (k = 0; k < deltlen[seg]; k++) { | |
259 mask[band] += delta ; | |
260 band++ ; | |
261 } | |
262 } | |
263 } | |
264 | |
0 | 265 /* compute bit allocation */ |
2967 | 266 |
0 | 267 i = start ; |
268 j = masktab[start] ; | |
269 do { | |
270 v=mask[j]; | |
271 v -= snroffset ; | |
272 v -= s->floor ; | |
273 if (v < 0) v = 0; | |
274 v &= 0x1fe0 ; | |
275 v += s->floor ; | |
276 | |
277 end1=bndtab[j] + bndsz[j]; | |
278 if (end1 > end) end1=end; | |
279 | |
280 for (k = i; k < end1; k++) { | |
281 address = (psd[i] - v) >> 5 ; | |
282 if (address < 0) address=0; | |
283 else if (address > 63) address=63; | |
284 bap[i] = baptab[address]; | |
285 i++; | |
286 } | |
287 } while (end > bndtab[j++]) ; | |
288 } | |
289 | |
290 typedef struct IComplex { | |
291 short re,im; | |
292 } IComplex; | |
293 | |
294 static void fft_init(int ln) | |
295 { | |
296 int i, j, m, n; | |
297 float alpha; | |
298 | |
299 n = 1 << ln; | |
300 | |
301 for(i=0;i<(n/2);i++) { | |
302 alpha = 2 * M_PI * (float)i / (float)n; | |
303 costab[i] = fix15(cos(alpha)); | |
304 sintab[i] = fix15(sin(alpha)); | |
305 } | |
306 | |
307 for(i=0;i<n;i++) { | |
308 m=0; | |
309 for(j=0;j<ln;j++) { | |
310 m |= ((i >> j) & 1) << (ln-j-1); | |
311 } | |
312 fft_rev[i]=m; | |
313 } | |
314 } | |
315 | |
316 /* butter fly op */ | |
317 #define BF(pre, pim, qre, qim, pre1, pim1, qre1, qim1) \ | |
318 {\ | |
319 int ax, ay, bx, by;\ | |
320 bx=pre1;\ | |
321 by=pim1;\ | |
322 ax=qre1;\ | |
323 ay=qim1;\ | |
324 pre = (bx + ax) >> 1;\ | |
325 pim = (by + ay) >> 1;\ | |
326 qre = (bx - ax) >> 1;\ | |
327 qim = (by - ay) >> 1;\ | |
328 } | |
329 | |
330 #define MUL16(a,b) ((a) * (b)) | |
331 | |
332 #define CMUL(pre, pim, are, aim, bre, bim) \ | |
333 {\ | |
334 pre = (MUL16(are, bre) - MUL16(aim, bim)) >> 15;\ | |
335 pim = (MUL16(are, bim) + MUL16(bre, aim)) >> 15;\ | |
336 } | |
337 | |
338 | |
339 /* do a 2^n point complex fft on 2^ln points. */ | |
340 static void fft(IComplex *z, int ln) | |
341 { | |
2979 | 342 int j, l, np, np2; |
343 int nblocks, nloops; | |
0 | 344 register IComplex *p,*q; |
345 int tmp_re, tmp_im; | |
346 | |
347 np = 1 << ln; | |
348 | |
349 /* reverse */ | |
350 for(j=0;j<np;j++) { | |
351 int k; | |
352 IComplex tmp; | |
353 k = fft_rev[j]; | |
354 if (k < j) { | |
355 tmp = z[k]; | |
356 z[k] = z[j]; | |
357 z[j] = tmp; | |
358 } | |
359 } | |
360 | |
361 /* pass 0 */ | |
362 | |
363 p=&z[0]; | |
364 j=(np >> 1); | |
365 do { | |
2967 | 366 BF(p[0].re, p[0].im, p[1].re, p[1].im, |
0 | 367 p[0].re, p[0].im, p[1].re, p[1].im); |
368 p+=2; | |
369 } while (--j != 0); | |
370 | |
371 /* pass 1 */ | |
372 | |
373 p=&z[0]; | |
374 j=np >> 2; | |
375 do { | |
2967 | 376 BF(p[0].re, p[0].im, p[2].re, p[2].im, |
0 | 377 p[0].re, p[0].im, p[2].re, p[2].im); |
2967 | 378 BF(p[1].re, p[1].im, p[3].re, p[3].im, |
0 | 379 p[1].re, p[1].im, p[3].im, -p[3].re); |
380 p+=4; | |
381 } while (--j != 0); | |
382 | |
383 /* pass 2 .. ln-1 */ | |
384 | |
385 nblocks = np >> 3; | |
386 nloops = 1 << 2; | |
387 np2 = np >> 1; | |
388 do { | |
389 p = z; | |
390 q = z + nloops; | |
391 for (j = 0; j < nblocks; ++j) { | |
392 | |
393 BF(p->re, p->im, q->re, q->im, | |
394 p->re, p->im, q->re, q->im); | |
2967 | 395 |
0 | 396 p++; |
397 q++; | |
398 for(l = nblocks; l < np2; l += nblocks) { | |
399 CMUL(tmp_re, tmp_im, costab[l], -sintab[l], q->re, q->im); | |
400 BF(p->re, p->im, q->re, q->im, | |
401 p->re, p->im, tmp_re, tmp_im); | |
402 p++; | |
403 q++; | |
404 } | |
405 p += nloops; | |
406 q += nloops; | |
407 } | |
408 nblocks = nblocks >> 1; | |
409 nloops = nloops << 1; | |
410 } while (nblocks != 0); | |
411 } | |
412 | |
413 /* do a 512 point mdct */ | |
1064 | 414 static void mdct512(int32_t *out, int16_t *in) |
0 | 415 { |
416 int i, re, im, re1, im1; | |
2967 | 417 int16_t rot[N]; |
0 | 418 IComplex x[N/4]; |
419 | |
420 /* shift to simplify computations */ | |
421 for(i=0;i<N/4;i++) | |
422 rot[i] = -in[i + 3*N/4]; | |
423 for(i=N/4;i<N;i++) | |
424 rot[i] = in[i - N/4]; | |
2967 | 425 |
0 | 426 /* pre rotation */ |
427 for(i=0;i<N/4;i++) { | |
428 re = ((int)rot[2*i] - (int)rot[N-1-2*i]) >> 1; | |
429 im = -((int)rot[N/2+2*i] - (int)rot[N/2-1-2*i]) >> 1; | |
430 CMUL(x[i].re, x[i].im, re, im, -xcos1[i], xsin1[i]); | |
431 } | |
432 | |
433 fft(x, MDCT_NBITS - 2); | |
2967 | 434 |
0 | 435 /* post rotation */ |
436 for(i=0;i<N/4;i++) { | |
437 re = x[i].re; | |
438 im = x[i].im; | |
439 CMUL(re1, im1, re, im, xsin1[i], xcos1[i]); | |
440 out[2*i] = im1; | |
441 out[N/2-1-2*i] = re1; | |
442 } | |
443 } | |
444 | |
445 /* XXX: use another norm ? */ | |
1064 | 446 static int calc_exp_diff(uint8_t *exp1, uint8_t *exp2, int n) |
0 | 447 { |
448 int sum, i; | |
449 sum = 0; | |
450 for(i=0;i<n;i++) { | |
451 sum += abs(exp1[i] - exp2[i]); | |
452 } | |
453 return sum; | |
454 } | |
455 | |
1064 | 456 static void compute_exp_strategy(uint8_t exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS], |
457 uint8_t exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2], | |
314 | 458 int ch, int is_lfe) |
0 | 459 { |
460 int i, j; | |
461 int exp_diff; | |
2967 | 462 |
0 | 463 /* estimate if the exponent variation & decide if they should be |
464 reused in the next frame */ | |
465 exp_strategy[0][ch] = EXP_NEW; | |
466 for(i=1;i<NB_BLOCKS;i++) { | |
467 exp_diff = calc_exp_diff(exp[i][ch], exp[i-1][ch], N/2); | |
2967 | 468 #ifdef DEBUG |
1602
fdb8244da1e5
av_log patch(2 of ?) by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1598
diff
changeset
|
469 av_log(NULL, AV_LOG_DEBUG, "exp_diff=%d\n", exp_diff); |
0 | 470 #endif |
471 if (exp_diff > EXP_DIFF_THRESHOLD) | |
472 exp_strategy[i][ch] = EXP_NEW; | |
473 else | |
474 exp_strategy[i][ch] = EXP_REUSE; | |
475 } | |
314 | 476 if (is_lfe) |
2979 | 477 return; |
314 | 478 |
0 | 479 /* now select the encoding strategy type : if exponents are often |
480 recoded, we use a coarse encoding */ | |
481 i = 0; | |
482 while (i < NB_BLOCKS) { | |
483 j = i + 1; | |
484 while (j < NB_BLOCKS && exp_strategy[j][ch] == EXP_REUSE) | |
485 j++; | |
486 switch(j - i) { | |
487 case 1: | |
488 exp_strategy[i][ch] = EXP_D45; | |
489 break; | |
490 case 2: | |
491 case 3: | |
492 exp_strategy[i][ch] = EXP_D25; | |
493 break; | |
494 default: | |
495 exp_strategy[i][ch] = EXP_D15; | |
496 break; | |
497 } | |
2979 | 498 i = j; |
0 | 499 } |
500 } | |
501 | |
502 /* set exp[i] to min(exp[i], exp1[i]) */ | |
1064 | 503 static void exponent_min(uint8_t exp[N/2], uint8_t exp1[N/2], int n) |
0 | 504 { |
505 int i; | |
506 | |
507 for(i=0;i<n;i++) { | |
508 if (exp1[i] < exp[i]) | |
509 exp[i] = exp1[i]; | |
510 } | |
511 } | |
2967 | 512 |
0 | 513 /* update the exponents so that they are the ones the decoder will |
514 decode. Return the number of bits used to code the exponents */ | |
2967 | 515 static int encode_exp(uint8_t encoded_exp[N/2], |
516 uint8_t exp[N/2], | |
0 | 517 int nb_exps, |
518 int exp_strategy) | |
519 { | |
2157
4478e603a8e3
simpler delta decreasing algorithm patch by (Jeff Muizelaar <jrmuizel at student dot cs dot uwaterloo dot ca>)
michael
parents:
1819
diff
changeset
|
520 int group_size, nb_groups, i, j, k, exp_min; |
1064 | 521 uint8_t exp1[N/2]; |
0 | 522 |
523 switch(exp_strategy) { | |
524 case EXP_D15: | |
525 group_size = 1; | |
526 break; | |
527 case EXP_D25: | |
528 group_size = 2; | |
529 break; | |
530 default: | |
531 case EXP_D45: | |
532 group_size = 4; | |
533 break; | |
534 } | |
535 nb_groups = ((nb_exps + (group_size * 3) - 4) / (3 * group_size)) * 3; | |
536 | |
537 /* for each group, compute the minimum exponent */ | |
538 exp1[0] = exp[0]; /* DC exponent is handled separately */ | |
539 k = 1; | |
540 for(i=1;i<=nb_groups;i++) { | |
541 exp_min = exp[k]; | |
542 assert(exp_min >= 0 && exp_min <= 24); | |
543 for(j=1;j<group_size;j++) { | |
544 if (exp[k+j] < exp_min) | |
545 exp_min = exp[k+j]; | |
546 } | |
547 exp1[i] = exp_min; | |
548 k += group_size; | |
549 } | |
550 | |
551 /* constraint for DC exponent */ | |
552 if (exp1[0] > 15) | |
553 exp1[0] = 15; | |
554 | |
2157
4478e603a8e3
simpler delta decreasing algorithm patch by (Jeff Muizelaar <jrmuizel at student dot cs dot uwaterloo dot ca>)
michael
parents:
1819
diff
changeset
|
555 /* Decrease the delta between each groups to within 2 |
4478e603a8e3
simpler delta decreasing algorithm patch by (Jeff Muizelaar <jrmuizel at student dot cs dot uwaterloo dot ca>)
michael
parents:
1819
diff
changeset
|
556 * so that they can be differentially encoded */ |
4478e603a8e3
simpler delta decreasing algorithm patch by (Jeff Muizelaar <jrmuizel at student dot cs dot uwaterloo dot ca>)
michael
parents:
1819
diff
changeset
|
557 for (i=1;i<=nb_groups;i++) |
2979 | 558 exp1[i] = FFMIN(exp1[i], exp1[i-1] + 2); |
2157
4478e603a8e3
simpler delta decreasing algorithm patch by (Jeff Muizelaar <jrmuizel at student dot cs dot uwaterloo dot ca>)
michael
parents:
1819
diff
changeset
|
559 for (i=nb_groups-1;i>=0;i--) |
2979 | 560 exp1[i] = FFMIN(exp1[i], exp1[i+1] + 2); |
2157
4478e603a8e3
simpler delta decreasing algorithm patch by (Jeff Muizelaar <jrmuizel at student dot cs dot uwaterloo dot ca>)
michael
parents:
1819
diff
changeset
|
561 |
0 | 562 /* now we have the exponent values the decoder will see */ |
563 encoded_exp[0] = exp1[0]; | |
564 k = 1; | |
565 for(i=1;i<=nb_groups;i++) { | |
566 for(j=0;j<group_size;j++) { | |
567 encoded_exp[k+j] = exp1[i]; | |
568 } | |
569 k += group_size; | |
570 } | |
2967 | 571 |
0 | 572 #if defined(DEBUG) |
1602
fdb8244da1e5
av_log patch(2 of ?) by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1598
diff
changeset
|
573 av_log(NULL, AV_LOG_DEBUG, "exponents: strategy=%d\n", exp_strategy); |
0 | 574 for(i=0;i<=nb_groups * group_size;i++) { |
1602
fdb8244da1e5
av_log patch(2 of ?) by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1598
diff
changeset
|
575 av_log(NULL, AV_LOG_DEBUG, "%d ", encoded_exp[i]); |
0 | 576 } |
1602
fdb8244da1e5
av_log patch(2 of ?) by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1598
diff
changeset
|
577 av_log(NULL, AV_LOG_DEBUG, "\n"); |
0 | 578 #endif |
579 | |
580 return 4 + (nb_groups / 3) * 7; | |
581 } | |
582 | |
583 /* return the size in bits taken by the mantissa */ | |
1064 | 584 static int compute_mantissa_size(AC3EncodeContext *s, uint8_t *m, int nb_coefs) |
0 | 585 { |
586 int bits, mant, i; | |
587 | |
588 bits = 0; | |
589 for(i=0;i<nb_coefs;i++) { | |
590 mant = m[i]; | |
591 switch(mant) { | |
592 case 0: | |
593 /* nothing */ | |
594 break; | |
595 case 1: | |
596 /* 3 mantissa in 5 bits */ | |
2967 | 597 if (s->mant1_cnt == 0) |
0 | 598 bits += 5; |
599 if (++s->mant1_cnt == 3) | |
600 s->mant1_cnt = 0; | |
601 break; | |
602 case 2: | |
603 /* 3 mantissa in 7 bits */ | |
2967 | 604 if (s->mant2_cnt == 0) |
0 | 605 bits += 7; |
606 if (++s->mant2_cnt == 3) | |
607 s->mant2_cnt = 0; | |
608 break; | |
609 case 3: | |
610 bits += 3; | |
611 break; | |
612 case 4: | |
613 /* 2 mantissa in 7 bits */ | |
614 if (s->mant4_cnt == 0) | |
615 bits += 7; | |
2967 | 616 if (++s->mant4_cnt == 2) |
0 | 617 s->mant4_cnt = 0; |
618 break; | |
619 case 14: | |
620 bits += 14; | |
621 break; | |
622 case 15: | |
623 bits += 16; | |
624 break; | |
625 default: | |
626 bits += mant - 1; | |
627 break; | |
628 } | |
629 } | |
630 return bits; | |
631 } | |
632 | |
633 | |
634 static int bit_alloc(AC3EncodeContext *s, | |
1064 | 635 uint8_t bap[NB_BLOCKS][AC3_MAX_CHANNELS][N/2], |
636 uint8_t encoded_exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2], | |
637 uint8_t exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS], | |
0 | 638 int frame_bits, int csnroffst, int fsnroffst) |
639 { | |
640 int i, ch; | |
641 | |
642 /* compute size */ | |
643 for(i=0;i<NB_BLOCKS;i++) { | |
644 s->mant1_cnt = 0; | |
645 s->mant2_cnt = 0; | |
646 s->mant4_cnt = 0; | |
314 | 647 for(ch=0;ch<s->nb_all_channels;ch++) { |
2967 | 648 ac3_parametric_bit_allocation(&s->bit_alloc, |
649 bap[i][ch], (int8_t *)encoded_exp[i][ch], | |
650 0, s->nb_coefs[ch], | |
651 (((csnroffst-15) << 4) + | |
652 fsnroffst) << 2, | |
782 | 653 fgaintab[s->fgaincod[ch]], |
654 ch == s->lfe_channel, | |
655 2, 0, NULL, NULL, NULL); | |
2967 | 656 frame_bits += compute_mantissa_size(s, bap[i][ch], |
0 | 657 s->nb_coefs[ch]); |
658 } | |
659 } | |
660 #if 0 | |
2967 | 661 printf("csnr=%d fsnr=%d frame_bits=%d diff=%d\n", |
662 csnroffst, fsnroffst, frame_bits, | |
0 | 663 16 * s->frame_size - ((frame_bits + 7) & ~7)); |
664 #endif | |
665 return 16 * s->frame_size - frame_bits; | |
666 } | |
667 | |
668 #define SNR_INC1 4 | |
669 | |
670 static int compute_bit_allocation(AC3EncodeContext *s, | |
1064 | 671 uint8_t bap[NB_BLOCKS][AC3_MAX_CHANNELS][N/2], |
672 uint8_t encoded_exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2], | |
673 uint8_t exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS], | |
0 | 674 int frame_bits) |
675 { | |
676 int i, ch; | |
677 int csnroffst, fsnroffst; | |
1064 | 678 uint8_t bap1[NB_BLOCKS][AC3_MAX_CHANNELS][N/2]; |
314 | 679 static int frame_bits_inc[8] = { 0, 0, 2, 2, 2, 4, 2, 4 }; |
0 | 680 |
681 /* init default parameters */ | |
682 s->sdecaycod = 2; | |
683 s->fdecaycod = 1; | |
684 s->sgaincod = 1; | |
685 s->dbkneecod = 2; | |
686 s->floorcod = 4; | |
2967 | 687 for(ch=0;ch<s->nb_all_channels;ch++) |
0 | 688 s->fgaincod[ch] = 4; |
2967 | 689 |
0 | 690 /* compute real values */ |
782 | 691 s->bit_alloc.fscod = s->fscod; |
692 s->bit_alloc.halfratecod = s->halfratecod; | |
693 s->bit_alloc.sdecay = sdecaytab[s->sdecaycod] >> s->halfratecod; | |
694 s->bit_alloc.fdecay = fdecaytab[s->fdecaycod] >> s->halfratecod; | |
695 s->bit_alloc.sgain = sgaintab[s->sgaincod]; | |
696 s->bit_alloc.dbknee = dbkneetab[s->dbkneecod]; | |
697 s->bit_alloc.floor = floortab[s->floorcod]; | |
2967 | 698 |
0 | 699 /* header size */ |
700 frame_bits += 65; | |
314 | 701 // if (s->acmod == 2) |
702 // frame_bits += 2; | |
703 frame_bits += frame_bits_inc[s->acmod]; | |
0 | 704 |
705 /* audio blocks */ | |
706 for(i=0;i<NB_BLOCKS;i++) { | |
314 | 707 frame_bits += s->nb_channels * 2 + 2; /* blksw * c, dithflag * c, dynrnge, cplstre */ |
2644
6ff5dc0dbaf0
While adding stereo rematrixing, I came across something that needs to
michael
parents:
2398
diff
changeset
|
708 if (s->acmod == 2) { |
314 | 709 frame_bits++; /* rematstr */ |
2644
6ff5dc0dbaf0
While adding stereo rematrixing, I came across something that needs to
michael
parents:
2398
diff
changeset
|
710 if(i==0) frame_bits += 4; |
6ff5dc0dbaf0
While adding stereo rematrixing, I came across something that needs to
michael
parents:
2398
diff
changeset
|
711 } |
314 | 712 frame_bits += 2 * s->nb_channels; /* chexpstr[2] * c */ |
2979 | 713 if (s->lfe) |
714 frame_bits++; /* lfeexpstr */ | |
0 | 715 for(ch=0;ch<s->nb_channels;ch++) { |
716 if (exp_strategy[i][ch] != EXP_REUSE) | |
314 | 717 frame_bits += 6 + 2; /* chbwcod[6], gainrng[2] */ |
0 | 718 } |
719 frame_bits++; /* baie */ | |
720 frame_bits++; /* snr */ | |
721 frame_bits += 2; /* delta / skip */ | |
722 } | |
723 frame_bits++; /* cplinu for block 0 */ | |
724 /* bit alloc info */ | |
314 | 725 /* sdcycod[2], fdcycod[2], sgaincod[2], dbpbcod[2], floorcod[3] */ |
726 /* csnroffset[6] */ | |
727 /* (fsnoffset[4] + fgaincod[4]) * c */ | |
728 frame_bits += 2*4 + 3 + 6 + s->nb_all_channels * (4 + 3); | |
0 | 729 |
1819
34cdcb221665
auxdatae, crcrs fix by (Jean-Francois Panisset <panisset at comcast dot net>)
michael
parents:
1602
diff
changeset
|
730 /* auxdatae, crcrsv */ |
34cdcb221665
auxdatae, crcrs fix by (Jean-Francois Panisset <panisset at comcast dot net>)
michael
parents:
1602
diff
changeset
|
731 frame_bits += 2; |
34cdcb221665
auxdatae, crcrs fix by (Jean-Francois Panisset <panisset at comcast dot net>)
michael
parents:
1602
diff
changeset
|
732 |
0 | 733 /* CRC */ |
734 frame_bits += 16; | |
735 | |
736 /* now the big work begins : do the bit allocation. Modify the snr | |
737 offset until we can pack everything in the requested frame size */ | |
738 | |
739 csnroffst = s->csnroffst; | |
2967 | 740 while (csnroffst >= 0 && |
2979 | 741 bit_alloc(s, bap, encoded_exp, exp_strategy, frame_bits, csnroffst, 0) < 0) |
742 csnroffst -= SNR_INC1; | |
0 | 743 if (csnroffst < 0) { |
3221 | 744 av_log(NULL, AV_LOG_ERROR, "Bit allocation failed, try increasing the bitrate, -ab 384 for example!\n"); |
2979 | 745 return -1; |
0 | 746 } |
2967 | 747 while ((csnroffst + SNR_INC1) <= 63 && |
748 bit_alloc(s, bap1, encoded_exp, exp_strategy, frame_bits, | |
0 | 749 csnroffst + SNR_INC1, 0) >= 0) { |
750 csnroffst += SNR_INC1; | |
751 memcpy(bap, bap1, sizeof(bap1)); | |
752 } | |
2967 | 753 while ((csnroffst + 1) <= 63 && |
0 | 754 bit_alloc(s, bap1, encoded_exp, exp_strategy, frame_bits, csnroffst + 1, 0) >= 0) { |
755 csnroffst++; | |
756 memcpy(bap, bap1, sizeof(bap1)); | |
757 } | |
758 | |
759 fsnroffst = 0; | |
2967 | 760 while ((fsnroffst + SNR_INC1) <= 15 && |
761 bit_alloc(s, bap1, encoded_exp, exp_strategy, frame_bits, | |
0 | 762 csnroffst, fsnroffst + SNR_INC1) >= 0) { |
763 fsnroffst += SNR_INC1; | |
764 memcpy(bap, bap1, sizeof(bap1)); | |
765 } | |
2967 | 766 while ((fsnroffst + 1) <= 15 && |
767 bit_alloc(s, bap1, encoded_exp, exp_strategy, frame_bits, | |
0 | 768 csnroffst, fsnroffst + 1) >= 0) { |
769 fsnroffst++; | |
770 memcpy(bap, bap1, sizeof(bap1)); | |
771 } | |
2967 | 772 |
0 | 773 s->csnroffst = csnroffst; |
314 | 774 for(ch=0;ch<s->nb_all_channels;ch++) |
0 | 775 s->fsnroffst[ch] = fsnroffst; |
776 #if defined(DEBUG_BITALLOC) | |
777 { | |
778 int j; | |
779 | |
780 for(i=0;i<6;i++) { | |
314 | 781 for(ch=0;ch<s->nb_all_channels;ch++) { |
0 | 782 printf("Block #%d Ch%d:\n", i, ch); |
783 printf("bap="); | |
784 for(j=0;j<s->nb_coefs[ch];j++) { | |
785 printf("%d ",bap[i][ch][j]); | |
786 } | |
787 printf("\n"); | |
788 } | |
789 } | |
790 } | |
791 #endif | |
792 return 0; | |
793 } | |
794 | |
782 | 795 void ac3_common_init(void) |
796 { | |
797 int i, j, k, l, v; | |
798 /* compute bndtab and masktab from bandsz */ | |
799 k = 0; | |
800 l = 0; | |
801 for(i=0;i<50;i++) { | |
802 bndtab[i] = l; | |
803 v = bndsz[i]; | |
804 for(j=0;j<v;j++) masktab[k++]=i; | |
805 l += v; | |
806 } | |
807 bndtab[50] = 0; | |
808 } | |
809 | |
810 | |
0 | 811 static int AC3_encode_init(AVCodecContext *avctx) |
812 { | |
813 int freq = avctx->sample_rate; | |
814 int bitrate = avctx->bit_rate; | |
815 int channels = avctx->channels; | |
816 AC3EncodeContext *s = avctx->priv_data; | |
782 | 817 int i, j, ch; |
0 | 818 float alpha; |
1064 | 819 static const uint8_t acmod_defs[6] = { |
2979 | 820 0x01, /* C */ |
821 0x02, /* L R */ | |
822 0x03, /* L C R */ | |
823 0x06, /* L R SL SR */ | |
824 0x07, /* L C R SL SR */ | |
825 0x07, /* L C R SL SR (+LFE) */ | |
314 | 826 }; |
0 | 827 |
828 avctx->frame_size = AC3_FRAME_SIZE; | |
2967 | 829 |
0 | 830 /* number of channels */ |
314 | 831 if (channels < 1 || channels > 6) |
2979 | 832 return -1; |
314 | 833 s->acmod = acmod_defs[channels - 1]; |
834 s->lfe = (channels == 6) ? 1 : 0; | |
835 s->nb_all_channels = channels; | |
836 s->nb_channels = channels > 5 ? 5 : channels; | |
837 s->lfe_channel = s->lfe ? 5 : -1; | |
0 | 838 |
839 /* frequency */ | |
840 for(i=0;i<3;i++) { | |
2967 | 841 for(j=0;j<3;j++) |
782 | 842 if ((ac3_freqs[j] >> i) == freq) |
0 | 843 goto found; |
844 } | |
845 return -1; | |
2967 | 846 found: |
0 | 847 s->sample_rate = freq; |
848 s->halfratecod = i; | |
849 s->fscod = j; | |
850 s->bsid = 8 + s->halfratecod; | |
851 s->bsmod = 0; /* complete main audio service */ | |
852 | |
853 /* bitrate & frame size */ | |
854 bitrate /= 1000; | |
855 for(i=0;i<19;i++) { | |
782 | 856 if ((ac3_bitratetab[i] >> s->halfratecod) == bitrate) |
0 | 857 break; |
858 } | |
859 if (i == 19) | |
860 return -1; | |
861 s->bit_rate = bitrate; | |
862 s->frmsizecod = i << 1; | |
863 s->frame_size_min = (bitrate * 1000 * AC3_FRAME_SIZE) / (freq * 16); | |
3246 | 864 s->bits_written = 0; |
865 s->samples_written = 0; | |
0 | 866 s->frame_size = s->frame_size_min; |
2967 | 867 |
0 | 868 /* bit allocation init */ |
869 for(ch=0;ch<s->nb_channels;ch++) { | |
870 /* bandwidth for each channel */ | |
871 /* XXX: should compute the bandwidth according to the frame | |
872 size, so that we avoid anoying high freq artefacts */ | |
873 s->chbwcod[ch] = 50; /* sample bandwidth as mpeg audio layer 2 table 0 */ | |
874 s->nb_coefs[ch] = ((s->chbwcod[ch] + 12) * 3) + 37; | |
875 } | |
314 | 876 if (s->lfe) { |
2979 | 877 s->nb_coefs[s->lfe_channel] = 7; /* fixed */ |
314 | 878 } |
0 | 879 /* initial snr offset */ |
880 s->csnroffst = 40; | |
881 | |
782 | 882 ac3_common_init(); |
0 | 883 |
884 /* mdct init */ | |
885 fft_init(MDCT_NBITS - 2); | |
886 for(i=0;i<N/4;i++) { | |
887 alpha = 2 * M_PI * (i + 1.0 / 8.0) / (float)N; | |
888 xcos1[i] = fix15(-cos(alpha)); | |
889 xsin1[i] = fix15(-sin(alpha)); | |
890 } | |
891 | |
925 | 892 avctx->coded_frame= avcodec_alloc_frame(); |
893 avctx->coded_frame->key_frame= 1; | |
0 | 894 |
895 return 0; | |
896 } | |
897 | |
898 /* output the AC3 frame header */ | |
899 static void output_frame_header(AC3EncodeContext *s, unsigned char *frame) | |
900 { | |
1522
79dddc5cd990
removed the obsolete and unused parameters of init_put_bits
alex
parents:
1408
diff
changeset
|
901 init_put_bits(&s->pb, frame, AC3_MAX_CODED_FRAME_SIZE); |
0 | 902 |
903 put_bits(&s->pb, 16, 0x0b77); /* frame header */ | |
904 put_bits(&s->pb, 16, 0); /* crc1: will be filled later */ | |
905 put_bits(&s->pb, 2, s->fscod); | |
906 put_bits(&s->pb, 6, s->frmsizecod + (s->frame_size - s->frame_size_min)); | |
907 put_bits(&s->pb, 5, s->bsid); | |
908 put_bits(&s->pb, 3, s->bsmod); | |
909 put_bits(&s->pb, 3, s->acmod); | |
314 | 910 if ((s->acmod & 0x01) && s->acmod != 0x01) |
2979 | 911 put_bits(&s->pb, 2, 1); /* XXX -4.5 dB */ |
314 | 912 if (s->acmod & 0x04) |
2979 | 913 put_bits(&s->pb, 2, 1); /* XXX -6 dB */ |
314 | 914 if (s->acmod == 0x02) |
0 | 915 put_bits(&s->pb, 2, 0); /* surround not indicated */ |
314 | 916 put_bits(&s->pb, 1, s->lfe); /* LFE */ |
0 | 917 put_bits(&s->pb, 5, 31); /* dialog norm: -31 db */ |
918 put_bits(&s->pb, 1, 0); /* no compression control word */ | |
919 put_bits(&s->pb, 1, 0); /* no lang code */ | |
920 put_bits(&s->pb, 1, 0); /* no audio production info */ | |
921 put_bits(&s->pb, 1, 0); /* no copyright */ | |
922 put_bits(&s->pb, 1, 1); /* original bitstream */ | |
923 put_bits(&s->pb, 1, 0); /* no time code 1 */ | |
924 put_bits(&s->pb, 1, 0); /* no time code 2 */ | |
925 put_bits(&s->pb, 1, 0); /* no addtional bit stream info */ | |
926 } | |
927 | |
928 /* symetric quantization on 'levels' levels */ | |
929 static inline int sym_quant(int c, int e, int levels) | |
930 { | |
931 int v; | |
932 | |
933 if (c >= 0) { | |
87 | 934 v = (levels * (c << e)) >> 24; |
935 v = (v + 1) >> 1; | |
0 | 936 v = (levels >> 1) + v; |
937 } else { | |
87 | 938 v = (levels * ((-c) << e)) >> 24; |
939 v = (v + 1) >> 1; | |
0 | 940 v = (levels >> 1) - v; |
941 } | |
942 assert (v >= 0 && v < levels); | |
943 return v; | |
944 } | |
945 | |
946 /* asymetric quantization on 2^qbits levels */ | |
947 static inline int asym_quant(int c, int e, int qbits) | |
948 { | |
949 int lshift, m, v; | |
950 | |
951 lshift = e + qbits - 24; | |
952 if (lshift >= 0) | |
953 v = c << lshift; | |
954 else | |
955 v = c >> (-lshift); | |
956 /* rounding */ | |
957 v = (v + 1) >> 1; | |
958 m = (1 << (qbits-1)); | |
959 if (v >= m) | |
960 v = m - 1; | |
961 assert(v >= -m); | |
962 return v & ((1 << qbits)-1); | |
963 } | |
964 | |
965 /* Output one audio block. There are NB_BLOCKS audio blocks in one AC3 | |
966 frame */ | |
967 static void output_audio_block(AC3EncodeContext *s, | |
1064 | 968 uint8_t exp_strategy[AC3_MAX_CHANNELS], |
969 uint8_t encoded_exp[AC3_MAX_CHANNELS][N/2], | |
970 uint8_t bap[AC3_MAX_CHANNELS][N/2], | |
971 int32_t mdct_coefs[AC3_MAX_CHANNELS][N/2], | |
972 int8_t global_exp[AC3_MAX_CHANNELS], | |
0 | 973 int block_num) |
974 { | |
1408
4d67eb341a0c
AC3 encoding patch ba (Ross Martin <ffmpeg at ross dot interwrx dot com>)
michaelni
parents:
1106
diff
changeset
|
975 int ch, nb_groups, group_size, i, baie, rbnd; |
1064 | 976 uint8_t *p; |
977 uint16_t qmant[AC3_MAX_CHANNELS][N/2]; | |
0 | 978 int exp0, exp1; |
979 int mant1_cnt, mant2_cnt, mant4_cnt; | |
1064 | 980 uint16_t *qmant1_ptr, *qmant2_ptr, *qmant4_ptr; |
0 | 981 int delta0, delta1, delta2; |
982 | |
2967 | 983 for(ch=0;ch<s->nb_channels;ch++) |
0 | 984 put_bits(&s->pb, 1, 0); /* 512 point MDCT */ |
2967 | 985 for(ch=0;ch<s->nb_channels;ch++) |
0 | 986 put_bits(&s->pb, 1, 1); /* no dither */ |
987 put_bits(&s->pb, 1, 0); /* no dynamic range */ | |
988 if (block_num == 0) { | |
989 /* for block 0, even if no coupling, we must say it. This is a | |
990 waste of bit :-) */ | |
991 put_bits(&s->pb, 1, 1); /* coupling strategy present */ | |
992 put_bits(&s->pb, 1, 0); /* no coupling strategy */ | |
993 } else { | |
994 put_bits(&s->pb, 1, 0); /* no new coupling strategy */ | |
995 } | |
996 | |
1408
4d67eb341a0c
AC3 encoding patch ba (Ross Martin <ffmpeg at ross dot interwrx dot com>)
michaelni
parents:
1106
diff
changeset
|
997 if (s->acmod == 2) |
4d67eb341a0c
AC3 encoding patch ba (Ross Martin <ffmpeg at ross dot interwrx dot com>)
michaelni
parents:
1106
diff
changeset
|
998 { |
2979 | 999 if(block_num==0) |
1000 { | |
1001 /* first block must define rematrixing (rematstr) */ | |
1002 put_bits(&s->pb, 1, 1); | |
2967 | 1003 |
2979 | 1004 /* dummy rematrixing rematflg(1:4)=0 */ |
1005 for (rbnd=0;rbnd<4;rbnd++) | |
1006 put_bits(&s->pb, 1, 0); | |
1007 } | |
1008 else | |
1009 { | |
1010 /* no matrixing (but should be used in the future) */ | |
1011 put_bits(&s->pb, 1, 0); | |
1012 } | |
1408
4d67eb341a0c
AC3 encoding patch ba (Ross Martin <ffmpeg at ross dot interwrx dot com>)
michaelni
parents:
1106
diff
changeset
|
1013 } |
0 | 1014 |
2967 | 1015 #if defined(DEBUG) |
0 | 1016 { |
1408
4d67eb341a0c
AC3 encoding patch ba (Ross Martin <ffmpeg at ross dot interwrx dot com>)
michaelni
parents:
1106
diff
changeset
|
1017 static int count = 0; |
1602
fdb8244da1e5
av_log patch(2 of ?) by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1598
diff
changeset
|
1018 av_log(NULL, AV_LOG_DEBUG, "Block #%d (%d)\n", block_num, count++); |
0 | 1019 } |
1020 #endif | |
1021 /* exponent strategy */ | |
1022 for(ch=0;ch<s->nb_channels;ch++) { | |
1023 put_bits(&s->pb, 2, exp_strategy[ch]); | |
1024 } | |
2967 | 1025 |
314 | 1026 if (s->lfe) { |
2979 | 1027 put_bits(&s->pb, 1, exp_strategy[s->lfe_channel]); |
314 | 1028 } |
1029 | |
0 | 1030 for(ch=0;ch<s->nb_channels;ch++) { |
1031 if (exp_strategy[ch] != EXP_REUSE) | |
1032 put_bits(&s->pb, 6, s->chbwcod[ch]); | |
1033 } | |
2967 | 1034 |
0 | 1035 /* exponents */ |
314 | 1036 for (ch = 0; ch < s->nb_all_channels; ch++) { |
0 | 1037 switch(exp_strategy[ch]) { |
1038 case EXP_REUSE: | |
1039 continue; | |
1040 case EXP_D15: | |
1041 group_size = 1; | |
1042 break; | |
1043 case EXP_D25: | |
1044 group_size = 2; | |
1045 break; | |
1046 default: | |
1047 case EXP_D45: | |
1048 group_size = 4; | |
1049 break; | |
1050 } | |
2979 | 1051 nb_groups = (s->nb_coefs[ch] + (group_size * 3) - 4) / (3 * group_size); |
0 | 1052 p = encoded_exp[ch]; |
1053 | |
1054 /* first exponent */ | |
1055 exp1 = *p++; | |
1056 put_bits(&s->pb, 4, exp1); | |
1057 | |
1058 /* next ones are delta encoded */ | |
1059 for(i=0;i<nb_groups;i++) { | |
1060 /* merge three delta in one code */ | |
1061 exp0 = exp1; | |
1062 exp1 = p[0]; | |
1063 p += group_size; | |
1064 delta0 = exp1 - exp0 + 2; | |
1065 | |
1066 exp0 = exp1; | |
1067 exp1 = p[0]; | |
1068 p += group_size; | |
1069 delta1 = exp1 - exp0 + 2; | |
1070 | |
1071 exp0 = exp1; | |
1072 exp1 = p[0]; | |
1073 p += group_size; | |
1074 delta2 = exp1 - exp0 + 2; | |
1075 | |
1076 put_bits(&s->pb, 7, ((delta0 * 5 + delta1) * 5) + delta2); | |
1077 } | |
1078 | |
2979 | 1079 if (ch != s->lfe_channel) |
1080 put_bits(&s->pb, 2, 0); /* no gain range info */ | |
0 | 1081 } |
1082 | |
1083 /* bit allocation info */ | |
1084 baie = (block_num == 0); | |
1085 put_bits(&s->pb, 1, baie); | |
1086 if (baie) { | |
1087 put_bits(&s->pb, 2, s->sdecaycod); | |
1088 put_bits(&s->pb, 2, s->fdecaycod); | |
1089 put_bits(&s->pb, 2, s->sgaincod); | |
1090 put_bits(&s->pb, 2, s->dbkneecod); | |
1091 put_bits(&s->pb, 3, s->floorcod); | |
1092 } | |
1093 | |
1094 /* snr offset */ | |
1095 put_bits(&s->pb, 1, baie); /* always present with bai */ | |
1096 if (baie) { | |
1097 put_bits(&s->pb, 6, s->csnroffst); | |
314 | 1098 for(ch=0;ch<s->nb_all_channels;ch++) { |
0 | 1099 put_bits(&s->pb, 4, s->fsnroffst[ch]); |
1100 put_bits(&s->pb, 3, s->fgaincod[ch]); | |
1101 } | |
1102 } | |
2967 | 1103 |
0 | 1104 put_bits(&s->pb, 1, 0); /* no delta bit allocation */ |
1105 put_bits(&s->pb, 1, 0); /* no data to skip */ | |
1106 | |
1107 /* mantissa encoding : we use two passes to handle the grouping. A | |
1108 one pass method may be faster, but it would necessitate to | |
1109 modify the output stream. */ | |
1110 | |
1111 /* first pass: quantize */ | |
1112 mant1_cnt = mant2_cnt = mant4_cnt = 0; | |
1113 qmant1_ptr = qmant2_ptr = qmant4_ptr = NULL; | |
1114 | |
314 | 1115 for (ch = 0; ch < s->nb_all_channels; ch++) { |
0 | 1116 int b, c, e, v; |
1117 | |
1118 for(i=0;i<s->nb_coefs[ch];i++) { | |
1119 c = mdct_coefs[ch][i]; | |
1120 e = encoded_exp[ch][i] - global_exp[ch]; | |
1121 b = bap[ch][i]; | |
1122 switch(b) { | |
1123 case 0: | |
1124 v = 0; | |
1125 break; | |
1126 case 1: | |
1127 v = sym_quant(c, e, 3); | |
1128 switch(mant1_cnt) { | |
1129 case 0: | |
1130 qmant1_ptr = &qmant[ch][i]; | |
1131 v = 9 * v; | |
1132 mant1_cnt = 1; | |
1133 break; | |
1134 case 1: | |
1135 *qmant1_ptr += 3 * v; | |
1136 mant1_cnt = 2; | |
1137 v = 128; | |
1138 break; | |
1139 default: | |
1140 *qmant1_ptr += v; | |
1141 mant1_cnt = 0; | |
1142 v = 128; | |
1143 break; | |
1144 } | |
1145 break; | |
1146 case 2: | |
1147 v = sym_quant(c, e, 5); | |
1148 switch(mant2_cnt) { | |
1149 case 0: | |
1150 qmant2_ptr = &qmant[ch][i]; | |
1151 v = 25 * v; | |
1152 mant2_cnt = 1; | |
1153 break; | |
1154 case 1: | |
1155 *qmant2_ptr += 5 * v; | |
1156 mant2_cnt = 2; | |
1157 v = 128; | |
1158 break; | |
1159 default: | |
1160 *qmant2_ptr += v; | |
1161 mant2_cnt = 0; | |
1162 v = 128; | |
1163 break; | |
1164 } | |
1165 break; | |
1166 case 3: | |
1167 v = sym_quant(c, e, 7); | |
1168 break; | |
1169 case 4: | |
1170 v = sym_quant(c, e, 11); | |
1171 switch(mant4_cnt) { | |
1172 case 0: | |
1173 qmant4_ptr = &qmant[ch][i]; | |
1174 v = 11 * v; | |
1175 mant4_cnt = 1; | |
1176 break; | |
1177 default: | |
1178 *qmant4_ptr += v; | |
1179 mant4_cnt = 0; | |
1180 v = 128; | |
1181 break; | |
1182 } | |
1183 break; | |
1184 case 5: | |
1185 v = sym_quant(c, e, 15); | |
1186 break; | |
1187 case 14: | |
1188 v = asym_quant(c, e, 14); | |
1189 break; | |
1190 case 15: | |
1191 v = asym_quant(c, e, 16); | |
1192 break; | |
1193 default: | |
1194 v = asym_quant(c, e, b - 1); | |
1195 break; | |
1196 } | |
1197 qmant[ch][i] = v; | |
1198 } | |
1199 } | |
1200 | |
1201 /* second pass : output the values */ | |
314 | 1202 for (ch = 0; ch < s->nb_all_channels; ch++) { |
0 | 1203 int b, q; |
2967 | 1204 |
0 | 1205 for(i=0;i<s->nb_coefs[ch];i++) { |
1206 q = qmant[ch][i]; | |
1207 b = bap[ch][i]; | |
1208 switch(b) { | |
1209 case 0: | |
1210 break; | |
1211 case 1: | |
2967 | 1212 if (q != 128) |
0 | 1213 put_bits(&s->pb, 5, q); |
1214 break; | |
1215 case 2: | |
2967 | 1216 if (q != 128) |
0 | 1217 put_bits(&s->pb, 7, q); |
1218 break; | |
1219 case 3: | |
1220 put_bits(&s->pb, 3, q); | |
1221 break; | |
1222 case 4: | |
1223 if (q != 128) | |
1224 put_bits(&s->pb, 7, q); | |
1225 break; | |
1226 case 14: | |
1227 put_bits(&s->pb, 14, q); | |
1228 break; | |
1229 case 15: | |
1230 put_bits(&s->pb, 16, q); | |
1231 break; | |
1232 default: | |
1233 put_bits(&s->pb, b - 1, q); | |
1234 break; | |
1235 } | |
1236 } | |
1237 } | |
1238 } | |
1239 | |
1240 #define CRC16_POLY ((1 << 0) | (1 << 2) | (1 << 15) | (1 << 16)) | |
1241 | |
1242 static unsigned int mul_poly(unsigned int a, unsigned int b, unsigned int poly) | |
1243 { | |
1244 unsigned int c; | |
1245 | |
1246 c = 0; | |
1247 while (a) { | |
1248 if (a & 1) | |
1249 c ^= b; | |
1250 a = a >> 1; | |
1251 b = b << 1; | |
1252 if (b & (1 << 16)) | |
1253 b ^= poly; | |
1254 } | |
1255 return c; | |
1256 } | |
1257 | |
1258 static unsigned int pow_poly(unsigned int a, unsigned int n, unsigned int poly) | |
1259 { | |
1260 unsigned int r; | |
1261 r = 1; | |
1262 while (n) { | |
1263 if (n & 1) | |
1264 r = mul_poly(r, a, poly); | |
1265 a = mul_poly(a, a, poly); | |
1266 n >>= 1; | |
1267 } | |
1268 return r; | |
1269 } | |
1270 | |
1271 | |
1272 /* compute log2(max(abs(tab[]))) */ | |
1064 | 1273 static int log2_tab(int16_t *tab, int n) |
0 | 1274 { |
1275 int i, v; | |
1276 | |
1277 v = 0; | |
1278 for(i=0;i<n;i++) { | |
1279 v |= abs(tab[i]); | |
1280 } | |
65 | 1281 return av_log2(v); |
0 | 1282 } |
1283 | |
1064 | 1284 static void lshift_tab(int16_t *tab, int n, int lshift) |
0 | 1285 { |
1286 int i; | |
1287 | |
1288 if (lshift > 0) { | |
1289 for(i=0;i<n;i++) { | |
1290 tab[i] <<= lshift; | |
1291 } | |
1292 } else if (lshift < 0) { | |
1293 lshift = -lshift; | |
1294 for(i=0;i<n;i++) { | |
1295 tab[i] >>= lshift; | |
1296 } | |
1297 } | |
1298 } | |
1299 | |
1300 /* fill the end of the frame and compute the two crcs */ | |
1301 static int output_frame_end(AC3EncodeContext *s) | |
1302 { | |
1303 int frame_size, frame_size_58, n, crc1, crc2, crc_inv; | |
1064 | 1304 uint8_t *frame; |
0 | 1305 |
1306 frame_size = s->frame_size; /* frame size in words */ | |
1307 /* align to 8 bits */ | |
1308 flush_put_bits(&s->pb); | |
1309 /* add zero bytes to reach the frame size */ | |
1310 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
|
1311 n = 2 * s->frame_size - (pbBufPtr(&s->pb) - frame) - 2; |
0 | 1312 assert(n >= 0); |
1408
4d67eb341a0c
AC3 encoding patch ba (Ross Martin <ffmpeg at ross dot interwrx dot com>)
michaelni
parents:
1106
diff
changeset
|
1313 if(n>0) |
4d67eb341a0c
AC3 encoding patch ba (Ross Martin <ffmpeg at ross dot interwrx dot com>)
michaelni
parents:
1106
diff
changeset
|
1314 memset(pbBufPtr(&s->pb), 0, n); |
2967 | 1315 |
0 | 1316 /* Now we must compute both crcs : this is not so easy for crc1 |
1317 because it is at the beginning of the data... */ | |
1318 frame_size_58 = (frame_size >> 1) + (frame_size >> 3); | |
3170 | 1319 crc1 = bswap_16(av_crc(av_crc8005, 0, frame + 4, 2 * frame_size_58 - 4)); |
0 | 1320 /* XXX: could precompute crc_inv */ |
1321 crc_inv = pow_poly((CRC16_POLY >> 1), (16 * frame_size_58) - 16, CRC16_POLY); | |
1322 crc1 = mul_poly(crc_inv, crc1, CRC16_POLY); | |
1323 frame[2] = crc1 >> 8; | |
1324 frame[3] = crc1; | |
2967 | 1325 |
3170 | 1326 crc2 = bswap_16(av_crc(av_crc8005, 0, frame + 2 * frame_size_58, (frame_size - frame_size_58) * 2 - 2)); |
0 | 1327 frame[2*frame_size - 2] = crc2 >> 8; |
1328 frame[2*frame_size - 1] = crc2; | |
1329 | |
1330 // printf("n=%d frame_size=%d\n", n, frame_size); | |
1331 return frame_size * 2; | |
1332 } | |
1333 | |
782 | 1334 static int AC3_encode_frame(AVCodecContext *avctx, |
1335 unsigned char *frame, int buf_size, void *data) | |
0 | 1336 { |
1337 AC3EncodeContext *s = avctx->priv_data; | |
2367
c353719836af
fix some type mismatches patch by (Jeff Muizelaar <muizelaar rogers com>)
michael
parents:
2157
diff
changeset
|
1338 int16_t *samples = data; |
0 | 1339 int i, j, k, v, ch; |
1064 | 1340 int16_t input_samples[N]; |
1341 int32_t mdct_coef[NB_BLOCKS][AC3_MAX_CHANNELS][N/2]; | |
1342 uint8_t exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2]; | |
1343 uint8_t exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS]; | |
1344 uint8_t encoded_exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2]; | |
1345 uint8_t bap[NB_BLOCKS][AC3_MAX_CHANNELS][N/2]; | |
1346 int8_t exp_samples[NB_BLOCKS][AC3_MAX_CHANNELS]; | |
0 | 1347 int frame_bits; |
1348 | |
1349 frame_bits = 0; | |
314 | 1350 for(ch=0;ch<s->nb_all_channels;ch++) { |
0 | 1351 /* fixed mdct to the six sub blocks & exponent computation */ |
1352 for(i=0;i<NB_BLOCKS;i++) { | |
1064 | 1353 int16_t *sptr; |
0 | 1354 int sinc; |
1355 | |
1356 /* compute input samples */ | |
1064 | 1357 memcpy(input_samples, s->last_samples[ch], N/2 * sizeof(int16_t)); |
314 | 1358 sinc = s->nb_all_channels; |
0 | 1359 sptr = samples + (sinc * (N/2) * i) + ch; |
1360 for(j=0;j<N/2;j++) { | |
1361 v = *sptr; | |
1362 input_samples[j + N/2] = v; | |
2967 | 1363 s->last_samples[ch][j] = v; |
0 | 1364 sptr += sinc; |
1365 } | |
1366 | |
1367 /* apply the MDCT window */ | |
1368 for(j=0;j<N/2;j++) { | |
2967 | 1369 input_samples[j] = MUL16(input_samples[j], |
0 | 1370 ac3_window[j]) >> 15; |
2967 | 1371 input_samples[N-j-1] = MUL16(input_samples[N-j-1], |
0 | 1372 ac3_window[j]) >> 15; |
1373 } | |
2967 | 1374 |
0 | 1375 /* Normalize the samples to use the maximum available |
1376 precision */ | |
1377 v = 14 - log2_tab(input_samples, N); | |
1378 if (v < 0) | |
1379 v = 0; | |
3258
8d42d21e570c
AC3 encoding volume fix, by Justin Ruggles jruggle earthlink net.
banan
parents:
3246
diff
changeset
|
1380 exp_samples[i][ch] = v - 9; |
0 | 1381 lshift_tab(input_samples, N, v); |
1382 | |
1383 /* do the MDCT */ | |
1384 mdct512(mdct_coef[i][ch], input_samples); | |
2967 | 1385 |
0 | 1386 /* compute "exponents". We take into account the |
1387 normalization there */ | |
1388 for(j=0;j<N/2;j++) { | |
1389 int e; | |
1390 v = abs(mdct_coef[i][ch][j]); | |
1391 if (v == 0) | |
1392 e = 24; | |
1393 else { | |
65 | 1394 e = 23 - av_log2(v) + exp_samples[i][ch]; |
0 | 1395 if (e >= 24) { |
1396 e = 24; | |
1397 mdct_coef[i][ch][j] = 0; | |
1398 } | |
1399 } | |
1400 exp[i][ch][j] = e; | |
1401 } | |
1402 } | |
2967 | 1403 |
314 | 1404 compute_exp_strategy(exp_strategy, exp, ch, ch == s->lfe_channel); |
0 | 1405 |
1406 /* compute the exponents as the decoder will see them. The | |
1407 EXP_REUSE case must be handled carefully : we select the | |
1408 min of the exponents */ | |
1409 i = 0; | |
1410 while (i < NB_BLOCKS) { | |
1411 j = i + 1; | |
1412 while (j < NB_BLOCKS && exp_strategy[j][ch] == EXP_REUSE) { | |
1413 exponent_min(exp[i][ch], exp[j][ch], s->nb_coefs[ch]); | |
1414 j++; | |
1415 } | |
1416 frame_bits += encode_exp(encoded_exp[i][ch], | |
2967 | 1417 exp[i][ch], s->nb_coefs[ch], |
0 | 1418 exp_strategy[i][ch]); |
1419 /* copy encoded exponents for reuse case */ | |
1420 for(k=i+1;k<j;k++) { | |
2967 | 1421 memcpy(encoded_exp[k][ch], encoded_exp[i][ch], |
1064 | 1422 s->nb_coefs[ch] * sizeof(uint8_t)); |
0 | 1423 } |
1424 i = j; | |
1425 } | |
1426 } | |
1427 | |
3246 | 1428 /* adjust for fractional frame sizes */ |
1429 while(s->bits_written >= s->bit_rate*1000 && s->samples_written >= s->sample_rate) { | |
1430 s->bits_written -= s->bit_rate*1000; | |
1431 s->samples_written -= s->sample_rate; | |
1432 } | |
1433 s->frame_size = s->frame_size_min + (s->bits_written * s->sample_rate < s->samples_written * s->bit_rate*1000); | |
1434 s->bits_written += s->frame_size * 16; | |
1435 s->samples_written += AC3_FRAME_SIZE; | |
1436 | |
0 | 1437 compute_bit_allocation(s, bap, encoded_exp, exp_strategy, frame_bits); |
1438 /* everything is known... let's output the frame */ | |
1439 output_frame_header(s, frame); | |
2967 | 1440 |
0 | 1441 for(i=0;i<NB_BLOCKS;i++) { |
2967 | 1442 output_audio_block(s, exp_strategy[i], encoded_exp[i], |
0 | 1443 bap[i], mdct_coef[i], exp_samples[i], i); |
1444 } | |
1445 return output_frame_end(s); | |
1446 } | |
1447 | |
925 | 1448 static int AC3_encode_close(AVCodecContext *avctx) |
1449 { | |
1450 av_freep(&avctx->coded_frame); | |
1014
48349e11c9b2
C99 initializers and kill warnings patch by (mru at users dot sourceforge dot net (Mns Rullgrd))
michaelni
parents:
925
diff
changeset
|
1451 return 0; |
925 | 1452 } |
1453 | |
0 | 1454 #if 0 |
1455 /*************************************************************************/ | |
1456 /* TEST */ | |
1457 | |
1458 #define FN (N/4) | |
1459 | |
1460 void fft_test(void) | |
1461 { | |
1462 IComplex in[FN], in1[FN]; | |
1463 int k, n, i; | |
1464 float sum_re, sum_im, a; | |
1465 | |
1466 /* FFT test */ | |
1467 | |
1468 for(i=0;i<FN;i++) { | |
1469 in[i].re = random() % 65535 - 32767; | |
1470 in[i].im = random() % 65535 - 32767; | |
1471 in1[i] = in[i]; | |
1472 } | |
1473 fft(in, 7); | |
1474 | |
1475 /* do it by hand */ | |
1476 for(k=0;k<FN;k++) { | |
1477 sum_re = 0; | |
1478 sum_im = 0; | |
1479 for(n=0;n<FN;n++) { | |
1480 a = -2 * M_PI * (n * k) / FN; | |
1481 sum_re += in1[n].re * cos(a) - in1[n].im * sin(a); | |
1482 sum_im += in1[n].re * sin(a) + in1[n].im * cos(a); | |
1483 } | |
2967 | 1484 printf("%3d: %6d,%6d %6.0f,%6.0f\n", |
1485 k, in[k].re, in[k].im, sum_re / FN, sum_im / FN); | |
0 | 1486 } |
1487 } | |
1488 | |
1489 void mdct_test(void) | |
1490 { | |
1064 | 1491 int16_t input[N]; |
1492 int32_t output[N/2]; | |
0 | 1493 float input1[N]; |
1494 float output1[N/2]; | |
1495 float s, a, err, e, emax; | |
1496 int i, k, n; | |
1497 | |
1498 for(i=0;i<N;i++) { | |
1499 input[i] = (random() % 65535 - 32767) * 9 / 10; | |
1500 input1[i] = input[i]; | |
1501 } | |
1502 | |
1503 mdct512(output, input); | |
2967 | 1504 |
0 | 1505 /* do it by hand */ |
1506 for(k=0;k<N/2;k++) { | |
1507 s = 0; | |
1508 for(n=0;n<N;n++) { | |
1509 a = (2*M_PI*(2*n+1+N/2)*(2*k+1) / (4 * N)); | |
1510 s += input1[n] * cos(a); | |
1511 } | |
1512 output1[k] = -2 * s / N; | |
1513 } | |
2967 | 1514 |
0 | 1515 err = 0; |
1516 emax = 0; | |
1517 for(i=0;i<N/2;i++) { | |
1518 printf("%3d: %7d %7.0f\n", i, output[i], output1[i]); | |
1519 e = output[i] - output1[i]; | |
1520 if (e > emax) | |
1521 emax = e; | |
1522 err += e * e; | |
1523 } | |
1524 printf("err2=%f emax=%f\n", err / (N/2), emax); | |
1525 } | |
1526 | |
1527 void test_ac3(void) | |
1528 { | |
1529 AC3EncodeContext ctx; | |
1530 unsigned char frame[AC3_MAX_CODED_FRAME_SIZE]; | |
1531 short samples[AC3_FRAME_SIZE]; | |
1532 int ret, i; | |
2967 | 1533 |
0 | 1534 AC3_encode_init(&ctx, 44100, 64000, 1); |
1535 | |
1536 fft_test(); | |
1537 mdct_test(); | |
1538 | |
1539 for(i=0;i<AC3_FRAME_SIZE;i++) | |
1540 samples[i] = (int)(sin(2*M_PI*i*1000.0/44100) * 10000); | |
1541 ret = AC3_encode_frame(&ctx, frame, samples); | |
1542 printf("ret=%d\n", ret); | |
1543 } | |
1544 #endif | |
1545 | |
1546 AVCodec ac3_encoder = { | |
1547 "ac3", | |
1548 CODEC_TYPE_AUDIO, | |
1549 CODEC_ID_AC3, | |
1550 sizeof(AC3EncodeContext), | |
1551 AC3_encode_init, | |
1552 AC3_encode_frame, | |
925 | 1553 AC3_encode_close, |
0 | 1554 NULL, |
1555 }; |