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 */
|
|
19 #include <stdlib.h>
|
|
20 #include <stdio.h>
|
|
21 #include <string.h>
|
|
22 #include <math.h>
|
|
23 #include "avcodec.h"
|
|
24
|
|
25 #include "ac3enc.h"
|
|
26 #include "ac3tab.h"
|
|
27
|
|
28 //#define DEBUG
|
|
29 //#define DEBUG_BITALLOC
|
|
30 #define NDEBUG
|
|
31 #include <assert.h>
|
|
32
|
|
33 #define MDCT_NBITS 9
|
|
34 #define N (1 << MDCT_NBITS)
|
|
35 #define NB_BLOCKS 6 /* number of PCM blocks inside an AC3 frame */
|
|
36
|
|
37 /* new exponents are sent if their Norm 1 exceed this number */
|
|
38 #define EXP_DIFF_THRESHOLD 1000
|
|
39
|
|
40 /* exponent encoding strategy */
|
|
41 #define EXP_REUSE 0
|
|
42 #define EXP_NEW 1
|
|
43
|
|
44 #define EXP_D15 1
|
|
45 #define EXP_D25 2
|
|
46 #define EXP_D45 3
|
|
47
|
|
48 static void fft_init(int ln);
|
|
49 static void ac3_crc_init(void);
|
|
50
|
|
51 static inline INT16 fix15(float a)
|
|
52 {
|
|
53 int v;
|
|
54 v = (int)(a * (float)(1 << 15));
|
|
55 if (v < -32767)
|
|
56 v = -32767;
|
|
57 else if (v > 32767)
|
|
58 v = 32767;
|
|
59 return v;
|
|
60 }
|
|
61
|
|
62 static inline int calc_lowcomp1(int a, int b0, int b1)
|
|
63 {
|
|
64 if ((b0 + 256) == b1) {
|
|
65 a = 384 ;
|
|
66 } else if (b0 > b1) {
|
|
67 a = a - 64;
|
|
68 if (a < 0) a=0;
|
|
69 }
|
|
70 return a;
|
|
71 }
|
|
72
|
|
73 static inline int calc_lowcomp(int a, int b0, int b1, int bin)
|
|
74 {
|
|
75 if (bin < 7) {
|
|
76 if ((b0 + 256) == b1) {
|
|
77 a = 384 ;
|
|
78 } else if (b0 > b1) {
|
|
79 a = a - 64;
|
|
80 if (a < 0) a=0;
|
|
81 }
|
|
82 } else if (bin < 20) {
|
|
83 if ((b0 + 256) == b1) {
|
|
84 a = 320 ;
|
|
85 } else if (b0 > b1) {
|
|
86 a= a - 64;
|
|
87 if (a < 0) a=0;
|
|
88 }
|
|
89 } else {
|
|
90 a = a - 128;
|
|
91 if (a < 0) a=0;
|
|
92 }
|
|
93 return a;
|
|
94 }
|
|
95
|
|
96 /* AC3 bit allocation. The algorithm is the one described in the AC3
|
|
97 spec with some optimizations because of our simplified encoding
|
|
98 assumptions. */
|
|
99 void parametric_bit_allocation(AC3EncodeContext *s, UINT8 *bap,
|
|
100 INT8 *exp, int start, int end,
|
|
101 int snroffset, int fgain)
|
|
102 {
|
|
103 int bin,i,j,k,end1,v,v1,bndstrt,bndend,lowcomp,begin;
|
|
104 int fastleak,slowleak,address,tmp;
|
|
105 INT16 psd[256]; /* scaled exponents */
|
|
106 INT16 bndpsd[50]; /* interpolated exponents */
|
|
107 INT16 excite[50]; /* excitation */
|
|
108 INT16 mask[50]; /* masking value */
|
|
109
|
|
110 /* exponent mapping to PSD */
|
|
111 for(bin=start;bin<end;bin++) {
|
|
112 psd[bin]=(3072 - (exp[bin] << 7));
|
|
113 }
|
|
114
|
|
115 /* PSD integration */
|
|
116 j=start;
|
|
117 k=masktab[start];
|
|
118 do {
|
|
119 v=psd[j];
|
|
120 j++;
|
|
121 end1=bndtab[k+1];
|
|
122 if (end1 > end) end1=end;
|
|
123 for(i=j;i<end1;i++) {
|
|
124 int c,adr;
|
|
125 /* logadd */
|
|
126 v1=psd[j];
|
|
127 c=v-v1;
|
|
128 if (c >= 0) {
|
|
129 adr=c >> 1;
|
|
130 if (adr > 255) adr=255;
|
|
131 v=v + latab[adr];
|
|
132 } else {
|
|
133 adr=(-c) >> 1;
|
|
134 if (adr > 255) adr=255;
|
|
135 v=v1 + latab[adr];
|
|
136 }
|
|
137 j++;
|
|
138 }
|
|
139 bndpsd[k]=v;
|
|
140 k++;
|
|
141 } while (end > bndtab[k]);
|
|
142
|
|
143 /* excitation function */
|
|
144 bndstrt = masktab[start];
|
|
145 bndend = masktab[end-1] + 1;
|
|
146
|
|
147 lowcomp = 0;
|
|
148 lowcomp = calc_lowcomp1(lowcomp, bndpsd[0], bndpsd[1]) ;
|
|
149 excite[0] = bndpsd[0] - fgain - lowcomp ;
|
|
150 lowcomp = calc_lowcomp1(lowcomp, bndpsd[1], bndpsd[2]) ;
|
|
151 excite[1] = bndpsd[1] - fgain - lowcomp ;
|
|
152 begin = 7 ;
|
|
153 for (bin = 2; bin < 7; bin++) {
|
|
154 lowcomp = calc_lowcomp1(lowcomp, bndpsd[bin], bndpsd[bin+1]) ;
|
|
155 fastleak = bndpsd[bin] - fgain ;
|
|
156 slowleak = bndpsd[bin] - s->sgain ;
|
|
157 excite[bin] = fastleak - lowcomp ;
|
|
158 if (bndpsd[bin] <= bndpsd[bin+1]) {
|
|
159 begin = bin + 1 ;
|
|
160 break ;
|
|
161 }
|
|
162 }
|
|
163
|
|
164 end1=bndend;
|
|
165 if (end1 > 22) end1=22;
|
|
166
|
|
167 for (bin = begin; bin < end1; bin++) {
|
|
168 lowcomp = calc_lowcomp(lowcomp, bndpsd[bin], bndpsd[bin+1], bin) ;
|
|
169
|
|
170 fastleak -= s->fdecay ;
|
|
171 v = bndpsd[bin] - fgain;
|
|
172 if (fastleak < v) fastleak = v;
|
|
173
|
|
174 slowleak -= s->sdecay ;
|
|
175 v = bndpsd[bin] - s->sgain;
|
|
176 if (slowleak < v) slowleak = v;
|
|
177
|
|
178 v=fastleak - lowcomp;
|
|
179 if (slowleak > v) v=slowleak;
|
|
180
|
|
181 excite[bin] = v;
|
|
182 }
|
|
183
|
|
184 for (bin = 22; bin < bndend; bin++) {
|
|
185 fastleak -= s->fdecay ;
|
|
186 v = bndpsd[bin] - fgain;
|
|
187 if (fastleak < v) fastleak = v;
|
|
188 slowleak -= s->sdecay ;
|
|
189 v = bndpsd[bin] - s->sgain;
|
|
190 if (slowleak < v) slowleak = v;
|
|
191
|
|
192 v=fastleak;
|
|
193 if (slowleak > v) v = slowleak;
|
|
194 excite[bin] = v;
|
|
195 }
|
|
196
|
|
197 /* compute masking curve */
|
|
198
|
|
199 for (bin = bndstrt; bin < bndend; bin++) {
|
|
200 v1 = excite[bin];
|
|
201 tmp = s->dbknee - bndpsd[bin];
|
|
202 if (tmp > 0) {
|
|
203 v1 += tmp >> 2;
|
|
204 }
|
|
205 v=hth[bin >> s->halfratecod][s->fscod];
|
|
206 if (v1 > v) v=v1;
|
|
207 mask[bin] = v;
|
|
208 }
|
|
209
|
|
210 /* compute bit allocation */
|
|
211
|
|
212 i = start ;
|
|
213 j = masktab[start] ;
|
|
214 do {
|
|
215 v=mask[j];
|
|
216 v -= snroffset ;
|
|
217 v -= s->floor ;
|
|
218 if (v < 0) v = 0;
|
|
219 v &= 0x1fe0 ;
|
|
220 v += s->floor ;
|
|
221
|
|
222 end1=bndtab[j] + bndsz[j];
|
|
223 if (end1 > end) end1=end;
|
|
224
|
|
225 for (k = i; k < end1; k++) {
|
|
226 address = (psd[i] - v) >> 5 ;
|
|
227 if (address < 0) address=0;
|
|
228 else if (address > 63) address=63;
|
|
229 bap[i] = baptab[address];
|
|
230 i++;
|
|
231 }
|
|
232 } while (end > bndtab[j++]) ;
|
|
233 }
|
|
234
|
|
235 typedef struct IComplex {
|
|
236 short re,im;
|
|
237 } IComplex;
|
|
238
|
|
239 static void fft_init(int ln)
|
|
240 {
|
|
241 int i, j, m, n;
|
|
242 float alpha;
|
|
243
|
|
244 n = 1 << ln;
|
|
245
|
|
246 for(i=0;i<(n/2);i++) {
|
|
247 alpha = 2 * M_PI * (float)i / (float)n;
|
|
248 costab[i] = fix15(cos(alpha));
|
|
249 sintab[i] = fix15(sin(alpha));
|
|
250 }
|
|
251
|
|
252 for(i=0;i<n;i++) {
|
|
253 m=0;
|
|
254 for(j=0;j<ln;j++) {
|
|
255 m |= ((i >> j) & 1) << (ln-j-1);
|
|
256 }
|
|
257 fft_rev[i]=m;
|
|
258 }
|
|
259 }
|
|
260
|
|
261 /* butter fly op */
|
|
262 #define BF(pre, pim, qre, qim, pre1, pim1, qre1, qim1) \
|
|
263 {\
|
|
264 int ax, ay, bx, by;\
|
|
265 bx=pre1;\
|
|
266 by=pim1;\
|
|
267 ax=qre1;\
|
|
268 ay=qim1;\
|
|
269 pre = (bx + ax) >> 1;\
|
|
270 pim = (by + ay) >> 1;\
|
|
271 qre = (bx - ax) >> 1;\
|
|
272 qim = (by - ay) >> 1;\
|
|
273 }
|
|
274
|
|
275 #define MUL16(a,b) ((a) * (b))
|
|
276
|
|
277 #define CMUL(pre, pim, are, aim, bre, bim) \
|
|
278 {\
|
|
279 pre = (MUL16(are, bre) - MUL16(aim, bim)) >> 15;\
|
|
280 pim = (MUL16(are, bim) + MUL16(bre, aim)) >> 15;\
|
|
281 }
|
|
282
|
|
283
|
|
284 /* do a 2^n point complex fft on 2^ln points. */
|
|
285 static void fft(IComplex *z, int ln)
|
|
286 {
|
|
287 int j, l, np, np2;
|
|
288 int nblocks, nloops;
|
|
289 register IComplex *p,*q;
|
|
290 int tmp_re, tmp_im;
|
|
291
|
|
292 np = 1 << ln;
|
|
293
|
|
294 /* reverse */
|
|
295 for(j=0;j<np;j++) {
|
|
296 int k;
|
|
297 IComplex tmp;
|
|
298 k = fft_rev[j];
|
|
299 if (k < j) {
|
|
300 tmp = z[k];
|
|
301 z[k] = z[j];
|
|
302 z[j] = tmp;
|
|
303 }
|
|
304 }
|
|
305
|
|
306 /* pass 0 */
|
|
307
|
|
308 p=&z[0];
|
|
309 j=(np >> 1);
|
|
310 do {
|
|
311 BF(p[0].re, p[0].im, p[1].re, p[1].im,
|
|
312 p[0].re, p[0].im, p[1].re, p[1].im);
|
|
313 p+=2;
|
|
314 } while (--j != 0);
|
|
315
|
|
316 /* pass 1 */
|
|
317
|
|
318 p=&z[0];
|
|
319 j=np >> 2;
|
|
320 do {
|
|
321 BF(p[0].re, p[0].im, p[2].re, p[2].im,
|
|
322 p[0].re, p[0].im, p[2].re, p[2].im);
|
|
323 BF(p[1].re, p[1].im, p[3].re, p[3].im,
|
|
324 p[1].re, p[1].im, p[3].im, -p[3].re);
|
|
325 p+=4;
|
|
326 } while (--j != 0);
|
|
327
|
|
328 /* pass 2 .. ln-1 */
|
|
329
|
|
330 nblocks = np >> 3;
|
|
331 nloops = 1 << 2;
|
|
332 np2 = np >> 1;
|
|
333 do {
|
|
334 p = z;
|
|
335 q = z + nloops;
|
|
336 for (j = 0; j < nblocks; ++j) {
|
|
337
|
|
338 BF(p->re, p->im, q->re, q->im,
|
|
339 p->re, p->im, q->re, q->im);
|
|
340
|
|
341 p++;
|
|
342 q++;
|
|
343 for(l = nblocks; l < np2; l += nblocks) {
|
|
344 CMUL(tmp_re, tmp_im, costab[l], -sintab[l], q->re, q->im);
|
|
345 BF(p->re, p->im, q->re, q->im,
|
|
346 p->re, p->im, tmp_re, tmp_im);
|
|
347 p++;
|
|
348 q++;
|
|
349 }
|
|
350 p += nloops;
|
|
351 q += nloops;
|
|
352 }
|
|
353 nblocks = nblocks >> 1;
|
|
354 nloops = nloops << 1;
|
|
355 } while (nblocks != 0);
|
|
356 }
|
|
357
|
|
358 /* do a 512 point mdct */
|
|
359 static void mdct512(INT32 *out, INT16 *in)
|
|
360 {
|
|
361 int i, re, im, re1, im1;
|
|
362 INT16 rot[N];
|
|
363 IComplex x[N/4];
|
|
364
|
|
365 /* shift to simplify computations */
|
|
366 for(i=0;i<N/4;i++)
|
|
367 rot[i] = -in[i + 3*N/4];
|
|
368 for(i=N/4;i<N;i++)
|
|
369 rot[i] = in[i - N/4];
|
|
370
|
|
371 /* pre rotation */
|
|
372 for(i=0;i<N/4;i++) {
|
|
373 re = ((int)rot[2*i] - (int)rot[N-1-2*i]) >> 1;
|
|
374 im = -((int)rot[N/2+2*i] - (int)rot[N/2-1-2*i]) >> 1;
|
|
375 CMUL(x[i].re, x[i].im, re, im, -xcos1[i], xsin1[i]);
|
|
376 }
|
|
377
|
|
378 fft(x, MDCT_NBITS - 2);
|
|
379
|
|
380 /* post rotation */
|
|
381 for(i=0;i<N/4;i++) {
|
|
382 re = x[i].re;
|
|
383 im = x[i].im;
|
|
384 CMUL(re1, im1, re, im, xsin1[i], xcos1[i]);
|
|
385 out[2*i] = im1;
|
|
386 out[N/2-1-2*i] = re1;
|
|
387 }
|
|
388 }
|
|
389
|
|
390 /* XXX: use another norm ? */
|
|
391 static int calc_exp_diff(UINT8 *exp1, UINT8 *exp2, int n)
|
|
392 {
|
|
393 int sum, i;
|
|
394 sum = 0;
|
|
395 for(i=0;i<n;i++) {
|
|
396 sum += abs(exp1[i] - exp2[i]);
|
|
397 }
|
|
398 return sum;
|
|
399 }
|
|
400
|
|
401 static void compute_exp_strategy(UINT8 exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS],
|
|
402 UINT8 exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
|
|
403 int ch)
|
|
404 {
|
|
405 int i, j;
|
|
406 int exp_diff;
|
|
407
|
|
408 /* estimate if the exponent variation & decide if they should be
|
|
409 reused in the next frame */
|
|
410 exp_strategy[0][ch] = EXP_NEW;
|
|
411 for(i=1;i<NB_BLOCKS;i++) {
|
|
412 exp_diff = calc_exp_diff(exp[i][ch], exp[i-1][ch], N/2);
|
|
413 #ifdef DEBUG
|
|
414 printf("exp_diff=%d\n", exp_diff);
|
|
415 #endif
|
|
416 if (exp_diff > EXP_DIFF_THRESHOLD)
|
|
417 exp_strategy[i][ch] = EXP_NEW;
|
|
418 else
|
|
419 exp_strategy[i][ch] = EXP_REUSE;
|
|
420 }
|
|
421 /* now select the encoding strategy type : if exponents are often
|
|
422 recoded, we use a coarse encoding */
|
|
423 i = 0;
|
|
424 while (i < NB_BLOCKS) {
|
|
425 j = i + 1;
|
|
426 while (j < NB_BLOCKS && exp_strategy[j][ch] == EXP_REUSE)
|
|
427 j++;
|
|
428 switch(j - i) {
|
|
429 case 1:
|
|
430 exp_strategy[i][ch] = EXP_D45;
|
|
431 break;
|
|
432 case 2:
|
|
433 case 3:
|
|
434 exp_strategy[i][ch] = EXP_D25;
|
|
435 break;
|
|
436 default:
|
|
437 exp_strategy[i][ch] = EXP_D15;
|
|
438 break;
|
|
439 }
|
|
440 i = j;
|
|
441 }
|
|
442 }
|
|
443
|
|
444 /* set exp[i] to min(exp[i], exp1[i]) */
|
|
445 static void exponent_min(UINT8 exp[N/2], UINT8 exp1[N/2], int n)
|
|
446 {
|
|
447 int i;
|
|
448
|
|
449 for(i=0;i<n;i++) {
|
|
450 if (exp1[i] < exp[i])
|
|
451 exp[i] = exp1[i];
|
|
452 }
|
|
453 }
|
|
454
|
|
455 /* update the exponents so that they are the ones the decoder will
|
|
456 decode. Return the number of bits used to code the exponents */
|
|
457 static int encode_exp(UINT8 encoded_exp[N/2],
|
|
458 UINT8 exp[N/2],
|
|
459 int nb_exps,
|
|
460 int exp_strategy)
|
|
461 {
|
|
462 int group_size, nb_groups, i, j, k, recurse, exp_min, delta;
|
|
463 UINT8 exp1[N/2];
|
|
464
|
|
465 switch(exp_strategy) {
|
|
466 case EXP_D15:
|
|
467 group_size = 1;
|
|
468 break;
|
|
469 case EXP_D25:
|
|
470 group_size = 2;
|
|
471 break;
|
|
472 default:
|
|
473 case EXP_D45:
|
|
474 group_size = 4;
|
|
475 break;
|
|
476 }
|
|
477 nb_groups = ((nb_exps + (group_size * 3) - 4) / (3 * group_size)) * 3;
|
|
478
|
|
479 /* for each group, compute the minimum exponent */
|
|
480 exp1[0] = exp[0]; /* DC exponent is handled separately */
|
|
481 k = 1;
|
|
482 for(i=1;i<=nb_groups;i++) {
|
|
483 exp_min = exp[k];
|
|
484 assert(exp_min >= 0 && exp_min <= 24);
|
|
485 for(j=1;j<group_size;j++) {
|
|
486 if (exp[k+j] < exp_min)
|
|
487 exp_min = exp[k+j];
|
|
488 }
|
|
489 exp1[i] = exp_min;
|
|
490 k += group_size;
|
|
491 }
|
|
492
|
|
493 /* constraint for DC exponent */
|
|
494 if (exp1[0] > 15)
|
|
495 exp1[0] = 15;
|
|
496
|
|
497 /* Iterate until the delta constraints between each groups are
|
|
498 satisfyed. I'm sure it is possible to find a better algorithm,
|
|
499 but I am lazy */
|
|
500 do {
|
|
501 recurse = 0;
|
|
502 for(i=1;i<=nb_groups;i++) {
|
|
503 delta = exp1[i] - exp1[i-1];
|
|
504 if (delta > 2) {
|
|
505 /* if delta too big, we encode a smaller exponent */
|
|
506 exp1[i] = exp1[i-1] + 2;
|
|
507 } else if (delta < -2) {
|
|
508 /* if delta is too small, we must decrease the previous
|
|
509 exponent, which means we must recurse */
|
|
510 recurse = 1;
|
|
511 exp1[i-1] = exp1[i] + 2;
|
|
512 }
|
|
513 }
|
|
514 } while (recurse);
|
|
515
|
|
516 /* now we have the exponent values the decoder will see */
|
|
517 encoded_exp[0] = exp1[0];
|
|
518 k = 1;
|
|
519 for(i=1;i<=nb_groups;i++) {
|
|
520 for(j=0;j<group_size;j++) {
|
|
521 encoded_exp[k+j] = exp1[i];
|
|
522 }
|
|
523 k += group_size;
|
|
524 }
|
|
525
|
|
526 #if defined(DEBUG)
|
|
527 printf("exponents: strategy=%d\n", exp_strategy);
|
|
528 for(i=0;i<=nb_groups * group_size;i++) {
|
|
529 printf("%d ", encoded_exp[i]);
|
|
530 }
|
|
531 printf("\n");
|
|
532 #endif
|
|
533
|
|
534 return 4 + (nb_groups / 3) * 7;
|
|
535 }
|
|
536
|
|
537 /* return the size in bits taken by the mantissa */
|
|
538 int compute_mantissa_size(AC3EncodeContext *s, UINT8 *m, int nb_coefs)
|
|
539 {
|
|
540 int bits, mant, i;
|
|
541
|
|
542 bits = 0;
|
|
543 for(i=0;i<nb_coefs;i++) {
|
|
544 mant = m[i];
|
|
545 switch(mant) {
|
|
546 case 0:
|
|
547 /* nothing */
|
|
548 break;
|
|
549 case 1:
|
|
550 /* 3 mantissa in 5 bits */
|
|
551 if (s->mant1_cnt == 0)
|
|
552 bits += 5;
|
|
553 if (++s->mant1_cnt == 3)
|
|
554 s->mant1_cnt = 0;
|
|
555 break;
|
|
556 case 2:
|
|
557 /* 3 mantissa in 7 bits */
|
|
558 if (s->mant2_cnt == 0)
|
|
559 bits += 7;
|
|
560 if (++s->mant2_cnt == 3)
|
|
561 s->mant2_cnt = 0;
|
|
562 break;
|
|
563 case 3:
|
|
564 bits += 3;
|
|
565 break;
|
|
566 case 4:
|
|
567 /* 2 mantissa in 7 bits */
|
|
568 if (s->mant4_cnt == 0)
|
|
569 bits += 7;
|
|
570 if (++s->mant4_cnt == 2)
|
|
571 s->mant4_cnt = 0;
|
|
572 break;
|
|
573 case 14:
|
|
574 bits += 14;
|
|
575 break;
|
|
576 case 15:
|
|
577 bits += 16;
|
|
578 break;
|
|
579 default:
|
|
580 bits += mant - 1;
|
|
581 break;
|
|
582 }
|
|
583 }
|
|
584 return bits;
|
|
585 }
|
|
586
|
|
587
|
|
588 static int bit_alloc(AC3EncodeContext *s,
|
|
589 UINT8 bap[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
|
|
590 UINT8 encoded_exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
|
|
591 UINT8 exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS],
|
|
592 int frame_bits, int csnroffst, int fsnroffst)
|
|
593 {
|
|
594 int i, ch;
|
|
595
|
|
596 /* compute size */
|
|
597 for(i=0;i<NB_BLOCKS;i++) {
|
|
598 s->mant1_cnt = 0;
|
|
599 s->mant2_cnt = 0;
|
|
600 s->mant4_cnt = 0;
|
|
601 for(ch=0;ch<s->nb_channels;ch++) {
|
|
602 parametric_bit_allocation(s, bap[i][ch], encoded_exp[i][ch],
|
|
603 0, s->nb_coefs[ch],
|
|
604 (((csnroffst-15) << 4) +
|
|
605 fsnroffst) << 2,
|
|
606 fgaintab[s->fgaincod[ch]]);
|
|
607 frame_bits += compute_mantissa_size(s, bap[i][ch],
|
|
608 s->nb_coefs[ch]);
|
|
609 }
|
|
610 }
|
|
611 #if 0
|
|
612 printf("csnr=%d fsnr=%d frame_bits=%d diff=%d\n",
|
|
613 csnroffst, fsnroffst, frame_bits,
|
|
614 16 * s->frame_size - ((frame_bits + 7) & ~7));
|
|
615 #endif
|
|
616 return 16 * s->frame_size - frame_bits;
|
|
617 }
|
|
618
|
|
619 #define SNR_INC1 4
|
|
620
|
|
621 static int compute_bit_allocation(AC3EncodeContext *s,
|
|
622 UINT8 bap[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
|
|
623 UINT8 encoded_exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
|
|
624 UINT8 exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS],
|
|
625 int frame_bits)
|
|
626 {
|
|
627 int i, ch;
|
|
628 int csnroffst, fsnroffst;
|
|
629 UINT8 bap1[NB_BLOCKS][AC3_MAX_CHANNELS][N/2];
|
|
630
|
|
631 /* init default parameters */
|
|
632 s->sdecaycod = 2;
|
|
633 s->fdecaycod = 1;
|
|
634 s->sgaincod = 1;
|
|
635 s->dbkneecod = 2;
|
|
636 s->floorcod = 4;
|
|
637 for(ch=0;ch<s->nb_channels;ch++)
|
|
638 s->fgaincod[ch] = 4;
|
|
639
|
|
640 /* compute real values */
|
|
641 s->sdecay = sdecaytab[s->sdecaycod] >> s->halfratecod;
|
|
642 s->fdecay = fdecaytab[s->fdecaycod] >> s->halfratecod;
|
|
643 s->sgain = sgaintab[s->sgaincod];
|
|
644 s->dbknee = dbkneetab[s->dbkneecod];
|
|
645 s->floor = floortab[s->floorcod];
|
|
646
|
|
647 /* header size */
|
|
648 frame_bits += 65;
|
|
649 if (s->acmod == 2)
|
|
650 frame_bits += 2;
|
|
651
|
|
652 /* audio blocks */
|
|
653 for(i=0;i<NB_BLOCKS;i++) {
|
|
654 frame_bits += s->nb_channels * 2 + 2;
|
|
655 if (s->acmod == 2)
|
|
656 frame_bits++;
|
|
657 frame_bits += 2 * s->nb_channels;
|
|
658 for(ch=0;ch<s->nb_channels;ch++) {
|
|
659 if (exp_strategy[i][ch] != EXP_REUSE)
|
|
660 frame_bits += 6 + 2;
|
|
661 }
|
|
662 frame_bits++; /* baie */
|
|
663 frame_bits++; /* snr */
|
|
664 frame_bits += 2; /* delta / skip */
|
|
665 }
|
|
666 frame_bits++; /* cplinu for block 0 */
|
|
667 /* bit alloc info */
|
|
668 frame_bits += 2*4 + 3 + 6 + s->nb_channels * (4 + 3);
|
|
669
|
|
670 /* CRC */
|
|
671 frame_bits += 16;
|
|
672
|
|
673 /* now the big work begins : do the bit allocation. Modify the snr
|
|
674 offset until we can pack everything in the requested frame size */
|
|
675
|
|
676 csnroffst = s->csnroffst;
|
|
677 while (csnroffst >= 0 &&
|
|
678 bit_alloc(s, bap, encoded_exp, exp_strategy, frame_bits, csnroffst, 0) < 0)
|
|
679 csnroffst -= SNR_INC1;
|
|
680 if (csnroffst < 0) {
|
|
681 fprintf(stderr, "Error !!!\n");
|
|
682 return -1;
|
|
683 }
|
|
684 while ((csnroffst + SNR_INC1) <= 63 &&
|
|
685 bit_alloc(s, bap1, encoded_exp, exp_strategy, frame_bits,
|
|
686 csnroffst + SNR_INC1, 0) >= 0) {
|
|
687 csnroffst += SNR_INC1;
|
|
688 memcpy(bap, bap1, sizeof(bap1));
|
|
689 }
|
|
690 while ((csnroffst + 1) <= 63 &&
|
|
691 bit_alloc(s, bap1, encoded_exp, exp_strategy, frame_bits, csnroffst + 1, 0) >= 0) {
|
|
692 csnroffst++;
|
|
693 memcpy(bap, bap1, sizeof(bap1));
|
|
694 }
|
|
695
|
|
696 fsnroffst = 0;
|
|
697 while ((fsnroffst + SNR_INC1) <= 15 &&
|
|
698 bit_alloc(s, bap1, encoded_exp, exp_strategy, frame_bits,
|
|
699 csnroffst, fsnroffst + SNR_INC1) >= 0) {
|
|
700 fsnroffst += SNR_INC1;
|
|
701 memcpy(bap, bap1, sizeof(bap1));
|
|
702 }
|
|
703 while ((fsnroffst + 1) <= 15 &&
|
|
704 bit_alloc(s, bap1, encoded_exp, exp_strategy, frame_bits,
|
|
705 csnroffst, fsnroffst + 1) >= 0) {
|
|
706 fsnroffst++;
|
|
707 memcpy(bap, bap1, sizeof(bap1));
|
|
708 }
|
|
709
|
|
710 s->csnroffst = csnroffst;
|
|
711 for(ch=0;ch<s->nb_channels;ch++)
|
|
712 s->fsnroffst[ch] = fsnroffst;
|
|
713 #if defined(DEBUG_BITALLOC)
|
|
714 {
|
|
715 int j;
|
|
716
|
|
717 for(i=0;i<6;i++) {
|
|
718 for(ch=0;ch<s->nb_channels;ch++) {
|
|
719 printf("Block #%d Ch%d:\n", i, ch);
|
|
720 printf("bap=");
|
|
721 for(j=0;j<s->nb_coefs[ch];j++) {
|
|
722 printf("%d ",bap[i][ch][j]);
|
|
723 }
|
|
724 printf("\n");
|
|
725 }
|
|
726 }
|
|
727 }
|
|
728 #endif
|
|
729 return 0;
|
|
730 }
|
|
731
|
|
732 static int AC3_encode_init(AVCodecContext *avctx)
|
|
733 {
|
|
734 int freq = avctx->sample_rate;
|
|
735 int bitrate = avctx->bit_rate;
|
|
736 int channels = avctx->channels;
|
|
737 AC3EncodeContext *s = avctx->priv_data;
|
|
738 int i, j, k, l, ch, v;
|
|
739 float alpha;
|
|
740 static unsigned short freqs[3] = { 48000, 44100, 32000 };
|
|
741
|
|
742 avctx->frame_size = AC3_FRAME_SIZE;
|
|
743 avctx->key_frame = 1; /* always key frame */
|
|
744
|
|
745 /* number of channels */
|
|
746 if (channels == 1)
|
|
747 s->acmod = 1;
|
|
748 else if (channels == 2)
|
|
749 s->acmod = 2;
|
|
750 else
|
|
751 return -1;
|
|
752 s->nb_channels = channels;
|
|
753
|
|
754 /* frequency */
|
|
755 for(i=0;i<3;i++) {
|
|
756 for(j=0;j<3;j++)
|
|
757 if ((freqs[j] >> i) == freq)
|
|
758 goto found;
|
|
759 }
|
|
760 return -1;
|
|
761 found:
|
|
762 s->sample_rate = freq;
|
|
763 s->halfratecod = i;
|
|
764 s->fscod = j;
|
|
765 s->bsid = 8 + s->halfratecod;
|
|
766 s->bsmod = 0; /* complete main audio service */
|
|
767
|
|
768 /* bitrate & frame size */
|
|
769 bitrate /= 1000;
|
|
770 for(i=0;i<19;i++) {
|
|
771 if ((bitratetab[i] >> s->halfratecod) == bitrate)
|
|
772 break;
|
|
773 }
|
|
774 if (i == 19)
|
|
775 return -1;
|
|
776 s->bit_rate = bitrate;
|
|
777 s->frmsizecod = i << 1;
|
|
778 s->frame_size_min = (bitrate * 1000 * AC3_FRAME_SIZE) / (freq * 16);
|
|
779 /* for now we do not handle fractional sizes */
|
|
780 s->frame_size = s->frame_size_min;
|
|
781
|
|
782 /* bit allocation init */
|
|
783 for(ch=0;ch<s->nb_channels;ch++) {
|
|
784 /* bandwidth for each channel */
|
|
785 /* XXX: should compute the bandwidth according to the frame
|
|
786 size, so that we avoid anoying high freq artefacts */
|
|
787 s->chbwcod[ch] = 50; /* sample bandwidth as mpeg audio layer 2 table 0 */
|
|
788 s->nb_coefs[ch] = ((s->chbwcod[ch] + 12) * 3) + 37;
|
|
789 }
|
|
790 /* initial snr offset */
|
|
791 s->csnroffst = 40;
|
|
792
|
|
793 /* compute bndtab and masktab from bandsz */
|
|
794 k = 0;
|
|
795 l = 0;
|
|
796 for(i=0;i<50;i++) {
|
|
797 bndtab[i] = l;
|
|
798 v = bndsz[i];
|
|
799 for(j=0;j<v;j++) masktab[k++]=i;
|
|
800 l += v;
|
|
801 }
|
|
802 bndtab[50] = 0;
|
|
803
|
|
804 /* mdct init */
|
|
805 fft_init(MDCT_NBITS - 2);
|
|
806 for(i=0;i<N/4;i++) {
|
|
807 alpha = 2 * M_PI * (i + 1.0 / 8.0) / (float)N;
|
|
808 xcos1[i] = fix15(-cos(alpha));
|
|
809 xsin1[i] = fix15(-sin(alpha));
|
|
810 }
|
|
811
|
|
812 ac3_crc_init();
|
|
813
|
|
814 return 0;
|
|
815 }
|
|
816
|
|
817 /* output the AC3 frame header */
|
|
818 static void output_frame_header(AC3EncodeContext *s, unsigned char *frame)
|
|
819 {
|
|
820 init_put_bits(&s->pb, frame, AC3_MAX_CODED_FRAME_SIZE, NULL, NULL);
|
|
821
|
|
822 put_bits(&s->pb, 16, 0x0b77); /* frame header */
|
|
823 put_bits(&s->pb, 16, 0); /* crc1: will be filled later */
|
|
824 put_bits(&s->pb, 2, s->fscod);
|
|
825 put_bits(&s->pb, 6, s->frmsizecod + (s->frame_size - s->frame_size_min));
|
|
826 put_bits(&s->pb, 5, s->bsid);
|
|
827 put_bits(&s->pb, 3, s->bsmod);
|
|
828 put_bits(&s->pb, 3, s->acmod);
|
|
829 if (s->acmod == 2) {
|
|
830 put_bits(&s->pb, 2, 0); /* surround not indicated */
|
|
831 }
|
|
832 put_bits(&s->pb, 1, 0); /* no LFE */
|
|
833 put_bits(&s->pb, 5, 31); /* dialog norm: -31 db */
|
|
834 put_bits(&s->pb, 1, 0); /* no compression control word */
|
|
835 put_bits(&s->pb, 1, 0); /* no lang code */
|
|
836 put_bits(&s->pb, 1, 0); /* no audio production info */
|
|
837 put_bits(&s->pb, 1, 0); /* no copyright */
|
|
838 put_bits(&s->pb, 1, 1); /* original bitstream */
|
|
839 put_bits(&s->pb, 1, 0); /* no time code 1 */
|
|
840 put_bits(&s->pb, 1, 0); /* no time code 2 */
|
|
841 put_bits(&s->pb, 1, 0); /* no addtional bit stream info */
|
|
842 }
|
|
843
|
|
844 /* symetric quantization on 'levels' levels */
|
|
845 static inline int sym_quant(int c, int e, int levels)
|
|
846 {
|
|
847 int v;
|
|
848
|
|
849 if (c >= 0) {
|
|
850 v = (levels * (c << e)) >> 25;
|
|
851 v = (levels >> 1) + v;
|
|
852 } else {
|
|
853 v = (levels * ((-c) << e)) >> 25;
|
|
854 v = (levels >> 1) - v;
|
|
855 }
|
|
856 assert (v >= 0 && v < levels);
|
|
857 return v;
|
|
858 }
|
|
859
|
|
860 /* asymetric quantization on 2^qbits levels */
|
|
861 static inline int asym_quant(int c, int e, int qbits)
|
|
862 {
|
|
863 int lshift, m, v;
|
|
864
|
|
865 lshift = e + qbits - 24;
|
|
866 if (lshift >= 0)
|
|
867 v = c << lshift;
|
|
868 else
|
|
869 v = c >> (-lshift);
|
|
870 /* rounding */
|
|
871 v = (v + 1) >> 1;
|
|
872 m = (1 << (qbits-1));
|
|
873 if (v >= m)
|
|
874 v = m - 1;
|
|
875 assert(v >= -m);
|
|
876 return v & ((1 << qbits)-1);
|
|
877 }
|
|
878
|
|
879 /* Output one audio block. There are NB_BLOCKS audio blocks in one AC3
|
|
880 frame */
|
|
881 static void output_audio_block(AC3EncodeContext *s,
|
|
882 UINT8 exp_strategy[AC3_MAX_CHANNELS],
|
|
883 UINT8 encoded_exp[AC3_MAX_CHANNELS][N/2],
|
|
884 UINT8 bap[AC3_MAX_CHANNELS][N/2],
|
|
885 INT32 mdct_coefs[AC3_MAX_CHANNELS][N/2],
|
|
886 INT8 global_exp[AC3_MAX_CHANNELS],
|
|
887 int block_num)
|
|
888 {
|
|
889 int ch, nb_groups, group_size, i, baie;
|
|
890 UINT8 *p;
|
|
891 UINT16 qmant[AC3_MAX_CHANNELS][N/2];
|
|
892 int exp0, exp1;
|
|
893 int mant1_cnt, mant2_cnt, mant4_cnt;
|
|
894 UINT16 *qmant1_ptr, *qmant2_ptr, *qmant4_ptr;
|
|
895 int delta0, delta1, delta2;
|
|
896
|
|
897 for(ch=0;ch<s->nb_channels;ch++)
|
|
898 put_bits(&s->pb, 1, 0); /* 512 point MDCT */
|
|
899 for(ch=0;ch<s->nb_channels;ch++)
|
|
900 put_bits(&s->pb, 1, 1); /* no dither */
|
|
901 put_bits(&s->pb, 1, 0); /* no dynamic range */
|
|
902 if (block_num == 0) {
|
|
903 /* for block 0, even if no coupling, we must say it. This is a
|
|
904 waste of bit :-) */
|
|
905 put_bits(&s->pb, 1, 1); /* coupling strategy present */
|
|
906 put_bits(&s->pb, 1, 0); /* no coupling strategy */
|
|
907 } else {
|
|
908 put_bits(&s->pb, 1, 0); /* no new coupling strategy */
|
|
909 }
|
|
910
|
|
911 if (s->acmod == 2) {
|
|
912 put_bits(&s->pb, 1, 0); /* no matrixing (but should be used in the future) */
|
|
913 }
|
|
914
|
|
915 #if defined(DEBUG)
|
|
916 {
|
|
917 static int count = 0;
|
|
918 printf("Block #%d (%d)\n", block_num, count++);
|
|
919 }
|
|
920 #endif
|
|
921 /* exponent strategy */
|
|
922 for(ch=0;ch<s->nb_channels;ch++) {
|
|
923 put_bits(&s->pb, 2, exp_strategy[ch]);
|
|
924 }
|
|
925
|
|
926 for(ch=0;ch<s->nb_channels;ch++) {
|
|
927 if (exp_strategy[ch] != EXP_REUSE)
|
|
928 put_bits(&s->pb, 6, s->chbwcod[ch]);
|
|
929 }
|
|
930
|
|
931 /* exponents */
|
|
932 for (ch = 0; ch < s->nb_channels; ch++) {
|
|
933 switch(exp_strategy[ch]) {
|
|
934 case EXP_REUSE:
|
|
935 continue;
|
|
936 case EXP_D15:
|
|
937 group_size = 1;
|
|
938 break;
|
|
939 case EXP_D25:
|
|
940 group_size = 2;
|
|
941 break;
|
|
942 default:
|
|
943 case EXP_D45:
|
|
944 group_size = 4;
|
|
945 break;
|
|
946 }
|
|
947 nb_groups = (s->nb_coefs[ch] + (group_size * 3) - 4) / (3 * group_size);
|
|
948 p = encoded_exp[ch];
|
|
949
|
|
950 /* first exponent */
|
|
951 exp1 = *p++;
|
|
952 put_bits(&s->pb, 4, exp1);
|
|
953
|
|
954 /* next ones are delta encoded */
|
|
955 for(i=0;i<nb_groups;i++) {
|
|
956 /* merge three delta in one code */
|
|
957 exp0 = exp1;
|
|
958 exp1 = p[0];
|
|
959 p += group_size;
|
|
960 delta0 = exp1 - exp0 + 2;
|
|
961
|
|
962 exp0 = exp1;
|
|
963 exp1 = p[0];
|
|
964 p += group_size;
|
|
965 delta1 = exp1 - exp0 + 2;
|
|
966
|
|
967 exp0 = exp1;
|
|
968 exp1 = p[0];
|
|
969 p += group_size;
|
|
970 delta2 = exp1 - exp0 + 2;
|
|
971
|
|
972 put_bits(&s->pb, 7, ((delta0 * 5 + delta1) * 5) + delta2);
|
|
973 }
|
|
974
|
|
975 put_bits(&s->pb, 2, 0); /* no gain range info */
|
|
976 }
|
|
977
|
|
978 /* bit allocation info */
|
|
979 baie = (block_num == 0);
|
|
980 put_bits(&s->pb, 1, baie);
|
|
981 if (baie) {
|
|
982 put_bits(&s->pb, 2, s->sdecaycod);
|
|
983 put_bits(&s->pb, 2, s->fdecaycod);
|
|
984 put_bits(&s->pb, 2, s->sgaincod);
|
|
985 put_bits(&s->pb, 2, s->dbkneecod);
|
|
986 put_bits(&s->pb, 3, s->floorcod);
|
|
987 }
|
|
988
|
|
989 /* snr offset */
|
|
990 put_bits(&s->pb, 1, baie); /* always present with bai */
|
|
991 if (baie) {
|
|
992 put_bits(&s->pb, 6, s->csnroffst);
|
|
993 for(ch=0;ch<s->nb_channels;ch++) {
|
|
994 put_bits(&s->pb, 4, s->fsnroffst[ch]);
|
|
995 put_bits(&s->pb, 3, s->fgaincod[ch]);
|
|
996 }
|
|
997 }
|
|
998
|
|
999 put_bits(&s->pb, 1, 0); /* no delta bit allocation */
|
|
1000 put_bits(&s->pb, 1, 0); /* no data to skip */
|
|
1001
|
|
1002 /* mantissa encoding : we use two passes to handle the grouping. A
|
|
1003 one pass method may be faster, but it would necessitate to
|
|
1004 modify the output stream. */
|
|
1005
|
|
1006 /* first pass: quantize */
|
|
1007 mant1_cnt = mant2_cnt = mant4_cnt = 0;
|
|
1008 qmant1_ptr = qmant2_ptr = qmant4_ptr = NULL;
|
|
1009
|
|
1010 for (ch = 0; ch < s->nb_channels; ch++) {
|
|
1011 int b, c, e, v;
|
|
1012
|
|
1013 for(i=0;i<s->nb_coefs[ch];i++) {
|
|
1014 c = mdct_coefs[ch][i];
|
|
1015 e = encoded_exp[ch][i] - global_exp[ch];
|
|
1016 b = bap[ch][i];
|
|
1017 switch(b) {
|
|
1018 case 0:
|
|
1019 v = 0;
|
|
1020 break;
|
|
1021 case 1:
|
|
1022 v = sym_quant(c, e, 3);
|
|
1023 switch(mant1_cnt) {
|
|
1024 case 0:
|
|
1025 qmant1_ptr = &qmant[ch][i];
|
|
1026 v = 9 * v;
|
|
1027 mant1_cnt = 1;
|
|
1028 break;
|
|
1029 case 1:
|
|
1030 *qmant1_ptr += 3 * v;
|
|
1031 mant1_cnt = 2;
|
|
1032 v = 128;
|
|
1033 break;
|
|
1034 default:
|
|
1035 *qmant1_ptr += v;
|
|
1036 mant1_cnt = 0;
|
|
1037 v = 128;
|
|
1038 break;
|
|
1039 }
|
|
1040 break;
|
|
1041 case 2:
|
|
1042 v = sym_quant(c, e, 5);
|
|
1043 switch(mant2_cnt) {
|
|
1044 case 0:
|
|
1045 qmant2_ptr = &qmant[ch][i];
|
|
1046 v = 25 * v;
|
|
1047 mant2_cnt = 1;
|
|
1048 break;
|
|
1049 case 1:
|
|
1050 *qmant2_ptr += 5 * v;
|
|
1051 mant2_cnt = 2;
|
|
1052 v = 128;
|
|
1053 break;
|
|
1054 default:
|
|
1055 *qmant2_ptr += v;
|
|
1056 mant2_cnt = 0;
|
|
1057 v = 128;
|
|
1058 break;
|
|
1059 }
|
|
1060 break;
|
|
1061 case 3:
|
|
1062 v = sym_quant(c, e, 7);
|
|
1063 break;
|
|
1064 case 4:
|
|
1065 v = sym_quant(c, e, 11);
|
|
1066 switch(mant4_cnt) {
|
|
1067 case 0:
|
|
1068 qmant4_ptr = &qmant[ch][i];
|
|
1069 v = 11 * v;
|
|
1070 mant4_cnt = 1;
|
|
1071 break;
|
|
1072 default:
|
|
1073 *qmant4_ptr += v;
|
|
1074 mant4_cnt = 0;
|
|
1075 v = 128;
|
|
1076 break;
|
|
1077 }
|
|
1078 break;
|
|
1079 case 5:
|
|
1080 v = sym_quant(c, e, 15);
|
|
1081 break;
|
|
1082 case 14:
|
|
1083 v = asym_quant(c, e, 14);
|
|
1084 break;
|
|
1085 case 15:
|
|
1086 v = asym_quant(c, e, 16);
|
|
1087 break;
|
|
1088 default:
|
|
1089 v = asym_quant(c, e, b - 1);
|
|
1090 break;
|
|
1091 }
|
|
1092 qmant[ch][i] = v;
|
|
1093 }
|
|
1094 }
|
|
1095
|
|
1096 /* second pass : output the values */
|
|
1097 for (ch = 0; ch < s->nb_channels; ch++) {
|
|
1098 int b, q;
|
|
1099
|
|
1100 for(i=0;i<s->nb_coefs[ch];i++) {
|
|
1101 q = qmant[ch][i];
|
|
1102 b = bap[ch][i];
|
|
1103 switch(b) {
|
|
1104 case 0:
|
|
1105 break;
|
|
1106 case 1:
|
|
1107 if (q != 128)
|
|
1108 put_bits(&s->pb, 5, q);
|
|
1109 break;
|
|
1110 case 2:
|
|
1111 if (q != 128)
|
|
1112 put_bits(&s->pb, 7, q);
|
|
1113 break;
|
|
1114 case 3:
|
|
1115 put_bits(&s->pb, 3, q);
|
|
1116 break;
|
|
1117 case 4:
|
|
1118 if (q != 128)
|
|
1119 put_bits(&s->pb, 7, q);
|
|
1120 break;
|
|
1121 case 14:
|
|
1122 put_bits(&s->pb, 14, q);
|
|
1123 break;
|
|
1124 case 15:
|
|
1125 put_bits(&s->pb, 16, q);
|
|
1126 break;
|
|
1127 default:
|
|
1128 put_bits(&s->pb, b - 1, q);
|
|
1129 break;
|
|
1130 }
|
|
1131 }
|
|
1132 }
|
|
1133 }
|
|
1134
|
|
1135 /* compute the ac3 crc */
|
|
1136
|
|
1137 #define CRC16_POLY ((1 << 0) | (1 << 2) | (1 << 15) | (1 << 16))
|
|
1138
|
|
1139 static void ac3_crc_init(void)
|
|
1140 {
|
|
1141 unsigned int c, n, k;
|
|
1142
|
|
1143 for(n=0;n<256;n++) {
|
|
1144 c = n << 8;
|
|
1145 for (k = 0; k < 8; k++) {
|
|
1146 if (c & (1 << 15))
|
|
1147 c = ((c << 1) & 0xffff) ^ (CRC16_POLY & 0xffff);
|
|
1148 else
|
|
1149 c = c << 1;
|
|
1150 }
|
|
1151 crc_table[n] = c;
|
|
1152 }
|
|
1153 }
|
|
1154
|
|
1155 static unsigned int ac3_crc(UINT8 *data, int n, unsigned int crc)
|
|
1156 {
|
|
1157 int i;
|
|
1158 for(i=0;i<n;i++) {
|
|
1159 crc = (crc_table[data[i] ^ (crc >> 8)] ^ (crc << 8)) & 0xffff;
|
|
1160 }
|
|
1161 return crc;
|
|
1162 }
|
|
1163
|
|
1164 static unsigned int mul_poly(unsigned int a, unsigned int b, unsigned int poly)
|
|
1165 {
|
|
1166 unsigned int c;
|
|
1167
|
|
1168 c = 0;
|
|
1169 while (a) {
|
|
1170 if (a & 1)
|
|
1171 c ^= b;
|
|
1172 a = a >> 1;
|
|
1173 b = b << 1;
|
|
1174 if (b & (1 << 16))
|
|
1175 b ^= poly;
|
|
1176 }
|
|
1177 return c;
|
|
1178 }
|
|
1179
|
|
1180 static unsigned int pow_poly(unsigned int a, unsigned int n, unsigned int poly)
|
|
1181 {
|
|
1182 unsigned int r;
|
|
1183 r = 1;
|
|
1184 while (n) {
|
|
1185 if (n & 1)
|
|
1186 r = mul_poly(r, a, poly);
|
|
1187 a = mul_poly(a, a, poly);
|
|
1188 n >>= 1;
|
|
1189 }
|
|
1190 return r;
|
|
1191 }
|
|
1192
|
|
1193
|
|
1194 /* compute log2(max(abs(tab[]))) */
|
|
1195 static int log2_tab(INT16 *tab, int n)
|
|
1196 {
|
|
1197 int i, v;
|
|
1198
|
|
1199 v = 0;
|
|
1200 for(i=0;i<n;i++) {
|
|
1201 v |= abs(tab[i]);
|
|
1202 }
|
|
1203 return log2(v);
|
|
1204 }
|
|
1205
|
|
1206 static void lshift_tab(INT16 *tab, int n, int lshift)
|
|
1207 {
|
|
1208 int i;
|
|
1209
|
|
1210 if (lshift > 0) {
|
|
1211 for(i=0;i<n;i++) {
|
|
1212 tab[i] <<= lshift;
|
|
1213 }
|
|
1214 } else if (lshift < 0) {
|
|
1215 lshift = -lshift;
|
|
1216 for(i=0;i<n;i++) {
|
|
1217 tab[i] >>= lshift;
|
|
1218 }
|
|
1219 }
|
|
1220 }
|
|
1221
|
|
1222 /* fill the end of the frame and compute the two crcs */
|
|
1223 static int output_frame_end(AC3EncodeContext *s)
|
|
1224 {
|
|
1225 int frame_size, frame_size_58, n, crc1, crc2, crc_inv;
|
|
1226 UINT8 *frame;
|
|
1227
|
|
1228 frame_size = s->frame_size; /* frame size in words */
|
|
1229 /* align to 8 bits */
|
|
1230 flush_put_bits(&s->pb);
|
|
1231 /* add zero bytes to reach the frame size */
|
|
1232 frame = s->pb.buf;
|
|
1233 n = 2 * s->frame_size - (s->pb.buf_ptr - frame) - 2;
|
|
1234 assert(n >= 0);
|
|
1235 memset(s->pb.buf_ptr, 0, n);
|
|
1236
|
|
1237 /* Now we must compute both crcs : this is not so easy for crc1
|
|
1238 because it is at the beginning of the data... */
|
|
1239 frame_size_58 = (frame_size >> 1) + (frame_size >> 3);
|
|
1240 crc1 = ac3_crc(frame + 4, (2 * frame_size_58) - 4, 0);
|
|
1241 /* XXX: could precompute crc_inv */
|
|
1242 crc_inv = pow_poly((CRC16_POLY >> 1), (16 * frame_size_58) - 16, CRC16_POLY);
|
|
1243 crc1 = mul_poly(crc_inv, crc1, CRC16_POLY);
|
|
1244 frame[2] = crc1 >> 8;
|
|
1245 frame[3] = crc1;
|
|
1246
|
|
1247 crc2 = ac3_crc(frame + 2 * frame_size_58, (frame_size - frame_size_58) * 2 - 2, 0);
|
|
1248 frame[2*frame_size - 2] = crc2 >> 8;
|
|
1249 frame[2*frame_size - 1] = crc2;
|
|
1250
|
|
1251 // printf("n=%d frame_size=%d\n", n, frame_size);
|
|
1252 return frame_size * 2;
|
|
1253 }
|
|
1254
|
|
1255 int AC3_encode_frame(AVCodecContext *avctx,
|
|
1256 unsigned char *frame, int buf_size, void *data)
|
|
1257 {
|
|
1258 AC3EncodeContext *s = avctx->priv_data;
|
|
1259 short *samples = data;
|
|
1260 int i, j, k, v, ch;
|
|
1261 INT16 input_samples[N];
|
|
1262 INT32 mdct_coef[NB_BLOCKS][AC3_MAX_CHANNELS][N/2];
|
|
1263 UINT8 exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2];
|
|
1264 UINT8 exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS];
|
|
1265 UINT8 encoded_exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2];
|
|
1266 UINT8 bap[NB_BLOCKS][AC3_MAX_CHANNELS][N/2];
|
|
1267 INT8 exp_samples[NB_BLOCKS][AC3_MAX_CHANNELS];
|
|
1268 int frame_bits;
|
|
1269
|
|
1270 frame_bits = 0;
|
|
1271 for(ch=0;ch<s->nb_channels;ch++) {
|
|
1272 /* fixed mdct to the six sub blocks & exponent computation */
|
|
1273 for(i=0;i<NB_BLOCKS;i++) {
|
|
1274 INT16 *sptr;
|
|
1275 int sinc;
|
|
1276
|
|
1277 /* compute input samples */
|
|
1278 memcpy(input_samples, s->last_samples[ch], N/2 * sizeof(INT16));
|
|
1279 sinc = s->nb_channels;
|
|
1280 sptr = samples + (sinc * (N/2) * i) + ch;
|
|
1281 for(j=0;j<N/2;j++) {
|
|
1282 v = *sptr;
|
|
1283 input_samples[j + N/2] = v;
|
|
1284 s->last_samples[ch][j] = v;
|
|
1285 sptr += sinc;
|
|
1286 }
|
|
1287
|
|
1288 /* apply the MDCT window */
|
|
1289 for(j=0;j<N/2;j++) {
|
|
1290 input_samples[j] = MUL16(input_samples[j],
|
|
1291 ac3_window[j]) >> 15;
|
|
1292 input_samples[N-j-1] = MUL16(input_samples[N-j-1],
|
|
1293 ac3_window[j]) >> 15;
|
|
1294 }
|
|
1295
|
|
1296 /* Normalize the samples to use the maximum available
|
|
1297 precision */
|
|
1298 v = 14 - log2_tab(input_samples, N);
|
|
1299 if (v < 0)
|
|
1300 v = 0;
|
|
1301 exp_samples[i][ch] = v - 8;
|
|
1302 lshift_tab(input_samples, N, v);
|
|
1303
|
|
1304 /* do the MDCT */
|
|
1305 mdct512(mdct_coef[i][ch], input_samples);
|
|
1306
|
|
1307 /* compute "exponents". We take into account the
|
|
1308 normalization there */
|
|
1309 for(j=0;j<N/2;j++) {
|
|
1310 int e;
|
|
1311 v = abs(mdct_coef[i][ch][j]);
|
|
1312 if (v == 0)
|
|
1313 e = 24;
|
|
1314 else {
|
|
1315 e = 23 - log2(v) + exp_samples[i][ch];
|
|
1316 if (e >= 24) {
|
|
1317 e = 24;
|
|
1318 mdct_coef[i][ch][j] = 0;
|
|
1319 }
|
|
1320 }
|
|
1321 exp[i][ch][j] = e;
|
|
1322 }
|
|
1323 }
|
|
1324
|
|
1325 compute_exp_strategy(exp_strategy, exp, ch);
|
|
1326
|
|
1327 /* compute the exponents as the decoder will see them. The
|
|
1328 EXP_REUSE case must be handled carefully : we select the
|
|
1329 min of the exponents */
|
|
1330 i = 0;
|
|
1331 while (i < NB_BLOCKS) {
|
|
1332 j = i + 1;
|
|
1333 while (j < NB_BLOCKS && exp_strategy[j][ch] == EXP_REUSE) {
|
|
1334 exponent_min(exp[i][ch], exp[j][ch], s->nb_coefs[ch]);
|
|
1335 j++;
|
|
1336 }
|
|
1337 frame_bits += encode_exp(encoded_exp[i][ch],
|
|
1338 exp[i][ch], s->nb_coefs[ch],
|
|
1339 exp_strategy[i][ch]);
|
|
1340 /* copy encoded exponents for reuse case */
|
|
1341 for(k=i+1;k<j;k++) {
|
|
1342 memcpy(encoded_exp[k][ch], encoded_exp[i][ch],
|
|
1343 s->nb_coefs[ch] * sizeof(UINT8));
|
|
1344 }
|
|
1345 i = j;
|
|
1346 }
|
|
1347 }
|
|
1348
|
|
1349 compute_bit_allocation(s, bap, encoded_exp, exp_strategy, frame_bits);
|
|
1350 /* everything is known... let's output the frame */
|
|
1351 output_frame_header(s, frame);
|
|
1352
|
|
1353 for(i=0;i<NB_BLOCKS;i++) {
|
|
1354 output_audio_block(s, exp_strategy[i], encoded_exp[i],
|
|
1355 bap[i], mdct_coef[i], exp_samples[i], i);
|
|
1356 }
|
|
1357 return output_frame_end(s);
|
|
1358 }
|
|
1359
|
|
1360 #if 0
|
|
1361 /*************************************************************************/
|
|
1362 /* TEST */
|
|
1363
|
|
1364 #define FN (N/4)
|
|
1365
|
|
1366 void fft_test(void)
|
|
1367 {
|
|
1368 IComplex in[FN], in1[FN];
|
|
1369 int k, n, i;
|
|
1370 float sum_re, sum_im, a;
|
|
1371
|
|
1372 /* FFT test */
|
|
1373
|
|
1374 for(i=0;i<FN;i++) {
|
|
1375 in[i].re = random() % 65535 - 32767;
|
|
1376 in[i].im = random() % 65535 - 32767;
|
|
1377 in1[i] = in[i];
|
|
1378 }
|
|
1379 fft(in, 7);
|
|
1380
|
|
1381 /* do it by hand */
|
|
1382 for(k=0;k<FN;k++) {
|
|
1383 sum_re = 0;
|
|
1384 sum_im = 0;
|
|
1385 for(n=0;n<FN;n++) {
|
|
1386 a = -2 * M_PI * (n * k) / FN;
|
|
1387 sum_re += in1[n].re * cos(a) - in1[n].im * sin(a);
|
|
1388 sum_im += in1[n].re * sin(a) + in1[n].im * cos(a);
|
|
1389 }
|
|
1390 printf("%3d: %6d,%6d %6.0f,%6.0f\n",
|
|
1391 k, in[k].re, in[k].im, sum_re / FN, sum_im / FN);
|
|
1392 }
|
|
1393 }
|
|
1394
|
|
1395 void mdct_test(void)
|
|
1396 {
|
|
1397 INT16 input[N];
|
|
1398 INT32 output[N/2];
|
|
1399 float input1[N];
|
|
1400 float output1[N/2];
|
|
1401 float s, a, err, e, emax;
|
|
1402 int i, k, n;
|
|
1403
|
|
1404 for(i=0;i<N;i++) {
|
|
1405 input[i] = (random() % 65535 - 32767) * 9 / 10;
|
|
1406 input1[i] = input[i];
|
|
1407 }
|
|
1408
|
|
1409 mdct512(output, input);
|
|
1410
|
|
1411 /* do it by hand */
|
|
1412 for(k=0;k<N/2;k++) {
|
|
1413 s = 0;
|
|
1414 for(n=0;n<N;n++) {
|
|
1415 a = (2*M_PI*(2*n+1+N/2)*(2*k+1) / (4 * N));
|
|
1416 s += input1[n] * cos(a);
|
|
1417 }
|
|
1418 output1[k] = -2 * s / N;
|
|
1419 }
|
|
1420
|
|
1421 err = 0;
|
|
1422 emax = 0;
|
|
1423 for(i=0;i<N/2;i++) {
|
|
1424 printf("%3d: %7d %7.0f\n", i, output[i], output1[i]);
|
|
1425 e = output[i] - output1[i];
|
|
1426 if (e > emax)
|
|
1427 emax = e;
|
|
1428 err += e * e;
|
|
1429 }
|
|
1430 printf("err2=%f emax=%f\n", err / (N/2), emax);
|
|
1431 }
|
|
1432
|
|
1433 void test_ac3(void)
|
|
1434 {
|
|
1435 AC3EncodeContext ctx;
|
|
1436 unsigned char frame[AC3_MAX_CODED_FRAME_SIZE];
|
|
1437 short samples[AC3_FRAME_SIZE];
|
|
1438 int ret, i;
|
|
1439
|
|
1440 AC3_encode_init(&ctx, 44100, 64000, 1);
|
|
1441
|
|
1442 fft_test();
|
|
1443 mdct_test();
|
|
1444
|
|
1445 for(i=0;i<AC3_FRAME_SIZE;i++)
|
|
1446 samples[i] = (int)(sin(2*M_PI*i*1000.0/44100) * 10000);
|
|
1447 ret = AC3_encode_frame(&ctx, frame, samples);
|
|
1448 printf("ret=%d\n", ret);
|
|
1449 }
|
|
1450 #endif
|
|
1451
|
|
1452 AVCodec ac3_encoder = {
|
|
1453 "ac3",
|
|
1454 CODEC_TYPE_AUDIO,
|
|
1455 CODEC_ID_AC3,
|
|
1456 sizeof(AC3EncodeContext),
|
|
1457 AC3_encode_init,
|
|
1458 AC3_encode_frame,
|
|
1459 NULL,
|
|
1460 };
|