Move ff_reverse in libavcodec to av_reverse in libavutil.
[libav.git] / libavcodec / ac3enc.c
1 /*
2 * The simplest AC-3 encoder
3 * Copyright (c) 2000 Fabrice Bellard
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22 /**
23 * @file libavcodec/ac3enc.c
24 * The simplest AC-3 encoder.
25 */
26 //#define DEBUG
27 //#define DEBUG_BITALLOC
28 #include "libavutil/crc.h"
29 #include "avcodec.h"
30 #include "libavutil/common.h" /* for av_reverse */
31 #include "put_bits.h"
32 #include "ac3.h"
33 #include "audioconvert.h"
34
35 typedef struct AC3EncodeContext {
36 PutBitContext pb;
37 int nb_channels;
38 int nb_all_channels;
39 int lfe_channel;
40 const uint8_t *channel_map;
41 int bit_rate;
42 unsigned int sample_rate;
43 unsigned int bitstream_id;
44 unsigned int frame_size_min; /* minimum frame size in case rounding is necessary */
45 unsigned int frame_size; /* current frame size in words */
46 unsigned int bits_written;
47 unsigned int samples_written;
48 int sr_shift;
49 unsigned int frame_size_code;
50 unsigned int sr_code; /* frequency */
51 unsigned int channel_mode;
52 int lfe;
53 unsigned int bitstream_mode;
54 short last_samples[AC3_MAX_CHANNELS][256];
55 unsigned int chbwcod[AC3_MAX_CHANNELS];
56 int nb_coefs[AC3_MAX_CHANNELS];
57
58 /* bitrate allocation control */
59 int slow_gain_code, slow_decay_code, fast_decay_code, db_per_bit_code, floor_code;
60 AC3BitAllocParameters bit_alloc;
61 int coarse_snr_offset;
62 int fast_gain_code[AC3_MAX_CHANNELS];
63 int fine_snr_offset[AC3_MAX_CHANNELS];
64 /* mantissa encoding */
65 int mant1_cnt, mant2_cnt, mant4_cnt;
66 } AC3EncodeContext;
67
68 static int16_t costab[64];
69 static int16_t sintab[64];
70 static int16_t xcos1[128];
71 static int16_t xsin1[128];
72
73 #define MDCT_NBITS 9
74 #define N (1 << MDCT_NBITS)
75
76 /* new exponents are sent if their Norm 1 exceed this number */
77 #define EXP_DIFF_THRESHOLD 1000
78
79 static inline int16_t fix15(float a)
80 {
81 int v;
82 v = (int)(a * (float)(1 << 15));
83 if (v < -32767)
84 v = -32767;
85 else if (v > 32767)
86 v = 32767;
87 return v;
88 }
89
90 typedef struct IComplex {
91 short re,im;
92 } IComplex;
93
94 static av_cold void fft_init(int ln)
95 {
96 int i, n;
97 float alpha;
98
99 n = 1 << ln;
100
101 for(i=0;i<(n/2);i++) {
102 alpha = 2 * M_PI * (float)i / (float)n;
103 costab[i] = fix15(cos(alpha));
104 sintab[i] = fix15(sin(alpha));
105 }
106 }
107
108 /* butter fly op */
109 #define BF(pre, pim, qre, qim, pre1, pim1, qre1, qim1) \
110 {\
111 int ax, ay, bx, by;\
112 bx=pre1;\
113 by=pim1;\
114 ax=qre1;\
115 ay=qim1;\
116 pre = (bx + ax) >> 1;\
117 pim = (by + ay) >> 1;\
118 qre = (bx - ax) >> 1;\
119 qim = (by - ay) >> 1;\
120 }
121
122 #define CMUL(pre, pim, are, aim, bre, bim) \
123 {\
124 pre = (MUL16(are, bre) - MUL16(aim, bim)) >> 15;\
125 pim = (MUL16(are, bim) + MUL16(bre, aim)) >> 15;\
126 }
127
128
129 /* do a 2^n point complex fft on 2^ln points. */
130 static void fft(IComplex *z, int ln)
131 {
132 int j, l, np, np2;
133 int nblocks, nloops;
134 register IComplex *p,*q;
135 int tmp_re, tmp_im;
136
137 np = 1 << ln;
138
139 /* reverse */
140 for(j=0;j<np;j++) {
141 int k = av_reverse[j] >> (8 - ln);
142 if (k < j)
143 FFSWAP(IComplex, z[k], z[j]);
144 }
145
146 /* pass 0 */
147
148 p=&z[0];
149 j=(np >> 1);
150 do {
151 BF(p[0].re, p[0].im, p[1].re, p[1].im,
152 p[0].re, p[0].im, p[1].re, p[1].im);
153 p+=2;
154 } while (--j != 0);
155
156 /* pass 1 */
157
158 p=&z[0];
159 j=np >> 2;
160 do {
161 BF(p[0].re, p[0].im, p[2].re, p[2].im,
162 p[0].re, p[0].im, p[2].re, p[2].im);
163 BF(p[1].re, p[1].im, p[3].re, p[3].im,
164 p[1].re, p[1].im, p[3].im, -p[3].re);
165 p+=4;
166 } while (--j != 0);
167
168 /* pass 2 .. ln-1 */
169
170 nblocks = np >> 3;
171 nloops = 1 << 2;
172 np2 = np >> 1;
173 do {
174 p = z;
175 q = z + nloops;
176 for (j = 0; j < nblocks; ++j) {
177
178 BF(p->re, p->im, q->re, q->im,
179 p->re, p->im, q->re, q->im);
180
181 p++;
182 q++;
183 for(l = nblocks; l < np2; l += nblocks) {
184 CMUL(tmp_re, tmp_im, costab[l], -sintab[l], q->re, q->im);
185 BF(p->re, p->im, q->re, q->im,
186 p->re, p->im, tmp_re, tmp_im);
187 p++;
188 q++;
189 }
190 p += nloops;
191 q += nloops;
192 }
193 nblocks = nblocks >> 1;
194 nloops = nloops << 1;
195 } while (nblocks != 0);
196 }
197
198 /* do a 512 point mdct */
199 static void mdct512(int32_t *out, int16_t *in)
200 {
201 int i, re, im, re1, im1;
202 int16_t rot[N];
203 IComplex x[N/4];
204
205 /* shift to simplify computations */
206 for(i=0;i<N/4;i++)
207 rot[i] = -in[i + 3*N/4];
208 for(i=N/4;i<N;i++)
209 rot[i] = in[i - N/4];
210
211 /* pre rotation */
212 for(i=0;i<N/4;i++) {
213 re = ((int)rot[2*i] - (int)rot[N-1-2*i]) >> 1;
214 im = -((int)rot[N/2+2*i] - (int)rot[N/2-1-2*i]) >> 1;
215 CMUL(x[i].re, x[i].im, re, im, -xcos1[i], xsin1[i]);
216 }
217
218 fft(x, MDCT_NBITS - 2);
219
220 /* post rotation */
221 for(i=0;i<N/4;i++) {
222 re = x[i].re;
223 im = x[i].im;
224 CMUL(re1, im1, re, im, xsin1[i], xcos1[i]);
225 out[2*i] = im1;
226 out[N/2-1-2*i] = re1;
227 }
228 }
229
230 /* XXX: use another norm ? */
231 static int calc_exp_diff(uint8_t *exp1, uint8_t *exp2, int n)
232 {
233 int sum, i;
234 sum = 0;
235 for(i=0;i<n;i++) {
236 sum += abs(exp1[i] - exp2[i]);
237 }
238 return sum;
239 }
240
241 static void compute_exp_strategy(uint8_t exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS],
242 uint8_t exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
243 int ch, int is_lfe)
244 {
245 int i, j;
246 int exp_diff;
247
248 /* estimate if the exponent variation & decide if they should be
249 reused in the next frame */
250 exp_strategy[0][ch] = EXP_NEW;
251 for(i=1;i<NB_BLOCKS;i++) {
252 exp_diff = calc_exp_diff(exp[i][ch], exp[i-1][ch], N/2);
253 dprintf(NULL, "exp_diff=%d\n", exp_diff);
254 if (exp_diff > EXP_DIFF_THRESHOLD)
255 exp_strategy[i][ch] = EXP_NEW;
256 else
257 exp_strategy[i][ch] = EXP_REUSE;
258 }
259 if (is_lfe)
260 return;
261
262 /* now select the encoding strategy type : if exponents are often
263 recoded, we use a coarse encoding */
264 i = 0;
265 while (i < NB_BLOCKS) {
266 j = i + 1;
267 while (j < NB_BLOCKS && exp_strategy[j][ch] == EXP_REUSE)
268 j++;
269 switch(j - i) {
270 case 1:
271 exp_strategy[i][ch] = EXP_D45;
272 break;
273 case 2:
274 case 3:
275 exp_strategy[i][ch] = EXP_D25;
276 break;
277 default:
278 exp_strategy[i][ch] = EXP_D15;
279 break;
280 }
281 i = j;
282 }
283 }
284
285 /* set exp[i] to min(exp[i], exp1[i]) */
286 static void exponent_min(uint8_t exp[N/2], uint8_t exp1[N/2], int n)
287 {
288 int i;
289
290 for(i=0;i<n;i++) {
291 if (exp1[i] < exp[i])
292 exp[i] = exp1[i];
293 }
294 }
295
296 /* update the exponents so that they are the ones the decoder will
297 decode. Return the number of bits used to code the exponents */
298 static int encode_exp(uint8_t encoded_exp[N/2],
299 uint8_t exp[N/2],
300 int nb_exps,
301 int exp_strategy)
302 {
303 int group_size, nb_groups, i, j, k, exp_min;
304 uint8_t exp1[N/2];
305
306 switch(exp_strategy) {
307 case EXP_D15:
308 group_size = 1;
309 break;
310 case EXP_D25:
311 group_size = 2;
312 break;
313 default:
314 case EXP_D45:
315 group_size = 4;
316 break;
317 }
318 nb_groups = ((nb_exps + (group_size * 3) - 4) / (3 * group_size)) * 3;
319
320 /* for each group, compute the minimum exponent */
321 exp1[0] = exp[0]; /* DC exponent is handled separately */
322 k = 1;
323 for(i=1;i<=nb_groups;i++) {
324 exp_min = exp[k];
325 assert(exp_min >= 0 && exp_min <= 24);
326 for(j=1;j<group_size;j++) {
327 if (exp[k+j] < exp_min)
328 exp_min = exp[k+j];
329 }
330 exp1[i] = exp_min;
331 k += group_size;
332 }
333
334 /* constraint for DC exponent */
335 if (exp1[0] > 15)
336 exp1[0] = 15;
337
338 /* Decrease the delta between each groups to within 2
339 * so that they can be differentially encoded */
340 for (i=1;i<=nb_groups;i++)
341 exp1[i] = FFMIN(exp1[i], exp1[i-1] + 2);
342 for (i=nb_groups-1;i>=0;i--)
343 exp1[i] = FFMIN(exp1[i], exp1[i+1] + 2);
344
345 /* now we have the exponent values the decoder will see */
346 encoded_exp[0] = exp1[0];
347 k = 1;
348 for(i=1;i<=nb_groups;i++) {
349 for(j=0;j<group_size;j++) {
350 encoded_exp[k+j] = exp1[i];
351 }
352 k += group_size;
353 }
354
355 #if defined(DEBUG)
356 av_log(NULL, AV_LOG_DEBUG, "exponents: strategy=%d\n", exp_strategy);
357 for(i=0;i<=nb_groups * group_size;i++) {
358 av_log(NULL, AV_LOG_DEBUG, "%d ", encoded_exp[i]);
359 }
360 av_log(NULL, AV_LOG_DEBUG, "\n");
361 #endif
362
363 return 4 + (nb_groups / 3) * 7;
364 }
365
366 /* return the size in bits taken by the mantissa */
367 static int compute_mantissa_size(AC3EncodeContext *s, uint8_t *m, int nb_coefs)
368 {
369 int bits, mant, i;
370
371 bits = 0;
372 for(i=0;i<nb_coefs;i++) {
373 mant = m[i];
374 switch(mant) {
375 case 0:
376 /* nothing */
377 break;
378 case 1:
379 /* 3 mantissa in 5 bits */
380 if (s->mant1_cnt == 0)
381 bits += 5;
382 if (++s->mant1_cnt == 3)
383 s->mant1_cnt = 0;
384 break;
385 case 2:
386 /* 3 mantissa in 7 bits */
387 if (s->mant2_cnt == 0)
388 bits += 7;
389 if (++s->mant2_cnt == 3)
390 s->mant2_cnt = 0;
391 break;
392 case 3:
393 bits += 3;
394 break;
395 case 4:
396 /* 2 mantissa in 7 bits */
397 if (s->mant4_cnt == 0)
398 bits += 7;
399 if (++s->mant4_cnt == 2)
400 s->mant4_cnt = 0;
401 break;
402 case 14:
403 bits += 14;
404 break;
405 case 15:
406 bits += 16;
407 break;
408 default:
409 bits += mant - 1;
410 break;
411 }
412 }
413 return bits;
414 }
415
416
417 static void bit_alloc_masking(AC3EncodeContext *s,
418 uint8_t encoded_exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
419 uint8_t exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS],
420 int16_t psd[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
421 int16_t mask[NB_BLOCKS][AC3_MAX_CHANNELS][50])
422 {
423 int blk, ch;
424 int16_t band_psd[NB_BLOCKS][AC3_MAX_CHANNELS][50];
425
426 for(blk=0; blk<NB_BLOCKS; blk++) {
427 for(ch=0;ch<s->nb_all_channels;ch++) {
428 if(exp_strategy[blk][ch] == EXP_REUSE) {
429 memcpy(psd[blk][ch], psd[blk-1][ch], (N/2)*sizeof(int16_t));
430 memcpy(mask[blk][ch], mask[blk-1][ch], 50*sizeof(int16_t));
431 } else {
432 ff_ac3_bit_alloc_calc_psd(encoded_exp[blk][ch], 0,
433 s->nb_coefs[ch],
434 psd[blk][ch], band_psd[blk][ch]);
435 ff_ac3_bit_alloc_calc_mask(&s->bit_alloc, band_psd[blk][ch],
436 0, s->nb_coefs[ch],
437 ff_ac3_fast_gain_tab[s->fast_gain_code[ch]],
438 ch == s->lfe_channel,
439 DBA_NONE, 0, NULL, NULL, NULL,
440 mask[blk][ch]);
441 }
442 }
443 }
444 }
445
446 static int bit_alloc(AC3EncodeContext *s,
447 int16_t mask[NB_BLOCKS][AC3_MAX_CHANNELS][50],
448 int16_t psd[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
449 uint8_t bap[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
450 int frame_bits, int coarse_snr_offset, int fine_snr_offset)
451 {
452 int i, ch;
453 int snr_offset;
454
455 snr_offset = (((coarse_snr_offset - 15) << 4) + fine_snr_offset) << 2;
456
457 /* compute size */
458 for(i=0;i<NB_BLOCKS;i++) {
459 s->mant1_cnt = 0;
460 s->mant2_cnt = 0;
461 s->mant4_cnt = 0;
462 for(ch=0;ch<s->nb_all_channels;ch++) {
463 ff_ac3_bit_alloc_calc_bap(mask[i][ch], psd[i][ch], 0,
464 s->nb_coefs[ch], snr_offset,
465 s->bit_alloc.floor, ff_ac3_bap_tab,
466 bap[i][ch]);
467 frame_bits += compute_mantissa_size(s, bap[i][ch],
468 s->nb_coefs[ch]);
469 }
470 }
471 #if 0
472 printf("csnr=%d fsnr=%d frame_bits=%d diff=%d\n",
473 coarse_snr_offset, fine_snr_offset, frame_bits,
474 16 * s->frame_size - ((frame_bits + 7) & ~7));
475 #endif
476 return 16 * s->frame_size - frame_bits;
477 }
478
479 #define SNR_INC1 4
480
481 static int compute_bit_allocation(AC3EncodeContext *s,
482 uint8_t bap[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
483 uint8_t encoded_exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
484 uint8_t exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS],
485 int frame_bits)
486 {
487 int i, ch;
488 int coarse_snr_offset, fine_snr_offset;
489 uint8_t bap1[NB_BLOCKS][AC3_MAX_CHANNELS][N/2];
490 int16_t psd[NB_BLOCKS][AC3_MAX_CHANNELS][N/2];
491 int16_t mask[NB_BLOCKS][AC3_MAX_CHANNELS][50];
492 static const int frame_bits_inc[8] = { 0, 0, 2, 2, 2, 4, 2, 4 };
493
494 /* init default parameters */
495 s->slow_decay_code = 2;
496 s->fast_decay_code = 1;
497 s->slow_gain_code = 1;
498 s->db_per_bit_code = 2;
499 s->floor_code = 4;
500 for(ch=0;ch<s->nb_all_channels;ch++)
501 s->fast_gain_code[ch] = 4;
502
503 /* compute real values */
504 s->bit_alloc.sr_code = s->sr_code;
505 s->bit_alloc.sr_shift = s->sr_shift;
506 s->bit_alloc.slow_decay = ff_ac3_slow_decay_tab[s->slow_decay_code] >> s->sr_shift;
507 s->bit_alloc.fast_decay = ff_ac3_fast_decay_tab[s->fast_decay_code] >> s->sr_shift;
508 s->bit_alloc.slow_gain = ff_ac3_slow_gain_tab[s->slow_gain_code];
509 s->bit_alloc.db_per_bit = ff_ac3_db_per_bit_tab[s->db_per_bit_code];
510 s->bit_alloc.floor = ff_ac3_floor_tab[s->floor_code];
511
512 /* header size */
513 frame_bits += 65;
514 // if (s->channel_mode == 2)
515 // frame_bits += 2;
516 frame_bits += frame_bits_inc[s->channel_mode];
517
518 /* audio blocks */
519 for(i=0;i<NB_BLOCKS;i++) {
520 frame_bits += s->nb_channels * 2 + 2; /* blksw * c, dithflag * c, dynrnge, cplstre */
521 if (s->channel_mode == AC3_CHMODE_STEREO) {
522 frame_bits++; /* rematstr */
523 if(i==0) frame_bits += 4;
524 }
525 frame_bits += 2 * s->nb_channels; /* chexpstr[2] * c */
526 if (s->lfe)
527 frame_bits++; /* lfeexpstr */
528 for(ch=0;ch<s->nb_channels;ch++) {
529 if (exp_strategy[i][ch] != EXP_REUSE)
530 frame_bits += 6 + 2; /* chbwcod[6], gainrng[2] */
531 }
532 frame_bits++; /* baie */
533 frame_bits++; /* snr */
534 frame_bits += 2; /* delta / skip */
535 }
536 frame_bits++; /* cplinu for block 0 */
537 /* bit alloc info */
538 /* sdcycod[2], fdcycod[2], sgaincod[2], dbpbcod[2], floorcod[3] */
539 /* csnroffset[6] */
540 /* (fsnoffset[4] + fgaincod[4]) * c */
541 frame_bits += 2*4 + 3 + 6 + s->nb_all_channels * (4 + 3);
542
543 /* auxdatae, crcrsv */
544 frame_bits += 2;
545
546 /* CRC */
547 frame_bits += 16;
548
549 /* calculate psd and masking curve before doing bit allocation */
550 bit_alloc_masking(s, encoded_exp, exp_strategy, psd, mask);
551
552 /* now the big work begins : do the bit allocation. Modify the snr
553 offset until we can pack everything in the requested frame size */
554
555 coarse_snr_offset = s->coarse_snr_offset;
556 while (coarse_snr_offset >= 0 &&
557 bit_alloc(s, mask, psd, bap, frame_bits, coarse_snr_offset, 0) < 0)
558 coarse_snr_offset -= SNR_INC1;
559 if (coarse_snr_offset < 0) {
560 av_log(NULL, AV_LOG_ERROR, "Bit allocation failed. Try increasing the bitrate.\n");
561 return -1;
562 }
563 while ((coarse_snr_offset + SNR_INC1) <= 63 &&
564 bit_alloc(s, mask, psd, bap1, frame_bits,
565 coarse_snr_offset + SNR_INC1, 0) >= 0) {
566 coarse_snr_offset += SNR_INC1;
567 memcpy(bap, bap1, sizeof(bap1));
568 }
569 while ((coarse_snr_offset + 1) <= 63 &&
570 bit_alloc(s, mask, psd, bap1, frame_bits, coarse_snr_offset + 1, 0) >= 0) {
571 coarse_snr_offset++;
572 memcpy(bap, bap1, sizeof(bap1));
573 }
574
575 fine_snr_offset = 0;
576 while ((fine_snr_offset + SNR_INC1) <= 15 &&
577 bit_alloc(s, mask, psd, bap1, frame_bits,
578 coarse_snr_offset, fine_snr_offset + SNR_INC1) >= 0) {
579 fine_snr_offset += SNR_INC1;
580 memcpy(bap, bap1, sizeof(bap1));
581 }
582 while ((fine_snr_offset + 1) <= 15 &&
583 bit_alloc(s, mask, psd, bap1, frame_bits,
584 coarse_snr_offset, fine_snr_offset + 1) >= 0) {
585 fine_snr_offset++;
586 memcpy(bap, bap1, sizeof(bap1));
587 }
588
589 s->coarse_snr_offset = coarse_snr_offset;
590 for(ch=0;ch<s->nb_all_channels;ch++)
591 s->fine_snr_offset[ch] = fine_snr_offset;
592 #if defined(DEBUG_BITALLOC)
593 {
594 int j;
595
596 for(i=0;i<6;i++) {
597 for(ch=0;ch<s->nb_all_channels;ch++) {
598 printf("Block #%d Ch%d:\n", i, ch);
599 printf("bap=");
600 for(j=0;j<s->nb_coefs[ch];j++) {
601 printf("%d ",bap[i][ch][j]);
602 }
603 printf("\n");
604 }
605 }
606 }
607 #endif
608 return 0;
609 }
610
611 static av_cold int set_channel_info(AC3EncodeContext *s, int channels,
612 int64_t *channel_layout)
613 {
614 int ch_layout;
615
616 if (channels < 1 || channels > AC3_MAX_CHANNELS)
617 return -1;
618 if ((uint64_t)*channel_layout > 0x7FF)
619 return -1;
620 ch_layout = *channel_layout;
621 if (!ch_layout)
622 ch_layout = avcodec_guess_channel_layout(channels, CODEC_ID_AC3, NULL);
623 if (avcodec_channel_layout_num_channels(ch_layout) != channels)
624 return -1;
625
626 s->lfe = !!(ch_layout & CH_LOW_FREQUENCY);
627 s->nb_all_channels = channels;
628 s->nb_channels = channels - s->lfe;
629 s->lfe_channel = s->lfe ? s->nb_channels : -1;
630 if (s->lfe)
631 ch_layout -= CH_LOW_FREQUENCY;
632
633 switch (ch_layout) {
634 case CH_LAYOUT_MONO: s->channel_mode = AC3_CHMODE_MONO; break;
635 case CH_LAYOUT_STEREO: s->channel_mode = AC3_CHMODE_STEREO; break;
636 case CH_LAYOUT_SURROUND: s->channel_mode = AC3_CHMODE_3F; break;
637 case CH_LAYOUT_2_1: s->channel_mode = AC3_CHMODE_2F1R; break;
638 case CH_LAYOUT_4POINT0: s->channel_mode = AC3_CHMODE_3F1R; break;
639 case CH_LAYOUT_QUAD:
640 case CH_LAYOUT_2_2: s->channel_mode = AC3_CHMODE_2F2R; break;
641 case CH_LAYOUT_5POINT0:
642 case CH_LAYOUT_5POINT0_BACK: s->channel_mode = AC3_CHMODE_3F2R; break;
643 default:
644 return -1;
645 }
646
647 s->channel_map = ff_ac3_enc_channel_map[s->channel_mode][s->lfe];
648 *channel_layout = ch_layout;
649 if (s->lfe)
650 *channel_layout |= CH_LOW_FREQUENCY;
651
652 return 0;
653 }
654
655 static av_cold int AC3_encode_init(AVCodecContext *avctx)
656 {
657 int freq = avctx->sample_rate;
658 int bitrate = avctx->bit_rate;
659 AC3EncodeContext *s = avctx->priv_data;
660 int i, j, ch;
661 float alpha;
662 int bw_code;
663
664 avctx->frame_size = AC3_FRAME_SIZE;
665
666 ac3_common_init();
667
668 if (!avctx->channel_layout) {
669 av_log(avctx, AV_LOG_WARNING, "No channel layout specified. The "
670 "encoder will guess the layout, but it "
671 "might be incorrect.\n");
672 }
673 if (set_channel_info(s, avctx->channels, &avctx->channel_layout)) {
674 av_log(avctx, AV_LOG_ERROR, "invalid channel layout\n");
675 return -1;
676 }
677
678 /* frequency */
679 for(i=0;i<3;i++) {
680 for(j=0;j<3;j++)
681 if ((ff_ac3_sample_rate_tab[j] >> i) == freq)
682 goto found;
683 }
684 return -1;
685 found:
686 s->sample_rate = freq;
687 s->sr_shift = i;
688 s->sr_code = j;
689 s->bitstream_id = 8 + s->sr_shift;
690 s->bitstream_mode = 0; /* complete main audio service */
691
692 /* bitrate & frame size */
693 for(i=0;i<19;i++) {
694 if ((ff_ac3_bitrate_tab[i] >> s->sr_shift)*1000 == bitrate)
695 break;
696 }
697 if (i == 19)
698 return -1;
699 s->bit_rate = bitrate;
700 s->frame_size_code = i << 1;
701 s->frame_size_min = ff_ac3_frame_size_tab[s->frame_size_code][s->sr_code];
702 s->bits_written = 0;
703 s->samples_written = 0;
704 s->frame_size = s->frame_size_min;
705
706 /* bit allocation init */
707 if(avctx->cutoff) {
708 /* calculate bandwidth based on user-specified cutoff frequency */
709 int cutoff = av_clip(avctx->cutoff, 1, s->sample_rate >> 1);
710 int fbw_coeffs = cutoff * 512 / s->sample_rate;
711 bw_code = av_clip((fbw_coeffs - 73) / 3, 0, 60);
712 } else {
713 /* use default bandwidth setting */
714 /* XXX: should compute the bandwidth according to the frame
715 size, so that we avoid annoying high frequency artifacts */
716 bw_code = 50;
717 }
718 for(ch=0;ch<s->nb_channels;ch++) {
719 /* bandwidth for each channel */
720 s->chbwcod[ch] = bw_code;
721 s->nb_coefs[ch] = bw_code * 3 + 73;
722 }
723 if (s->lfe) {
724 s->nb_coefs[s->lfe_channel] = 7; /* fixed */
725 }
726 /* initial snr offset */
727 s->coarse_snr_offset = 40;
728
729 /* mdct init */
730 fft_init(MDCT_NBITS - 2);
731 for(i=0;i<N/4;i++) {
732 alpha = 2 * M_PI * (i + 1.0 / 8.0) / (float)N;
733 xcos1[i] = fix15(-cos(alpha));
734 xsin1[i] = fix15(-sin(alpha));
735 }
736
737 avctx->coded_frame= avcodec_alloc_frame();
738 avctx->coded_frame->key_frame= 1;
739
740 return 0;
741 }
742
743 /* output the AC-3 frame header */
744 static void output_frame_header(AC3EncodeContext *s, unsigned char *frame)
745 {
746 init_put_bits(&s->pb, frame, AC3_MAX_CODED_FRAME_SIZE);
747
748 put_bits(&s->pb, 16, 0x0b77); /* frame header */
749 put_bits(&s->pb, 16, 0); /* crc1: will be filled later */
750 put_bits(&s->pb, 2, s->sr_code);
751 put_bits(&s->pb, 6, s->frame_size_code + (s->frame_size - s->frame_size_min));
752 put_bits(&s->pb, 5, s->bitstream_id);
753 put_bits(&s->pb, 3, s->bitstream_mode);
754 put_bits(&s->pb, 3, s->channel_mode);
755 if ((s->channel_mode & 0x01) && s->channel_mode != AC3_CHMODE_MONO)
756 put_bits(&s->pb, 2, 1); /* XXX -4.5 dB */
757 if (s->channel_mode & 0x04)
758 put_bits(&s->pb, 2, 1); /* XXX -6 dB */
759 if (s->channel_mode == AC3_CHMODE_STEREO)
760 put_bits(&s->pb, 2, 0); /* surround not indicated */
761 put_bits(&s->pb, 1, s->lfe); /* LFE */
762 put_bits(&s->pb, 5, 31); /* dialog norm: -31 db */
763 put_bits(&s->pb, 1, 0); /* no compression control word */
764 put_bits(&s->pb, 1, 0); /* no lang code */
765 put_bits(&s->pb, 1, 0); /* no audio production info */
766 put_bits(&s->pb, 1, 0); /* no copyright */
767 put_bits(&s->pb, 1, 1); /* original bitstream */
768 put_bits(&s->pb, 1, 0); /* no time code 1 */
769 put_bits(&s->pb, 1, 0); /* no time code 2 */
770 put_bits(&s->pb, 1, 0); /* no additional bit stream info */
771 }
772
773 /* symetric quantization on 'levels' levels */
774 static inline int sym_quant(int c, int e, int levels)
775 {
776 int v;
777
778 if (c >= 0) {
779 v = (levels * (c << e)) >> 24;
780 v = (v + 1) >> 1;
781 v = (levels >> 1) + v;
782 } else {
783 v = (levels * ((-c) << e)) >> 24;
784 v = (v + 1) >> 1;
785 v = (levels >> 1) - v;
786 }
787 assert (v >= 0 && v < levels);
788 return v;
789 }
790
791 /* asymetric quantization on 2^qbits levels */
792 static inline int asym_quant(int c, int e, int qbits)
793 {
794 int lshift, m, v;
795
796 lshift = e + qbits - 24;
797 if (lshift >= 0)
798 v = c << lshift;
799 else
800 v = c >> (-lshift);
801 /* rounding */
802 v = (v + 1) >> 1;
803 m = (1 << (qbits-1));
804 if (v >= m)
805 v = m - 1;
806 assert(v >= -m);
807 return v & ((1 << qbits)-1);
808 }
809
810 /* Output one audio block. There are NB_BLOCKS audio blocks in one AC-3
811 frame */
812 static void output_audio_block(AC3EncodeContext *s,
813 uint8_t exp_strategy[AC3_MAX_CHANNELS],
814 uint8_t encoded_exp[AC3_MAX_CHANNELS][N/2],
815 uint8_t bap[AC3_MAX_CHANNELS][N/2],
816 int32_t mdct_coefs[AC3_MAX_CHANNELS][N/2],
817 int8_t global_exp[AC3_MAX_CHANNELS],
818 int block_num)
819 {
820 int ch, nb_groups, group_size, i, baie, rbnd;
821 uint8_t *p;
822 uint16_t qmant[AC3_MAX_CHANNELS][N/2];
823 int exp0, exp1;
824 int mant1_cnt, mant2_cnt, mant4_cnt;
825 uint16_t *qmant1_ptr, *qmant2_ptr, *qmant4_ptr;
826 int delta0, delta1, delta2;
827
828 for(ch=0;ch<s->nb_channels;ch++)
829 put_bits(&s->pb, 1, 0); /* 512 point MDCT */
830 for(ch=0;ch<s->nb_channels;ch++)
831 put_bits(&s->pb, 1, 1); /* no dither */
832 put_bits(&s->pb, 1, 0); /* no dynamic range */
833 if (block_num == 0) {
834 /* for block 0, even if no coupling, we must say it. This is a
835 waste of bit :-) */
836 put_bits(&s->pb, 1, 1); /* coupling strategy present */
837 put_bits(&s->pb, 1, 0); /* no coupling strategy */
838 } else {
839 put_bits(&s->pb, 1, 0); /* no new coupling strategy */
840 }
841
842 if (s->channel_mode == AC3_CHMODE_STEREO)
843 {
844 if(block_num==0)
845 {
846 /* first block must define rematrixing (rematstr) */
847 put_bits(&s->pb, 1, 1);
848
849 /* dummy rematrixing rematflg(1:4)=0 */
850 for (rbnd=0;rbnd<4;rbnd++)
851 put_bits(&s->pb, 1, 0);
852 }
853 else
854 {
855 /* no matrixing (but should be used in the future) */
856 put_bits(&s->pb, 1, 0);
857 }
858 }
859
860 #if defined(DEBUG)
861 {
862 static int count = 0;
863 av_log(NULL, AV_LOG_DEBUG, "Block #%d (%d)\n", block_num, count++);
864 }
865 #endif
866 /* exponent strategy */
867 for(ch=0;ch<s->nb_channels;ch++) {
868 put_bits(&s->pb, 2, exp_strategy[ch]);
869 }
870
871 if (s->lfe) {
872 put_bits(&s->pb, 1, exp_strategy[s->lfe_channel]);
873 }
874
875 for(ch=0;ch<s->nb_channels;ch++) {
876 if (exp_strategy[ch] != EXP_REUSE)
877 put_bits(&s->pb, 6, s->chbwcod[ch]);
878 }
879
880 /* exponents */
881 for (ch = 0; ch < s->nb_all_channels; ch++) {
882 switch(exp_strategy[ch]) {
883 case EXP_REUSE:
884 continue;
885 case EXP_D15:
886 group_size = 1;
887 break;
888 case EXP_D25:
889 group_size = 2;
890 break;
891 default:
892 case EXP_D45:
893 group_size = 4;
894 break;
895 }
896 nb_groups = (s->nb_coefs[ch] + (group_size * 3) - 4) / (3 * group_size);
897 p = encoded_exp[ch];
898
899 /* first exponent */
900 exp1 = *p++;
901 put_bits(&s->pb, 4, exp1);
902
903 /* next ones are delta encoded */
904 for(i=0;i<nb_groups;i++) {
905 /* merge three delta in one code */
906 exp0 = exp1;
907 exp1 = p[0];
908 p += group_size;
909 delta0 = exp1 - exp0 + 2;
910
911 exp0 = exp1;
912 exp1 = p[0];
913 p += group_size;
914 delta1 = exp1 - exp0 + 2;
915
916 exp0 = exp1;
917 exp1 = p[0];
918 p += group_size;
919 delta2 = exp1 - exp0 + 2;
920
921 put_bits(&s->pb, 7, ((delta0 * 5 + delta1) * 5) + delta2);
922 }
923
924 if (ch != s->lfe_channel)
925 put_bits(&s->pb, 2, 0); /* no gain range info */
926 }
927
928 /* bit allocation info */
929 baie = (block_num == 0);
930 put_bits(&s->pb, 1, baie);
931 if (baie) {
932 put_bits(&s->pb, 2, s->slow_decay_code);
933 put_bits(&s->pb, 2, s->fast_decay_code);
934 put_bits(&s->pb, 2, s->slow_gain_code);
935 put_bits(&s->pb, 2, s->db_per_bit_code);
936 put_bits(&s->pb, 3, s->floor_code);
937 }
938
939 /* snr offset */
940 put_bits(&s->pb, 1, baie); /* always present with bai */
941 if (baie) {
942 put_bits(&s->pb, 6, s->coarse_snr_offset);
943 for(ch=0;ch<s->nb_all_channels;ch++) {
944 put_bits(&s->pb, 4, s->fine_snr_offset[ch]);
945 put_bits(&s->pb, 3, s->fast_gain_code[ch]);
946 }
947 }
948
949 put_bits(&s->pb, 1, 0); /* no delta bit allocation */
950 put_bits(&s->pb, 1, 0); /* no data to skip */
951
952 /* mantissa encoding : we use two passes to handle the grouping. A
953 one pass method may be faster, but it would necessitate to
954 modify the output stream. */
955
956 /* first pass: quantize */
957 mant1_cnt = mant2_cnt = mant4_cnt = 0;
958 qmant1_ptr = qmant2_ptr = qmant4_ptr = NULL;
959
960 for (ch = 0; ch < s->nb_all_channels; ch++) {
961 int b, c, e, v;
962
963 for(i=0;i<s->nb_coefs[ch];i++) {
964 c = mdct_coefs[ch][i];
965 e = encoded_exp[ch][i] - global_exp[ch];
966 b = bap[ch][i];
967 switch(b) {
968 case 0:
969 v = 0;
970 break;
971 case 1:
972 v = sym_quant(c, e, 3);
973 switch(mant1_cnt) {
974 case 0:
975 qmant1_ptr = &qmant[ch][i];
976 v = 9 * v;
977 mant1_cnt = 1;
978 break;
979 case 1:
980 *qmant1_ptr += 3 * v;
981 mant1_cnt = 2;
982 v = 128;
983 break;
984 default:
985 *qmant1_ptr += v;
986 mant1_cnt = 0;
987 v = 128;
988 break;
989 }
990 break;
991 case 2:
992 v = sym_quant(c, e, 5);
993 switch(mant2_cnt) {
994 case 0:
995 qmant2_ptr = &qmant[ch][i];
996 v = 25 * v;
997 mant2_cnt = 1;
998 break;
999 case 1:
1000 *qmant2_ptr += 5 * v;
1001 mant2_cnt = 2;
1002 v = 128;
1003 break;
1004 default:
1005 *qmant2_ptr += v;
1006 mant2_cnt = 0;
1007 v = 128;
1008 break;
1009 }
1010 break;
1011 case 3:
1012 v = sym_quant(c, e, 7);
1013 break;
1014 case 4:
1015 v = sym_quant(c, e, 11);
1016 switch(mant4_cnt) {
1017 case 0:
1018 qmant4_ptr = &qmant[ch][i];
1019 v = 11 * v;
1020 mant4_cnt = 1;
1021 break;
1022 default:
1023 *qmant4_ptr += v;
1024 mant4_cnt = 0;
1025 v = 128;
1026 break;
1027 }
1028 break;
1029 case 5:
1030 v = sym_quant(c, e, 15);
1031 break;
1032 case 14:
1033 v = asym_quant(c, e, 14);
1034 break;
1035 case 15:
1036 v = asym_quant(c, e, 16);
1037 break;
1038 default:
1039 v = asym_quant(c, e, b - 1);
1040 break;
1041 }
1042 qmant[ch][i] = v;
1043 }
1044 }
1045
1046 /* second pass : output the values */
1047 for (ch = 0; ch < s->nb_all_channels; ch++) {
1048 int b, q;
1049
1050 for(i=0;i<s->nb_coefs[ch];i++) {
1051 q = qmant[ch][i];
1052 b = bap[ch][i];
1053 switch(b) {
1054 case 0:
1055 break;
1056 case 1:
1057 if (q != 128)
1058 put_bits(&s->pb, 5, q);
1059 break;
1060 case 2:
1061 if (q != 128)
1062 put_bits(&s->pb, 7, q);
1063 break;
1064 case 3:
1065 put_bits(&s->pb, 3, q);
1066 break;
1067 case 4:
1068 if (q != 128)
1069 put_bits(&s->pb, 7, q);
1070 break;
1071 case 14:
1072 put_bits(&s->pb, 14, q);
1073 break;
1074 case 15:
1075 put_bits(&s->pb, 16, q);
1076 break;
1077 default:
1078 put_bits(&s->pb, b - 1, q);
1079 break;
1080 }
1081 }
1082 }
1083 }
1084
1085 #define CRC16_POLY ((1 << 0) | (1 << 2) | (1 << 15) | (1 << 16))
1086
1087 static unsigned int mul_poly(unsigned int a, unsigned int b, unsigned int poly)
1088 {
1089 unsigned int c;
1090
1091 c = 0;
1092 while (a) {
1093 if (a & 1)
1094 c ^= b;
1095 a = a >> 1;
1096 b = b << 1;
1097 if (b & (1 << 16))
1098 b ^= poly;
1099 }
1100 return c;
1101 }
1102
1103 static unsigned int pow_poly(unsigned int a, unsigned int n, unsigned int poly)
1104 {
1105 unsigned int r;
1106 r = 1;
1107 while (n) {
1108 if (n & 1)
1109 r = mul_poly(r, a, poly);
1110 a = mul_poly(a, a, poly);
1111 n >>= 1;
1112 }
1113 return r;
1114 }
1115
1116
1117 /* compute log2(max(abs(tab[]))) */
1118 static int log2_tab(int16_t *tab, int n)
1119 {
1120 int i, v;
1121
1122 v = 0;
1123 for(i=0;i<n;i++) {
1124 v |= abs(tab[i]);
1125 }
1126 return av_log2(v);
1127 }
1128
1129 static void lshift_tab(int16_t *tab, int n, int lshift)
1130 {
1131 int i;
1132
1133 if (lshift > 0) {
1134 for(i=0;i<n;i++) {
1135 tab[i] <<= lshift;
1136 }
1137 } else if (lshift < 0) {
1138 lshift = -lshift;
1139 for(i=0;i<n;i++) {
1140 tab[i] >>= lshift;
1141 }
1142 }
1143 }
1144
1145 /* fill the end of the frame and compute the two crcs */
1146 static int output_frame_end(AC3EncodeContext *s)
1147 {
1148 int frame_size, frame_size_58, n, crc1, crc2, crc_inv;
1149 uint8_t *frame;
1150
1151 frame_size = s->frame_size; /* frame size in words */
1152 /* align to 8 bits */
1153 flush_put_bits(&s->pb);
1154 /* add zero bytes to reach the frame size */
1155 frame = s->pb.buf;
1156 n = 2 * s->frame_size - (put_bits_ptr(&s->pb) - frame) - 2;
1157 assert(n >= 0);
1158 if(n>0)
1159 memset(put_bits_ptr(&s->pb), 0, n);
1160
1161 /* Now we must compute both crcs : this is not so easy for crc1
1162 because it is at the beginning of the data... */
1163 frame_size_58 = (frame_size >> 1) + (frame_size >> 3);
1164 crc1 = bswap_16(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0,
1165 frame + 4, 2 * frame_size_58 - 4));
1166 /* XXX: could precompute crc_inv */
1167 crc_inv = pow_poly((CRC16_POLY >> 1), (16 * frame_size_58) - 16, CRC16_POLY);
1168 crc1 = mul_poly(crc_inv, crc1, CRC16_POLY);
1169 AV_WB16(frame+2,crc1);
1170
1171 crc2 = bswap_16(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0,
1172 frame + 2 * frame_size_58,
1173 (frame_size - frame_size_58) * 2 - 2));
1174 AV_WB16(frame+2*frame_size-2,crc2);
1175
1176 // printf("n=%d frame_size=%d\n", n, frame_size);
1177 return frame_size * 2;
1178 }
1179
1180 static int AC3_encode_frame(AVCodecContext *avctx,
1181 unsigned char *frame, int buf_size, void *data)
1182 {
1183 AC3EncodeContext *s = avctx->priv_data;
1184 int16_t *samples = data;
1185 int i, j, k, v, ch;
1186 int16_t input_samples[N];
1187 int32_t mdct_coef[NB_BLOCKS][AC3_MAX_CHANNELS][N/2];
1188 uint8_t exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2];
1189 uint8_t exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS];
1190 uint8_t encoded_exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2];
1191 uint8_t bap[NB_BLOCKS][AC3_MAX_CHANNELS][N/2];
1192 int8_t exp_samples[NB_BLOCKS][AC3_MAX_CHANNELS];
1193 int frame_bits;
1194
1195 frame_bits = 0;
1196 for(ch=0;ch<s->nb_all_channels;ch++) {
1197 int ich = s->channel_map[ch];
1198 /* fixed mdct to the six sub blocks & exponent computation */
1199 for(i=0;i<NB_BLOCKS;i++) {
1200 int16_t *sptr;
1201 int sinc;
1202
1203 /* compute input samples */
1204 memcpy(input_samples, s->last_samples[ich], N/2 * sizeof(int16_t));
1205 sinc = s->nb_all_channels;
1206 sptr = samples + (sinc * (N/2) * i) + ich;
1207 for(j=0;j<N/2;j++) {
1208 v = *sptr;
1209 input_samples[j + N/2] = v;
1210 s->last_samples[ich][j] = v;
1211 sptr += sinc;
1212 }
1213
1214 /* apply the MDCT window */
1215 for(j=0;j<N/2;j++) {
1216 input_samples[j] = MUL16(input_samples[j],
1217 ff_ac3_window[j]) >> 15;
1218 input_samples[N-j-1] = MUL16(input_samples[N-j-1],
1219 ff_ac3_window[j]) >> 15;
1220 }
1221
1222 /* Normalize the samples to use the maximum available
1223 precision */
1224 v = 14 - log2_tab(input_samples, N);
1225 if (v < 0)
1226 v = 0;
1227 exp_samples[i][ch] = v - 9;
1228 lshift_tab(input_samples, N, v);
1229
1230 /* do the MDCT */
1231 mdct512(mdct_coef[i][ch], input_samples);
1232
1233 /* compute "exponents". We take into account the
1234 normalization there */
1235 for(j=0;j<N/2;j++) {
1236 int e;
1237 v = abs(mdct_coef[i][ch][j]);
1238 if (v == 0)
1239 e = 24;
1240 else {
1241 e = 23 - av_log2(v) + exp_samples[i][ch];
1242 if (e >= 24) {
1243 e = 24;
1244 mdct_coef[i][ch][j] = 0;
1245 }
1246 }
1247 exp[i][ch][j] = e;
1248 }
1249 }
1250
1251 compute_exp_strategy(exp_strategy, exp, ch, ch == s->lfe_channel);
1252
1253 /* compute the exponents as the decoder will see them. The
1254 EXP_REUSE case must be handled carefully : we select the
1255 min of the exponents */
1256 i = 0;
1257 while (i < NB_BLOCKS) {
1258 j = i + 1;
1259 while (j < NB_BLOCKS && exp_strategy[j][ch] == EXP_REUSE) {
1260 exponent_min(exp[i][ch], exp[j][ch], s->nb_coefs[ch]);
1261 j++;
1262 }
1263 frame_bits += encode_exp(encoded_exp[i][ch],
1264 exp[i][ch], s->nb_coefs[ch],
1265 exp_strategy[i][ch]);
1266 /* copy encoded exponents for reuse case */
1267 for(k=i+1;k<j;k++) {
1268 memcpy(encoded_exp[k][ch], encoded_exp[i][ch],
1269 s->nb_coefs[ch] * sizeof(uint8_t));
1270 }
1271 i = j;
1272 }
1273 }
1274
1275 /* adjust for fractional frame sizes */
1276 while(s->bits_written >= s->bit_rate && s->samples_written >= s->sample_rate) {
1277 s->bits_written -= s->bit_rate;
1278 s->samples_written -= s->sample_rate;
1279 }
1280 s->frame_size = s->frame_size_min + (s->bits_written * s->sample_rate < s->samples_written * s->bit_rate);
1281 s->bits_written += s->frame_size * 16;
1282 s->samples_written += AC3_FRAME_SIZE;
1283
1284 compute_bit_allocation(s, bap, encoded_exp, exp_strategy, frame_bits);
1285 /* everything is known... let's output the frame */
1286 output_frame_header(s, frame);
1287
1288 for(i=0;i<NB_BLOCKS;i++) {
1289 output_audio_block(s, exp_strategy[i], encoded_exp[i],
1290 bap[i], mdct_coef[i], exp_samples[i], i);
1291 }
1292 return output_frame_end(s);
1293 }
1294
1295 static av_cold int AC3_encode_close(AVCodecContext *avctx)
1296 {
1297 av_freep(&avctx->coded_frame);
1298 return 0;
1299 }
1300
1301 #if 0
1302 /*************************************************************************/
1303 /* TEST */
1304
1305 #undef random
1306 #define FN (N/4)
1307
1308 void fft_test(void)
1309 {
1310 IComplex in[FN], in1[FN];
1311 int k, n, i;
1312 float sum_re, sum_im, a;
1313
1314 /* FFT test */
1315
1316 for(i=0;i<FN;i++) {
1317 in[i].re = random() % 65535 - 32767;
1318 in[i].im = random() % 65535 - 32767;
1319 in1[i] = in[i];
1320 }
1321 fft(in, 7);
1322
1323 /* do it by hand */
1324 for(k=0;k<FN;k++) {
1325 sum_re = 0;
1326 sum_im = 0;
1327 for(n=0;n<FN;n++) {
1328 a = -2 * M_PI * (n * k) / FN;
1329 sum_re += in1[n].re * cos(a) - in1[n].im * sin(a);
1330 sum_im += in1[n].re * sin(a) + in1[n].im * cos(a);
1331 }
1332 printf("%3d: %6d,%6d %6.0f,%6.0f\n",
1333 k, in[k].re, in[k].im, sum_re / FN, sum_im / FN);
1334 }
1335 }
1336
1337 void mdct_test(void)
1338 {
1339 int16_t input[N];
1340 int32_t output[N/2];
1341 float input1[N];
1342 float output1[N/2];
1343 float s, a, err, e, emax;
1344 int i, k, n;
1345
1346 for(i=0;i<N;i++) {
1347 input[i] = (random() % 65535 - 32767) * 9 / 10;
1348 input1[i] = input[i];
1349 }
1350
1351 mdct512(output, input);
1352
1353 /* do it by hand */
1354 for(k=0;k<N/2;k++) {
1355 s = 0;
1356 for(n=0;n<N;n++) {
1357 a = (2*M_PI*(2*n+1+N/2)*(2*k+1) / (4 * N));
1358 s += input1[n] * cos(a);
1359 }
1360 output1[k] = -2 * s / N;
1361 }
1362
1363 err = 0;
1364 emax = 0;
1365 for(i=0;i<N/2;i++) {
1366 printf("%3d: %7d %7.0f\n", i, output[i], output1[i]);
1367 e = output[i] - output1[i];
1368 if (e > emax)
1369 emax = e;
1370 err += e * e;
1371 }
1372 printf("err2=%f emax=%f\n", err / (N/2), emax);
1373 }
1374
1375 void test_ac3(void)
1376 {
1377 AC3EncodeContext ctx;
1378 unsigned char frame[AC3_MAX_CODED_FRAME_SIZE];
1379 short samples[AC3_FRAME_SIZE];
1380 int ret, i;
1381
1382 AC3_encode_init(&ctx, 44100, 64000, 1);
1383
1384 fft_test();
1385 mdct_test();
1386
1387 for(i=0;i<AC3_FRAME_SIZE;i++)
1388 samples[i] = (int)(sin(2*M_PI*i*1000.0/44100) * 10000);
1389 ret = AC3_encode_frame(&ctx, frame, samples);
1390 printf("ret=%d\n", ret);
1391 }
1392 #endif
1393
1394 AVCodec ac3_encoder = {
1395 "ac3",
1396 CODEC_TYPE_AUDIO,
1397 CODEC_ID_AC3,
1398 sizeof(AC3EncodeContext),
1399 AC3_encode_init,
1400 AC3_encode_frame,
1401 AC3_encode_close,
1402 NULL,
1403 .sample_fmts = (const enum SampleFormat[]){SAMPLE_FMT_S16,SAMPLE_FMT_NONE},
1404 .long_name = NULL_IF_CONFIG_SMALL("ATSC A/52A (AC-3)"),
1405 .channel_layouts = (const int64_t[]){
1406 CH_LAYOUT_MONO,
1407 CH_LAYOUT_STEREO,
1408 CH_LAYOUT_2_1,
1409 CH_LAYOUT_SURROUND,
1410 CH_LAYOUT_2_2,
1411 CH_LAYOUT_QUAD,
1412 CH_LAYOUT_4POINT0,
1413 CH_LAYOUT_5POINT0,
1414 CH_LAYOUT_5POINT0_BACK,
1415 (CH_LAYOUT_MONO | CH_LOW_FREQUENCY),
1416 (CH_LAYOUT_STEREO | CH_LOW_FREQUENCY),
1417 (CH_LAYOUT_2_1 | CH_LOW_FREQUENCY),
1418 (CH_LAYOUT_SURROUND | CH_LOW_FREQUENCY),
1419 (CH_LAYOUT_2_2 | CH_LOW_FREQUENCY),
1420 (CH_LAYOUT_QUAD | CH_LOW_FREQUENCY),
1421 (CH_LAYOUT_4POINT0 | CH_LOW_FREQUENCY),
1422 CH_LAYOUT_5POINT1,
1423 CH_LAYOUT_5POINT1_BACK,
1424 0 },
1425 };