?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Normalize triplets in gain_val_tab[][] so gain_exp_tab[] can be just an
[ffmpeg-lucabe.git] / libavcodec / flacenc.c
blob1bba6f84cd50c602754721210f834746071cf657
1 /**
2 * FLAC audio encoder
3 * Copyright (c) 2006 Justin Ruggles <justin.ruggles@gmail.com>
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 #include "libavutil/crc.h"
23 #include "libavutil/lls.h"
24 #include "avcodec.h"
25 #include "bitstream.h"
26 #include "dsputil.h"
27 #include "golomb.h"
28 #include "lpc.h"
29
30 #define FLAC_MAX_CH 8
31 #define FLAC_MIN_BLOCKSIZE 16
32 #define FLAC_MAX_BLOCKSIZE 65535
33
34 #define FLAC_SUBFRAME_CONSTANT 0
35 #define FLAC_SUBFRAME_VERBATIM 1
36 #define FLAC_SUBFRAME_FIXED 8
37 #define FLAC_SUBFRAME_LPC 32
38
39 #define FLAC_CHMODE_NOT_STEREO 0
40 #define FLAC_CHMODE_LEFT_RIGHT 1
41 #define FLAC_CHMODE_LEFT_SIDE 8
42 #define FLAC_CHMODE_RIGHT_SIDE 9
43 #define FLAC_CHMODE_MID_SIDE 10
44
45 #define FLAC_STREAMINFO_SIZE 34
46
47 #define MAX_FIXED_ORDER 4
48 #define MAX_PARTITION_ORDER 8
49 #define MAX_PARTITIONS (1 << MAX_PARTITION_ORDER)
50 #define MAX_LPC_PRECISION 15
51 #define MAX_LPC_SHIFT 15
52 #define MAX_RICE_PARAM 14
53
54 typedef struct CompressionOptions {
55 int compression_level;
56 int block_time_ms;
57 int use_lpc;
58 int lpc_coeff_precision;
59 int min_prediction_order;
60 int max_prediction_order;
61 int prediction_order_method;
62 int min_partition_order;
63 int max_partition_order;
64 } CompressionOptions;
65
66 typedef struct RiceContext {
67 int porder;
68 int params[MAX_PARTITIONS];
69 } RiceContext;
70
71 typedef struct FlacSubframe {
72 int type;
73 int type_code;
74 int obits;
75 int order;
76 int32_t coefs[MAX_LPC_ORDER];
77 int shift;
78 RiceContext rc;
79 int32_t samples[FLAC_MAX_BLOCKSIZE];
80 int32_t residual[FLAC_MAX_BLOCKSIZE+1];
81 } FlacSubframe;
82
83 typedef struct FlacFrame {
84 FlacSubframe subframes[FLAC_MAX_CH];
85 int blocksize;
86 int bs_code[2];
87 uint8_t crc8;
88 int ch_mode;
89 } FlacFrame;
90
91 typedef struct FlacEncodeContext {
92 PutBitContext pb;
93 int channels;
94 int ch_code;
95 int samplerate;
96 int sr_code[2];
97 int max_framesize;
98 uint32_t frame_count;
99 FlacFrame frame;
100 CompressionOptions options;
101 AVCodecContext *avctx;
102 DSPContext dsp;
103 } FlacEncodeContext;
104
105 static const int flac_samplerates[16] = {
106 0, 0, 0, 0,
107 8000, 16000, 22050, 24000, 32000, 44100, 48000, 96000,
108 0, 0, 0, 0
109 };
110
111 static const int flac_blocksizes[16] = {
112 0,
113 192,
114 576, 1152, 2304, 4608,
115 0, 0,
116 256, 512, 1024, 2048, 4096, 8192, 16384, 32768
117 };
118
119 /**
120 * Writes streaminfo metadata block to byte array
121 */
122 static void write_streaminfo(FlacEncodeContext *s, uint8_t *header)
123 {
124 PutBitContext pb;
125
126 memset(header, 0, FLAC_STREAMINFO_SIZE);
127 init_put_bits(&pb, header, FLAC_STREAMINFO_SIZE);
128
129 /* streaminfo metadata block */
130 put_bits(&pb, 16, s->avctx->frame_size);
131 put_bits(&pb, 16, s->avctx->frame_size);
132 put_bits(&pb, 24, 0);
133 put_bits(&pb, 24, s->max_framesize);
134 put_bits(&pb, 20, s->samplerate);
135 put_bits(&pb, 3, s->channels-1);
136 put_bits(&pb, 5, 15); /* bits per sample - 1 */
137 flush_put_bits(&pb);
138 /* total samples = 0 */
139 /* MD5 signature = 0 */
140 }
141
142 /**
143 * Sets blocksize based on samplerate
144 * Chooses the closest predefined blocksize >= BLOCK_TIME_MS milliseconds
145 */
146 static int select_blocksize(int samplerate, int block_time_ms)
147 {
148 int i;
149 int target;
150 int blocksize;
151
152 assert(samplerate > 0);
153 blocksize = flac_blocksizes[1];
154 target = (samplerate * block_time_ms) / 1000;
155 for(i=0; i<16; i++) {
156 if(target >= flac_blocksizes[i] && flac_blocksizes[i] > blocksize) {
157 blocksize = flac_blocksizes[i];
158 }
159 }
160 return blocksize;
161 }
162
163 static av_cold int flac_encode_init(AVCodecContext *avctx)
164 {
165 int freq = avctx->sample_rate;
166 int channels = avctx->channels;
167 FlacEncodeContext *s = avctx->priv_data;
168 int i, level;
169 uint8_t *streaminfo;
170
171 s->avctx = avctx;
172
173 dsputil_init(&s->dsp, avctx);
174
175 if(avctx->sample_fmt != SAMPLE_FMT_S16) {
176 return -1;
177 }
178
179 if(channels < 1 || channels > FLAC_MAX_CH) {
180 return -1;
181 }
182 s->channels = channels;
183 s->ch_code = s->channels-1;
184
185 /* find samplerate in table */
186 if(freq < 1)
187 return -1;
188 for(i=4; i<12; i++) {
189 if(freq == flac_samplerates[i]) {
190 s->samplerate = flac_samplerates[i];
191 s->sr_code[0] = i;
192 s->sr_code[1] = 0;
193 break;
194 }
195 }
196 /* if not in table, samplerate is non-standard */
197 if(i == 12) {
198 if(freq % 1000 == 0 && freq < 255000) {
199 s->sr_code[0] = 12;
200 s->sr_code[1] = freq / 1000;
201 } else if(freq % 10 == 0 && freq < 655350) {
202 s->sr_code[0] = 14;
203 s->sr_code[1] = freq / 10;
204 } else if(freq < 65535) {
205 s->sr_code[0] = 13;
206 s->sr_code[1] = freq;
207 } else {
208 return -1;
209 }
210 s->samplerate = freq;
211 }
212
213 /* set compression option defaults based on avctx->compression_level */
214 if(avctx->compression_level < 0) {
215 s->options.compression_level = 5;
216 } else {
217 s->options.compression_level = avctx->compression_level;
218 }
219 av_log(avctx, AV_LOG_DEBUG, " compression: %d\n", s->options.compression_level);
220
221 level= s->options.compression_level;
222 if(level > 12) {
223 av_log(avctx, AV_LOG_ERROR, "invalid compression level: %d\n",
224 s->options.compression_level);
225 return -1;
226 }
227
228 s->options.block_time_ms = ((int[]){ 27, 27, 27,105,105,105,105,105,105,105,105,105,105})[level];
229 s->options.use_lpc = ((int[]){ 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1})[level];
230 s->options.min_prediction_order= ((int[]){ 2, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1})[level];
231 s->options.max_prediction_order= ((int[]){ 3, 4, 4, 6, 8, 8, 8, 8, 12, 12, 12, 32, 32})[level];
232 s->options.prediction_order_method = ((int[]){ ORDER_METHOD_EST, ORDER_METHOD_EST, ORDER_METHOD_EST,
233 ORDER_METHOD_EST, ORDER_METHOD_EST, ORDER_METHOD_EST,
234 ORDER_METHOD_4LEVEL, ORDER_METHOD_LOG, ORDER_METHOD_4LEVEL,
235 ORDER_METHOD_LOG, ORDER_METHOD_SEARCH, ORDER_METHOD_LOG,
236 ORDER_METHOD_SEARCH})[level];
237 s->options.min_partition_order = ((int[]){ 2, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0})[level];
238 s->options.max_partition_order = ((int[]){ 2, 2, 3, 3, 3, 8, 8, 8, 8, 8, 8, 8, 8})[level];
239
240 /* set compression option overrides from AVCodecContext */
241 if(avctx->use_lpc >= 0) {
242 s->options.use_lpc = av_clip(avctx->use_lpc, 0, 11);
243 }
244 if(s->options.use_lpc == 1)
245 av_log(avctx, AV_LOG_DEBUG, " use lpc: Levinson-Durbin recursion with Welch window\n");
246 else if(s->options.use_lpc > 1)
247 av_log(avctx, AV_LOG_DEBUG, " use lpc: Cholesky factorization\n");
248
249 if(avctx->min_prediction_order >= 0) {
250 if(s->options.use_lpc) {
251 if(avctx->min_prediction_order < MIN_LPC_ORDER ||
252 avctx->min_prediction_order > MAX_LPC_ORDER) {
253 av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n",
254 avctx->min_prediction_order);
255 return -1;
256 }
257 } else {
258 if(avctx->min_prediction_order > MAX_FIXED_ORDER) {
259 av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n",
260 avctx->min_prediction_order);
261 return -1;
262 }
263 }
264 s->options.min_prediction_order = avctx->min_prediction_order;
265 }
266 if(avctx->max_prediction_order >= 0) {
267 if(s->options.use_lpc) {
268 if(avctx->max_prediction_order < MIN_LPC_ORDER ||
269 avctx->max_prediction_order > MAX_LPC_ORDER) {
270 av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n",
271 avctx->max_prediction_order);
272 return -1;
273 }
274 } else {
275 if(avctx->max_prediction_order > MAX_FIXED_ORDER) {
276 av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n",
277 avctx->max_prediction_order);
278 return -1;
279 }
280 }
281 s->options.max_prediction_order = avctx->max_prediction_order;
282 }
283 if(s->options.max_prediction_order < s->options.min_prediction_order) {
284 av_log(avctx, AV_LOG_ERROR, "invalid prediction orders: min=%d max=%d\n",
285 s->options.min_prediction_order, s->options.max_prediction_order);
286 return -1;
287 }
288 av_log(avctx, AV_LOG_DEBUG, " prediction order: %d, %d\n",
289 s->options.min_prediction_order, s->options.max_prediction_order);
290
291 if(avctx->prediction_order_method >= 0) {
292 if(avctx->prediction_order_method > ORDER_METHOD_LOG) {
293 av_log(avctx, AV_LOG_ERROR, "invalid prediction order method: %d\n",
294 avctx->prediction_order_method);
295 return -1;
296 }
297 s->options.prediction_order_method = avctx->prediction_order_method;
298 }
299 switch(s->options.prediction_order_method) {
300 case ORDER_METHOD_EST: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
301 "estimate"); break;
302 case ORDER_METHOD_2LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
303 "2-level"); break;
304 case ORDER_METHOD_4LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
305 "4-level"); break;
306 case ORDER_METHOD_8LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
307 "8-level"); break;
308 case ORDER_METHOD_SEARCH: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
309 "full search"); break;
310 case ORDER_METHOD_LOG: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
311 "log search"); break;
312 }
313
314 if(avctx->min_partition_order >= 0) {
315 if(avctx->min_partition_order > MAX_PARTITION_ORDER) {
316 av_log(avctx, AV_LOG_ERROR, "invalid min partition order: %d\n",
317 avctx->min_partition_order);
318 return -1;
319 }
320 s->options.min_partition_order = avctx->min_partition_order;
321 }
322 if(avctx->max_partition_order >= 0) {
323 if(avctx->max_partition_order > MAX_PARTITION_ORDER) {
324 av_log(avctx, AV_LOG_ERROR, "invalid max partition order: %d\n",
325 avctx->max_partition_order);
326 return -1;
327 }
328 s->options.max_partition_order = avctx->max_partition_order;
329 }
330 if(s->options.max_partition_order < s->options.min_partition_order) {
331 av_log(avctx, AV_LOG_ERROR, "invalid partition orders: min=%d max=%d\n",
332 s->options.min_partition_order, s->options.max_partition_order);
333 return -1;
334 }
335 av_log(avctx, AV_LOG_DEBUG, " partition order: %d, %d\n",
336 s->options.min_partition_order, s->options.max_partition_order);
337
338 if(avctx->frame_size > 0) {
339 if(avctx->frame_size < FLAC_MIN_BLOCKSIZE ||
340 avctx->frame_size > FLAC_MAX_BLOCKSIZE) {
341 av_log(avctx, AV_LOG_ERROR, "invalid block size: %d\n",
342 avctx->frame_size);
343 return -1;
344 }
345 } else {
346 s->avctx->frame_size = select_blocksize(s->samplerate, s->options.block_time_ms);
347 }
348 av_log(avctx, AV_LOG_DEBUG, " block size: %d\n", s->avctx->frame_size);
349
350 /* set LPC precision */
351 if(avctx->lpc_coeff_precision > 0) {
352 if(avctx->lpc_coeff_precision > MAX_LPC_PRECISION) {
353 av_log(avctx, AV_LOG_ERROR, "invalid lpc coeff precision: %d\n",
354 avctx->lpc_coeff_precision);
355 return -1;
356 }
357 s->options.lpc_coeff_precision = avctx->lpc_coeff_precision;
358 } else {
359 /* default LPC precision */
360 s->options.lpc_coeff_precision = 15;
361 }
362 av_log(avctx, AV_LOG_DEBUG, " lpc precision: %d\n",
363 s->options.lpc_coeff_precision);
364
365 /* set maximum encoded frame size in verbatim mode */
366 if(s->channels == 2) {
367 s->max_framesize = 14 + ((s->avctx->frame_size * 33 + 7) >> 3);
368 } else {
369 s->max_framesize = 14 + (s->avctx->frame_size * s->channels * 2);
370 }
371
372 streaminfo = av_malloc(FLAC_STREAMINFO_SIZE);
373 write_streaminfo(s, streaminfo);
374 avctx->extradata = streaminfo;
375 avctx->extradata_size = FLAC_STREAMINFO_SIZE;
376
377 s->frame_count = 0;
378
379 avctx->coded_frame = avcodec_alloc_frame();
380 avctx->coded_frame->key_frame = 1;
381
382 return 0;
383 }
384
385 static void init_frame(FlacEncodeContext *s)
386 {
387 int i, ch;
388 FlacFrame *frame;
389
390 frame = &s->frame;
391
392 for(i=0; i<16; i++) {
393 if(s->avctx->frame_size == flac_blocksizes[i]) {
394 frame->blocksize = flac_blocksizes[i];
395 frame->bs_code[0] = i;
396 frame->bs_code[1] = 0;
397 break;
398 }
399 }
400 if(i == 16) {
401 frame->blocksize = s->avctx->frame_size;
402 if(frame->blocksize <= 256) {
403 frame->bs_code[0] = 6;
404 frame->bs_code[1] = frame->blocksize-1;
405 } else {
406 frame->bs_code[0] = 7;
407 frame->bs_code[1] = frame->blocksize-1;
408 }
409 }
410
411 for(ch=0; ch<s->channels; ch++) {
412 frame->subframes[ch].obits = 16;
413 }
414 }
415
416 /**
417 * Copy channel-interleaved input samples into separate subframes
418 */
419 static void copy_samples(FlacEncodeContext *s, int16_t *samples)
420 {
421 int i, j, ch;
422 FlacFrame *frame;
423
424 frame = &s->frame;
425 for(i=0,j=0; i<frame->blocksize; i++) {
426 for(ch=0; ch<s->channels; ch++,j++) {
427 frame->subframes[ch].samples[i] = samples[j];
428 }
429 }
430 }
431
432
433 #define rice_encode_count(sum, n, k) (((n)*((k)+1))+((sum-(n>>1))>>(k)))
434
435 /**
436 * Solve for d/dk(rice_encode_count) = n-((sum-(n>>1))>>(k+1)) = 0
437 */
438 static int find_optimal_param(uint32_t sum, int n)
439 {
440 int k;
441 uint32_t sum2;
442
443 if(sum <= n>>1)
444 return 0;
445 sum2 = sum-(n>>1);
446 k = av_log2(n<256 ? FASTDIV(sum2,n) : sum2/n);
447 return FFMIN(k, MAX_RICE_PARAM);
448 }
449
450 static uint32_t calc_optimal_rice_params(RiceContext *rc, int porder,
451 uint32_t *sums, int n, int pred_order)
452 {
453 int i;
454 int k, cnt, part;
455 uint32_t all_bits;
456
457 part = (1 << porder);
458 all_bits = 4 * part;
459
460 cnt = (n >> porder) - pred_order;
461 for(i=0; i<part; i++) {
462 k = find_optimal_param(sums[i], cnt);
463 rc->params[i] = k;
464 all_bits += rice_encode_count(sums[i], cnt, k);
465 cnt = n >> porder;
466 }
467
468 rc->porder = porder;
469
470 return all_bits;
471 }
472
473 static void calc_sums(int pmin, int pmax, uint32_t *data, int n, int pred_order,
474 uint32_t sums[][MAX_PARTITIONS])
475 {
476 int i, j;
477 int parts;
478 uint32_t *res, *res_end;
479
480 /* sums for highest level */
481 parts = (1 << pmax);
482 res = &data[pred_order];
483 res_end = &data[n >> pmax];
484 for(i=0; i<parts; i++) {
485 uint32_t sum = 0;
486 while(res < res_end){
487 sum += *(res++);
488 }
489 sums[pmax][i] = sum;
490 res_end+= n >> pmax;
491 }
492 /* sums for lower levels */
493 for(i=pmax-1; i>=pmin; i--) {
494 parts = (1 << i);
495 for(j=0; j<parts; j++) {
496 sums[i][j] = sums[i+1][2*j] + sums[i+1][2*j+1];
497 }
498 }
499 }
500
501 static uint32_t calc_rice_params(RiceContext *rc, int pmin, int pmax,
502 int32_t *data, int n, int pred_order)
503 {
504 int i;
505 uint32_t bits[MAX_PARTITION_ORDER+1];
506 int opt_porder;
507 RiceContext tmp_rc;
508 uint32_t *udata;
509 uint32_t sums[MAX_PARTITION_ORDER+1][MAX_PARTITIONS];
510
511 assert(pmin >= 0 && pmin <= MAX_PARTITION_ORDER);
512 assert(pmax >= 0 && pmax <= MAX_PARTITION_ORDER);
513 assert(pmin <= pmax);
514
515 udata = av_malloc(n * sizeof(uint32_t));
516 for(i=0; i<n; i++) {
517 udata[i] = (2*data[i]) ^ (data[i]>>31);
518 }
519
520 calc_sums(pmin, pmax, udata, n, pred_order, sums);
521
522 opt_porder = pmin;
523 bits[pmin] = UINT32_MAX;
524 for(i=pmin; i<=pmax; i++) {
525 bits[i] = calc_optimal_rice_params(&tmp_rc, i, sums[i], n, pred_order);
526 if(bits[i] <= bits[opt_porder]) {
527 opt_porder = i;
528 *rc= tmp_rc;
529 }
530 }
531
532 av_freep(&udata);
533 return bits[opt_porder];
534 }
535
536 static int get_max_p_order(int max_porder, int n, int order)
537 {
538 int porder = FFMIN(max_porder, av_log2(n^(n-1)));
539 if(order > 0)
540 porder = FFMIN(porder, av_log2(n/order));
541 return porder;
542 }
543
544 static uint32_t calc_rice_params_fixed(RiceContext *rc, int pmin, int pmax,
545 int32_t *data, int n, int pred_order,
546 int bps)
547 {
548 uint32_t bits;
549 pmin = get_max_p_order(pmin, n, pred_order);
550 pmax = get_max_p_order(pmax, n, pred_order);
551 bits = pred_order*bps + 6;
552 bits += calc_rice_params(rc, pmin, pmax, data, n, pred_order);
553 return bits;
554 }
555
556 static uint32_t calc_rice_params_lpc(RiceContext *rc, int pmin, int pmax,
557 int32_t *data, int n, int pred_order,
558 int bps, int precision)
559 {
560 uint32_t bits;
561 pmin = get_max_p_order(pmin, n, pred_order);
562 pmax = get_max_p_order(pmax, n, pred_order);
563 bits = pred_order*bps + 4 + 5 + pred_order*precision + 6;
564 bits += calc_rice_params(rc, pmin, pmax, data, n, pred_order);
565 return bits;
566 }
567
568 /**
569 * Apply Welch window function to audio block
570 */
571 static void apply_welch_window(const int32_t *data, int len, double *w_data)
572 {
573 int i, n2;
574 double w;
575 double c;
576
577 assert(!(len&1)); //the optimization in r11881 does not support odd len
578 //if someone wants odd len extend the change in r11881
579
580 n2 = (len >> 1);
581 c = 2.0 / (len - 1.0);
582
583 w_data+=n2;
584 data+=n2;
585 for(i=0; i<n2; i++) {
586 w = c - n2 + i;
587 w = 1.0 - (w * w);
588 w_data[-i-1] = data[-i-1] * w;
589 w_data[+i ] = data[+i ] * w;
590 }
591 }
592
593 /**
594 * Calculates autocorrelation data from audio samples
595 * A Welch window function is applied before calculation.
596 */
597 void ff_flac_compute_autocorr(const int32_t *data, int len, int lag,
598 double *autoc)
599 {
600 int i, j;
601 double tmp[len + lag + 1];
602 double *data1= tmp + lag;
603
604 apply_welch_window(data, len, data1);
605
606 for(j=0; j<lag; j++)
607 data1[j-lag]= 0.0;
608 data1[len] = 0.0;
609
610 for(j=0; j<lag; j+=2){
611 double sum0 = 1.0, sum1 = 1.0;
612 for(i=0; i<len; i++){
613 sum0 += data1[i] * data1[i-j];
614 sum1 += data1[i] * data1[i-j-1];
615 }
616 autoc[j ] = sum0;
617 autoc[j+1] = sum1;
618 }
619
620 if(j==lag){
621 double sum = 1.0;
622 for(i=0; i<len; i+=2){
623 sum += data1[i ] * data1[i-j ]
624 + data1[i+1] * data1[i-j+1];
625 }
626 autoc[j] = sum;
627 }
628 }
629
630
631 static void encode_residual_verbatim(int32_t *res, int32_t *smp, int n)
632 {
633 assert(n > 0);
634 memcpy(res, smp, n * sizeof(int32_t));
635 }
636
637 static void encode_residual_fixed(int32_t *res, const int32_t *smp, int n,
638 int order)
639 {
640 int i;
641
642 for(i=0; i<order; i++) {
643 res[i] = smp[i];
644 }
645
646 if(order==0){
647 for(i=order; i<n; i++)
648 res[i]= smp[i];
649 }else if(order==1){
650 for(i=order; i<n; i++)
651 res[i]= smp[i] - smp[i-1];
652 }else if(order==2){
653 int a = smp[order-1] - smp[order-2];
654 for(i=order; i<n; i+=2) {
655 int b = smp[i] - smp[i-1];
656 res[i]= b - a;
657 a = smp[i+1] - smp[i];
658 res[i+1]= a - b;
659 }
660 }else if(order==3){
661 int a = smp[order-1] - smp[order-2];
662 int c = smp[order-1] - 2*smp[order-2] + smp[order-3];
663 for(i=order; i<n; i+=2) {
664 int b = smp[i] - smp[i-1];
665 int d = b - a;
666 res[i]= d - c;
667 a = smp[i+1] - smp[i];
668 c = a - b;
669 res[i+1]= c - d;
670 }
671 }else{
672 int a = smp[order-1] - smp[order-2];
673 int c = smp[order-1] - 2*smp[order-2] + smp[order-3];
674 int e = smp[order-1] - 3*smp[order-2] + 3*smp[order-3] - smp[order-4];
675 for(i=order; i<n; i+=2) {
676 int b = smp[i] - smp[i-1];
677 int d = b - a;
678 int f = d - c;
679 res[i]= f - e;
680 a = smp[i+1] - smp[i];
681 c = a - b;
682 e = c - d;
683 res[i+1]= e - f;
684 }
685 }
686 }
687
688 #define LPC1(x) {\
689 int c = coefs[(x)-1];\
690 p0 += c*s;\
691 s = smp[i-(x)+1];\
692 p1 += c*s;\
693 }
694
695 static av_always_inline void encode_residual_lpc_unrolled(
696 int32_t *res, const int32_t *smp, int n,
697 int order, const int32_t *coefs, int shift, int big)
698 {
699 int i;
700 for(i=order; i<n; i+=2) {
701 int s = smp[i-order];
702 int p0 = 0, p1 = 0;
703 if(big) {
704 switch(order) {
705 case 32: LPC1(32)
706 case 31: LPC1(31)
707 case 30: LPC1(30)
708 case 29: LPC1(29)
709 case 28: LPC1(28)
710 case 27: LPC1(27)
711 case 26: LPC1(26)
712 case 25: LPC1(25)
713 case 24: LPC1(24)
714 case 23: LPC1(23)
715 case 22: LPC1(22)
716 case 21: LPC1(21)
717 case 20: LPC1(20)
718 case 19: LPC1(19)
719 case 18: LPC1(18)
720 case 17: LPC1(17)
721 case 16: LPC1(16)
722 case 15: LPC1(15)
723 case 14: LPC1(14)
724 case 13: LPC1(13)
725 case 12: LPC1(12)
726 case 11: LPC1(11)
727 case 10: LPC1(10)
728 case 9: LPC1( 9)
729 LPC1( 8)
730 LPC1( 7)
731 LPC1( 6)
732 LPC1( 5)
733 LPC1( 4)
734 LPC1( 3)
735 LPC1( 2)
736 LPC1( 1)
737 }
738 } else {
739 switch(order) {
740 case 8: LPC1( 8)
741 case 7: LPC1( 7)
742 case 6: LPC1( 6)
743 case 5: LPC1( 5)
744 case 4: LPC1( 4)
745 case 3: LPC1( 3)
746 case 2: LPC1( 2)
747 case 1: LPC1( 1)
748 }
749 }
750 res[i ] = smp[i ] - (p0 >> shift);
751 res[i+1] = smp[i+1] - (p1 >> shift);
752 }
753 }
754
755 static void encode_residual_lpc(int32_t *res, const int32_t *smp, int n,
756 int order, const int32_t *coefs, int shift)
757 {
758 int i;
759 for(i=0; i<order; i++) {
760 res[i] = smp[i];
761 }
762 #ifdef CONFIG_SMALL
763 for(i=order; i<n; i+=2) {
764 int j;
765 int s = smp[i];
766 int p0 = 0, p1 = 0;
767 for(j=0; j<order; j++) {
768 int c = coefs[j];
769 p1 += c*s;
770 s = smp[i-j-1];
771 p0 += c*s;
772 }
773 res[i ] = smp[i ] - (p0 >> shift);
774 res[i+1] = smp[i+1] - (p1 >> shift);
775 }
776 #else
777 switch(order) {
778 case 1: encode_residual_lpc_unrolled(res, smp, n, 1, coefs, shift, 0); break;
779 case 2: encode_residual_lpc_unrolled(res, smp, n, 2, coefs, shift, 0); break;
780 case 3: encode_residual_lpc_unrolled(res, smp, n, 3, coefs, shift, 0); break;
781 case 4: encode_residual_lpc_unrolled(res, smp, n, 4, coefs, shift, 0); break;
782 case 5: encode_residual_lpc_unrolled(res, smp, n, 5, coefs, shift, 0); break;
783 case 6: encode_residual_lpc_unrolled(res, smp, n, 6, coefs, shift, 0); break;
784 case 7: encode_residual_lpc_unrolled(res, smp, n, 7, coefs, shift, 0); break;
785 case 8: encode_residual_lpc_unrolled(res, smp, n, 8, coefs, shift, 0); break;
786 default: encode_residual_lpc_unrolled(res, smp, n, order, coefs, shift, 1); break;
787 }
788 #endif
789 }
790
791 static int encode_residual(FlacEncodeContext *ctx, int ch)
792 {
793 int i, n;
794 int min_order, max_order, opt_order, precision, omethod;
795 int min_porder, max_porder;
796 FlacFrame *frame;
797 FlacSubframe *sub;
798 int32_t coefs[MAX_LPC_ORDER][MAX_LPC_ORDER];
799 int shift[MAX_LPC_ORDER];
800 int32_t *res, *smp;
801
802 frame = &ctx->frame;
803 sub = &frame->subframes[ch];
804 res = sub->residual;
805 smp = sub->samples;
806 n = frame->blocksize;
807
808 /* CONSTANT */
809 for(i=1; i<n; i++) {
810 if(smp[i] != smp[0]) break;
811 }
812 if(i == n) {
813 sub->type = sub->type_code = FLAC_SUBFRAME_CONSTANT;
814 res[0] = smp[0];
815 return sub->obits;
816 }
817
818 /* VERBATIM */
819 if(n < 5) {
820 sub->type = sub->type_code = FLAC_SUBFRAME_VERBATIM;
821 encode_residual_verbatim(res, smp, n);
822 return sub->obits * n;
823 }
824
825 min_order = ctx->options.min_prediction_order;
826 max_order = ctx->options.max_prediction_order;
827 min_porder = ctx->options.min_partition_order;
828 max_porder = ctx->options.max_partition_order;
829 precision = ctx->options.lpc_coeff_precision;
830 omethod = ctx->options.prediction_order_method;
831
832 /* FIXED */
833 if(!ctx->options.use_lpc || max_order == 0 || (n <= max_order)) {
834 uint32_t bits[MAX_FIXED_ORDER+1];
835 if(max_order > MAX_FIXED_ORDER) max_order = MAX_FIXED_ORDER;
836 opt_order = 0;
837 bits[0] = UINT32_MAX;
838 for(i=min_order; i<=max_order; i++) {
839 encode_residual_fixed(res, smp, n, i);
840 bits[i] = calc_rice_params_fixed(&sub->rc, min_porder, max_porder, res,
841 n, i, sub->obits);
842 if(bits[i] < bits[opt_order]) {
843 opt_order = i;
844 }
845 }
846 sub->order = opt_order;
847 sub->type = FLAC_SUBFRAME_FIXED;
848 sub->type_code = sub->type | sub->order;
849 if(sub->order != max_order) {
850 encode_residual_fixed(res, smp, n, sub->order);
851 return calc_rice_params_fixed(&sub->rc, min_porder, max_porder, res, n,
852 sub->order, sub->obits);
853 }
854 return bits[sub->order];
855 }
856
857 /* LPC */
858 opt_order = ff_lpc_calc_coefs(&ctx->dsp, smp, n, min_order, max_order,
859 precision, coefs, shift, ctx->options.use_lpc,
860 omethod, MAX_LPC_SHIFT, 0);
861
862 if(omethod == ORDER_METHOD_2LEVEL ||
863 omethod == ORDER_METHOD_4LEVEL ||
864 omethod == ORDER_METHOD_8LEVEL) {
865 int levels = 1 << omethod;
866 uint32_t bits[levels];
867 int order;
868 int opt_index = levels-1;
869 opt_order = max_order-1;
870 bits[opt_index] = UINT32_MAX;
871 for(i=levels-1; i>=0; i--) {
872 order = min_order + (((max_order-min_order+1) * (i+1)) / levels)-1;
873 if(order < 0) order = 0;
874 encode_residual_lpc(res, smp, n, order+1, coefs[order], shift[order]);
875 bits[i] = calc_rice_params_lpc(&sub->rc, min_porder, max_porder,
876 res, n, order+1, sub->obits, precision);
877 if(bits[i] < bits[opt_index]) {
878 opt_index = i;
879 opt_order = order;
880 }
881 }
882 opt_order++;
883 } else if(omethod == ORDER_METHOD_SEARCH) {
884 // brute-force optimal order search
885 uint32_t bits[MAX_LPC_ORDER];
886 opt_order = 0;
887 bits[0] = UINT32_MAX;
888 for(i=min_order-1; i<max_order; i++) {
889 encode_residual_lpc(res, smp, n, i+1, coefs[i], shift[i]);
890 bits[i] = calc_rice_params_lpc(&sub->rc, min_porder, max_porder,
891 res, n, i+1, sub->obits, precision);
892 if(bits[i] < bits[opt_order]) {
893 opt_order = i;
894 }
895 }
896 opt_order++;
897 } else if(omethod == ORDER_METHOD_LOG) {
898 uint32_t bits[MAX_LPC_ORDER];
899 int step;
900
901 opt_order= min_order - 1 + (max_order-min_order)/3;
902 memset(bits, -1, sizeof(bits));
903
904 for(step=16 ;step; step>>=1){
905 int last= opt_order;
906 for(i=last-step; i<=last+step; i+= step){
907 if(i<min_order-1 || i>=max_order || bits[i] < UINT32_MAX)
908 continue;
909 encode_residual_lpc(res, smp, n, i+1, coefs[i], shift[i]);
910 bits[i] = calc_rice_params_lpc(&sub->rc, min_porder, max_porder,
911 res, n, i+1, sub->obits, precision);
912 if(bits[i] < bits[opt_order])
913 opt_order= i;
914 }
915 }
916 opt_order++;
917 }
918
919 sub->order = opt_order;
920 sub->type = FLAC_SUBFRAME_LPC;
921 sub->type_code = sub->type | (sub->order-1);
922 sub->shift = shift[sub->order-1];
923 for(i=0; i<sub->order; i++) {
924 sub->coefs[i] = coefs[sub->order-1][i];
925 }
926 encode_residual_lpc(res, smp, n, sub->order, sub->coefs, sub->shift);
927 return calc_rice_params_lpc(&sub->rc, min_porder, max_porder, res, n, sub->order,
928 sub->obits, precision);
929 }
930
931 static int encode_residual_v(FlacEncodeContext *ctx, int ch)
932 {
933 int i, n;
934 FlacFrame *frame;
935 FlacSubframe *sub;
936 int32_t *res, *smp;
937
938 frame = &ctx->frame;
939 sub = &frame->subframes[ch];
940 res = sub->residual;
941 smp = sub->samples;
942 n = frame->blocksize;
943
944 /* CONSTANT */
945 for(i=1; i<n; i++) {
946 if(smp[i] != smp[0]) break;
947 }
948 if(i == n) {
949 sub->type = sub->type_code = FLAC_SUBFRAME_CONSTANT;
950 res[0] = smp[0];
951 return sub->obits;
952 }
953
954 /* VERBATIM */
955 sub->type = sub->type_code = FLAC_SUBFRAME_VERBATIM;
956 encode_residual_verbatim(res, smp, n);
957 return sub->obits * n;
958 }
959
960 static int estimate_stereo_mode(int32_t *left_ch, int32_t *right_ch, int n)
961 {
962 int i, best;
963 int32_t lt, rt;
964 uint64_t sum[4];
965 uint64_t score[4];
966 int k;
967
968 /* calculate sum of 2nd order residual for each channel */
969 sum[0] = sum[1] = sum[2] = sum[3] = 0;
970 for(i=2; i<n; i++) {
971 lt = left_ch[i] - 2*left_ch[i-1] + left_ch[i-2];
972 rt = right_ch[i] - 2*right_ch[i-1] + right_ch[i-2];
973 sum[2] += FFABS((lt + rt) >> 1);
974 sum[3] += FFABS(lt - rt);
975 sum[0] += FFABS(lt);
976 sum[1] += FFABS(rt);
977 }
978 /* estimate bit counts */
979 for(i=0; i<4; i++) {
980 k = find_optimal_param(2*sum[i], n);
981 sum[i] = rice_encode_count(2*sum[i], n, k);
982 }
983
984 /* calculate score for each mode */
985 score[0] = sum[0] + sum[1];
986 score[1] = sum[0] + sum[3];
987 score[2] = sum[1] + sum[3];
988 score[3] = sum[2] + sum[3];
989
990 /* return mode with lowest score */
991 best = 0;
992 for(i=1; i<4; i++) {
993 if(score[i] < score[best]) {
994 best = i;
995 }
996 }
997 if(best == 0) {
998 return FLAC_CHMODE_LEFT_RIGHT;
999 } else if(best == 1) {
1000 return FLAC_CHMODE_LEFT_SIDE;
1001 } else if(best == 2) {
1002 return FLAC_CHMODE_RIGHT_SIDE;
1003 } else {
1004 return FLAC_CHMODE_MID_SIDE;
1005 }
1006 }
1007
1008 /**
1009 * Perform stereo channel decorrelation
1010 */
1011 static void channel_decorrelation(FlacEncodeContext *ctx)
1012 {
1013 FlacFrame *frame;
1014 int32_t *left, *right;
1015 int i, n;
1016
1017 frame = &ctx->frame;
1018 n = frame->blocksize;
1019 left = frame->subframes[0].samples;
1020 right = frame->subframes[1].samples;
1021
1022 if(ctx->channels != 2) {
1023 frame->ch_mode = FLAC_CHMODE_NOT_STEREO;
1024 return;
1025 }
1026
1027 frame->ch_mode = estimate_stereo_mode(left, right, n);
1028
1029 /* perform decorrelation and adjust bits-per-sample */
1030 if(frame->ch_mode == FLAC_CHMODE_LEFT_RIGHT) {
1031 return;
1032 }
1033 if(frame->ch_mode == FLAC_CHMODE_MID_SIDE) {
1034 int32_t tmp;
1035 for(i=0; i<n; i++) {
1036 tmp = left[i];
1037 left[i] = (tmp + right[i]) >> 1;
1038 right[i] = tmp - right[i];
1039 }
1040 frame->subframes[1].obits++;
1041 } else if(frame->ch_mode == FLAC_CHMODE_LEFT_SIDE) {
1042 for(i=0; i<n; i++) {
1043 right[i] = left[i] - right[i];
1044 }
1045 frame->subframes[1].obits++;
1046 } else {
1047 for(i=0; i<n; i++) {
1048 left[i] -= right[i];
1049 }
1050 frame->subframes[0].obits++;
1051 }
1052 }
1053
1054 static void write_utf8(PutBitContext *pb, uint32_t val)
1055 {
1056 uint8_t tmp;
1057 PUT_UTF8(val, tmp, put_bits(pb, 8, tmp);)
1058 }
1059
1060 static void output_frame_header(FlacEncodeContext *s)
1061 {
1062 FlacFrame *frame;
1063 int crc;
1064
1065 frame = &s->frame;
1066
1067 put_bits(&s->pb, 16, 0xFFF8);
1068 put_bits(&s->pb, 4, frame->bs_code[0]);
1069 put_bits(&s->pb, 4, s->sr_code[0]);
1070 if(frame->ch_mode == FLAC_CHMODE_NOT_STEREO) {
1071 put_bits(&s->pb, 4, s->ch_code);
1072 } else {
1073 put_bits(&s->pb, 4, frame->ch_mode);
1074 }
1075 put_bits(&s->pb, 3, 4); /* bits-per-sample code */
1076 put_bits(&s->pb, 1, 0);
1077 write_utf8(&s->pb, s->frame_count);
1078 if(frame->bs_code[0] == 6) {
1079 put_bits(&s->pb, 8, frame->bs_code[1]);
1080 } else if(frame->bs_code[0] == 7) {
1081 put_bits(&s->pb, 16, frame->bs_code[1]);
1082 }
1083 if(s->sr_code[0] == 12) {
1084 put_bits(&s->pb, 8, s->sr_code[1]);
1085 } else if(s->sr_code[0] > 12) {
1086 put_bits(&s->pb, 16, s->sr_code[1]);
1087 }
1088 flush_put_bits(&s->pb);
1089 crc = av_crc(av_crc_get_table(AV_CRC_8_ATM), 0,
1090 s->pb.buf, put_bits_count(&s->pb)>>3);
1091 put_bits(&s->pb, 8, crc);
1092 }
1093
1094 static void output_subframe_constant(FlacEncodeContext *s, int ch)
1095 {
1096 FlacSubframe *sub;
1097 int32_t res;
1098
1099 sub = &s->frame.subframes[ch];
1100 res = sub->residual[0];
1101 put_sbits(&s->pb, sub->obits, res);
1102 }
1103
1104 static void output_subframe_verbatim(FlacEncodeContext *s, int ch)
1105 {
1106 int i;
1107 FlacFrame *frame;
1108 FlacSubframe *sub;
1109 int32_t res;
1110
1111 frame = &s->frame;
1112 sub = &frame->subframes[ch];
1113
1114 for(i=0; i<frame->blocksize; i++) {
1115 res = sub->residual[i];
1116 put_sbits(&s->pb, sub->obits, res);
1117 }
1118 }
1119
1120 static void output_residual(FlacEncodeContext *ctx, int ch)
1121 {
1122 int i, j, p, n, parts;
1123 int k, porder, psize, res_cnt;
1124 FlacFrame *frame;
1125 FlacSubframe *sub;
1126 int32_t *res;
1127
1128 frame = &ctx->frame;
1129 sub = &frame->subframes[ch];
1130 res = sub->residual;
1131 n = frame->blocksize;
1132
1133 /* rice-encoded block */
1134 put_bits(&ctx->pb, 2, 0);
1135
1136 /* partition order */
1137 porder = sub->rc.porder;
1138 psize = n >> porder;
1139 parts = (1 << porder);
1140 put_bits(&ctx->pb, 4, porder);
1141 res_cnt = psize - sub->order;
1142
1143 /* residual */
1144 j = sub->order;
1145 for(p=0; p<parts; p++) {
1146 k = sub->rc.params[p];
1147 put_bits(&ctx->pb, 4, k);
1148 if(p == 1) res_cnt = psize;
1149 for(i=0; i<res_cnt && j<n; i++, j++) {
1150 set_sr_golomb_flac(&ctx->pb, res[j], k, INT32_MAX, 0);
1151 }
1152 }
1153 }
1154
1155 static void output_subframe_fixed(FlacEncodeContext *ctx, int ch)
1156 {
1157 int i;
1158 FlacFrame *frame;
1159 FlacSubframe *sub;
1160
1161 frame = &ctx->frame;
1162 sub = &frame->subframes[ch];
1163
1164 /* warm-up samples */
1165 for(i=0; i<sub->order; i++) {
1166 put_sbits(&ctx->pb, sub->obits, sub->residual[i]);
1167 }
1168
1169 /* residual */
1170 output_residual(ctx, ch);
1171 }
1172
1173 static void output_subframe_lpc(FlacEncodeContext *ctx, int ch)
1174 {
1175 int i, cbits;
1176 FlacFrame *frame;
1177 FlacSubframe *sub;
1178
1179 frame = &ctx->frame;
1180 sub = &frame->subframes[ch];
1181
1182 /* warm-up samples */
1183 for(i=0; i<sub->order; i++) {
1184 put_sbits(&ctx->pb, sub->obits, sub->residual[i]);
1185 }
1186
1187 /* LPC coefficients */
1188 cbits = ctx->options.lpc_coeff_precision;
1189 put_bits(&ctx->pb, 4, cbits-1);
1190 put_sbits(&ctx->pb, 5, sub->shift);
1191 for(i=0; i<sub->order; i++) {
1192 put_sbits(&ctx->pb, cbits, sub->coefs[i]);
1193 }
1194
1195 /* residual */
1196 output_residual(ctx, ch);
1197 }
1198
1199 static void output_subframes(FlacEncodeContext *s)
1200 {
1201 FlacFrame *frame;
1202 FlacSubframe *sub;
1203 int ch;
1204
1205 frame = &s->frame;
1206
1207 for(ch=0; ch<s->channels; ch++) {
1208 sub = &frame->subframes[ch];
1209
1210 /* subframe header */
1211 put_bits(&s->pb, 1, 0);
1212 put_bits(&s->pb, 6, sub->type_code);
1213 put_bits(&s->pb, 1, 0); /* no wasted bits */
1214
1215 /* subframe */
1216 if(sub->type == FLAC_SUBFRAME_CONSTANT) {
1217 output_subframe_constant(s, ch);
1218 } else if(sub->type == FLAC_SUBFRAME_VERBATIM) {
1219 output_subframe_verbatim(s, ch);
1220 } else if(sub->type == FLAC_SUBFRAME_FIXED) {
1221 output_subframe_fixed(s, ch);
1222 } else if(sub->type == FLAC_SUBFRAME_LPC) {
1223 output_subframe_lpc(s, ch);
1224 }
1225 }
1226 }
1227
1228 static void output_frame_footer(FlacEncodeContext *s)
1229 {
1230 int crc;
1231 flush_put_bits(&s->pb);
1232 crc = bswap_16(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0,
1233 s->pb.buf, put_bits_count(&s->pb)>>3));
1234 put_bits(&s->pb, 16, crc);
1235 flush_put_bits(&s->pb);
1236 }
1237
1238 static int flac_encode_frame(AVCodecContext *avctx, uint8_t *frame,
1239 int buf_size, void *data)
1240 {
1241 int ch;
1242 FlacEncodeContext *s;
1243 int16_t *samples = data;
1244 int out_bytes;
1245 int reencoded=0;
1246
1247 s = avctx->priv_data;
1248
1249 if(buf_size < s->max_framesize*2) {
1250 av_log(avctx, AV_LOG_ERROR, "output buffer too small\n");
1251 return 0;
1252 }
1253
1254 init_frame(s);
1255
1256 copy_samples(s, samples);
1257
1258 channel_decorrelation(s);
1259
1260 for(ch=0; ch<s->channels; ch++) {
1261 encode_residual(s, ch);
1262 }
1263
1264 write_frame:
1265 init_put_bits(&s->pb, frame, buf_size);
1266 output_frame_header(s);
1267 output_subframes(s);
1268 output_frame_footer(s);
1269 out_bytes = put_bits_count(&s->pb) >> 3;
1270
1271 if(out_bytes > s->max_framesize) {
1272 if(reencoded) {
1273 /* still too large. must be an error. */
1274 av_log(avctx, AV_LOG_ERROR, "error encoding frame\n");
1275 return -1;
1276 }
1277
1278 /* frame too large. use verbatim mode */
1279 for(ch=0; ch<s->channels; ch++) {
1280 encode_residual_v(s, ch);
1281 }
1282 reencoded = 1;
1283 goto write_frame;
1284 }
1285
1286 s->frame_count++;
1287 return out_bytes;
1288 }
1289
1290 static av_cold int flac_encode_close(AVCodecContext *avctx)
1291 {
1292 av_freep(&avctx->extradata);
1293 avctx->extradata_size = 0;
1294 av_freep(&avctx->coded_frame);
1295 return 0;
1296 }
1297
1298 AVCodec flac_encoder = {
1299 "flac",
1300 CODEC_TYPE_AUDIO,
1301 CODEC_ID_FLAC,
1302 sizeof(FlacEncodeContext),
1303 flac_encode_init,
1304 flac_encode_frame,
1305 flac_encode_close,
1306 NULL,
1307 .capabilities = CODEC_CAP_SMALL_LAST_FRAME,
1308 .sample_fmts = (enum SampleFormat[]){SAMPLE_FMT_S16,SAMPLE_FMT_NONE},
1309 .long_name = NULL_IF_CONFIG_SMALL("FLAC (Free Lossless Audio Codec)"),
1310 };
lucabe's unfinished and experimental patches for ffmpeg