flacenc: Allow more flexible shift calculation in LPC.
[libav.git] / libavcodec / flacenc.c
1 /**
2 * FLAC audio encoder
3 * Copyright (c) 2006 Justin Ruggles <jruggle@earthlink.net>
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
29 #define FLAC_MAX_CH 8
30 #define FLAC_MIN_BLOCKSIZE 16
31 #define FLAC_MAX_BLOCKSIZE 65535
32
33 #define FLAC_SUBFRAME_CONSTANT 0
34 #define FLAC_SUBFRAME_VERBATIM 1
35 #define FLAC_SUBFRAME_FIXED 8
36 #define FLAC_SUBFRAME_LPC 32
37
38 #define FLAC_CHMODE_NOT_STEREO 0
39 #define FLAC_CHMODE_LEFT_RIGHT 1
40 #define FLAC_CHMODE_LEFT_SIDE 8
41 #define FLAC_CHMODE_RIGHT_SIDE 9
42 #define FLAC_CHMODE_MID_SIDE 10
43
44 #define ORDER_METHOD_EST 0
45 #define ORDER_METHOD_2LEVEL 1
46 #define ORDER_METHOD_4LEVEL 2
47 #define ORDER_METHOD_8LEVEL 3
48 #define ORDER_METHOD_SEARCH 4
49 #define ORDER_METHOD_LOG 5
50
51 #define FLAC_STREAMINFO_SIZE 34
52
53 #define MIN_LPC_ORDER 1
54 #define MAX_LPC_ORDER 32
55 #define MAX_FIXED_ORDER 4
56 #define MAX_PARTITION_ORDER 8
57 #define MAX_PARTITIONS (1 << MAX_PARTITION_ORDER)
58 #define MAX_LPC_PRECISION 15
59 #define MAX_LPC_SHIFT 15
60 #define MAX_RICE_PARAM 14
61
62 typedef struct CompressionOptions {
63 int compression_level;
64 int block_time_ms;
65 int use_lpc;
66 int lpc_coeff_precision;
67 int min_prediction_order;
68 int max_prediction_order;
69 int prediction_order_method;
70 int min_partition_order;
71 int max_partition_order;
72 } CompressionOptions;
73
74 typedef struct RiceContext {
75 int porder;
76 int params[MAX_PARTITIONS];
77 } RiceContext;
78
79 typedef struct FlacSubframe {
80 int type;
81 int type_code;
82 int obits;
83 int order;
84 int32_t coefs[MAX_LPC_ORDER];
85 int shift;
86 RiceContext rc;
87 int32_t samples[FLAC_MAX_BLOCKSIZE];
88 int32_t residual[FLAC_MAX_BLOCKSIZE+1];
89 } FlacSubframe;
90
91 typedef struct FlacFrame {
92 FlacSubframe subframes[FLAC_MAX_CH];
93 int blocksize;
94 int bs_code[2];
95 uint8_t crc8;
96 int ch_mode;
97 } FlacFrame;
98
99 typedef struct FlacEncodeContext {
100 PutBitContext pb;
101 int channels;
102 int ch_code;
103 int samplerate;
104 int sr_code[2];
105 int max_framesize;
106 uint32_t frame_count;
107 FlacFrame frame;
108 CompressionOptions options;
109 AVCodecContext *avctx;
110 DSPContext dsp;
111 } FlacEncodeContext;
112
113 static const int flac_samplerates[16] = {
114 0, 0, 0, 0,
115 8000, 16000, 22050, 24000, 32000, 44100, 48000, 96000,
116 0, 0, 0, 0
117 };
118
119 static const int flac_blocksizes[16] = {
120 0,
121 192,
122 576, 1152, 2304, 4608,
123 0, 0,
124 256, 512, 1024, 2048, 4096, 8192, 16384, 32768
125 };
126
127 /**
128 * Writes streaminfo metadata block to byte array
129 */
130 static void write_streaminfo(FlacEncodeContext *s, uint8_t *header)
131 {
132 PutBitContext pb;
133
134 memset(header, 0, FLAC_STREAMINFO_SIZE);
135 init_put_bits(&pb, header, FLAC_STREAMINFO_SIZE);
136
137 /* streaminfo metadata block */
138 put_bits(&pb, 16, s->avctx->frame_size);
139 put_bits(&pb, 16, s->avctx->frame_size);
140 put_bits(&pb, 24, 0);
141 put_bits(&pb, 24, s->max_framesize);
142 put_bits(&pb, 20, s->samplerate);
143 put_bits(&pb, 3, s->channels-1);
144 put_bits(&pb, 5, 15); /* bits per sample - 1 */
145 flush_put_bits(&pb);
146 /* total samples = 0 */
147 /* MD5 signature = 0 */
148 }
149
150 /**
151 * Sets blocksize based on samplerate
152 * Chooses the closest predefined blocksize >= BLOCK_TIME_MS milliseconds
153 */
154 static int select_blocksize(int samplerate, int block_time_ms)
155 {
156 int i;
157 int target;
158 int blocksize;
159
160 assert(samplerate > 0);
161 blocksize = flac_blocksizes[1];
162 target = (samplerate * block_time_ms) / 1000;
163 for(i=0; i<16; i++) {
164 if(target >= flac_blocksizes[i] && flac_blocksizes[i] > blocksize) {
165 blocksize = flac_blocksizes[i];
166 }
167 }
168 return blocksize;
169 }
170
171 static av_cold int flac_encode_init(AVCodecContext *avctx)
172 {
173 int freq = avctx->sample_rate;
174 int channels = avctx->channels;
175 FlacEncodeContext *s = avctx->priv_data;
176 int i, level;
177 uint8_t *streaminfo;
178
179 s->avctx = avctx;
180
181 dsputil_init(&s->dsp, avctx);
182
183 if(avctx->sample_fmt != SAMPLE_FMT_S16) {
184 return -1;
185 }
186
187 if(channels < 1 || channels > FLAC_MAX_CH) {
188 return -1;
189 }
190 s->channels = channels;
191 s->ch_code = s->channels-1;
192
193 /* find samplerate in table */
194 if(freq < 1)
195 return -1;
196 for(i=4; i<12; i++) {
197 if(freq == flac_samplerates[i]) {
198 s->samplerate = flac_samplerates[i];
199 s->sr_code[0] = i;
200 s->sr_code[1] = 0;
201 break;
202 }
203 }
204 /* if not in table, samplerate is non-standard */
205 if(i == 12) {
206 if(freq % 1000 == 0 && freq < 255000) {
207 s->sr_code[0] = 12;
208 s->sr_code[1] = freq / 1000;
209 } else if(freq % 10 == 0 && freq < 655350) {
210 s->sr_code[0] = 14;
211 s->sr_code[1] = freq / 10;
212 } else if(freq < 65535) {
213 s->sr_code[0] = 13;
214 s->sr_code[1] = freq;
215 } else {
216 return -1;
217 }
218 s->samplerate = freq;
219 }
220
221 /* set compression option defaults based on avctx->compression_level */
222 if(avctx->compression_level < 0) {
223 s->options.compression_level = 5;
224 } else {
225 s->options.compression_level = avctx->compression_level;
226 }
227 av_log(avctx, AV_LOG_DEBUG, " compression: %d\n", s->options.compression_level);
228
229 level= s->options.compression_level;
230 if(level > 12) {
231 av_log(avctx, AV_LOG_ERROR, "invalid compression level: %d\n",
232 s->options.compression_level);
233 return -1;
234 }
235
236 s->options.block_time_ms = ((int[]){ 27, 27, 27,105,105,105,105,105,105,105,105,105,105})[level];
237 s->options.use_lpc = ((int[]){ 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1})[level];
238 s->options.min_prediction_order= ((int[]){ 2, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1})[level];
239 s->options.max_prediction_order= ((int[]){ 3, 4, 4, 6, 8, 8, 8, 8, 12, 12, 12, 32, 32})[level];
240 s->options.prediction_order_method = ((int[]){ ORDER_METHOD_EST, ORDER_METHOD_EST, ORDER_METHOD_EST,
241 ORDER_METHOD_EST, ORDER_METHOD_EST, ORDER_METHOD_EST,
242 ORDER_METHOD_4LEVEL, ORDER_METHOD_LOG, ORDER_METHOD_4LEVEL,
243 ORDER_METHOD_LOG, ORDER_METHOD_SEARCH, ORDER_METHOD_LOG,
244 ORDER_METHOD_SEARCH})[level];
245 s->options.min_partition_order = ((int[]){ 2, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0})[level];
246 s->options.max_partition_order = ((int[]){ 2, 2, 3, 3, 3, 8, 8, 8, 8, 8, 8, 8, 8})[level];
247
248 /* set compression option overrides from AVCodecContext */
249 if(avctx->use_lpc >= 0) {
250 s->options.use_lpc = av_clip(avctx->use_lpc, 0, 11);
251 }
252 if(s->options.use_lpc == 1)
253 av_log(avctx, AV_LOG_DEBUG, " use lpc: Levinson-Durbin recursion with Welch window\n");
254 else if(s->options.use_lpc > 1)
255 av_log(avctx, AV_LOG_DEBUG, " use lpc: Cholesky factorization\n");
256
257 if(avctx->min_prediction_order >= 0) {
258 if(s->options.use_lpc) {
259 if(avctx->min_prediction_order < MIN_LPC_ORDER ||
260 avctx->min_prediction_order > MAX_LPC_ORDER) {
261 av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n",
262 avctx->min_prediction_order);
263 return -1;
264 }
265 } else {
266 if(avctx->min_prediction_order > MAX_FIXED_ORDER) {
267 av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n",
268 avctx->min_prediction_order);
269 return -1;
270 }
271 }
272 s->options.min_prediction_order = avctx->min_prediction_order;
273 }
274 if(avctx->max_prediction_order >= 0) {
275 if(s->options.use_lpc) {
276 if(avctx->max_prediction_order < MIN_LPC_ORDER ||
277 avctx->max_prediction_order > MAX_LPC_ORDER) {
278 av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n",
279 avctx->max_prediction_order);
280 return -1;
281 }
282 } else {
283 if(avctx->max_prediction_order > MAX_FIXED_ORDER) {
284 av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n",
285 avctx->max_prediction_order);
286 return -1;
287 }
288 }
289 s->options.max_prediction_order = avctx->max_prediction_order;
290 }
291 if(s->options.max_prediction_order < s->options.min_prediction_order) {
292 av_log(avctx, AV_LOG_ERROR, "invalid prediction orders: min=%d max=%d\n",
293 s->options.min_prediction_order, s->options.max_prediction_order);
294 return -1;
295 }
296 av_log(avctx, AV_LOG_DEBUG, " prediction order: %d, %d\n",
297 s->options.min_prediction_order, s->options.max_prediction_order);
298
299 if(avctx->prediction_order_method >= 0) {
300 if(avctx->prediction_order_method > ORDER_METHOD_LOG) {
301 av_log(avctx, AV_LOG_ERROR, "invalid prediction order method: %d\n",
302 avctx->prediction_order_method);
303 return -1;
304 }
305 s->options.prediction_order_method = avctx->prediction_order_method;
306 }
307 switch(s->options.prediction_order_method) {
308 case ORDER_METHOD_EST: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
309 "estimate"); break;
310 case ORDER_METHOD_2LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
311 "2-level"); break;
312 case ORDER_METHOD_4LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
313 "4-level"); break;
314 case ORDER_METHOD_8LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
315 "8-level"); break;
316 case ORDER_METHOD_SEARCH: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
317 "full search"); break;
318 case ORDER_METHOD_LOG: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
319 "log search"); break;
320 }
321
322 if(avctx->min_partition_order >= 0) {
323 if(avctx->min_partition_order > MAX_PARTITION_ORDER) {
324 av_log(avctx, AV_LOG_ERROR, "invalid min partition order: %d\n",
325 avctx->min_partition_order);
326 return -1;
327 }
328 s->options.min_partition_order = avctx->min_partition_order;
329 }
330 if(avctx->max_partition_order >= 0) {
331 if(avctx->max_partition_order > MAX_PARTITION_ORDER) {
332 av_log(avctx, AV_LOG_ERROR, "invalid max partition order: %d\n",
333 avctx->max_partition_order);
334 return -1;
335 }
336 s->options.max_partition_order = avctx->max_partition_order;
337 }
338 if(s->options.max_partition_order < s->options.min_partition_order) {
339 av_log(avctx, AV_LOG_ERROR, "invalid partition orders: min=%d max=%d\n",
340 s->options.min_partition_order, s->options.max_partition_order);
341 return -1;
342 }
343 av_log(avctx, AV_LOG_DEBUG, " partition order: %d, %d\n",
344 s->options.min_partition_order, s->options.max_partition_order);
345
346 if(avctx->frame_size > 0) {
347 if(avctx->frame_size < FLAC_MIN_BLOCKSIZE ||
348 avctx->frame_size > FLAC_MAX_BLOCKSIZE) {
349 av_log(avctx, AV_LOG_ERROR, "invalid block size: %d\n",
350 avctx->frame_size);
351 return -1;
352 }
353 } else {
354 s->avctx->frame_size = select_blocksize(s->samplerate, s->options.block_time_ms);
355 }
356 av_log(avctx, AV_LOG_DEBUG, " block size: %d\n", s->avctx->frame_size);
357
358 /* set LPC precision */
359 if(avctx->lpc_coeff_precision > 0) {
360 if(avctx->lpc_coeff_precision > MAX_LPC_PRECISION) {
361 av_log(avctx, AV_LOG_ERROR, "invalid lpc coeff precision: %d\n",
362 avctx->lpc_coeff_precision);
363 return -1;
364 }
365 s->options.lpc_coeff_precision = avctx->lpc_coeff_precision;
366 } else {
367 /* default LPC precision */
368 s->options.lpc_coeff_precision = 15;
369 }
370 av_log(avctx, AV_LOG_DEBUG, " lpc precision: %d\n",
371 s->options.lpc_coeff_precision);
372
373 /* set maximum encoded frame size in verbatim mode */
374 if(s->channels == 2) {
375 s->max_framesize = 14 + ((s->avctx->frame_size * 33 + 7) >> 3);
376 } else {
377 s->max_framesize = 14 + (s->avctx->frame_size * s->channels * 2);
378 }
379
380 streaminfo = av_malloc(FLAC_STREAMINFO_SIZE);
381 write_streaminfo(s, streaminfo);
382 avctx->extradata = streaminfo;
383 avctx->extradata_size = FLAC_STREAMINFO_SIZE;
384
385 s->frame_count = 0;
386
387 avctx->coded_frame = avcodec_alloc_frame();
388 avctx->coded_frame->key_frame = 1;
389
390 return 0;
391 }
392
393 static void init_frame(FlacEncodeContext *s)
394 {
395 int i, ch;
396 FlacFrame *frame;
397
398 frame = &s->frame;
399
400 for(i=0; i<16; i++) {
401 if(s->avctx->frame_size == flac_blocksizes[i]) {
402 frame->blocksize = flac_blocksizes[i];
403 frame->bs_code[0] = i;
404 frame->bs_code[1] = 0;
405 break;
406 }
407 }
408 if(i == 16) {
409 frame->blocksize = s->avctx->frame_size;
410 if(frame->blocksize <= 256) {
411 frame->bs_code[0] = 6;
412 frame->bs_code[1] = frame->blocksize-1;
413 } else {
414 frame->bs_code[0] = 7;
415 frame->bs_code[1] = frame->blocksize-1;
416 }
417 }
418
419 for(ch=0; ch<s->channels; ch++) {
420 frame->subframes[ch].obits = 16;
421 }
422 }
423
424 /**
425 * Copy channel-interleaved input samples into separate subframes
426 */
427 static void copy_samples(FlacEncodeContext *s, int16_t *samples)
428 {
429 int i, j, ch;
430 FlacFrame *frame;
431
432 frame = &s->frame;
433 for(i=0,j=0; i<frame->blocksize; i++) {
434 for(ch=0; ch<s->channels; ch++,j++) {
435 frame->subframes[ch].samples[i] = samples[j];
436 }
437 }
438 }
439
440
441 #define rice_encode_count(sum, n, k) (((n)*((k)+1))+((sum-(n>>1))>>(k)))
442
443 /**
444 * Solve for d/dk(rice_encode_count) = n-((sum-(n>>1))>>(k+1)) = 0
445 */
446 static int find_optimal_param(uint32_t sum, int n)
447 {
448 int k;
449 uint32_t sum2;
450
451 if(sum <= n>>1)
452 return 0;
453 sum2 = sum-(n>>1);
454 k = av_log2(n<256 ? FASTDIV(sum2,n) : sum2/n);
455 return FFMIN(k, MAX_RICE_PARAM);
456 }
457
458 static uint32_t calc_optimal_rice_params(RiceContext *rc, int porder,
459 uint32_t *sums, int n, int pred_order)
460 {
461 int i;
462 int k, cnt, part;
463 uint32_t all_bits;
464
465 part = (1 << porder);
466 all_bits = 4 * part;
467
468 cnt = (n >> porder) - pred_order;
469 for(i=0; i<part; i++) {
470 k = find_optimal_param(sums[i], cnt);
471 rc->params[i] = k;
472 all_bits += rice_encode_count(sums[i], cnt, k);
473 cnt = n >> porder;
474 }
475
476 rc->porder = porder;
477
478 return all_bits;
479 }
480
481 static void calc_sums(int pmin, int pmax, uint32_t *data, int n, int pred_order,
482 uint32_t sums[][MAX_PARTITIONS])
483 {
484 int i, j;
485 int parts;
486 uint32_t *res, *res_end;
487
488 /* sums for highest level */
489 parts = (1 << pmax);
490 res = &data[pred_order];
491 res_end = &data[n >> pmax];
492 for(i=0; i<parts; i++) {
493 uint32_t sum = 0;
494 while(res < res_end){
495 sum += *(res++);
496 }
497 sums[pmax][i] = sum;
498 res_end+= n >> pmax;
499 }
500 /* sums for lower levels */
501 for(i=pmax-1; i>=pmin; i--) {
502 parts = (1 << i);
503 for(j=0; j<parts; j++) {
504 sums[i][j] = sums[i+1][2*j] + sums[i+1][2*j+1];
505 }
506 }
507 }
508
509 static uint32_t calc_rice_params(RiceContext *rc, int pmin, int pmax,
510 int32_t *data, int n, int pred_order)
511 {
512 int i;
513 uint32_t bits[MAX_PARTITION_ORDER+1];
514 int opt_porder;
515 RiceContext tmp_rc;
516 uint32_t *udata;
517 uint32_t sums[MAX_PARTITION_ORDER+1][MAX_PARTITIONS];
518
519 assert(pmin >= 0 && pmin <= MAX_PARTITION_ORDER);
520 assert(pmax >= 0 && pmax <= MAX_PARTITION_ORDER);
521 assert(pmin <= pmax);
522
523 udata = av_malloc(n * sizeof(uint32_t));
524 for(i=0; i<n; i++) {
525 udata[i] = (2*data[i]) ^ (data[i]>>31);
526 }
527
528 calc_sums(pmin, pmax, udata, n, pred_order, sums);
529
530 opt_porder = pmin;
531 bits[pmin] = UINT32_MAX;
532 for(i=pmin; i<=pmax; i++) {
533 bits[i] = calc_optimal_rice_params(&tmp_rc, i, sums[i], n, pred_order);
534 if(bits[i] <= bits[opt_porder]) {
535 opt_porder = i;
536 *rc= tmp_rc;
537 }
538 }
539
540 av_freep(&udata);
541 return bits[opt_porder];
542 }
543
544 static int get_max_p_order(int max_porder, int n, int order)
545 {
546 int porder = FFMIN(max_porder, av_log2(n^(n-1)));
547 if(order > 0)
548 porder = FFMIN(porder, av_log2(n/order));
549 return porder;
550 }
551
552 static uint32_t calc_rice_params_fixed(RiceContext *rc, int pmin, int pmax,
553 int32_t *data, int n, int pred_order,
554 int bps)
555 {
556 uint32_t bits;
557 pmin = get_max_p_order(pmin, n, pred_order);
558 pmax = get_max_p_order(pmax, n, pred_order);
559 bits = pred_order*bps + 6;
560 bits += calc_rice_params(rc, pmin, pmax, data, n, pred_order);
561 return bits;
562 }
563
564 static uint32_t calc_rice_params_lpc(RiceContext *rc, int pmin, int pmax,
565 int32_t *data, int n, int pred_order,
566 int bps, int precision)
567 {
568 uint32_t bits;
569 pmin = get_max_p_order(pmin, n, pred_order);
570 pmax = get_max_p_order(pmax, n, pred_order);
571 bits = pred_order*bps + 4 + 5 + pred_order*precision + 6;
572 bits += calc_rice_params(rc, pmin, pmax, data, n, pred_order);
573 return bits;
574 }
575
576 /**
577 * Apply Welch window function to audio block
578 */
579 static void apply_welch_window(const int32_t *data, int len, double *w_data)
580 {
581 int i, n2;
582 double w;
583 double c;
584
585 assert(!(len&1)); //the optimization in r11881 does not support odd len
586 //if someone wants odd len extend the change in r11881
587
588 n2 = (len >> 1);
589 c = 2.0 / (len - 1.0);
590
591 w_data+=n2;
592 data+=n2;
593 for(i=0; i<n2; i++) {
594 w = c - n2 + i;
595 w = 1.0 - (w * w);
596 w_data[-i-1] = data[-i-1] * w;
597 w_data[+i ] = data[+i ] * w;
598 }
599 }
600
601 /**
602 * Calculates autocorrelation data from audio samples
603 * A Welch window function is applied before calculation.
604 */
605 void ff_flac_compute_autocorr(const int32_t *data, int len, int lag,
606 double *autoc)
607 {
608 int i, j;
609 double tmp[len + lag + 1];
610 double *data1= tmp + lag;
611
612 apply_welch_window(data, len, data1);
613
614 for(j=0; j<lag; j++)
615 data1[j-lag]= 0.0;
616 data1[len] = 0.0;
617
618 for(j=0; j<lag; j+=2){
619 double sum0 = 1.0, sum1 = 1.0;
620 for(i=0; i<len; i++){
621 sum0 += data1[i] * data1[i-j];
622 sum1 += data1[i] * data1[i-j-1];
623 }
624 autoc[j ] = sum0;
625 autoc[j+1] = sum1;
626 }
627
628 if(j==lag){
629 double sum = 1.0;
630 for(i=0; i<len; i+=2){
631 sum += data1[i ] * data1[i-j ]
632 + data1[i+1] * data1[i-j+1];
633 }
634 autoc[j] = sum;
635 }
636 }
637
638 /**
639 * Levinson-Durbin recursion.
640 * Produces LPC coefficients from autocorrelation data.
641 */
642 static void compute_lpc_coefs(const double *autoc, int max_order,
643 double lpc[][MAX_LPC_ORDER], double *ref)
644 {
645 int i, j, i2;
646 double r, err, tmp;
647 double lpc_tmp[MAX_LPC_ORDER];
648
649 for(i=0; i<max_order; i++) lpc_tmp[i] = 0;
650 err = autoc[0];
651
652 for(i=0; i<max_order; i++) {
653 r = -autoc[i+1];
654 for(j=0; j<i; j++) {
655 r -= lpc_tmp[j] * autoc[i-j];
656 }
657 r /= err;
658 ref[i] = fabs(r);
659
660 err *= 1.0 - (r * r);
661
662 i2 = (i >> 1);
663 lpc_tmp[i] = r;
664 for(j=0; j<i2; j++) {
665 tmp = lpc_tmp[j];
666 lpc_tmp[j] += r * lpc_tmp[i-1-j];
667 lpc_tmp[i-1-j] += r * tmp;
668 }
669 if(i & 1) {
670 lpc_tmp[j] += lpc_tmp[j] * r;
671 }
672
673 for(j=0; j<=i; j++) {
674 lpc[i][j] = -lpc_tmp[j];
675 }
676 }
677 }
678
679 /**
680 * Quantize LPC coefficients
681 */
682 static void quantize_lpc_coefs(double *lpc_in, int order, int precision,
683 int32_t *lpc_out, int *shift, int max_shift, int zero_shift)
684 {
685 int i;
686 double cmax, error;
687 int32_t qmax;
688 int sh;
689
690 /* define maximum levels */
691 qmax = (1 << (precision - 1)) - 1;
692
693 /* find maximum coefficient value */
694 cmax = 0.0;
695 for(i=0; i<order; i++) {
696 cmax= FFMAX(cmax, fabs(lpc_in[i]));
697 }
698
699 /* if maximum value quantizes to zero, return all zeros */
700 if(cmax * (1 << max_shift) < 1.0) {
701 *shift = zero_shift;
702 memset(lpc_out, 0, sizeof(int32_t) * order);
703 return;
704 }
705
706 /* calculate level shift which scales max coeff to available bits */
707 sh = max_shift;
708 while((cmax * (1 << sh) > qmax) && (sh > 0)) {
709 sh--;
710 }
711
712 /* since negative shift values are unsupported in decoder, scale down
713 coefficients instead */
714 if(sh == 0 && cmax > qmax) {
715 double scale = ((double)qmax) / cmax;
716 for(i=0; i<order; i++) {
717 lpc_in[i] *= scale;
718 }
719 }
720
721 /* output quantized coefficients and level shift */
722 error=0;
723 for(i=0; i<order; i++) {
724 error += lpc_in[i] * (1 << sh);
725 lpc_out[i] = av_clip(lrintf(error), -qmax, qmax);
726 error -= lpc_out[i];
727 }
728 *shift = sh;
729 }
730
731 static int estimate_best_order(double *ref, int max_order)
732 {
733 int i, est;
734
735 est = 1;
736 for(i=max_order-1; i>=0; i--) {
737 if(ref[i] > 0.10) {
738 est = i+1;
739 break;
740 }
741 }
742 return est;
743 }
744
745 /**
746 * Calculate LPC coefficients for multiple orders
747 */
748 static int lpc_calc_coefs(DSPContext *s,
749 const int32_t *samples, int blocksize, int max_order,
750 int precision, int32_t coefs[][MAX_LPC_ORDER],
751 int *shift, int use_lpc, int omethod, int max_shift, int zero_shift)
752 {
753 double autoc[MAX_LPC_ORDER+1];
754 double ref[MAX_LPC_ORDER];
755 double lpc[MAX_LPC_ORDER][MAX_LPC_ORDER];
756 int i, j, pass;
757 int opt_order;
758
759 assert(max_order >= MIN_LPC_ORDER && max_order <= MAX_LPC_ORDER);
760
761 if(use_lpc == 1){
762 s->flac_compute_autocorr(samples, blocksize, max_order, autoc);
763
764 compute_lpc_coefs(autoc, max_order, lpc, ref);
765 }else{
766 LLSModel m[2];
767 double var[MAX_LPC_ORDER+1], weight;
768
769 for(pass=0; pass<use_lpc-1; pass++){
770 av_init_lls(&m[pass&1], max_order);
771
772 weight=0;
773 for(i=max_order; i<blocksize; i++){
774 for(j=0; j<=max_order; j++)
775 var[j]= samples[i-j];
776
777 if(pass){
778 double eval, inv, rinv;
779 eval= av_evaluate_lls(&m[(pass-1)&1], var+1, max_order-1);
780 eval= (512>>pass) + fabs(eval - var[0]);
781 inv = 1/eval;
782 rinv = sqrt(inv);
783 for(j=0; j<=max_order; j++)
784 var[j] *= rinv;
785 weight += inv;
786 }else
787 weight++;
788
789 av_update_lls(&m[pass&1], var, 1.0);
790 }
791 av_solve_lls(&m[pass&1], 0.001, 0);
792 }
793
794 for(i=0; i<max_order; i++){
795 for(j=0; j<max_order; j++)
796 lpc[i][j]= m[(pass-1)&1].coeff[i][j];
797 ref[i]= sqrt(m[(pass-1)&1].variance[i] / weight) * (blocksize - max_order) / 4000;
798 }
799 for(i=max_order-1; i>0; i--)
800 ref[i] = ref[i-1] - ref[i];
801 }
802 opt_order = max_order;
803
804 if(omethod == ORDER_METHOD_EST) {
805 opt_order = estimate_best_order(ref, max_order);
806 i = opt_order-1;
807 quantize_lpc_coefs(lpc[i], i+1, precision, coefs[i], &shift[i], max_shift, zero_shift);
808 } else {
809 for(i=0; i<max_order; i++) {
810 quantize_lpc_coefs(lpc[i], i+1, precision, coefs[i], &shift[i], max_shift, zero_shift);
811 }
812 }
813
814 return opt_order;
815 }
816
817
818 static void encode_residual_verbatim(int32_t *res, int32_t *smp, int n)
819 {
820 assert(n > 0);
821 memcpy(res, smp, n * sizeof(int32_t));
822 }
823
824 static void encode_residual_fixed(int32_t *res, const int32_t *smp, int n,
825 int order)
826 {
827 int i;
828
829 for(i=0; i<order; i++) {
830 res[i] = smp[i];
831 }
832
833 if(order==0){
834 for(i=order; i<n; i++)
835 res[i]= smp[i];
836 }else if(order==1){
837 for(i=order; i<n; i++)
838 res[i]= smp[i] - smp[i-1];
839 }else if(order==2){
840 int a = smp[order-1] - smp[order-2];
841 for(i=order; i<n; i+=2) {
842 int b = smp[i] - smp[i-1];
843 res[i]= b - a;
844 a = smp[i+1] - smp[i];
845 res[i+1]= a - b;
846 }
847 }else if(order==3){
848 int a = smp[order-1] - smp[order-2];
849 int c = smp[order-1] - 2*smp[order-2] + smp[order-3];
850 for(i=order; i<n; i+=2) {
851 int b = smp[i] - smp[i-1];
852 int d = b - a;
853 res[i]= d - c;
854 a = smp[i+1] - smp[i];
855 c = a - b;
856 res[i+1]= c - d;
857 }
858 }else{
859 int a = smp[order-1] - smp[order-2];
860 int c = smp[order-1] - 2*smp[order-2] + smp[order-3];
861 int e = smp[order-1] - 3*smp[order-2] + 3*smp[order-3] - smp[order-4];
862 for(i=order; i<n; i+=2) {
863 int b = smp[i] - smp[i-1];
864 int d = b - a;
865 int f = d - c;
866 res[i]= f - e;
867 a = smp[i+1] - smp[i];
868 c = a - b;
869 e = c - d;
870 res[i+1]= e - f;
871 }
872 }
873 }
874
875 #define LPC1(x) {\
876 int c = coefs[(x)-1];\
877 p0 += c*s;\
878 s = smp[i-(x)+1];\
879 p1 += c*s;\
880 }
881
882 static av_always_inline void encode_residual_lpc_unrolled(
883 int32_t *res, const int32_t *smp, int n,
884 int order, const int32_t *coefs, int shift, int big)
885 {
886 int i;
887 for(i=order; i<n; i+=2) {
888 int s = smp[i-order];
889 int p0 = 0, p1 = 0;
890 if(big) {
891 switch(order) {
892 case 32: LPC1(32)
893 case 31: LPC1(31)
894 case 30: LPC1(30)
895 case 29: LPC1(29)
896 case 28: LPC1(28)
897 case 27: LPC1(27)
898 case 26: LPC1(26)
899 case 25: LPC1(25)
900 case 24: LPC1(24)
901 case 23: LPC1(23)
902 case 22: LPC1(22)
903 case 21: LPC1(21)
904 case 20: LPC1(20)
905 case 19: LPC1(19)
906 case 18: LPC1(18)
907 case 17: LPC1(17)
908 case 16: LPC1(16)
909 case 15: LPC1(15)
910 case 14: LPC1(14)
911 case 13: LPC1(13)
912 case 12: LPC1(12)
913 case 11: LPC1(11)
914 case 10: LPC1(10)
915 case 9: LPC1( 9)
916 LPC1( 8)
917 LPC1( 7)
918 LPC1( 6)
919 LPC1( 5)
920 LPC1( 4)
921 LPC1( 3)
922 LPC1( 2)
923 LPC1( 1)
924 }
925 } else {
926 switch(order) {
927 case 8: LPC1( 8)
928 case 7: LPC1( 7)
929 case 6: LPC1( 6)
930 case 5: LPC1( 5)
931 case 4: LPC1( 4)
932 case 3: LPC1( 3)
933 case 2: LPC1( 2)
934 case 1: LPC1( 1)
935 }
936 }
937 res[i ] = smp[i ] - (p0 >> shift);
938 res[i+1] = smp[i+1] - (p1 >> shift);
939 }
940 }
941
942 static void encode_residual_lpc(int32_t *res, const int32_t *smp, int n,
943 int order, const int32_t *coefs, int shift)
944 {
945 int i;
946 for(i=0; i<order; i++) {
947 res[i] = smp[i];
948 }
949 #ifdef CONFIG_SMALL
950 for(i=order; i<n; i+=2) {
951 int j;
952 int s = smp[i];
953 int p0 = 0, p1 = 0;
954 for(j=0; j<order; j++) {
955 int c = coefs[j];
956 p1 += c*s;
957 s = smp[i-j-1];
958 p0 += c*s;
959 }
960 res[i ] = smp[i ] - (p0 >> shift);
961 res[i+1] = smp[i+1] - (p1 >> shift);
962 }
963 #else
964 switch(order) {
965 case 1: encode_residual_lpc_unrolled(res, smp, n, 1, coefs, shift, 0); break;
966 case 2: encode_residual_lpc_unrolled(res, smp, n, 2, coefs, shift, 0); break;
967 case 3: encode_residual_lpc_unrolled(res, smp, n, 3, coefs, shift, 0); break;
968 case 4: encode_residual_lpc_unrolled(res, smp, n, 4, coefs, shift, 0); break;
969 case 5: encode_residual_lpc_unrolled(res, smp, n, 5, coefs, shift, 0); break;
970 case 6: encode_residual_lpc_unrolled(res, smp, n, 6, coefs, shift, 0); break;
971 case 7: encode_residual_lpc_unrolled(res, smp, n, 7, coefs, shift, 0); break;
972 case 8: encode_residual_lpc_unrolled(res, smp, n, 8, coefs, shift, 0); break;
973 default: encode_residual_lpc_unrolled(res, smp, n, order, coefs, shift, 1); break;
974 }
975 #endif
976 }
977
978 static int encode_residual(FlacEncodeContext *ctx, int ch)
979 {
980 int i, n;
981 int min_order, max_order, opt_order, precision, omethod;
982 int min_porder, max_porder;
983 FlacFrame *frame;
984 FlacSubframe *sub;
985 int32_t coefs[MAX_LPC_ORDER][MAX_LPC_ORDER];
986 int shift[MAX_LPC_ORDER];
987 int32_t *res, *smp;
988
989 frame = &ctx->frame;
990 sub = &frame->subframes[ch];
991 res = sub->residual;
992 smp = sub->samples;
993 n = frame->blocksize;
994
995 /* CONSTANT */
996 for(i=1; i<n; i++) {
997 if(smp[i] != smp[0]) break;
998 }
999 if(i == n) {
1000 sub->type = sub->type_code = FLAC_SUBFRAME_CONSTANT;
1001 res[0] = smp[0];
1002 return sub->obits;
1003 }
1004
1005 /* VERBATIM */
1006 if(n < 5) {
1007 sub->type = sub->type_code = FLAC_SUBFRAME_VERBATIM;
1008 encode_residual_verbatim(res, smp, n);
1009 return sub->obits * n;
1010 }
1011
1012 min_order = ctx->options.min_prediction_order;
1013 max_order = ctx->options.max_prediction_order;
1014 min_porder = ctx->options.min_partition_order;
1015 max_porder = ctx->options.max_partition_order;
1016 precision = ctx->options.lpc_coeff_precision;
1017 omethod = ctx->options.prediction_order_method;
1018
1019 /* FIXED */
1020 if(!ctx->options.use_lpc || max_order == 0 || (n <= max_order)) {
1021 uint32_t bits[MAX_FIXED_ORDER+1];
1022 if(max_order > MAX_FIXED_ORDER) max_order = MAX_FIXED_ORDER;
1023 opt_order = 0;
1024 bits[0] = UINT32_MAX;
1025 for(i=min_order; i<=max_order; i++) {
1026 encode_residual_fixed(res, smp, n, i);
1027 bits[i] = calc_rice_params_fixed(&sub->rc, min_porder, max_porder, res,
1028 n, i, sub->obits);
1029 if(bits[i] < bits[opt_order]) {
1030 opt_order = i;
1031 }
1032 }
1033 sub->order = opt_order;
1034 sub->type = FLAC_SUBFRAME_FIXED;
1035 sub->type_code = sub->type | sub->order;
1036 if(sub->order != max_order) {
1037 encode_residual_fixed(res, smp, n, sub->order);
1038 return calc_rice_params_fixed(&sub->rc, min_porder, max_porder, res, n,
1039 sub->order, sub->obits);
1040 }
1041 return bits[sub->order];
1042 }
1043
1044 /* LPC */
1045 opt_order = lpc_calc_coefs(&ctx->dsp, smp, n, max_order, precision, coefs,
1046 shift, ctx->options.use_lpc, omethod, MAX_LPC_SHIFT, 0);
1047
1048 if(omethod == ORDER_METHOD_2LEVEL ||
1049 omethod == ORDER_METHOD_4LEVEL ||
1050 omethod == ORDER_METHOD_8LEVEL) {
1051 int levels = 1 << omethod;
1052 uint32_t bits[levels];
1053 int order;
1054 int opt_index = levels-1;
1055 opt_order = max_order-1;
1056 bits[opt_index] = UINT32_MAX;
1057 for(i=levels-1; i>=0; i--) {
1058 order = min_order + (((max_order-min_order+1) * (i+1)) / levels)-1;
1059 if(order < 0) order = 0;
1060 encode_residual_lpc(res, smp, n, order+1, coefs[order], shift[order]);
1061 bits[i] = calc_rice_params_lpc(&sub->rc, min_porder, max_porder,
1062 res, n, order+1, sub->obits, precision);
1063 if(bits[i] < bits[opt_index]) {
1064 opt_index = i;
1065 opt_order = order;
1066 }
1067 }
1068 opt_order++;
1069 } else if(omethod == ORDER_METHOD_SEARCH) {
1070 // brute-force optimal order search
1071 uint32_t bits[MAX_LPC_ORDER];
1072 opt_order = 0;
1073 bits[0] = UINT32_MAX;
1074 for(i=min_order-1; i<max_order; i++) {
1075 encode_residual_lpc(res, smp, n, i+1, coefs[i], shift[i]);
1076 bits[i] = calc_rice_params_lpc(&sub->rc, min_porder, max_porder,
1077 res, n, i+1, sub->obits, precision);
1078 if(bits[i] < bits[opt_order]) {
1079 opt_order = i;
1080 }
1081 }
1082 opt_order++;
1083 } else if(omethod == ORDER_METHOD_LOG) {
1084 uint32_t bits[MAX_LPC_ORDER];
1085 int step;
1086
1087 opt_order= min_order - 1 + (max_order-min_order)/3;
1088 memset(bits, -1, sizeof(bits));
1089
1090 for(step=16 ;step; step>>=1){
1091 int last= opt_order;
1092 for(i=last-step; i<=last+step; i+= step){
1093 if(i<min_order-1 || i>=max_order || bits[i] < UINT32_MAX)
1094 continue;
1095 encode_residual_lpc(res, smp, n, i+1, coefs[i], shift[i]);
1096 bits[i] = calc_rice_params_lpc(&sub->rc, min_porder, max_porder,
1097 res, n, i+1, sub->obits, precision);
1098 if(bits[i] < bits[opt_order])
1099 opt_order= i;
1100 }
1101 }
1102 opt_order++;
1103 }
1104
1105 sub->order = opt_order;
1106 sub->type = FLAC_SUBFRAME_LPC;
1107 sub->type_code = sub->type | (sub->order-1);
1108 sub->shift = shift[sub->order-1];
1109 for(i=0; i<sub->order; i++) {
1110 sub->coefs[i] = coefs[sub->order-1][i];
1111 }
1112 encode_residual_lpc(res, smp, n, sub->order, sub->coefs, sub->shift);
1113 return calc_rice_params_lpc(&sub->rc, min_porder, max_porder, res, n, sub->order,
1114 sub->obits, precision);
1115 }
1116
1117 static int encode_residual_v(FlacEncodeContext *ctx, int ch)
1118 {
1119 int i, n;
1120 FlacFrame *frame;
1121 FlacSubframe *sub;
1122 int32_t *res, *smp;
1123
1124 frame = &ctx->frame;
1125 sub = &frame->subframes[ch];
1126 res = sub->residual;
1127 smp = sub->samples;
1128 n = frame->blocksize;
1129
1130 /* CONSTANT */
1131 for(i=1; i<n; i++) {
1132 if(smp[i] != smp[0]) break;
1133 }
1134 if(i == n) {
1135 sub->type = sub->type_code = FLAC_SUBFRAME_CONSTANT;
1136 res[0] = smp[0];
1137 return sub->obits;
1138 }
1139
1140 /* VERBATIM */
1141 sub->type = sub->type_code = FLAC_SUBFRAME_VERBATIM;
1142 encode_residual_verbatim(res, smp, n);
1143 return sub->obits * n;
1144 }
1145
1146 static int estimate_stereo_mode(int32_t *left_ch, int32_t *right_ch, int n)
1147 {
1148 int i, best;
1149 int32_t lt, rt;
1150 uint64_t sum[4];
1151 uint64_t score[4];
1152 int k;
1153
1154 /* calculate sum of 2nd order residual for each channel */
1155 sum[0] = sum[1] = sum[2] = sum[3] = 0;
1156 for(i=2; i<n; i++) {
1157 lt = left_ch[i] - 2*left_ch[i-1] + left_ch[i-2];
1158 rt = right_ch[i] - 2*right_ch[i-1] + right_ch[i-2];
1159 sum[2] += FFABS((lt + rt) >> 1);
1160 sum[3] += FFABS(lt - rt);
1161 sum[0] += FFABS(lt);
1162 sum[1] += FFABS(rt);
1163 }
1164 /* estimate bit counts */
1165 for(i=0; i<4; i++) {
1166 k = find_optimal_param(2*sum[i], n);
1167 sum[i] = rice_encode_count(2*sum[i], n, k);
1168 }
1169
1170 /* calculate score for each mode */
1171 score[0] = sum[0] + sum[1];
1172 score[1] = sum[0] + sum[3];
1173 score[2] = sum[1] + sum[3];
1174 score[3] = sum[2] + sum[3];
1175
1176 /* return mode with lowest score */
1177 best = 0;
1178 for(i=1; i<4; i++) {
1179 if(score[i] < score[best]) {
1180 best = i;
1181 }
1182 }
1183 if(best == 0) {
1184 return FLAC_CHMODE_LEFT_RIGHT;
1185 } else if(best == 1) {
1186 return FLAC_CHMODE_LEFT_SIDE;
1187 } else if(best == 2) {
1188 return FLAC_CHMODE_RIGHT_SIDE;
1189 } else {
1190 return FLAC_CHMODE_MID_SIDE;
1191 }
1192 }
1193
1194 /**
1195 * Perform stereo channel decorrelation
1196 */
1197 static void channel_decorrelation(FlacEncodeContext *ctx)
1198 {
1199 FlacFrame *frame;
1200 int32_t *left, *right;
1201 int i, n;
1202
1203 frame = &ctx->frame;
1204 n = frame->blocksize;
1205 left = frame->subframes[0].samples;
1206 right = frame->subframes[1].samples;
1207
1208 if(ctx->channels != 2) {
1209 frame->ch_mode = FLAC_CHMODE_NOT_STEREO;
1210 return;
1211 }
1212
1213 frame->ch_mode = estimate_stereo_mode(left, right, n);
1214
1215 /* perform decorrelation and adjust bits-per-sample */
1216 if(frame->ch_mode == FLAC_CHMODE_LEFT_RIGHT) {
1217 return;
1218 }
1219 if(frame->ch_mode == FLAC_CHMODE_MID_SIDE) {
1220 int32_t tmp;
1221 for(i=0; i<n; i++) {
1222 tmp = left[i];
1223 left[i] = (tmp + right[i]) >> 1;
1224 right[i] = tmp - right[i];
1225 }
1226 frame->subframes[1].obits++;
1227 } else if(frame->ch_mode == FLAC_CHMODE_LEFT_SIDE) {
1228 for(i=0; i<n; i++) {
1229 right[i] = left[i] - right[i];
1230 }
1231 frame->subframes[1].obits++;
1232 } else {
1233 for(i=0; i<n; i++) {
1234 left[i] -= right[i];
1235 }
1236 frame->subframes[0].obits++;
1237 }
1238 }
1239
1240 static void write_utf8(PutBitContext *pb, uint32_t val)
1241 {
1242 uint8_t tmp;
1243 PUT_UTF8(val, tmp, put_bits(pb, 8, tmp);)
1244 }
1245
1246 static void output_frame_header(FlacEncodeContext *s)
1247 {
1248 FlacFrame *frame;
1249 int crc;
1250
1251 frame = &s->frame;
1252
1253 put_bits(&s->pb, 16, 0xFFF8);
1254 put_bits(&s->pb, 4, frame->bs_code[0]);
1255 put_bits(&s->pb, 4, s->sr_code[0]);
1256 if(frame->ch_mode == FLAC_CHMODE_NOT_STEREO) {
1257 put_bits(&s->pb, 4, s->ch_code);
1258 } else {
1259 put_bits(&s->pb, 4, frame->ch_mode);
1260 }
1261 put_bits(&s->pb, 3, 4); /* bits-per-sample code */
1262 put_bits(&s->pb, 1, 0);
1263 write_utf8(&s->pb, s->frame_count);
1264 if(frame->bs_code[0] == 6) {
1265 put_bits(&s->pb, 8, frame->bs_code[1]);
1266 } else if(frame->bs_code[0] == 7) {
1267 put_bits(&s->pb, 16, frame->bs_code[1]);
1268 }
1269 if(s->sr_code[0] == 12) {
1270 put_bits(&s->pb, 8, s->sr_code[1]);
1271 } else if(s->sr_code[0] > 12) {
1272 put_bits(&s->pb, 16, s->sr_code[1]);
1273 }
1274 flush_put_bits(&s->pb);
1275 crc = av_crc(av_crc_get_table(AV_CRC_8_ATM), 0,
1276 s->pb.buf, put_bits_count(&s->pb)>>3);
1277 put_bits(&s->pb, 8, crc);
1278 }
1279
1280 static void output_subframe_constant(FlacEncodeContext *s, int ch)
1281 {
1282 FlacSubframe *sub;
1283 int32_t res;
1284
1285 sub = &s->frame.subframes[ch];
1286 res = sub->residual[0];
1287 put_sbits(&s->pb, sub->obits, res);
1288 }
1289
1290 static void output_subframe_verbatim(FlacEncodeContext *s, int ch)
1291 {
1292 int i;
1293 FlacFrame *frame;
1294 FlacSubframe *sub;
1295 int32_t res;
1296
1297 frame = &s->frame;
1298 sub = &frame->subframes[ch];
1299
1300 for(i=0; i<frame->blocksize; i++) {
1301 res = sub->residual[i];
1302 put_sbits(&s->pb, sub->obits, res);
1303 }
1304 }
1305
1306 static void output_residual(FlacEncodeContext *ctx, int ch)
1307 {
1308 int i, j, p, n, parts;
1309 int k, porder, psize, res_cnt;
1310 FlacFrame *frame;
1311 FlacSubframe *sub;
1312 int32_t *res;
1313
1314 frame = &ctx->frame;
1315 sub = &frame->subframes[ch];
1316 res = sub->residual;
1317 n = frame->blocksize;
1318
1319 /* rice-encoded block */
1320 put_bits(&ctx->pb, 2, 0);
1321
1322 /* partition order */
1323 porder = sub->rc.porder;
1324 psize = n >> porder;
1325 parts = (1 << porder);
1326 put_bits(&ctx->pb, 4, porder);
1327 res_cnt = psize - sub->order;
1328
1329 /* residual */
1330 j = sub->order;
1331 for(p=0; p<parts; p++) {
1332 k = sub->rc.params[p];
1333 put_bits(&ctx->pb, 4, k);
1334 if(p == 1) res_cnt = psize;
1335 for(i=0; i<res_cnt && j<n; i++, j++) {
1336 set_sr_golomb_flac(&ctx->pb, res[j], k, INT32_MAX, 0);
1337 }
1338 }
1339 }
1340
1341 static void output_subframe_fixed(FlacEncodeContext *ctx, int ch)
1342 {
1343 int i;
1344 FlacFrame *frame;
1345 FlacSubframe *sub;
1346
1347 frame = &ctx->frame;
1348 sub = &frame->subframes[ch];
1349
1350 /* warm-up samples */
1351 for(i=0; i<sub->order; i++) {
1352 put_sbits(&ctx->pb, sub->obits, sub->residual[i]);
1353 }
1354
1355 /* residual */
1356 output_residual(ctx, ch);
1357 }
1358
1359 static void output_subframe_lpc(FlacEncodeContext *ctx, int ch)
1360 {
1361 int i, cbits;
1362 FlacFrame *frame;
1363 FlacSubframe *sub;
1364
1365 frame = &ctx->frame;
1366 sub = &frame->subframes[ch];
1367
1368 /* warm-up samples */
1369 for(i=0; i<sub->order; i++) {
1370 put_sbits(&ctx->pb, sub->obits, sub->residual[i]);
1371 }
1372
1373 /* LPC coefficients */
1374 cbits = ctx->options.lpc_coeff_precision;
1375 put_bits(&ctx->pb, 4, cbits-1);
1376 put_sbits(&ctx->pb, 5, sub->shift);
1377 for(i=0; i<sub->order; i++) {
1378 put_sbits(&ctx->pb, cbits, sub->coefs[i]);
1379 }
1380
1381 /* residual */
1382 output_residual(ctx, ch);
1383 }
1384
1385 static void output_subframes(FlacEncodeContext *s)
1386 {
1387 FlacFrame *frame;
1388 FlacSubframe *sub;
1389 int ch;
1390
1391 frame = &s->frame;
1392
1393 for(ch=0; ch<s->channels; ch++) {
1394 sub = &frame->subframes[ch];
1395
1396 /* subframe header */
1397 put_bits(&s->pb, 1, 0);
1398 put_bits(&s->pb, 6, sub->type_code);
1399 put_bits(&s->pb, 1, 0); /* no wasted bits */
1400
1401 /* subframe */
1402 if(sub->type == FLAC_SUBFRAME_CONSTANT) {
1403 output_subframe_constant(s, ch);
1404 } else if(sub->type == FLAC_SUBFRAME_VERBATIM) {
1405 output_subframe_verbatim(s, ch);
1406 } else if(sub->type == FLAC_SUBFRAME_FIXED) {
1407 output_subframe_fixed(s, ch);
1408 } else if(sub->type == FLAC_SUBFRAME_LPC) {
1409 output_subframe_lpc(s, ch);
1410 }
1411 }
1412 }
1413
1414 static void output_frame_footer(FlacEncodeContext *s)
1415 {
1416 int crc;
1417 flush_put_bits(&s->pb);
1418 crc = bswap_16(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0,
1419 s->pb.buf, put_bits_count(&s->pb)>>3));
1420 put_bits(&s->pb, 16, crc);
1421 flush_put_bits(&s->pb);
1422 }
1423
1424 static int flac_encode_frame(AVCodecContext *avctx, uint8_t *frame,
1425 int buf_size, void *data)
1426 {
1427 int ch;
1428 FlacEncodeContext *s;
1429 int16_t *samples = data;
1430 int out_bytes;
1431
1432 s = avctx->priv_data;
1433
1434 init_frame(s);
1435
1436 copy_samples(s, samples);
1437
1438 channel_decorrelation(s);
1439
1440 for(ch=0; ch<s->channels; ch++) {
1441 encode_residual(s, ch);
1442 }
1443 init_put_bits(&s->pb, frame, buf_size);
1444 output_frame_header(s);
1445 output_subframes(s);
1446 output_frame_footer(s);
1447 out_bytes = put_bits_count(&s->pb) >> 3;
1448
1449 if(out_bytes > s->max_framesize || out_bytes >= buf_size) {
1450 /* frame too large. use verbatim mode */
1451 for(ch=0; ch<s->channels; ch++) {
1452 encode_residual_v(s, ch);
1453 }
1454 init_put_bits(&s->pb, frame, buf_size);
1455 output_frame_header(s);
1456 output_subframes(s);
1457 output_frame_footer(s);
1458 out_bytes = put_bits_count(&s->pb) >> 3;
1459
1460 if(out_bytes > s->max_framesize || out_bytes >= buf_size) {
1461 /* still too large. must be an error. */
1462 av_log(avctx, AV_LOG_ERROR, "error encoding frame\n");
1463 return -1;
1464 }
1465 }
1466
1467 s->frame_count++;
1468 return out_bytes;
1469 }
1470
1471 static av_cold int flac_encode_close(AVCodecContext *avctx)
1472 {
1473 av_freep(&avctx->extradata);
1474 avctx->extradata_size = 0;
1475 av_freep(&avctx->coded_frame);
1476 return 0;
1477 }
1478
1479 AVCodec flac_encoder = {
1480 "flac",
1481 CODEC_TYPE_AUDIO,
1482 CODEC_ID_FLAC,
1483 sizeof(FlacEncodeContext),
1484 flac_encode_init,
1485 flac_encode_frame,
1486 flac_encode_close,
1487 NULL,
1488 .capabilities = CODEC_CAP_SMALL_LAST_FRAME,
1489 .sample_fmts = (enum SampleFormat[]){SAMPLE_FMT_S16,SAMPLE_FMT_NONE},
1490 .long_name = NULL_IF_CONFIG_SMALL("FLAC (Free Lossless Audio Codec)"),
1491 };