3 * Copyright (c) 2006 Justin Ruggles <jruggle@earthlink.net>
5 * This library is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU Lesser General Public
7 * License as published by the Free Software Foundation; either
8 * version 2 of the License, or (at your option) any later version.
10 * This library is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Lesser General Public License for more details.
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "bitstream.h"
26 #define FLAC_MIN_BLOCKSIZE 16
27 #define FLAC_MAX_BLOCKSIZE 65535
29 #define FLAC_SUBFRAME_CONSTANT 0
30 #define FLAC_SUBFRAME_VERBATIM 1
31 #define FLAC_SUBFRAME_FIXED 8
32 #define FLAC_SUBFRAME_LPC 32
34 #define FLAC_CHMODE_NOT_STEREO 0
35 #define FLAC_CHMODE_LEFT_RIGHT 1
36 #define FLAC_CHMODE_LEFT_SIDE 8
37 #define FLAC_CHMODE_RIGHT_SIDE 9
38 #define FLAC_CHMODE_MID_SIDE 10
40 #define ORDER_METHOD_EST 0
41 #define ORDER_METHOD_2LEVEL 1
42 #define ORDER_METHOD_4LEVEL 2
43 #define ORDER_METHOD_8LEVEL 3
44 #define ORDER_METHOD_SEARCH 4
46 #define FLAC_STREAMINFO_SIZE 34
48 #define MIN_LPC_ORDER 1
49 #define MAX_LPC_ORDER 32
50 #define MAX_FIXED_ORDER 4
51 #define MAX_PARTITION_ORDER 8
52 #define MAX_PARTITIONS (1 << MAX_PARTITION_ORDER)
53 #define MAX_LPC_PRECISION 15
54 #define MAX_LPC_SHIFT 15
55 #define MAX_RICE_PARAM 14
57 typedef struct CompressionOptions {
58 int compression_level;
61 int lpc_coeff_precision;
62 int min_prediction_order;
63 int max_prediction_order;
64 int prediction_order_method;
65 int min_partition_order;
66 int max_partition_order;
69 typedef struct RiceContext {
71 int params[MAX_PARTITIONS];
74 typedef struct FlacSubframe {
79 int32_t coefs[MAX_LPC_ORDER];
82 int32_t samples[FLAC_MAX_BLOCKSIZE];
83 int32_t residual[FLAC_MAX_BLOCKSIZE];
86 typedef struct FlacFrame {
87 FlacSubframe subframes[FLAC_MAX_CH];
94 typedef struct FlacEncodeContext {
102 uint32_t frame_count;
104 CompressionOptions options;
105 AVCodecContext *avctx;
108 static const int flac_samplerates[16] = {
110 8000, 16000, 22050, 24000, 32000, 44100, 48000, 96000,
114 static const int flac_blocksizes[16] = {
117 576, 1152, 2304, 4608,
119 256, 512, 1024, 2048, 4096, 8192, 16384, 32768
123 * Writes streaminfo metadata block to byte array
125 static void write_streaminfo(FlacEncodeContext *s, uint8_t *header)
129 memset(header, 0, FLAC_STREAMINFO_SIZE);
130 init_put_bits(&pb, header, FLAC_STREAMINFO_SIZE);
132 /* streaminfo metadata block */
133 put_bits(&pb, 16, s->blocksize);
134 put_bits(&pb, 16, s->blocksize);
135 put_bits(&pb, 24, 0);
136 put_bits(&pb, 24, s->max_framesize);
137 put_bits(&pb, 20, s->samplerate);
138 put_bits(&pb, 3, s->channels-1);
139 put_bits(&pb, 5, 15); /* bits per sample - 1 */
141 /* total samples = 0 */
142 /* MD5 signature = 0 */
146 * Sets blocksize based on samplerate
147 * Chooses the closest predefined blocksize >= BLOCK_TIME_MS milliseconds
149 static int select_blocksize(int samplerate, int block_time_ms)
155 assert(samplerate > 0);
156 blocksize = flac_blocksizes[1];
157 target = (samplerate * block_time_ms) / 1000;
158 for(i=0; i<16; i++) {
159 if(target >= flac_blocksizes[i] && flac_blocksizes[i] > blocksize) {
160 blocksize = flac_blocksizes[i];
166 static int flac_encode_init(AVCodecContext *avctx)
168 int freq = avctx->sample_rate;
169 int channels = avctx->channels;
170 FlacEncodeContext *s = avctx->priv_data;
176 if(avctx->sample_fmt != SAMPLE_FMT_S16) {
180 if(channels < 1 || channels > FLAC_MAX_CH) {
183 s->channels = channels;
184 s->ch_code = s->channels-1;
186 /* find samplerate in table */
189 for(i=4; i<12; i++) {
190 if(freq == flac_samplerates[i]) {
191 s->samplerate = flac_samplerates[i];
197 /* if not in table, samplerate is non-standard */
199 if(freq % 1000 == 0 && freq < 255000) {
201 s->sr_code[1] = freq / 1000;
202 } else if(freq % 10 == 0 && freq < 655350) {
204 s->sr_code[1] = freq / 10;
205 } else if(freq < 65535) {
207 s->sr_code[1] = freq;
211 s->samplerate = freq;
214 /* set compression option defaults based on avctx->compression_level */
215 if(avctx->compression_level < 0) {
216 s->options.compression_level = 5;
218 s->options.compression_level = avctx->compression_level;
220 av_log(avctx, AV_LOG_DEBUG, " compression: %d\n", s->options.compression_level);
222 level= s->options.compression_level;
224 av_log(avctx, AV_LOG_ERROR, "invalid compression level: %d\n",
225 s->options.compression_level);
229 s->options.block_time_ms = ((int[]){ 27, 27, 27,105,105,105})[level];
230 s->options.use_lpc = ((int[]){ 0, 0, 0, 1, 1, 1})[level];
231 s->options.min_prediction_order= ((int[]){ 2, 0, 0, 1, 1, 1})[level];
232 s->options.max_prediction_order= ((int[]){ 3, 4, 4, 6, 8, 8})[level];
233 s->options.prediction_order_method = ORDER_METHOD_EST;
234 s->options.min_partition_order = ((int[]){ 2, 2, 0, 0, 0, 0})[level];
235 s->options.max_partition_order = ((int[]){ 2, 2, 3, 3, 3, 8})[level];
237 /* set compression option overrides from AVCodecContext */
238 if(avctx->use_lpc >= 0) {
239 s->options.use_lpc = !!avctx->use_lpc;
241 av_log(avctx, AV_LOG_DEBUG, " use lpc: %s\n",
242 s->options.use_lpc? "yes" : "no");
244 if(avctx->min_prediction_order >= 0) {
245 if(s->options.use_lpc) {
246 if(avctx->min_prediction_order < MIN_LPC_ORDER ||
247 avctx->min_prediction_order > MAX_LPC_ORDER) {
248 av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n",
249 avctx->min_prediction_order);
253 if(avctx->min_prediction_order > MAX_FIXED_ORDER) {
254 av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n",
255 avctx->min_prediction_order);
259 s->options.min_prediction_order = avctx->min_prediction_order;
261 if(avctx->max_prediction_order >= 0) {
262 if(s->options.use_lpc) {
263 if(avctx->max_prediction_order < MIN_LPC_ORDER ||
264 avctx->max_prediction_order > MAX_LPC_ORDER) {
265 av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n",
266 avctx->max_prediction_order);
270 if(avctx->max_prediction_order > MAX_FIXED_ORDER) {
271 av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n",
272 avctx->max_prediction_order);
276 s->options.max_prediction_order = avctx->max_prediction_order;
278 if(s->options.max_prediction_order < s->options.min_prediction_order) {
279 av_log(avctx, AV_LOG_ERROR, "invalid prediction orders: min=%d max=%d\n",
280 s->options.min_prediction_order, s->options.max_prediction_order);
283 av_log(avctx, AV_LOG_DEBUG, " prediction order: %d, %d\n",
284 s->options.min_prediction_order, s->options.max_prediction_order);
286 if(avctx->prediction_order_method >= 0) {
287 if(avctx->prediction_order_method > ORDER_METHOD_SEARCH) {
288 av_log(avctx, AV_LOG_ERROR, "invalid prediction order method: %d\n",
289 avctx->prediction_order_method);
292 s->options.prediction_order_method = avctx->prediction_order_method;
294 switch(avctx->prediction_order_method) {
295 case ORDER_METHOD_EST: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
297 case ORDER_METHOD_2LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
299 case ORDER_METHOD_4LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
301 case ORDER_METHOD_8LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
303 case ORDER_METHOD_SEARCH: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
304 "full search"); break;
307 if(avctx->min_partition_order >= 0) {
308 if(avctx->min_partition_order > MAX_PARTITION_ORDER) {
309 av_log(avctx, AV_LOG_ERROR, "invalid min partition order: %d\n",
310 avctx->min_partition_order);
313 s->options.min_partition_order = avctx->min_partition_order;
315 if(avctx->max_partition_order >= 0) {
316 if(avctx->max_partition_order > MAX_PARTITION_ORDER) {
317 av_log(avctx, AV_LOG_ERROR, "invalid max partition order: %d\n",
318 avctx->max_partition_order);
321 s->options.max_partition_order = avctx->max_partition_order;
323 if(s->options.max_partition_order < s->options.min_partition_order) {
324 av_log(avctx, AV_LOG_ERROR, "invalid partition orders: min=%d max=%d\n",
325 s->options.min_partition_order, s->options.max_partition_order);
328 av_log(avctx, AV_LOG_DEBUG, " partition order: %d, %d\n",
329 s->options.min_partition_order, s->options.max_partition_order);
331 if(avctx->frame_size > 0) {
332 if(avctx->frame_size < FLAC_MIN_BLOCKSIZE ||
333 avctx->frame_size > FLAC_MIN_BLOCKSIZE) {
334 av_log(avctx, AV_LOG_ERROR, "invalid block size: %d\n",
338 s->blocksize = avctx->frame_size;
340 s->blocksize = select_blocksize(s->samplerate, s->options.block_time_ms);
341 avctx->frame_size = s->blocksize;
343 av_log(avctx, AV_LOG_DEBUG, " block size: %d\n", s->blocksize);
345 /* set LPC precision */
346 if(avctx->lpc_coeff_precision > 0) {
347 if(avctx->lpc_coeff_precision > MAX_LPC_PRECISION) {
348 av_log(avctx, AV_LOG_ERROR, "invalid lpc coeff precision: %d\n",
349 avctx->lpc_coeff_precision);
352 s->options.lpc_coeff_precision = avctx->lpc_coeff_precision;
354 /* select LPC precision based on block size */
355 if( s->blocksize <= 192) s->options.lpc_coeff_precision = 7;
356 else if(s->blocksize <= 384) s->options.lpc_coeff_precision = 8;
357 else if(s->blocksize <= 576) s->options.lpc_coeff_precision = 9;
358 else if(s->blocksize <= 1152) s->options.lpc_coeff_precision = 10;
359 else if(s->blocksize <= 2304) s->options.lpc_coeff_precision = 11;
360 else if(s->blocksize <= 4608) s->options.lpc_coeff_precision = 12;
361 else if(s->blocksize <= 8192) s->options.lpc_coeff_precision = 13;
362 else if(s->blocksize <= 16384) s->options.lpc_coeff_precision = 14;
363 else s->options.lpc_coeff_precision = 15;
365 av_log(avctx, AV_LOG_DEBUG, " lpc precision: %d\n",
366 s->options.lpc_coeff_precision);
368 /* set maximum encoded frame size in verbatim mode */
369 if(s->channels == 2) {
370 s->max_framesize = 14 + ((s->blocksize * 33 + 7) >> 3);
372 s->max_framesize = 14 + (s->blocksize * s->channels * 2);
375 streaminfo = av_malloc(FLAC_STREAMINFO_SIZE);
376 write_streaminfo(s, streaminfo);
377 avctx->extradata = streaminfo;
378 avctx->extradata_size = FLAC_STREAMINFO_SIZE;
382 avctx->coded_frame = avcodec_alloc_frame();
383 avctx->coded_frame->key_frame = 1;
388 static void init_frame(FlacEncodeContext *s)
395 for(i=0; i<16; i++) {
396 if(s->blocksize == flac_blocksizes[i]) {
397 frame->blocksize = flac_blocksizes[i];
398 frame->bs_code[0] = i;
399 frame->bs_code[1] = 0;
404 frame->blocksize = s->blocksize;
405 if(frame->blocksize <= 256) {
406 frame->bs_code[0] = 6;
407 frame->bs_code[1] = frame->blocksize-1;
409 frame->bs_code[0] = 7;
410 frame->bs_code[1] = frame->blocksize-1;
414 for(ch=0; ch<s->channels; ch++) {
415 frame->subframes[ch].obits = 16;
420 * Copy channel-interleaved input samples into separate subframes
422 static void copy_samples(FlacEncodeContext *s, int16_t *samples)
428 for(i=0,j=0; i<frame->blocksize; i++) {
429 for(ch=0; ch<s->channels; ch++,j++) {
430 frame->subframes[ch].samples[i] = samples[j];
436 #define rice_encode_count(sum, n, k) (((n)*((k)+1))+((sum-(n>>1))>>(k)))
438 static int find_optimal_param(uint32_t sum, int n)
441 uint32_t nbits[MAX_RICE_PARAM+1];
444 nbits[0] = UINT32_MAX;
445 for(k=0; k<=MAX_RICE_PARAM; k++) {
446 nbits[k] = rice_encode_count(sum, n, k);
447 if(nbits[k] < nbits[k_opt]) {
454 static uint32_t calc_optimal_rice_params(RiceContext *rc, int porder,
455 uint32_t *sums, int n, int pred_order)
461 part = (1 << porder);
464 cnt = (n >> porder) - pred_order;
465 for(i=0; i<part; i++) {
466 if(i == 1) cnt = (n >> porder);
467 k = find_optimal_param(sums[i], cnt);
469 all_bits += rice_encode_count(sums[i], cnt, k);
471 all_bits += (4 * part);
478 static void calc_sums(int pmin, int pmax, uint32_t *data, int n, int pred_order,
479 uint32_t sums[][MAX_PARTITIONS])
483 uint32_t *res, *res_end;
485 /* sums for highest level */
487 res = &data[pred_order];
488 res_end = &data[n >> pmax];
489 for(i=0; i<parts; i++) {
491 while(res < res_end){
492 sums[pmax][i] += *(res++);
496 /* sums for lower levels */
497 for(i=pmax-1; i>=pmin; i--) {
499 for(j=0; j<parts; j++) {
500 sums[i][j] = sums[i+1][2*j] + sums[i+1][2*j+1];
505 static uint32_t calc_rice_params(RiceContext *rc, int pmin, int pmax,
506 int32_t *data, int n, int pred_order)
509 uint32_t bits[MAX_PARTITION_ORDER+1];
513 uint32_t sums[MAX_PARTITION_ORDER+1][MAX_PARTITIONS];
515 assert(pmin >= 0 && pmin <= MAX_PARTITION_ORDER);
516 assert(pmax >= 0 && pmax <= MAX_PARTITION_ORDER);
517 assert(pmin <= pmax);
519 udata = av_malloc(n * sizeof(uint32_t));
521 udata[i] = (2*data[i]) ^ (data[i]>>31);
524 calc_sums(pmin, pmax, udata, n, pred_order, sums);
527 bits[pmin] = UINT32_MAX;
528 for(i=pmin; i<=pmax; i++) {
529 bits[i] = calc_optimal_rice_params(&tmp_rc, i, sums[i], n, pred_order);
530 if(bits[i] <= bits[opt_porder]) {
537 return bits[opt_porder];
540 static uint32_t calc_rice_params_fixed(RiceContext *rc, int pmin, int pmax,
541 int32_t *data, int n, int pred_order,
545 bits = pred_order*bps + 6;
546 bits += calc_rice_params(rc, pmin, pmax, data, n, pred_order);
550 static uint32_t calc_rice_params_lpc(RiceContext *rc, int pmin, int pmax,
551 int32_t *data, int n, int pred_order,
552 int bps, int precision)
555 bits = pred_order*bps + 4 + 5 + pred_order*precision + 6;
556 bits += calc_rice_params(rc, pmin, pmax, data, n, pred_order);
561 * Apply Welch window function to audio block
563 static void apply_welch_window(const int32_t *data, int len, double *w_data)
570 c = 2.0 / (len - 1.0);
571 for(i=0; i<n2; i++) {
574 w_data[i] = data[i] * w;
575 w_data[len-1-i] = data[len-1-i] * w;
580 * Calculates autocorrelation data from audio samples
581 * A Welch window function is applied before calculation.
583 static void compute_autocorr(const int32_t *data, int len, int lag,
587 double tmp[len + lag];
588 double *data1= tmp + lag;
590 apply_welch_window(data, len, data1);
592 for(i=0; i<lag; i++){
597 for(i=0; i<len; i++){
598 for(lag_ptr= i-lag; lag_ptr<=i; lag_ptr++){
599 autoc[i-lag_ptr] += data1[i] * data1[lag_ptr];
605 * Levinson-Durbin recursion.
606 * Produces LPC coefficients from autocorrelation data.
608 static void compute_lpc_coefs(const double *autoc, int max_order,
609 double lpc[][MAX_LPC_ORDER], double *ref)
613 double lpc_tmp[MAX_LPC_ORDER];
615 for(i=0; i<max_order; i++) lpc_tmp[i] = 0;
618 for(i=0; i<max_order; i++) {
621 r -= lpc_tmp[j] * autoc[i-j];
626 err *= 1.0 - (r * r);
630 for(j=0; j<i2; j++) {
632 lpc_tmp[j] += r * lpc_tmp[i-1-j];
633 lpc_tmp[i-1-j] += r * tmp;
636 lpc_tmp[j] += lpc_tmp[j] * r;
639 for(j=0; j<=i; j++) {
640 lpc[i][j] = -lpc_tmp[j];
646 * Quantize LPC coefficients
648 static void quantize_lpc_coefs(double *lpc_in, int order, int precision,
649 int32_t *lpc_out, int *shift)
656 /* define maximum levels */
657 qmax = (1 << (precision - 1)) - 1;
659 /* find maximum coefficient value */
661 for(i=0; i<order; i++) {
662 cmax= FFMAX(cmax, fabs(lpc_in[i]));
665 /* if maximum value quantizes to zero, return all zeros */
666 if(cmax * (1 << MAX_LPC_SHIFT) < 1.0) {
668 memset(lpc_out, 0, sizeof(int32_t) * order);
672 /* calculate level shift which scales max coeff to available bits */
674 while((cmax * (1 << sh) > qmax) && (sh > 0)) {
678 /* since negative shift values are unsupported in decoder, scale down
679 coefficients instead */
680 if(sh == 0 && cmax > qmax) {
681 double scale = ((double)qmax) / cmax;
682 for(i=0; i<order; i++) {
687 /* output quantized coefficients and level shift */
688 for(i=0; i<order; i++) {
689 lpc_out[i] = (int32_t)(lpc_in[i] * (1 << sh));
694 static int estimate_best_order(double *ref, int max_order)
699 for(i=max_order-1; i>=0; i--) {
709 * Calculate LPC coefficients for multiple orders
711 static int lpc_calc_coefs(const int32_t *samples, int blocksize, int max_order,
712 int precision, int32_t coefs[][MAX_LPC_ORDER],
715 double autoc[MAX_LPC_ORDER+1];
716 double ref[MAX_LPC_ORDER];
717 double lpc[MAX_LPC_ORDER][MAX_LPC_ORDER];
721 assert(max_order >= MIN_LPC_ORDER && max_order <= MAX_LPC_ORDER);
723 compute_autocorr(samples, blocksize, max_order+1, autoc);
725 compute_lpc_coefs(autoc, max_order, lpc, ref);
727 opt_order = estimate_best_order(ref, max_order);
730 quantize_lpc_coefs(lpc[i], i+1, precision, coefs[i], &shift[i]);
736 static void encode_residual_verbatim(int32_t *res, int32_t *smp, int n)
739 memcpy(res, smp, n * sizeof(int32_t));
742 static void encode_residual_fixed(int32_t *res, const int32_t *smp, int n,
747 for(i=0; i<order; i++) {
752 for(i=order; i<n; i++)
755 for(i=order; i<n; i++)
756 res[i]= smp[i] - smp[i-1];
758 for(i=order; i<n; i++)
759 res[i]= smp[i] - 2*smp[i-1] + smp[i-2];
761 for(i=order; i<n; i++)
762 res[i]= smp[i] - 3*smp[i-1] + 3*smp[i-2] - smp[i-3];
764 for(i=order; i<n; i++)
765 res[i]= smp[i] - 4*smp[i-1] + 6*smp[i-2] - 4*smp[i-3] + smp[i-4];
769 static void encode_residual_lpc(int32_t *res, const int32_t *smp, int n,
770 int order, const int32_t *coefs, int shift)
775 for(i=0; i<order; i++) {
778 for(i=order; i<n; i++) {
780 for(j=0; j<order; j++) {
781 pred += coefs[j] * smp[i-j-1];
783 res[i] = smp[i] - (pred >> shift);
787 static int get_max_p_order(int max_porder, int n, int order)
789 int porder, max_parts;
791 for(porder = max_porder; porder > 0; porder--) {
792 max_parts = (1 << porder);
793 if(!(n % max_parts) && (n > max_parts*order)) {
800 static int encode_residual(FlacEncodeContext *ctx, int ch)
803 int min_order, max_order, opt_order, precision;
804 int porder, min_porder, max_porder;
807 int32_t coefs[MAX_LPC_ORDER][MAX_LPC_ORDER];
808 int shift[MAX_LPC_ORDER];
812 sub = &frame->subframes[ch];
815 n = frame->blocksize;
819 if(smp[i] != smp[0]) break;
822 sub->type = sub->type_code = FLAC_SUBFRAME_CONSTANT;
829 sub->type = sub->type_code = FLAC_SUBFRAME_VERBATIM;
830 encode_residual_verbatim(res, smp, n);
831 return sub->obits * n;
834 min_order = ctx->options.min_prediction_order;
835 max_order = ctx->options.max_prediction_order;
836 min_porder = ctx->options.min_partition_order;
837 max_porder = ctx->options.max_partition_order;
838 precision = ctx->options.lpc_coeff_precision;
841 if(!ctx->options.use_lpc || max_order == 0 || (n <= max_order)) {
842 uint32_t bits[MAX_FIXED_ORDER+1];
843 if(max_order > MAX_FIXED_ORDER) max_order = MAX_FIXED_ORDER;
845 bits[0] = UINT32_MAX;
846 for(i=min_order; i<=max_order; i++) {
847 encode_residual_fixed(res, smp, n, i);
848 porder = get_max_p_order(max_porder, n, i);
849 bits[i] = calc_rice_params_fixed(&sub->rc, min_porder, porder, res,
851 if(bits[i] < bits[opt_order]) {
855 sub->order = opt_order;
856 sub->type = FLAC_SUBFRAME_FIXED;
857 sub->type_code = sub->type | sub->order;
858 if(sub->order != max_order) {
859 encode_residual_fixed(res, smp, n, sub->order);
860 porder = get_max_p_order(max_porder, n, sub->order);
861 return calc_rice_params_fixed(&sub->rc, min_porder, porder, res, n,
862 sub->order, sub->obits);
864 return bits[sub->order];
868 sub->order = lpc_calc_coefs(smp, n, max_order, precision, coefs, shift);
869 sub->type = FLAC_SUBFRAME_LPC;
870 sub->type_code = sub->type | (sub->order-1);
871 sub->shift = shift[sub->order-1];
872 for(i=0; i<sub->order; i++) {
873 sub->coefs[i] = coefs[sub->order-1][i];
875 porder = get_max_p_order(max_porder, n, sub->order);
876 encode_residual_lpc(res, smp, n, sub->order, sub->coefs, sub->shift);
877 return calc_rice_params_lpc(&sub->rc, 0, porder, res, n, sub->order,
878 sub->obits, precision);
881 static int encode_residual_v(FlacEncodeContext *ctx, int ch)
889 sub = &frame->subframes[ch];
892 n = frame->blocksize;
896 if(smp[i] != smp[0]) break;
899 sub->type = sub->type_code = FLAC_SUBFRAME_CONSTANT;
905 sub->type = sub->type_code = FLAC_SUBFRAME_VERBATIM;
906 encode_residual_verbatim(res, smp, n);
907 return sub->obits * n;
910 static int estimate_stereo_mode(int32_t *left_ch, int32_t *right_ch, int n)
918 /* calculate sum of 2nd order residual for each channel */
919 sum[0] = sum[1] = sum[2] = sum[3] = 0;
921 lt = left_ch[i] - 2*left_ch[i-1] + left_ch[i-2];
922 rt = right_ch[i] - 2*right_ch[i-1] + right_ch[i-2];
923 sum[2] += ABS((lt + rt) >> 1);
924 sum[3] += ABS(lt - rt);
928 /* estimate bit counts */
930 k = find_optimal_param(2*sum[i], n);
931 sum[i] = rice_encode_count(2*sum[i], n, k);
934 /* calculate score for each mode */
935 score[0] = sum[0] + sum[1];
936 score[1] = sum[0] + sum[3];
937 score[2] = sum[1] + sum[3];
938 score[3] = sum[2] + sum[3];
940 /* return mode with lowest score */
943 if(score[i] < score[best]) {
948 return FLAC_CHMODE_LEFT_RIGHT;
949 } else if(best == 1) {
950 return FLAC_CHMODE_LEFT_SIDE;
951 } else if(best == 2) {
952 return FLAC_CHMODE_RIGHT_SIDE;
954 return FLAC_CHMODE_MID_SIDE;
959 * Perform stereo channel decorrelation
961 static void channel_decorrelation(FlacEncodeContext *ctx)
964 int32_t *left, *right;
968 n = frame->blocksize;
969 left = frame->subframes[0].samples;
970 right = frame->subframes[1].samples;
972 if(ctx->channels != 2) {
973 frame->ch_mode = FLAC_CHMODE_NOT_STEREO;
977 frame->ch_mode = estimate_stereo_mode(left, right, n);
979 /* perform decorrelation and adjust bits-per-sample */
980 if(frame->ch_mode == FLAC_CHMODE_LEFT_RIGHT) {
983 if(frame->ch_mode == FLAC_CHMODE_MID_SIDE) {
987 left[i] = (tmp + right[i]) >> 1;
988 right[i] = tmp - right[i];
990 frame->subframes[1].obits++;
991 } else if(frame->ch_mode == FLAC_CHMODE_LEFT_SIDE) {
993 right[i] = left[i] - right[i];
995 frame->subframes[1].obits++;
1000 frame->subframes[0].obits++;
1004 static void put_sbits(PutBitContext *pb, int bits, int32_t val)
1006 assert(bits >= 0 && bits <= 31);
1008 put_bits(pb, bits, val & ((1<<bits)-1));
1011 static void write_utf8(PutBitContext *pb, uint32_t val)
1016 put_bits(pb, 8, val);
1020 bytes= (av_log2(val)+4) / 5;
1021 shift = (bytes - 1) * 6;
1022 put_bits(pb, 8, (256 - (256>>bytes)) | (val >> shift));
1025 put_bits(pb, 8, 0x80 | ((val >> shift) & 0x3F));
1029 static void output_frame_header(FlacEncodeContext *s)
1036 put_bits(&s->pb, 16, 0xFFF8);
1037 put_bits(&s->pb, 4, frame->bs_code[0]);
1038 put_bits(&s->pb, 4, s->sr_code[0]);
1039 if(frame->ch_mode == FLAC_CHMODE_NOT_STEREO) {
1040 put_bits(&s->pb, 4, s->ch_code);
1042 put_bits(&s->pb, 4, frame->ch_mode);
1044 put_bits(&s->pb, 3, 4); /* bits-per-sample code */
1045 put_bits(&s->pb, 1, 0);
1046 write_utf8(&s->pb, s->frame_count);
1047 if(frame->bs_code[0] == 6) {
1048 put_bits(&s->pb, 8, frame->bs_code[1]);
1049 } else if(frame->bs_code[0] == 7) {
1050 put_bits(&s->pb, 16, frame->bs_code[1]);
1052 if(s->sr_code[0] == 12) {
1053 put_bits(&s->pb, 8, s->sr_code[1]);
1054 } else if(s->sr_code[0] > 12) {
1055 put_bits(&s->pb, 16, s->sr_code[1]);
1057 flush_put_bits(&s->pb);
1058 crc = av_crc(av_crc07, 0, s->pb.buf, put_bits_count(&s->pb)>>3);
1059 put_bits(&s->pb, 8, crc);
1062 static void output_subframe_constant(FlacEncodeContext *s, int ch)
1067 sub = &s->frame.subframes[ch];
1068 res = sub->residual[0];
1069 put_sbits(&s->pb, sub->obits, res);
1072 static void output_subframe_verbatim(FlacEncodeContext *s, int ch)
1080 sub = &frame->subframes[ch];
1082 for(i=0; i<frame->blocksize; i++) {
1083 res = sub->residual[i];
1084 put_sbits(&s->pb, sub->obits, res);
1088 static void output_residual(FlacEncodeContext *ctx, int ch)
1090 int i, j, p, n, parts;
1091 int k, porder, psize, res_cnt;
1096 frame = &ctx->frame;
1097 sub = &frame->subframes[ch];
1098 res = sub->residual;
1099 n = frame->blocksize;
1101 /* rice-encoded block */
1102 put_bits(&ctx->pb, 2, 0);
1104 /* partition order */
1105 porder = sub->rc.porder;
1106 psize = n >> porder;
1107 parts = (1 << porder);
1108 put_bits(&ctx->pb, 4, porder);
1109 res_cnt = psize - sub->order;
1113 for(p=0; p<parts; p++) {
1114 k = sub->rc.params[p];
1115 put_bits(&ctx->pb, 4, k);
1116 if(p == 1) res_cnt = psize;
1117 for(i=0; i<res_cnt && j<n; i++, j++) {
1118 set_sr_golomb_flac(&ctx->pb, res[j], k, INT32_MAX, 0);
1123 static void output_subframe_fixed(FlacEncodeContext *ctx, int ch)
1129 frame = &ctx->frame;
1130 sub = &frame->subframes[ch];
1132 /* warm-up samples */
1133 for(i=0; i<sub->order; i++) {
1134 put_sbits(&ctx->pb, sub->obits, sub->residual[i]);
1138 output_residual(ctx, ch);
1141 static void output_subframe_lpc(FlacEncodeContext *ctx, int ch)
1147 frame = &ctx->frame;
1148 sub = &frame->subframes[ch];
1150 /* warm-up samples */
1151 for(i=0; i<sub->order; i++) {
1152 put_sbits(&ctx->pb, sub->obits, sub->residual[i]);
1155 /* LPC coefficients */
1156 cbits = ctx->options.lpc_coeff_precision;
1157 put_bits(&ctx->pb, 4, cbits-1);
1158 put_sbits(&ctx->pb, 5, sub->shift);
1159 for(i=0; i<sub->order; i++) {
1160 put_sbits(&ctx->pb, cbits, sub->coefs[i]);
1164 output_residual(ctx, ch);
1167 static void output_subframes(FlacEncodeContext *s)
1175 for(ch=0; ch<s->channels; ch++) {
1176 sub = &frame->subframes[ch];
1178 /* subframe header */
1179 put_bits(&s->pb, 1, 0);
1180 put_bits(&s->pb, 6, sub->type_code);
1181 put_bits(&s->pb, 1, 0); /* no wasted bits */
1184 if(sub->type == FLAC_SUBFRAME_CONSTANT) {
1185 output_subframe_constant(s, ch);
1186 } else if(sub->type == FLAC_SUBFRAME_VERBATIM) {
1187 output_subframe_verbatim(s, ch);
1188 } else if(sub->type == FLAC_SUBFRAME_FIXED) {
1189 output_subframe_fixed(s, ch);
1190 } else if(sub->type == FLAC_SUBFRAME_LPC) {
1191 output_subframe_lpc(s, ch);
1196 static void output_frame_footer(FlacEncodeContext *s)
1199 flush_put_bits(&s->pb);
1200 crc = bswap_16(av_crc(av_crc8005, 0, s->pb.buf, put_bits_count(&s->pb)>>3));
1201 put_bits(&s->pb, 16, crc);
1202 flush_put_bits(&s->pb);
1205 static int flac_encode_frame(AVCodecContext *avctx, uint8_t *frame,
1206 int buf_size, void *data)
1209 FlacEncodeContext *s;
1210 int16_t *samples = data;
1213 s = avctx->priv_data;
1215 s->blocksize = avctx->frame_size;
1218 copy_samples(s, samples);
1220 channel_decorrelation(s);
1222 for(ch=0; ch<s->channels; ch++) {
1223 encode_residual(s, ch);
1225 init_put_bits(&s->pb, frame, buf_size);
1226 output_frame_header(s);
1227 output_subframes(s);
1228 output_frame_footer(s);
1229 out_bytes = put_bits_count(&s->pb) >> 3;
1231 if(out_bytes > s->max_framesize || out_bytes >= buf_size) {
1232 /* frame too large. use verbatim mode */
1233 for(ch=0; ch<s->channels; ch++) {
1234 encode_residual_v(s, ch);
1236 init_put_bits(&s->pb, frame, buf_size);
1237 output_frame_header(s);
1238 output_subframes(s);
1239 output_frame_footer(s);
1240 out_bytes = put_bits_count(&s->pb) >> 3;
1242 if(out_bytes > s->max_framesize || out_bytes >= buf_size) {
1243 /* still too large. must be an error. */
1244 av_log(avctx, AV_LOG_ERROR, "error encoding frame\n");
1253 static int flac_encode_close(AVCodecContext *avctx)
1255 av_freep(&avctx->extradata);
1256 avctx->extradata_size = 0;
1257 av_freep(&avctx->coded_frame);
1261 AVCodec flac_encoder = {
1265 sizeof(FlacEncodeContext),
1270 .capabilities = CODEC_CAP_SMALL_LAST_FRAME,