3 * Copyright (c) 2006 Justin Ruggles <justin.ruggles@gmail.com>
5 * This file is part of FFmpeg.
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.
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.
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
22 #include "libavutil/crc.h"
23 #include "libavutil/lls.h"
25 #include "bitstream.h"
31 #define FLAC_MIN_BLOCKSIZE 16
32 #define FLAC_MAX_BLOCKSIZE 65535
34 #define FLAC_SUBFRAME_CONSTANT 0
35 #define FLAC_SUBFRAME_VERBATIM 1
36 #define FLAC_SUBFRAME_FIXED 8
37 #define FLAC_SUBFRAME_LPC 32
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
45 #define FLAC_STREAMINFO_SIZE 34
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
54 typedef struct CompressionOptions {
55 int compression_level;
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;
66 typedef struct RiceContext {
68 int params[MAX_PARTITIONS];
71 typedef struct FlacSubframe {
76 int32_t coefs[MAX_LPC_ORDER];
79 int32_t samples[FLAC_MAX_BLOCKSIZE];
80 int32_t residual[FLAC_MAX_BLOCKSIZE+1];
83 typedef struct FlacFrame {
84 FlacSubframe subframes[FLAC_MAX_CH];
91 typedef struct FlacEncodeContext {
100 CompressionOptions options;
101 AVCodecContext *avctx;
105 static const int flac_samplerates[16] = {
107 8000, 16000, 22050, 24000, 32000, 44100, 48000, 96000,
111 static const int flac_blocksizes[16] = {
114 576, 1152, 2304, 4608,
116 256, 512, 1024, 2048, 4096, 8192, 16384, 32768
120 * Writes streaminfo metadata block to byte array
122 static void write_streaminfo(FlacEncodeContext *s, uint8_t *header)
126 memset(header, 0, FLAC_STREAMINFO_SIZE);
127 init_put_bits(&pb, header, FLAC_STREAMINFO_SIZE);
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 */
138 /* total samples = 0 */
139 /* MD5 signature = 0 */
143 * Sets blocksize based on samplerate
144 * Chooses the closest predefined blocksize >= BLOCK_TIME_MS milliseconds
146 static int select_blocksize(int samplerate, int block_time_ms)
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];
163 static av_cold int flac_encode_init(AVCodecContext *avctx)
165 int freq = avctx->sample_rate;
166 int channels = avctx->channels;
167 FlacEncodeContext *s = avctx->priv_data;
173 dsputil_init(&s->dsp, avctx);
175 if(avctx->sample_fmt != SAMPLE_FMT_S16) {
179 if(channels < 1 || channels > FLAC_MAX_CH) {
182 s->channels = channels;
183 s->ch_code = s->channels-1;
185 /* find samplerate in table */
188 for(i=4; i<12; i++) {
189 if(freq == flac_samplerates[i]) {
190 s->samplerate = flac_samplerates[i];
196 /* if not in table, samplerate is non-standard */
198 if(freq % 1000 == 0 && freq < 255000) {
200 s->sr_code[1] = freq / 1000;
201 } else if(freq % 10 == 0 && freq < 655350) {
203 s->sr_code[1] = freq / 10;
204 } else if(freq < 65535) {
206 s->sr_code[1] = freq;
210 s->samplerate = freq;
213 /* set compression option defaults based on avctx->compression_level */
214 if(avctx->compression_level < 0) {
215 s->options.compression_level = 5;
217 s->options.compression_level = avctx->compression_level;
219 av_log(avctx, AV_LOG_DEBUG, " compression: %d\n", s->options.compression_level);
221 level= s->options.compression_level;
223 av_log(avctx, AV_LOG_ERROR, "invalid compression level: %d\n",
224 s->options.compression_level);
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];
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);
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");
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);
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);
264 s->options.min_prediction_order = avctx->min_prediction_order;
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);
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);
281 s->options.max_prediction_order = avctx->max_prediction_order;
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);
288 av_log(avctx, AV_LOG_DEBUG, " prediction order: %d, %d\n",
289 s->options.min_prediction_order, s->options.max_prediction_order);
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);
297 s->options.prediction_order_method = avctx->prediction_order_method;
299 switch(s->options.prediction_order_method) {
300 case ORDER_METHOD_EST: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
302 case ORDER_METHOD_2LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
304 case ORDER_METHOD_4LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
306 case ORDER_METHOD_8LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
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;
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);
320 s->options.min_partition_order = avctx->min_partition_order;
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);
328 s->options.max_partition_order = avctx->max_partition_order;
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);
335 av_log(avctx, AV_LOG_DEBUG, " partition order: %d, %d\n",
336 s->options.min_partition_order, s->options.max_partition_order);
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",
346 s->avctx->frame_size = select_blocksize(s->samplerate, s->options.block_time_ms);
348 av_log(avctx, AV_LOG_DEBUG, " block size: %d\n", s->avctx->frame_size);
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);
357 s->options.lpc_coeff_precision = avctx->lpc_coeff_precision;
359 /* default LPC precision */
360 s->options.lpc_coeff_precision = 15;
362 av_log(avctx, AV_LOG_DEBUG, " lpc precision: %d\n",
363 s->options.lpc_coeff_precision);
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);
369 s->max_framesize = 14 + (s->avctx->frame_size * s->channels * 2);
372 streaminfo = av_malloc(FLAC_STREAMINFO_SIZE);
373 write_streaminfo(s, streaminfo);
374 avctx->extradata = streaminfo;
375 avctx->extradata_size = FLAC_STREAMINFO_SIZE;
379 avctx->coded_frame = avcodec_alloc_frame();
380 avctx->coded_frame->key_frame = 1;
385 static void init_frame(FlacEncodeContext *s)
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;
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;
406 frame->bs_code[0] = 7;
407 frame->bs_code[1] = frame->blocksize-1;
411 for(ch=0; ch<s->channels; ch++) {
412 frame->subframes[ch].obits = 16;
417 * Copy channel-interleaved input samples into separate subframes
419 static void copy_samples(FlacEncodeContext *s, int16_t *samples)
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];
433 #define rice_encode_count(sum, n, k) (((n)*((k)+1))+((sum-(n>>1))>>(k)))
436 * Solve for d/dk(rice_encode_count) = n-((sum-(n>>1))>>(k+1)) = 0
438 static int find_optimal_param(uint32_t sum, int n)
446 k = av_log2(n<256 ? FASTDIV(sum2,n) : sum2/n);
447 return FFMIN(k, MAX_RICE_PARAM);
450 static uint32_t calc_optimal_rice_params(RiceContext *rc, int porder,
451 uint32_t *sums, int n, int pred_order)
457 part = (1 << porder);
460 cnt = (n >> porder) - pred_order;
461 for(i=0; i<part; i++) {
462 k = find_optimal_param(sums[i], cnt);
464 all_bits += rice_encode_count(sums[i], cnt, k);
473 static void calc_sums(int pmin, int pmax, uint32_t *data, int n, int pred_order,
474 uint32_t sums[][MAX_PARTITIONS])
478 uint32_t *res, *res_end;
480 /* sums for highest level */
482 res = &data[pred_order];
483 res_end = &data[n >> pmax];
484 for(i=0; i<parts; i++) {
486 while(res < res_end){
492 /* sums for lower levels */
493 for(i=pmax-1; i>=pmin; i--) {
495 for(j=0; j<parts; j++) {
496 sums[i][j] = sums[i+1][2*j] + sums[i+1][2*j+1];
501 static uint32_t calc_rice_params(RiceContext *rc, int pmin, int pmax,
502 int32_t *data, int n, int pred_order)
505 uint32_t bits[MAX_PARTITION_ORDER+1];
509 uint32_t sums[MAX_PARTITION_ORDER+1][MAX_PARTITIONS];
511 assert(pmin >= 0 && pmin <= MAX_PARTITION_ORDER);
512 assert(pmax >= 0 && pmax <= MAX_PARTITION_ORDER);
513 assert(pmin <= pmax);
515 udata = av_malloc(n * sizeof(uint32_t));
517 udata[i] = (2*data[i]) ^ (data[i]>>31);
520 calc_sums(pmin, pmax, udata, n, pred_order, sums);
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]) {
533 return bits[opt_porder];
536 static int get_max_p_order(int max_porder, int n, int order)
538 int porder = FFMIN(max_porder, av_log2(n^(n-1)));
540 porder = FFMIN(porder, av_log2(n/order));
544 static uint32_t calc_rice_params_fixed(RiceContext *rc, int pmin, int pmax,
545 int32_t *data, int n, int pred_order,
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);
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)
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);
569 * Apply Welch window function to audio block
571 static void apply_welch_window(const int32_t *data, int len, double *w_data)
577 assert(!(len&1)); //the optimization in r11881 does not support odd len
578 //if someone wants odd len extend the change in r11881
581 c = 2.0 / (len - 1.0);
585 for(i=0; i<n2; i++) {
588 w_data[-i-1] = data[-i-1] * w;
589 w_data[+i ] = data[+i ] * w;
594 * Calculates autocorrelation data from audio samples
595 * A Welch window function is applied before calculation.
597 void ff_flac_compute_autocorr(const int32_t *data, int len, int lag,
601 double tmp[len + lag + 1];
602 double *data1= tmp + lag;
604 apply_welch_window(data, len, data1);
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];
622 for(i=0; i<len; i+=2){
623 sum += data1[i ] * data1[i-j ]
624 + data1[i+1] * data1[i-j+1];
631 static void encode_residual_verbatim(int32_t *res, int32_t *smp, int n)
634 memcpy(res, smp, n * sizeof(int32_t));
637 static void encode_residual_fixed(int32_t *res, const int32_t *smp, int n,
642 for(i=0; i<order; i++) {
647 for(i=order; i<n; i++)
650 for(i=order; i<n; i++)
651 res[i]= smp[i] - smp[i-1];
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];
657 a = smp[i+1] - smp[i];
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];
667 a = smp[i+1] - smp[i];
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];
680 a = smp[i+1] - smp[i];
689 int c = coefs[(x)-1];\
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)
700 for(i=order; i<n; i+=2) {
701 int s = smp[i-order];
750 res[i ] = smp[i ] - (p0 >> shift);
751 res[i+1] = smp[i+1] - (p1 >> shift);
755 static void encode_residual_lpc(int32_t *res, const int32_t *smp, int n,
756 int order, const int32_t *coefs, int shift)
759 for(i=0; i<order; i++) {
763 for(i=order; i<n; i+=2) {
767 for(j=0; j<order; j++) {
773 res[i ] = smp[i ] - (p0 >> shift);
774 res[i+1] = smp[i+1] - (p1 >> shift);
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;
791 static int encode_residual(FlacEncodeContext *ctx, int ch)
794 int min_order, max_order, opt_order, precision, omethod;
795 int min_porder, max_porder;
798 int32_t coefs[MAX_LPC_ORDER][MAX_LPC_ORDER];
799 int shift[MAX_LPC_ORDER];
803 sub = &frame->subframes[ch];
806 n = frame->blocksize;
810 if(smp[i] != smp[0]) break;
813 sub->type = sub->type_code = FLAC_SUBFRAME_CONSTANT;
820 sub->type = sub->type_code = FLAC_SUBFRAME_VERBATIM;
821 encode_residual_verbatim(res, smp, n);
822 return sub->obits * n;
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;
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;
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,
842 if(bits[i] < bits[opt_order]) {
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);
854 return bits[sub->order];
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);
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];
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]) {
883 } else if(omethod == ORDER_METHOD_SEARCH) {
884 // brute-force optimal order search
885 uint32_t bits[MAX_LPC_ORDER];
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]) {
897 } else if(omethod == ORDER_METHOD_LOG) {
898 uint32_t bits[MAX_LPC_ORDER];
901 opt_order= min_order - 1 + (max_order-min_order)/3;
902 memset(bits, -1, sizeof(bits));
904 for(step=16 ;step; step>>=1){
906 for(i=last-step; i<=last+step; i+= step){
907 if(i<min_order-1 || i>=max_order || bits[i] < UINT32_MAX)
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])
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];
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);
931 static int encode_residual_v(FlacEncodeContext *ctx, int ch)
939 sub = &frame->subframes[ch];
942 n = frame->blocksize;
946 if(smp[i] != smp[0]) break;
949 sub->type = sub->type_code = FLAC_SUBFRAME_CONSTANT;
955 sub->type = sub->type_code = FLAC_SUBFRAME_VERBATIM;
956 encode_residual_verbatim(res, smp, n);
957 return sub->obits * n;
960 static int estimate_stereo_mode(int32_t *left_ch, int32_t *right_ch, int n)
968 /* calculate sum of 2nd order residual for each channel */
969 sum[0] = sum[1] = sum[2] = sum[3] = 0;
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);
978 /* estimate bit counts */
980 k = find_optimal_param(2*sum[i], n);
981 sum[i] = rice_encode_count(2*sum[i], n, k);
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];
990 /* return mode with lowest score */
993 if(score[i] < score[best]) {
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;
1004 return FLAC_CHMODE_MID_SIDE;
1009 * Perform stereo channel decorrelation
1011 static void channel_decorrelation(FlacEncodeContext *ctx)
1014 int32_t *left, *right;
1017 frame = &ctx->frame;
1018 n = frame->blocksize;
1019 left = frame->subframes[0].samples;
1020 right = frame->subframes[1].samples;
1022 if(ctx->channels != 2) {
1023 frame->ch_mode = FLAC_CHMODE_NOT_STEREO;
1027 frame->ch_mode = estimate_stereo_mode(left, right, n);
1029 /* perform decorrelation and adjust bits-per-sample */
1030 if(frame->ch_mode == FLAC_CHMODE_LEFT_RIGHT) {
1033 if(frame->ch_mode == FLAC_CHMODE_MID_SIDE) {
1035 for(i=0; i<n; i++) {
1037 left[i] = (tmp + right[i]) >> 1;
1038 right[i] = tmp - right[i];
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];
1045 frame->subframes[1].obits++;
1047 for(i=0; i<n; i++) {
1048 left[i] -= right[i];
1050 frame->subframes[0].obits++;
1054 static void write_utf8(PutBitContext *pb, uint32_t val)
1057 PUT_UTF8(val, tmp, put_bits(pb, 8, tmp);)
1060 static void output_frame_header(FlacEncodeContext *s)
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);
1073 put_bits(&s->pb, 4, frame->ch_mode);
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]);
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]);
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);
1094 static void output_subframe_constant(FlacEncodeContext *s, int ch)
1099 sub = &s->frame.subframes[ch];
1100 res = sub->residual[0];
1101 put_sbits(&s->pb, sub->obits, res);
1104 static void output_subframe_verbatim(FlacEncodeContext *s, int ch)
1112 sub = &frame->subframes[ch];
1114 for(i=0; i<frame->blocksize; i++) {
1115 res = sub->residual[i];
1116 put_sbits(&s->pb, sub->obits, res);
1120 static void output_residual(FlacEncodeContext *ctx, int ch)
1122 int i, j, p, n, parts;
1123 int k, porder, psize, res_cnt;
1128 frame = &ctx->frame;
1129 sub = &frame->subframes[ch];
1130 res = sub->residual;
1131 n = frame->blocksize;
1133 /* rice-encoded block */
1134 put_bits(&ctx->pb, 2, 0);
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;
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);
1155 static void output_subframe_fixed(FlacEncodeContext *ctx, int ch)
1161 frame = &ctx->frame;
1162 sub = &frame->subframes[ch];
1164 /* warm-up samples */
1165 for(i=0; i<sub->order; i++) {
1166 put_sbits(&ctx->pb, sub->obits, sub->residual[i]);
1170 output_residual(ctx, ch);
1173 static void output_subframe_lpc(FlacEncodeContext *ctx, int ch)
1179 frame = &ctx->frame;
1180 sub = &frame->subframes[ch];
1182 /* warm-up samples */
1183 for(i=0; i<sub->order; i++) {
1184 put_sbits(&ctx->pb, sub->obits, sub->residual[i]);
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]);
1196 output_residual(ctx, ch);
1199 static void output_subframes(FlacEncodeContext *s)
1207 for(ch=0; ch<s->channels; ch++) {
1208 sub = &frame->subframes[ch];
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 */
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);
1228 static void output_frame_footer(FlacEncodeContext *s)
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);
1238 static int flac_encode_frame(AVCodecContext *avctx, uint8_t *frame,
1239 int buf_size, void *data)
1242 FlacEncodeContext *s;
1243 int16_t *samples = data;
1247 s = avctx->priv_data;
1249 if(buf_size < s->max_framesize*2) {
1250 av_log(avctx, AV_LOG_ERROR, "output buffer too small\n");
1256 copy_samples(s, samples);
1258 channel_decorrelation(s);
1260 for(ch=0; ch<s->channels; ch++) {
1261 encode_residual(s, ch);
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;
1271 if(out_bytes > s->max_framesize) {
1273 /* still too large. must be an error. */
1274 av_log(avctx, AV_LOG_ERROR, "error encoding frame\n");
1278 /* frame too large. use verbatim mode */
1279 for(ch=0; ch<s->channels; ch++) {
1280 encode_residual_v(s, ch);
1290 static av_cold int flac_encode_close(AVCodecContext *avctx)
1292 av_freep(&avctx->extradata);
1293 avctx->extradata_size = 0;
1294 av_freep(&avctx->coded_frame);
1298 AVCodec flac_encoder = {
1302 sizeof(FlacEncodeContext),
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)"),