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/md5.h"
32 #define FLAC_SUBFRAME_CONSTANT 0
33 #define FLAC_SUBFRAME_VERBATIM 1
34 #define FLAC_SUBFRAME_FIXED 8
35 #define FLAC_SUBFRAME_LPC 32
37 #define MAX_FIXED_ORDER 4
38 #define MAX_PARTITION_ORDER 8
39 #define MAX_PARTITIONS (1 << MAX_PARTITION_ORDER)
40 #define MAX_LPC_PRECISION 15
41 #define MAX_LPC_SHIFT 15
42 #define MAX_RICE_PARAM 14
44 typedef struct CompressionOptions {
45 int compression_level;
48 int lpc_coeff_precision;
49 int min_prediction_order;
50 int max_prediction_order;
51 int prediction_order_method;
52 int min_partition_order;
53 int max_partition_order;
56 typedef struct RiceContext {
58 int params[MAX_PARTITIONS];
61 typedef struct FlacSubframe {
66 int32_t coefs[MAX_LPC_ORDER];
69 int32_t samples[FLAC_MAX_BLOCKSIZE];
70 int32_t residual[FLAC_MAX_BLOCKSIZE+1];
73 typedef struct FlacFrame {
74 FlacSubframe subframes[FLAC_MAX_CHANNELS];
81 typedef struct FlacEncodeContext {
89 int max_encoded_framesize;
91 uint64_t sample_count;
94 CompressionOptions options;
95 AVCodecContext *avctx;
101 * Writes streaminfo metadata block to byte array
103 static void write_streaminfo(FlacEncodeContext *s, uint8_t *header)
107 memset(header, 0, FLAC_STREAMINFO_SIZE);
108 init_put_bits(&pb, header, FLAC_STREAMINFO_SIZE);
110 /* streaminfo metadata block */
111 put_bits(&pb, 16, s->max_blocksize);
112 put_bits(&pb, 16, s->max_blocksize);
113 put_bits(&pb, 24, s->min_framesize);
114 put_bits(&pb, 24, s->max_framesize);
115 put_bits(&pb, 20, s->samplerate);
116 put_bits(&pb, 3, s->channels-1);
117 put_bits(&pb, 5, 15); /* bits per sample - 1 */
118 /* write 36-bit sample count in 2 put_bits() calls */
119 put_bits(&pb, 24, (s->sample_count & 0xFFFFFF000LL) >> 12);
120 put_bits(&pb, 12, s->sample_count & 0x000000FFFLL);
122 memcpy(&header[18], s->md5sum, 16);
126 * Sets blocksize based on samplerate
127 * Chooses the closest predefined blocksize >= BLOCK_TIME_MS milliseconds
129 static int select_blocksize(int samplerate, int block_time_ms)
135 assert(samplerate > 0);
136 blocksize = ff_flac_blocksize_table[1];
137 target = (samplerate * block_time_ms) / 1000;
138 for(i=0; i<16; i++) {
139 if(target >= ff_flac_blocksize_table[i] && ff_flac_blocksize_table[i] > blocksize) {
140 blocksize = ff_flac_blocksize_table[i];
146 static av_cold int flac_encode_init(AVCodecContext *avctx)
148 int freq = avctx->sample_rate;
149 int channels = avctx->channels;
150 FlacEncodeContext *s = avctx->priv_data;
156 dsputil_init(&s->dsp, avctx);
158 if(avctx->sample_fmt != SAMPLE_FMT_S16) {
162 if(channels < 1 || channels > FLAC_MAX_CHANNELS) {
165 s->channels = channels;
167 /* find samplerate in table */
170 for(i=4; i<12; i++) {
171 if(freq == ff_flac_sample_rate_table[i]) {
172 s->samplerate = ff_flac_sample_rate_table[i];
178 /* if not in table, samplerate is non-standard */
180 if(freq % 1000 == 0 && freq < 255000) {
182 s->sr_code[1] = freq / 1000;
183 } else if(freq % 10 == 0 && freq < 655350) {
185 s->sr_code[1] = freq / 10;
186 } else if(freq < 65535) {
188 s->sr_code[1] = freq;
192 s->samplerate = freq;
195 /* set compression option defaults based on avctx->compression_level */
196 if(avctx->compression_level < 0) {
197 s->options.compression_level = 5;
199 s->options.compression_level = avctx->compression_level;
201 av_log(avctx, AV_LOG_DEBUG, " compression: %d\n", s->options.compression_level);
203 level= s->options.compression_level;
205 av_log(avctx, AV_LOG_ERROR, "invalid compression level: %d\n",
206 s->options.compression_level);
210 s->options.block_time_ms = ((int[]){ 27, 27, 27,105,105,105,105,105,105,105,105,105,105})[level];
211 s->options.use_lpc = ((int[]){ 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1})[level];
212 s->options.min_prediction_order= ((int[]){ 2, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1})[level];
213 s->options.max_prediction_order= ((int[]){ 3, 4, 4, 6, 8, 8, 8, 8, 12, 12, 12, 32, 32})[level];
214 s->options.prediction_order_method = ((int[]){ ORDER_METHOD_EST, ORDER_METHOD_EST, ORDER_METHOD_EST,
215 ORDER_METHOD_EST, ORDER_METHOD_EST, ORDER_METHOD_EST,
216 ORDER_METHOD_4LEVEL, ORDER_METHOD_LOG, ORDER_METHOD_4LEVEL,
217 ORDER_METHOD_LOG, ORDER_METHOD_SEARCH, ORDER_METHOD_LOG,
218 ORDER_METHOD_SEARCH})[level];
219 s->options.min_partition_order = ((int[]){ 2, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0})[level];
220 s->options.max_partition_order = ((int[]){ 2, 2, 3, 3, 3, 8, 8, 8, 8, 8, 8, 8, 8})[level];
222 /* set compression option overrides from AVCodecContext */
223 if(avctx->use_lpc >= 0) {
224 s->options.use_lpc = av_clip(avctx->use_lpc, 0, 11);
226 if(s->options.use_lpc == 1)
227 av_log(avctx, AV_LOG_DEBUG, " use lpc: Levinson-Durbin recursion with Welch window\n");
228 else if(s->options.use_lpc > 1)
229 av_log(avctx, AV_LOG_DEBUG, " use lpc: Cholesky factorization\n");
231 if(avctx->min_prediction_order >= 0) {
232 if(s->options.use_lpc) {
233 if(avctx->min_prediction_order < MIN_LPC_ORDER ||
234 avctx->min_prediction_order > MAX_LPC_ORDER) {
235 av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n",
236 avctx->min_prediction_order);
240 if(avctx->min_prediction_order > MAX_FIXED_ORDER) {
241 av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n",
242 avctx->min_prediction_order);
246 s->options.min_prediction_order = avctx->min_prediction_order;
248 if(avctx->max_prediction_order >= 0) {
249 if(s->options.use_lpc) {
250 if(avctx->max_prediction_order < MIN_LPC_ORDER ||
251 avctx->max_prediction_order > MAX_LPC_ORDER) {
252 av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n",
253 avctx->max_prediction_order);
257 if(avctx->max_prediction_order > MAX_FIXED_ORDER) {
258 av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n",
259 avctx->max_prediction_order);
263 s->options.max_prediction_order = avctx->max_prediction_order;
265 if(s->options.max_prediction_order < s->options.min_prediction_order) {
266 av_log(avctx, AV_LOG_ERROR, "invalid prediction orders: min=%d max=%d\n",
267 s->options.min_prediction_order, s->options.max_prediction_order);
270 av_log(avctx, AV_LOG_DEBUG, " prediction order: %d, %d\n",
271 s->options.min_prediction_order, s->options.max_prediction_order);
273 if(avctx->prediction_order_method >= 0) {
274 if(avctx->prediction_order_method > ORDER_METHOD_LOG) {
275 av_log(avctx, AV_LOG_ERROR, "invalid prediction order method: %d\n",
276 avctx->prediction_order_method);
279 s->options.prediction_order_method = avctx->prediction_order_method;
281 switch(s->options.prediction_order_method) {
282 case ORDER_METHOD_EST: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
284 case ORDER_METHOD_2LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
286 case ORDER_METHOD_4LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
288 case ORDER_METHOD_8LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
290 case ORDER_METHOD_SEARCH: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
291 "full search"); break;
292 case ORDER_METHOD_LOG: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
293 "log search"); break;
296 if(avctx->min_partition_order >= 0) {
297 if(avctx->min_partition_order > MAX_PARTITION_ORDER) {
298 av_log(avctx, AV_LOG_ERROR, "invalid min partition order: %d\n",
299 avctx->min_partition_order);
302 s->options.min_partition_order = avctx->min_partition_order;
304 if(avctx->max_partition_order >= 0) {
305 if(avctx->max_partition_order > MAX_PARTITION_ORDER) {
306 av_log(avctx, AV_LOG_ERROR, "invalid max partition order: %d\n",
307 avctx->max_partition_order);
310 s->options.max_partition_order = avctx->max_partition_order;
312 if(s->options.max_partition_order < s->options.min_partition_order) {
313 av_log(avctx, AV_LOG_ERROR, "invalid partition orders: min=%d max=%d\n",
314 s->options.min_partition_order, s->options.max_partition_order);
317 av_log(avctx, AV_LOG_DEBUG, " partition order: %d, %d\n",
318 s->options.min_partition_order, s->options.max_partition_order);
320 if(avctx->frame_size > 0) {
321 if(avctx->frame_size < FLAC_MIN_BLOCKSIZE ||
322 avctx->frame_size > FLAC_MAX_BLOCKSIZE) {
323 av_log(avctx, AV_LOG_ERROR, "invalid block size: %d\n",
328 s->avctx->frame_size = select_blocksize(s->samplerate, s->options.block_time_ms);
330 s->max_blocksize = s->avctx->frame_size;
331 av_log(avctx, AV_LOG_DEBUG, " block size: %d\n", s->avctx->frame_size);
333 /* set LPC precision */
334 if(avctx->lpc_coeff_precision > 0) {
335 if(avctx->lpc_coeff_precision > MAX_LPC_PRECISION) {
336 av_log(avctx, AV_LOG_ERROR, "invalid lpc coeff precision: %d\n",
337 avctx->lpc_coeff_precision);
340 s->options.lpc_coeff_precision = avctx->lpc_coeff_precision;
342 /* default LPC precision */
343 s->options.lpc_coeff_precision = 15;
345 av_log(avctx, AV_LOG_DEBUG, " lpc precision: %d\n",
346 s->options.lpc_coeff_precision);
348 /* set maximum encoded frame size in verbatim mode */
349 s->max_framesize = ff_flac_get_max_frame_size(s->avctx->frame_size,
352 /* initialize MD5 context */
353 s->md5ctx = av_malloc(av_md5_size);
355 return AVERROR_NOMEM;
356 av_md5_init(s->md5ctx);
358 streaminfo = av_malloc(FLAC_STREAMINFO_SIZE);
359 write_streaminfo(s, streaminfo);
360 avctx->extradata = streaminfo;
361 avctx->extradata_size = FLAC_STREAMINFO_SIZE;
364 s->min_framesize = s->max_framesize;
366 avctx->coded_frame = avcodec_alloc_frame();
367 avctx->coded_frame->key_frame = 1;
372 static void init_frame(FlacEncodeContext *s)
379 for(i=0; i<16; i++) {
380 if(s->avctx->frame_size == ff_flac_blocksize_table[i]) {
381 frame->blocksize = ff_flac_blocksize_table[i];
382 frame->bs_code[0] = i;
383 frame->bs_code[1] = 0;
388 frame->blocksize = s->avctx->frame_size;
389 if(frame->blocksize <= 256) {
390 frame->bs_code[0] = 6;
391 frame->bs_code[1] = frame->blocksize-1;
393 frame->bs_code[0] = 7;
394 frame->bs_code[1] = frame->blocksize-1;
398 for(ch=0; ch<s->channels; ch++) {
399 frame->subframes[ch].obits = 16;
404 * Copy channel-interleaved input samples into separate subframes
406 static void copy_samples(FlacEncodeContext *s, int16_t *samples)
412 for(i=0,j=0; i<frame->blocksize; i++) {
413 for(ch=0; ch<s->channels; ch++,j++) {
414 frame->subframes[ch].samples[i] = samples[j];
420 #define rice_encode_count(sum, n, k) (((n)*((k)+1))+((sum-(n>>1))>>(k)))
423 * Solve for d/dk(rice_encode_count) = n-((sum-(n>>1))>>(k+1)) = 0
425 static int find_optimal_param(uint32_t sum, int n)
433 k = av_log2(n<256 ? FASTDIV(sum2,n) : sum2/n);
434 return FFMIN(k, MAX_RICE_PARAM);
437 static uint32_t calc_optimal_rice_params(RiceContext *rc, int porder,
438 uint32_t *sums, int n, int pred_order)
444 part = (1 << porder);
447 cnt = (n >> porder) - pred_order;
448 for(i=0; i<part; i++) {
449 k = find_optimal_param(sums[i], cnt);
451 all_bits += rice_encode_count(sums[i], cnt, k);
460 static void calc_sums(int pmin, int pmax, uint32_t *data, int n, int pred_order,
461 uint32_t sums[][MAX_PARTITIONS])
465 uint32_t *res, *res_end;
467 /* sums for highest level */
469 res = &data[pred_order];
470 res_end = &data[n >> pmax];
471 for(i=0; i<parts; i++) {
473 while(res < res_end){
479 /* sums for lower levels */
480 for(i=pmax-1; i>=pmin; i--) {
482 for(j=0; j<parts; j++) {
483 sums[i][j] = sums[i+1][2*j] + sums[i+1][2*j+1];
488 static uint32_t calc_rice_params(RiceContext *rc, int pmin, int pmax,
489 int32_t *data, int n, int pred_order)
492 uint32_t bits[MAX_PARTITION_ORDER+1];
496 uint32_t sums[MAX_PARTITION_ORDER+1][MAX_PARTITIONS];
498 assert(pmin >= 0 && pmin <= MAX_PARTITION_ORDER);
499 assert(pmax >= 0 && pmax <= MAX_PARTITION_ORDER);
500 assert(pmin <= pmax);
502 udata = av_malloc(n * sizeof(uint32_t));
504 udata[i] = (2*data[i]) ^ (data[i]>>31);
507 calc_sums(pmin, pmax, udata, n, pred_order, sums);
510 bits[pmin] = UINT32_MAX;
511 for(i=pmin; i<=pmax; i++) {
512 bits[i] = calc_optimal_rice_params(&tmp_rc, i, sums[i], n, pred_order);
513 if(bits[i] <= bits[opt_porder]) {
520 return bits[opt_porder];
523 static int get_max_p_order(int max_porder, int n, int order)
525 int porder = FFMIN(max_porder, av_log2(n^(n-1)));
527 porder = FFMIN(porder, av_log2(n/order));
531 static uint32_t calc_rice_params_fixed(RiceContext *rc, int pmin, int pmax,
532 int32_t *data, int n, int pred_order,
536 pmin = get_max_p_order(pmin, n, pred_order);
537 pmax = get_max_p_order(pmax, n, pred_order);
538 bits = pred_order*bps + 6;
539 bits += calc_rice_params(rc, pmin, pmax, data, n, pred_order);
543 static uint32_t calc_rice_params_lpc(RiceContext *rc, int pmin, int pmax,
544 int32_t *data, int n, int pred_order,
545 int bps, int precision)
548 pmin = get_max_p_order(pmin, n, pred_order);
549 pmax = get_max_p_order(pmax, n, pred_order);
550 bits = pred_order*bps + 4 + 5 + pred_order*precision + 6;
551 bits += calc_rice_params(rc, pmin, pmax, data, n, pred_order);
555 static void encode_residual_verbatim(int32_t *res, int32_t *smp, int n)
558 memcpy(res, smp, n * sizeof(int32_t));
561 static void encode_residual_fixed(int32_t *res, const int32_t *smp, int n,
566 for(i=0; i<order; i++) {
571 for(i=order; i<n; i++)
574 for(i=order; i<n; i++)
575 res[i]= smp[i] - smp[i-1];
577 int a = smp[order-1] - smp[order-2];
578 for(i=order; i<n; i+=2) {
579 int b = smp[i] - smp[i-1];
581 a = smp[i+1] - smp[i];
585 int a = smp[order-1] - smp[order-2];
586 int c = smp[order-1] - 2*smp[order-2] + smp[order-3];
587 for(i=order; i<n; i+=2) {
588 int b = smp[i] - smp[i-1];
591 a = smp[i+1] - smp[i];
596 int a = smp[order-1] - smp[order-2];
597 int c = smp[order-1] - 2*smp[order-2] + smp[order-3];
598 int e = smp[order-1] - 3*smp[order-2] + 3*smp[order-3] - smp[order-4];
599 for(i=order; i<n; i+=2) {
600 int b = smp[i] - smp[i-1];
604 a = smp[i+1] - smp[i];
613 int c = coefs[(x)-1];\
619 static av_always_inline void encode_residual_lpc_unrolled(
620 int32_t *res, const int32_t *smp, int n,
621 int order, const int32_t *coefs, int shift, int big)
624 for(i=order; i<n; i+=2) {
625 int s = smp[i-order];
674 res[i ] = smp[i ] - (p0 >> shift);
675 res[i+1] = smp[i+1] - (p1 >> shift);
679 static void encode_residual_lpc(int32_t *res, const int32_t *smp, int n,
680 int order, const int32_t *coefs, int shift)
683 for(i=0; i<order; i++) {
687 for(i=order; i<n; i+=2) {
691 for(j=0; j<order; j++) {
697 res[i ] = smp[i ] - (p0 >> shift);
698 res[i+1] = smp[i+1] - (p1 >> shift);
702 case 1: encode_residual_lpc_unrolled(res, smp, n, 1, coefs, shift, 0); break;
703 case 2: encode_residual_lpc_unrolled(res, smp, n, 2, coefs, shift, 0); break;
704 case 3: encode_residual_lpc_unrolled(res, smp, n, 3, coefs, shift, 0); break;
705 case 4: encode_residual_lpc_unrolled(res, smp, n, 4, coefs, shift, 0); break;
706 case 5: encode_residual_lpc_unrolled(res, smp, n, 5, coefs, shift, 0); break;
707 case 6: encode_residual_lpc_unrolled(res, smp, n, 6, coefs, shift, 0); break;
708 case 7: encode_residual_lpc_unrolled(res, smp, n, 7, coefs, shift, 0); break;
709 case 8: encode_residual_lpc_unrolled(res, smp, n, 8, coefs, shift, 0); break;
710 default: encode_residual_lpc_unrolled(res, smp, n, order, coefs, shift, 1); break;
715 static int encode_residual(FlacEncodeContext *ctx, int ch)
718 int min_order, max_order, opt_order, precision, omethod;
719 int min_porder, max_porder;
722 int32_t coefs[MAX_LPC_ORDER][MAX_LPC_ORDER];
723 int shift[MAX_LPC_ORDER];
727 sub = &frame->subframes[ch];
730 n = frame->blocksize;
734 if(smp[i] != smp[0]) break;
737 sub->type = sub->type_code = FLAC_SUBFRAME_CONSTANT;
744 sub->type = sub->type_code = FLAC_SUBFRAME_VERBATIM;
745 encode_residual_verbatim(res, smp, n);
746 return sub->obits * n;
749 min_order = ctx->options.min_prediction_order;
750 max_order = ctx->options.max_prediction_order;
751 min_porder = ctx->options.min_partition_order;
752 max_porder = ctx->options.max_partition_order;
753 precision = ctx->options.lpc_coeff_precision;
754 omethod = ctx->options.prediction_order_method;
757 if(!ctx->options.use_lpc || max_order == 0 || (n <= max_order)) {
758 uint32_t bits[MAX_FIXED_ORDER+1];
759 if(max_order > MAX_FIXED_ORDER) max_order = MAX_FIXED_ORDER;
761 bits[0] = UINT32_MAX;
762 for(i=min_order; i<=max_order; i++) {
763 encode_residual_fixed(res, smp, n, i);
764 bits[i] = calc_rice_params_fixed(&sub->rc, min_porder, max_porder, res,
766 if(bits[i] < bits[opt_order]) {
770 sub->order = opt_order;
771 sub->type = FLAC_SUBFRAME_FIXED;
772 sub->type_code = sub->type | sub->order;
773 if(sub->order != max_order) {
774 encode_residual_fixed(res, smp, n, sub->order);
775 return calc_rice_params_fixed(&sub->rc, min_porder, max_porder, res, n,
776 sub->order, sub->obits);
778 return bits[sub->order];
782 opt_order = ff_lpc_calc_coefs(&ctx->dsp, smp, n, min_order, max_order,
783 precision, coefs, shift, ctx->options.use_lpc,
784 omethod, MAX_LPC_SHIFT, 0);
786 if(omethod == ORDER_METHOD_2LEVEL ||
787 omethod == ORDER_METHOD_4LEVEL ||
788 omethod == ORDER_METHOD_8LEVEL) {
789 int levels = 1 << omethod;
790 uint32_t bits[levels];
792 int opt_index = levels-1;
793 opt_order = max_order-1;
794 bits[opt_index] = UINT32_MAX;
795 for(i=levels-1; i>=0; i--) {
796 order = min_order + (((max_order-min_order+1) * (i+1)) / levels)-1;
797 if(order < 0) order = 0;
798 encode_residual_lpc(res, smp, n, order+1, coefs[order], shift[order]);
799 bits[i] = calc_rice_params_lpc(&sub->rc, min_porder, max_porder,
800 res, n, order+1, sub->obits, precision);
801 if(bits[i] < bits[opt_index]) {
807 } else if(omethod == ORDER_METHOD_SEARCH) {
808 // brute-force optimal order search
809 uint32_t bits[MAX_LPC_ORDER];
811 bits[0] = UINT32_MAX;
812 for(i=min_order-1; i<max_order; i++) {
813 encode_residual_lpc(res, smp, n, i+1, coefs[i], shift[i]);
814 bits[i] = calc_rice_params_lpc(&sub->rc, min_porder, max_porder,
815 res, n, i+1, sub->obits, precision);
816 if(bits[i] < bits[opt_order]) {
821 } else if(omethod == ORDER_METHOD_LOG) {
822 uint32_t bits[MAX_LPC_ORDER];
825 opt_order= min_order - 1 + (max_order-min_order)/3;
826 memset(bits, -1, sizeof(bits));
828 for(step=16 ;step; step>>=1){
830 for(i=last-step; i<=last+step; i+= step){
831 if(i<min_order-1 || i>=max_order || bits[i] < UINT32_MAX)
833 encode_residual_lpc(res, smp, n, i+1, coefs[i], shift[i]);
834 bits[i] = calc_rice_params_lpc(&sub->rc, min_porder, max_porder,
835 res, n, i+1, sub->obits, precision);
836 if(bits[i] < bits[opt_order])
843 sub->order = opt_order;
844 sub->type = FLAC_SUBFRAME_LPC;
845 sub->type_code = sub->type | (sub->order-1);
846 sub->shift = shift[sub->order-1];
847 for(i=0; i<sub->order; i++) {
848 sub->coefs[i] = coefs[sub->order-1][i];
850 encode_residual_lpc(res, smp, n, sub->order, sub->coefs, sub->shift);
851 return calc_rice_params_lpc(&sub->rc, min_porder, max_porder, res, n, sub->order,
852 sub->obits, precision);
855 static int encode_residual_v(FlacEncodeContext *ctx, int ch)
863 sub = &frame->subframes[ch];
866 n = frame->blocksize;
870 if(smp[i] != smp[0]) break;
873 sub->type = sub->type_code = FLAC_SUBFRAME_CONSTANT;
879 sub->type = sub->type_code = FLAC_SUBFRAME_VERBATIM;
880 encode_residual_verbatim(res, smp, n);
881 return sub->obits * n;
884 static int estimate_stereo_mode(int32_t *left_ch, int32_t *right_ch, int n)
892 /* calculate sum of 2nd order residual for each channel */
893 sum[0] = sum[1] = sum[2] = sum[3] = 0;
895 lt = left_ch[i] - 2*left_ch[i-1] + left_ch[i-2];
896 rt = right_ch[i] - 2*right_ch[i-1] + right_ch[i-2];
897 sum[2] += FFABS((lt + rt) >> 1);
898 sum[3] += FFABS(lt - rt);
902 /* estimate bit counts */
904 k = find_optimal_param(2*sum[i], n);
905 sum[i] = rice_encode_count(2*sum[i], n, k);
908 /* calculate score for each mode */
909 score[0] = sum[0] + sum[1];
910 score[1] = sum[0] + sum[3];
911 score[2] = sum[1] + sum[3];
912 score[3] = sum[2] + sum[3];
914 /* return mode with lowest score */
917 if(score[i] < score[best]) {
922 return FLAC_CHMODE_INDEPENDENT;
923 } else if(best == 1) {
924 return FLAC_CHMODE_LEFT_SIDE;
925 } else if(best == 2) {
926 return FLAC_CHMODE_RIGHT_SIDE;
928 return FLAC_CHMODE_MID_SIDE;
933 * Perform stereo channel decorrelation
935 static void channel_decorrelation(FlacEncodeContext *ctx)
938 int32_t *left, *right;
942 n = frame->blocksize;
943 left = frame->subframes[0].samples;
944 right = frame->subframes[1].samples;
946 if(ctx->channels != 2) {
947 frame->ch_mode = FLAC_CHMODE_INDEPENDENT;
951 frame->ch_mode = estimate_stereo_mode(left, right, n);
953 /* perform decorrelation and adjust bits-per-sample */
954 if(frame->ch_mode == FLAC_CHMODE_INDEPENDENT) {
957 if(frame->ch_mode == FLAC_CHMODE_MID_SIDE) {
961 left[i] = (tmp + right[i]) >> 1;
962 right[i] = tmp - right[i];
964 frame->subframes[1].obits++;
965 } else if(frame->ch_mode == FLAC_CHMODE_LEFT_SIDE) {
967 right[i] = left[i] - right[i];
969 frame->subframes[1].obits++;
974 frame->subframes[0].obits++;
978 static void write_utf8(PutBitContext *pb, uint32_t val)
981 PUT_UTF8(val, tmp, put_bits(pb, 8, tmp);)
984 static void output_frame_header(FlacEncodeContext *s)
991 put_bits(&s->pb, 16, 0xFFF8);
992 put_bits(&s->pb, 4, frame->bs_code[0]);
993 put_bits(&s->pb, 4, s->sr_code[0]);
994 if(frame->ch_mode == FLAC_CHMODE_INDEPENDENT) {
995 put_bits(&s->pb, 4, s->channels-1);
997 put_bits(&s->pb, 4, frame->ch_mode);
999 put_bits(&s->pb, 3, 4); /* bits-per-sample code */
1000 put_bits(&s->pb, 1, 0);
1001 write_utf8(&s->pb, s->frame_count);
1002 if(frame->bs_code[0] == 6) {
1003 put_bits(&s->pb, 8, frame->bs_code[1]);
1004 } else if(frame->bs_code[0] == 7) {
1005 put_bits(&s->pb, 16, frame->bs_code[1]);
1007 if(s->sr_code[0] == 12) {
1008 put_bits(&s->pb, 8, s->sr_code[1]);
1009 } else if(s->sr_code[0] > 12) {
1010 put_bits(&s->pb, 16, s->sr_code[1]);
1012 flush_put_bits(&s->pb);
1013 crc = av_crc(av_crc_get_table(AV_CRC_8_ATM), 0,
1014 s->pb.buf, put_bits_count(&s->pb)>>3);
1015 put_bits(&s->pb, 8, crc);
1018 static void output_subframe_constant(FlacEncodeContext *s, int ch)
1023 sub = &s->frame.subframes[ch];
1024 res = sub->residual[0];
1025 put_sbits(&s->pb, sub->obits, res);
1028 static void output_subframe_verbatim(FlacEncodeContext *s, int ch)
1036 sub = &frame->subframes[ch];
1038 for(i=0; i<frame->blocksize; i++) {
1039 res = sub->residual[i];
1040 put_sbits(&s->pb, sub->obits, res);
1044 static void output_residual(FlacEncodeContext *ctx, int ch)
1046 int i, j, p, n, parts;
1047 int k, porder, psize, res_cnt;
1052 frame = &ctx->frame;
1053 sub = &frame->subframes[ch];
1054 res = sub->residual;
1055 n = frame->blocksize;
1057 /* rice-encoded block */
1058 put_bits(&ctx->pb, 2, 0);
1060 /* partition order */
1061 porder = sub->rc.porder;
1062 psize = n >> porder;
1063 parts = (1 << porder);
1064 put_bits(&ctx->pb, 4, porder);
1065 res_cnt = psize - sub->order;
1069 for(p=0; p<parts; p++) {
1070 k = sub->rc.params[p];
1071 put_bits(&ctx->pb, 4, k);
1072 if(p == 1) res_cnt = psize;
1073 for(i=0; i<res_cnt && j<n; i++, j++) {
1074 set_sr_golomb_flac(&ctx->pb, res[j], k, INT32_MAX, 0);
1079 static void output_subframe_fixed(FlacEncodeContext *ctx, int ch)
1085 frame = &ctx->frame;
1086 sub = &frame->subframes[ch];
1088 /* warm-up samples */
1089 for(i=0; i<sub->order; i++) {
1090 put_sbits(&ctx->pb, sub->obits, sub->residual[i]);
1094 output_residual(ctx, ch);
1097 static void output_subframe_lpc(FlacEncodeContext *ctx, int ch)
1103 frame = &ctx->frame;
1104 sub = &frame->subframes[ch];
1106 /* warm-up samples */
1107 for(i=0; i<sub->order; i++) {
1108 put_sbits(&ctx->pb, sub->obits, sub->residual[i]);
1111 /* LPC coefficients */
1112 cbits = ctx->options.lpc_coeff_precision;
1113 put_bits(&ctx->pb, 4, cbits-1);
1114 put_sbits(&ctx->pb, 5, sub->shift);
1115 for(i=0; i<sub->order; i++) {
1116 put_sbits(&ctx->pb, cbits, sub->coefs[i]);
1120 output_residual(ctx, ch);
1123 static void output_subframes(FlacEncodeContext *s)
1131 for(ch=0; ch<s->channels; ch++) {
1132 sub = &frame->subframes[ch];
1134 /* subframe header */
1135 put_bits(&s->pb, 1, 0);
1136 put_bits(&s->pb, 6, sub->type_code);
1137 put_bits(&s->pb, 1, 0); /* no wasted bits */
1140 if(sub->type == FLAC_SUBFRAME_CONSTANT) {
1141 output_subframe_constant(s, ch);
1142 } else if(sub->type == FLAC_SUBFRAME_VERBATIM) {
1143 output_subframe_verbatim(s, ch);
1144 } else if(sub->type == FLAC_SUBFRAME_FIXED) {
1145 output_subframe_fixed(s, ch);
1146 } else if(sub->type == FLAC_SUBFRAME_LPC) {
1147 output_subframe_lpc(s, ch);
1152 static void output_frame_footer(FlacEncodeContext *s)
1155 flush_put_bits(&s->pb);
1156 crc = bswap_16(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0,
1157 s->pb.buf, put_bits_count(&s->pb)>>3));
1158 put_bits(&s->pb, 16, crc);
1159 flush_put_bits(&s->pb);
1162 static void update_md5_sum(FlacEncodeContext *s, int16_t *samples)
1166 for(i = 0; i < s->frame.blocksize*s->channels; i++) {
1167 int16_t smp = le2me_16(samples[i]);
1168 av_md5_update(s->md5ctx, (uint8_t *)&smp, 2);
1171 av_md5_update(s->md5ctx, (uint8_t *)samples, s->frame.blocksize*s->channels*2);
1175 static int flac_encode_frame(AVCodecContext *avctx, uint8_t *frame,
1176 int buf_size, void *data)
1179 FlacEncodeContext *s;
1180 int16_t *samples = data;
1184 s = avctx->priv_data;
1186 if(buf_size < s->max_framesize*2) {
1187 av_log(avctx, AV_LOG_ERROR, "output buffer too small\n");
1191 /* when the last block is reached, update the header in extradata */
1193 s->max_framesize = s->max_encoded_framesize;
1194 av_md5_final(s->md5ctx, s->md5sum);
1195 write_streaminfo(s, avctx->extradata);
1201 copy_samples(s, samples);
1203 channel_decorrelation(s);
1205 for(ch=0; ch<s->channels; ch++) {
1206 encode_residual(s, ch);
1210 init_put_bits(&s->pb, frame, buf_size);
1211 output_frame_header(s);
1212 output_subframes(s);
1213 output_frame_footer(s);
1214 out_bytes = put_bits_count(&s->pb) >> 3;
1216 if(out_bytes > s->max_framesize) {
1218 /* still too large. must be an error. */
1219 av_log(avctx, AV_LOG_ERROR, "error encoding frame\n");
1223 /* frame too large. use verbatim mode */
1224 for(ch=0; ch<s->channels; ch++) {
1225 encode_residual_v(s, ch);
1232 s->sample_count += avctx->frame_size;
1233 update_md5_sum(s, samples);
1234 if (out_bytes > s->max_encoded_framesize)
1235 s->max_encoded_framesize = out_bytes;
1236 if (out_bytes < s->min_framesize)
1237 s->min_framesize = out_bytes;
1242 static av_cold int flac_encode_close(AVCodecContext *avctx)
1244 if (avctx->priv_data) {
1245 FlacEncodeContext *s = avctx->priv_data;
1246 av_freep(&s->md5ctx);
1248 av_freep(&avctx->extradata);
1249 avctx->extradata_size = 0;
1250 av_freep(&avctx->coded_frame);
1254 AVCodec flac_encoder = {
1258 sizeof(FlacEncodeContext),
1263 .capabilities = CODEC_CAP_SMALL_LAST_FRAME | CODEC_CAP_DELAY,
1264 .sample_fmts = (const enum SampleFormat[]){SAMPLE_FMT_S16,SAMPLE_FMT_NONE},
1265 .long_name = NULL_IF_CONFIG_SMALL("FLAC (Free Lossless Audio Codec)"),