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/avassert.h"
23 #include "libavutil/crc.h"
24 #include "libavutil/md5.h"
25 #include "libavutil/opt.h"
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 MAX_FIXED_ORDER 4
40 #define MAX_PARTITION_ORDER 8
41 #define MAX_PARTITIONS (1 << MAX_PARTITION_ORDER)
42 #define MAX_LPC_PRECISION 15
43 #define MAX_LPC_SHIFT 15
44 #define MAX_RICE_PARAM 14
46 typedef struct CompressionOptions {
47 int compression_level;
49 enum FFLPCType lpc_type;
51 int lpc_coeff_precision;
52 int min_prediction_order;
53 int max_prediction_order;
54 int prediction_order_method;
55 int min_partition_order;
56 int max_partition_order;
60 typedef struct RiceContext {
62 int params[MAX_PARTITIONS];
65 typedef struct FlacSubframe {
70 int32_t coefs[MAX_LPC_ORDER];
73 int32_t samples[FLAC_MAX_BLOCKSIZE];
74 int32_t residual[FLAC_MAX_BLOCKSIZE+1];
77 typedef struct FlacFrame {
78 FlacSubframe subframes[FLAC_MAX_CHANNELS];
86 typedef struct FlacEncodeContext {
95 int max_encoded_framesize;
97 uint64_t sample_count;
100 CompressionOptions options;
101 AVCodecContext *avctx;
103 struct AVMD5 *md5ctx;
108 * Write streaminfo metadata block to byte array.
110 static void write_streaminfo(FlacEncodeContext *s, uint8_t *header)
114 memset(header, 0, FLAC_STREAMINFO_SIZE);
115 init_put_bits(&pb, header, FLAC_STREAMINFO_SIZE);
117 /* streaminfo metadata block */
118 put_bits(&pb, 16, s->max_blocksize);
119 put_bits(&pb, 16, s->max_blocksize);
120 put_bits(&pb, 24, s->min_framesize);
121 put_bits(&pb, 24, s->max_framesize);
122 put_bits(&pb, 20, s->samplerate);
123 put_bits(&pb, 3, s->channels-1);
124 put_bits(&pb, 5, 15); /* bits per sample - 1 */
125 /* write 36-bit sample count in 2 put_bits() calls */
126 put_bits(&pb, 24, (s->sample_count & 0xFFFFFF000LL) >> 12);
127 put_bits(&pb, 12, s->sample_count & 0x000000FFFLL);
129 memcpy(&header[18], s->md5sum, 16);
134 * Set blocksize based on samplerate.
135 * Choose the closest predefined blocksize >= BLOCK_TIME_MS milliseconds.
137 static int select_blocksize(int samplerate, int block_time_ms)
143 av_assert0(samplerate > 0);
144 blocksize = ff_flac_blocksize_table[1];
145 target = (samplerate * block_time_ms) / 1000;
146 for (i = 0; i < 16; i++) {
147 if (target >= ff_flac_blocksize_table[i] &&
148 ff_flac_blocksize_table[i] > blocksize) {
149 blocksize = ff_flac_blocksize_table[i];
156 static av_cold void dprint_compression_options(FlacEncodeContext *s)
158 AVCodecContext *avctx = s->avctx;
159 CompressionOptions *opt = &s->options;
161 av_log(avctx, AV_LOG_DEBUG, " compression: %d\n", opt->compression_level);
163 switch (opt->lpc_type) {
164 case FF_LPC_TYPE_NONE:
165 av_log(avctx, AV_LOG_DEBUG, " lpc type: None\n");
167 case FF_LPC_TYPE_FIXED:
168 av_log(avctx, AV_LOG_DEBUG, " lpc type: Fixed pre-defined coefficients\n");
170 case FF_LPC_TYPE_LEVINSON:
171 av_log(avctx, AV_LOG_DEBUG, " lpc type: Levinson-Durbin recursion with Welch window\n");
173 case FF_LPC_TYPE_CHOLESKY:
174 av_log(avctx, AV_LOG_DEBUG, " lpc type: Cholesky factorization, %d pass%s\n",
175 opt->lpc_passes, opt->lpc_passes == 1 ? "" : "es");
179 av_log(avctx, AV_LOG_DEBUG, " prediction order: %d, %d\n",
180 opt->min_prediction_order, opt->max_prediction_order);
182 switch (opt->prediction_order_method) {
183 case ORDER_METHOD_EST:
184 av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "estimate");
186 case ORDER_METHOD_2LEVEL:
187 av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "2-level");
189 case ORDER_METHOD_4LEVEL:
190 av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "4-level");
192 case ORDER_METHOD_8LEVEL:
193 av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "8-level");
195 case ORDER_METHOD_SEARCH:
196 av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "full search");
198 case ORDER_METHOD_LOG:
199 av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "log search");
204 av_log(avctx, AV_LOG_DEBUG, " partition order: %d, %d\n",
205 opt->min_partition_order, opt->max_partition_order);
207 av_log(avctx, AV_LOG_DEBUG, " block size: %d\n", avctx->frame_size);
209 av_log(avctx, AV_LOG_DEBUG, " lpc precision: %d\n",
210 opt->lpc_coeff_precision);
214 static av_cold int flac_encode_init(AVCodecContext *avctx)
216 int freq = avctx->sample_rate;
217 int channels = avctx->channels;
218 FlacEncodeContext *s = avctx->priv_data;
224 if (avctx->sample_fmt != AV_SAMPLE_FMT_S16)
227 if (channels < 1 || channels > FLAC_MAX_CHANNELS)
229 s->channels = channels;
231 /* find samplerate in table */
234 for (i = 4; i < 12; i++) {
235 if (freq == ff_flac_sample_rate_table[i]) {
236 s->samplerate = ff_flac_sample_rate_table[i];
242 /* if not in table, samplerate is non-standard */
244 if (freq % 1000 == 0 && freq < 255000) {
246 s->sr_code[1] = freq / 1000;
247 } else if (freq % 10 == 0 && freq < 655350) {
249 s->sr_code[1] = freq / 10;
250 } else if (freq < 65535) {
252 s->sr_code[1] = freq;
256 s->samplerate = freq;
259 /* set compression option defaults based on avctx->compression_level */
260 if (avctx->compression_level < 0)
261 s->options.compression_level = 5;
263 s->options.compression_level = avctx->compression_level;
265 level = s->options.compression_level;
267 av_log(avctx, AV_LOG_ERROR, "invalid compression level: %d\n",
268 s->options.compression_level);
272 s->options.block_time_ms = ((int[]){ 27, 27, 27,105,105,105,105,105,105,105,105,105,105})[level];
274 if (s->options.lpc_type == FF_LPC_TYPE_DEFAULT)
275 s->options.lpc_type = ((int[]){ FF_LPC_TYPE_FIXED, FF_LPC_TYPE_FIXED, FF_LPC_TYPE_FIXED,
276 FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON,
277 FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON,
278 FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON,
279 FF_LPC_TYPE_LEVINSON})[level];
281 s->options.min_prediction_order = ((int[]){ 2, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1})[level];
282 s->options.max_prediction_order = ((int[]){ 3, 4, 4, 6, 8, 8, 8, 8, 12, 12, 12, 32, 32})[level];
284 if (s->options.prediction_order_method < 0)
285 s->options.prediction_order_method = ((int[]){ ORDER_METHOD_EST, ORDER_METHOD_EST, ORDER_METHOD_EST,
286 ORDER_METHOD_EST, ORDER_METHOD_EST, ORDER_METHOD_EST,
287 ORDER_METHOD_4LEVEL, ORDER_METHOD_LOG, ORDER_METHOD_4LEVEL,
288 ORDER_METHOD_LOG, ORDER_METHOD_SEARCH, ORDER_METHOD_LOG,
289 ORDER_METHOD_SEARCH})[level];
291 if (s->options.min_partition_order > s->options.max_partition_order) {
292 av_log(avctx, AV_LOG_ERROR, "invalid partition orders: min=%d max=%d\n",
293 s->options.min_partition_order, s->options.max_partition_order);
294 return AVERROR(EINVAL);
296 if (s->options.min_partition_order < 0)
297 s->options.min_partition_order = ((int[]){ 2, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0})[level];
298 if (s->options.max_partition_order < 0)
299 s->options.max_partition_order = ((int[]){ 2, 2, 3, 3, 3, 8, 8, 8, 8, 8, 8, 8, 8})[level];
301 if (s->options.lpc_type == FF_LPC_TYPE_NONE) {
302 s->options.min_prediction_order = 0;
303 } else if (avctx->min_prediction_order >= 0) {
304 if (s->options.lpc_type == FF_LPC_TYPE_FIXED) {
305 if (avctx->min_prediction_order > MAX_FIXED_ORDER) {
306 av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n",
307 avctx->min_prediction_order);
310 } else if (avctx->min_prediction_order < MIN_LPC_ORDER ||
311 avctx->min_prediction_order > MAX_LPC_ORDER) {
312 av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n",
313 avctx->min_prediction_order);
316 s->options.min_prediction_order = avctx->min_prediction_order;
318 if (s->options.lpc_type == FF_LPC_TYPE_NONE) {
319 s->options.max_prediction_order = 0;
320 } else if (avctx->max_prediction_order >= 0) {
321 if (s->options.lpc_type == FF_LPC_TYPE_FIXED) {
322 if (avctx->max_prediction_order > MAX_FIXED_ORDER) {
323 av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n",
324 avctx->max_prediction_order);
327 } else if (avctx->max_prediction_order < MIN_LPC_ORDER ||
328 avctx->max_prediction_order > MAX_LPC_ORDER) {
329 av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n",
330 avctx->max_prediction_order);
333 s->options.max_prediction_order = avctx->max_prediction_order;
335 if (s->options.max_prediction_order < s->options.min_prediction_order) {
336 av_log(avctx, AV_LOG_ERROR, "invalid prediction orders: min=%d max=%d\n",
337 s->options.min_prediction_order, s->options.max_prediction_order);
341 if (avctx->frame_size > 0) {
342 if (avctx->frame_size < FLAC_MIN_BLOCKSIZE ||
343 avctx->frame_size > FLAC_MAX_BLOCKSIZE) {
344 av_log(avctx, AV_LOG_ERROR, "invalid block size: %d\n",
349 s->avctx->frame_size = select_blocksize(s->samplerate, s->options.block_time_ms);
351 s->max_blocksize = s->avctx->frame_size;
353 /* set maximum encoded frame size in verbatim mode */
354 s->max_framesize = ff_flac_get_max_frame_size(s->avctx->frame_size,
357 /* initialize MD5 context */
358 s->md5ctx = av_md5_alloc();
360 return AVERROR(ENOMEM);
361 av_md5_init(s->md5ctx);
363 streaminfo = av_malloc(FLAC_STREAMINFO_SIZE);
365 return AVERROR(ENOMEM);
366 write_streaminfo(s, streaminfo);
367 avctx->extradata = streaminfo;
368 avctx->extradata_size = FLAC_STREAMINFO_SIZE;
371 s->min_framesize = s->max_framesize;
373 #if FF_API_OLD_ENCODE_AUDIO
374 avctx->coded_frame = avcodec_alloc_frame();
375 if (!avctx->coded_frame)
376 return AVERROR(ENOMEM);
380 avctx->channel_layout != (AV_CH_LAYOUT_STEREO|AV_CH_FRONT_CENTER) ||
382 avctx->channel_layout != AV_CH_LAYOUT_2_2 &&
383 avctx->channel_layout != AV_CH_LAYOUT_QUAD ||
385 avctx->channel_layout != AV_CH_LAYOUT_5POINT0 &&
386 avctx->channel_layout != AV_CH_LAYOUT_5POINT0_BACK ||
388 avctx->channel_layout != AV_CH_LAYOUT_5POINT1 &&
389 avctx->channel_layout != AV_CH_LAYOUT_5POINT1_BACK) {
390 if (avctx->channel_layout) {
391 av_log(avctx, AV_LOG_ERROR, "Channel layout not supported by Flac, "
392 "output stream will have incorrect "
393 "channel layout.\n");
395 av_log(avctx, AV_LOG_WARNING, "No channel layout specified. The encoder "
396 "will use Flac channel layout for "
397 "%d channels.\n", channels);
401 ret = ff_lpc_init(&s->lpc_ctx, avctx->frame_size,
402 s->options.max_prediction_order, FF_LPC_TYPE_LEVINSON);
404 dprint_compression_options(s);
410 static void init_frame(FlacEncodeContext *s, int nb_samples)
417 for (i = 0; i < 16; i++) {
418 if (nb_samples == ff_flac_blocksize_table[i]) {
419 frame->blocksize = ff_flac_blocksize_table[i];
420 frame->bs_code[0] = i;
421 frame->bs_code[1] = 0;
426 frame->blocksize = nb_samples;
427 if (frame->blocksize <= 256) {
428 frame->bs_code[0] = 6;
429 frame->bs_code[1] = frame->blocksize-1;
431 frame->bs_code[0] = 7;
432 frame->bs_code[1] = frame->blocksize-1;
436 for (ch = 0; ch < s->channels; ch++)
437 frame->subframes[ch].obits = 16;
439 frame->verbatim_only = 0;
444 * Copy channel-interleaved input samples into separate subframes.
446 static void copy_samples(FlacEncodeContext *s, const int16_t *samples)
452 for (i = 0, j = 0; i < frame->blocksize; i++)
453 for (ch = 0; ch < s->channels; ch++, j++)
454 frame->subframes[ch].samples[i] = samples[j];
458 static int rice_count_exact(int32_t *res, int n, int k)
463 for (i = 0; i < n; i++) {
464 int32_t v = -2 * res[i] - 1;
466 count += (v >> k) + 1 + k;
472 static int subframe_count_exact(FlacEncodeContext *s, FlacSubframe *sub,
475 int p, porder, psize;
479 /* subframe header */
483 if (sub->type == FLAC_SUBFRAME_CONSTANT) {
485 } else if (sub->type == FLAC_SUBFRAME_VERBATIM) {
486 count += s->frame.blocksize * sub->obits;
488 /* warm-up samples */
489 count += pred_order * sub->obits;
491 /* LPC coefficients */
492 if (sub->type == FLAC_SUBFRAME_LPC)
493 count += 4 + 5 + pred_order * s->options.lpc_coeff_precision;
495 /* rice-encoded block */
498 /* partition order */
499 porder = sub->rc.porder;
500 psize = s->frame.blocksize >> porder;
506 for (p = 0; p < 1 << porder; p++) {
507 int k = sub->rc.params[p];
509 count += rice_count_exact(&sub->residual[i], part_end - i, k);
511 part_end = FFMIN(s->frame.blocksize, part_end + psize);
519 #define rice_encode_count(sum, n, k) (((n)*((k)+1))+((sum-(n>>1))>>(k)))
522 * Solve for d/dk(rice_encode_count) = n-((sum-(n>>1))>>(k+1)) = 0.
524 static int find_optimal_param(uint32_t sum, int n)
531 sum2 = sum - (n >> 1);
532 k = av_log2(n < 256 ? FASTDIV(sum2, n) : sum2 / n);
533 return FFMIN(k, MAX_RICE_PARAM);
537 static uint32_t calc_optimal_rice_params(RiceContext *rc, int porder,
538 uint32_t *sums, int n, int pred_order)
544 part = (1 << porder);
547 cnt = (n >> porder) - pred_order;
548 for (i = 0; i < part; i++) {
549 k = find_optimal_param(sums[i], cnt);
551 all_bits += rice_encode_count(sums[i], cnt, k);
561 static void calc_sums(int pmin, int pmax, uint32_t *data, int n, int pred_order,
562 uint32_t sums[][MAX_PARTITIONS])
566 uint32_t *res, *res_end;
568 /* sums for highest level */
570 res = &data[pred_order];
571 res_end = &data[n >> pmax];
572 for (i = 0; i < parts; i++) {
574 while (res < res_end)
577 res_end += n >> pmax;
579 /* sums for lower levels */
580 for (i = pmax - 1; i >= pmin; i--) {
582 for (j = 0; j < parts; j++)
583 sums[i][j] = sums[i+1][2*j] + sums[i+1][2*j+1];
588 static uint32_t calc_rice_params(RiceContext *rc, int pmin, int pmax,
589 int32_t *data, int n, int pred_order)
592 uint32_t bits[MAX_PARTITION_ORDER+1];
596 uint32_t sums[MAX_PARTITION_ORDER+1][MAX_PARTITIONS];
598 av_assert1(pmin >= 0 && pmin <= MAX_PARTITION_ORDER);
599 av_assert1(pmax >= 0 && pmax <= MAX_PARTITION_ORDER);
600 av_assert1(pmin <= pmax);
602 udata = av_malloc(n * sizeof(uint32_t));
603 for (i = 0; i < n; i++)
604 udata[i] = (2*data[i]) ^ (data[i]>>31);
606 calc_sums(pmin, pmax, udata, n, pred_order, sums);
609 bits[pmin] = UINT32_MAX;
610 for (i = pmin; i <= pmax; i++) {
611 bits[i] = calc_optimal_rice_params(&tmp_rc, i, sums[i], n, pred_order);
612 if (bits[i] <= bits[opt_porder]) {
619 return bits[opt_porder];
623 static int get_max_p_order(int max_porder, int n, int order)
625 int porder = FFMIN(max_porder, av_log2(n^(n-1)));
627 porder = FFMIN(porder, av_log2(n/order));
632 static uint32_t find_subframe_rice_params(FlacEncodeContext *s,
633 FlacSubframe *sub, int pred_order)
635 int pmin = get_max_p_order(s->options.min_partition_order,
636 s->frame.blocksize, pred_order);
637 int pmax = get_max_p_order(s->options.max_partition_order,
638 s->frame.blocksize, pred_order);
640 uint32_t bits = 8 + pred_order * sub->obits + 2 + 4;
641 if (sub->type == FLAC_SUBFRAME_LPC)
642 bits += 4 + 5 + pred_order * s->options.lpc_coeff_precision;
643 bits += calc_rice_params(&sub->rc, pmin, pmax, sub->residual,
644 s->frame.blocksize, pred_order);
649 static void encode_residual_fixed(int32_t *res, const int32_t *smp, int n,
654 for (i = 0; i < order; i++)
658 for (i = order; i < n; i++)
660 } else if (order == 1) {
661 for (i = order; i < n; i++)
662 res[i] = smp[i] - smp[i-1];
663 } else if (order == 2) {
664 int a = smp[order-1] - smp[order-2];
665 for (i = order; i < n; i += 2) {
666 int b = smp[i ] - smp[i-1];
668 a = smp[i+1] - smp[i ];
671 } else if (order == 3) {
672 int a = smp[order-1] - smp[order-2];
673 int c = smp[order-1] - 2*smp[order-2] + smp[order-3];
674 for (i = order; i < n; i += 2) {
675 int b = smp[i ] - smp[i-1];
678 a = smp[i+1] - smp[i ];
683 int a = smp[order-1] - smp[order-2];
684 int c = smp[order-1] - 2*smp[order-2] + smp[order-3];
685 int e = smp[order-1] - 3*smp[order-2] + 3*smp[order-3] - smp[order-4];
686 for (i = order; i < n; i += 2) {
687 int b = smp[i ] - smp[i-1];
691 a = smp[i+1] - smp[i ];
701 int c = coefs[(x)-1];\
707 static av_always_inline void encode_residual_lpc_unrolled(int32_t *res,
708 const int32_t *smp, int n, int order,
709 const int32_t *coefs, int shift, int big)
712 for (i = order; i < n; i += 2) {
713 int s = smp[i-order];
762 res[i ] = smp[i ] - (p0 >> shift);
763 res[i+1] = smp[i+1] - (p1 >> shift);
768 static void encode_residual_lpc(int32_t *res, const int32_t *smp, int n,
769 int order, const int32_t *coefs, int shift)
772 for (i = 0; i < order; i++)
775 for (i = order; i < n; i += 2) {
779 for (j = 0; j < order; j++) {
785 res[i ] = smp[i ] - (p0 >> shift);
786 res[i+1] = smp[i+1] - (p1 >> shift);
790 case 1: encode_residual_lpc_unrolled(res, smp, n, 1, coefs, shift, 0); break;
791 case 2: encode_residual_lpc_unrolled(res, smp, n, 2, coefs, shift, 0); break;
792 case 3: encode_residual_lpc_unrolled(res, smp, n, 3, coefs, shift, 0); break;
793 case 4: encode_residual_lpc_unrolled(res, smp, n, 4, coefs, shift, 0); break;
794 case 5: encode_residual_lpc_unrolled(res, smp, n, 5, coefs, shift, 0); break;
795 case 6: encode_residual_lpc_unrolled(res, smp, n, 6, coefs, shift, 0); break;
796 case 7: encode_residual_lpc_unrolled(res, smp, n, 7, coefs, shift, 0); break;
797 case 8: encode_residual_lpc_unrolled(res, smp, n, 8, coefs, shift, 0); break;
798 default: encode_residual_lpc_unrolled(res, smp, n, order, coefs, shift, 1); break;
804 static int encode_residual_ch(FlacEncodeContext *s, int ch)
807 int min_order, max_order, opt_order, omethod;
810 int32_t coefs[MAX_LPC_ORDER][MAX_LPC_ORDER];
811 int shift[MAX_LPC_ORDER];
815 sub = &frame->subframes[ch];
818 n = frame->blocksize;
821 for (i = 1; i < n; i++)
825 sub->type = sub->type_code = FLAC_SUBFRAME_CONSTANT;
827 return subframe_count_exact(s, sub, 0);
831 if (frame->verbatim_only || n < 5) {
832 sub->type = sub->type_code = FLAC_SUBFRAME_VERBATIM;
833 memcpy(res, smp, n * sizeof(int32_t));
834 return subframe_count_exact(s, sub, 0);
837 min_order = s->options.min_prediction_order;
838 max_order = s->options.max_prediction_order;
839 omethod = s->options.prediction_order_method;
842 sub->type = FLAC_SUBFRAME_FIXED;
843 if (s->options.lpc_type == FF_LPC_TYPE_NONE ||
844 s->options.lpc_type == FF_LPC_TYPE_FIXED || n <= max_order) {
845 uint32_t bits[MAX_FIXED_ORDER+1];
846 if (max_order > MAX_FIXED_ORDER)
847 max_order = MAX_FIXED_ORDER;
849 bits[0] = UINT32_MAX;
850 for (i = min_order; i <= max_order; i++) {
851 encode_residual_fixed(res, smp, n, i);
852 bits[i] = find_subframe_rice_params(s, sub, i);
853 if (bits[i] < bits[opt_order])
856 sub->order = opt_order;
857 sub->type_code = sub->type | sub->order;
858 if (sub->order != max_order) {
859 encode_residual_fixed(res, smp, n, sub->order);
860 find_subframe_rice_params(s, sub, sub->order);
862 return subframe_count_exact(s, sub, sub->order);
866 sub->type = FLAC_SUBFRAME_LPC;
867 opt_order = ff_lpc_calc_coefs(&s->lpc_ctx, smp, n, min_order, max_order,
868 s->options.lpc_coeff_precision, coefs, shift, s->options.lpc_type,
869 s->options.lpc_passes, omethod,
872 if (omethod == ORDER_METHOD_2LEVEL ||
873 omethod == ORDER_METHOD_4LEVEL ||
874 omethod == ORDER_METHOD_8LEVEL) {
875 int levels = 1 << omethod;
876 uint32_t bits[1 << ORDER_METHOD_8LEVEL];
878 int opt_index = levels-1;
879 opt_order = max_order-1;
880 bits[opt_index] = UINT32_MAX;
881 for (i = levels-1; i >= 0; i--) {
882 order = min_order + (((max_order-min_order+1) * (i+1)) / levels)-1;
885 encode_residual_lpc(res, smp, n, order+1, coefs[order], shift[order]);
886 bits[i] = find_subframe_rice_params(s, sub, order+1);
887 if (bits[i] < bits[opt_index]) {
893 } else if (omethod == ORDER_METHOD_SEARCH) {
894 // brute-force optimal order search
895 uint32_t bits[MAX_LPC_ORDER];
897 bits[0] = UINT32_MAX;
898 for (i = min_order-1; i < max_order; i++) {
899 encode_residual_lpc(res, smp, n, i+1, coefs[i], shift[i]);
900 bits[i] = find_subframe_rice_params(s, sub, i+1);
901 if (bits[i] < bits[opt_order])
905 } else if (omethod == ORDER_METHOD_LOG) {
906 uint32_t bits[MAX_LPC_ORDER];
909 opt_order = min_order - 1 + (max_order-min_order)/3;
910 memset(bits, -1, sizeof(bits));
912 for (step = 16; step; step >>= 1) {
913 int last = opt_order;
914 for (i = last-step; i <= last+step; i += step) {
915 if (i < min_order-1 || i >= max_order || bits[i] < UINT32_MAX)
917 encode_residual_lpc(res, smp, n, i+1, coefs[i], shift[i]);
918 bits[i] = find_subframe_rice_params(s, sub, i+1);
919 if (bits[i] < bits[opt_order])
926 sub->order = opt_order;
927 sub->type_code = sub->type | (sub->order-1);
928 sub->shift = shift[sub->order-1];
929 for (i = 0; i < sub->order; i++)
930 sub->coefs[i] = coefs[sub->order-1][i];
932 encode_residual_lpc(res, smp, n, sub->order, sub->coefs, sub->shift);
934 find_subframe_rice_params(s, sub, sub->order);
936 return subframe_count_exact(s, sub, sub->order);
940 static int count_frame_header(FlacEncodeContext *s)
942 uint8_t av_unused tmp;
948 <1> Blocking strategy
949 <4> Block size in inter-channel samples
951 <4> Channel assignment
952 <3> Sample size in bits
957 /* coded frame number */
958 PUT_UTF8(s->frame_count, tmp, count += 8;)
960 /* explicit block size */
961 if (s->frame.bs_code[0] == 6)
963 else if (s->frame.bs_code[0] == 7)
966 /* explicit sample rate */
967 count += ((s->sr_code[0] == 12) + (s->sr_code[0] > 12)) * 8;
969 /* frame header CRC-8 */
976 static int encode_frame(FlacEncodeContext *s)
980 count = count_frame_header(s);
982 for (ch = 0; ch < s->channels; ch++)
983 count += encode_residual_ch(s, ch);
985 count += (8 - (count & 7)) & 7; // byte alignment
986 count += 16; // CRC-16
992 static int estimate_stereo_mode(int32_t *left_ch, int32_t *right_ch, int n)
1000 /* calculate sum of 2nd order residual for each channel */
1001 sum[0] = sum[1] = sum[2] = sum[3] = 0;
1002 for (i = 2; i < n; i++) {
1003 lt = left_ch[i] - 2*left_ch[i-1] + left_ch[i-2];
1004 rt = right_ch[i] - 2*right_ch[i-1] + right_ch[i-2];
1005 sum[2] += FFABS((lt + rt) >> 1);
1006 sum[3] += FFABS(lt - rt);
1007 sum[0] += FFABS(lt);
1008 sum[1] += FFABS(rt);
1010 /* estimate bit counts */
1011 for (i = 0; i < 4; i++) {
1012 k = find_optimal_param(2 * sum[i], n);
1013 sum[i] = rice_encode_count( 2 * sum[i], n, k);
1016 /* calculate score for each mode */
1017 score[0] = sum[0] + sum[1];
1018 score[1] = sum[0] + sum[3];
1019 score[2] = sum[1] + sum[3];
1020 score[3] = sum[2] + sum[3];
1022 /* return mode with lowest score */
1024 for (i = 1; i < 4; i++)
1025 if (score[i] < score[best])
1033 * Perform stereo channel decorrelation.
1035 static void channel_decorrelation(FlacEncodeContext *s)
1038 int32_t *left, *right;
1042 n = frame->blocksize;
1043 left = frame->subframes[0].samples;
1044 right = frame->subframes[1].samples;
1046 if (s->channels != 2) {
1047 frame->ch_mode = FLAC_CHMODE_INDEPENDENT;
1051 if (s->options.ch_mode < 0)
1052 frame->ch_mode = estimate_stereo_mode(left, right, n);
1054 frame->ch_mode = s->options.ch_mode;
1056 /* perform decorrelation and adjust bits-per-sample */
1057 if (frame->ch_mode == FLAC_CHMODE_INDEPENDENT)
1059 if (frame->ch_mode == FLAC_CHMODE_MID_SIDE) {
1061 for (i = 0; i < n; i++) {
1063 left[i] = (tmp + right[i]) >> 1;
1064 right[i] = tmp - right[i];
1066 frame->subframes[1].obits++;
1067 } else if (frame->ch_mode == FLAC_CHMODE_LEFT_SIDE) {
1068 for (i = 0; i < n; i++)
1069 right[i] = left[i] - right[i];
1070 frame->subframes[1].obits++;
1072 for (i = 0; i < n; i++)
1073 left[i] -= right[i];
1074 frame->subframes[0].obits++;
1079 static void write_utf8(PutBitContext *pb, uint32_t val)
1082 PUT_UTF8(val, tmp, put_bits(pb, 8, tmp);)
1086 static void write_frame_header(FlacEncodeContext *s)
1093 put_bits(&s->pb, 16, 0xFFF8);
1094 put_bits(&s->pb, 4, frame->bs_code[0]);
1095 put_bits(&s->pb, 4, s->sr_code[0]);
1097 if (frame->ch_mode == FLAC_CHMODE_INDEPENDENT)
1098 put_bits(&s->pb, 4, s->channels-1);
1100 put_bits(&s->pb, 4, frame->ch_mode + FLAC_MAX_CHANNELS - 1);
1102 put_bits(&s->pb, 3, 4); /* bits-per-sample code */
1103 put_bits(&s->pb, 1, 0);
1104 write_utf8(&s->pb, s->frame_count);
1106 if (frame->bs_code[0] == 6)
1107 put_bits(&s->pb, 8, frame->bs_code[1]);
1108 else if (frame->bs_code[0] == 7)
1109 put_bits(&s->pb, 16, frame->bs_code[1]);
1111 if (s->sr_code[0] == 12)
1112 put_bits(&s->pb, 8, s->sr_code[1]);
1113 else if (s->sr_code[0] > 12)
1114 put_bits(&s->pb, 16, s->sr_code[1]);
1116 flush_put_bits(&s->pb);
1117 crc = av_crc(av_crc_get_table(AV_CRC_8_ATM), 0, s->pb.buf,
1118 put_bits_count(&s->pb) >> 3);
1119 put_bits(&s->pb, 8, crc);
1123 static void write_subframes(FlacEncodeContext *s)
1127 for (ch = 0; ch < s->channels; ch++) {
1128 FlacSubframe *sub = &s->frame.subframes[ch];
1129 int i, p, porder, psize;
1131 int32_t *res = sub->residual;
1132 int32_t *frame_end = &sub->residual[s->frame.blocksize];
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 put_sbits(&s->pb, sub->obits, res[0]);
1142 } else if (sub->type == FLAC_SUBFRAME_VERBATIM) {
1143 while (res < frame_end)
1144 put_sbits(&s->pb, sub->obits, *res++);
1146 /* warm-up samples */
1147 for (i = 0; i < sub->order; i++)
1148 put_sbits(&s->pb, sub->obits, *res++);
1150 /* LPC coefficients */
1151 if (sub->type == FLAC_SUBFRAME_LPC) {
1152 int cbits = s->options.lpc_coeff_precision;
1153 put_bits( &s->pb, 4, cbits-1);
1154 put_sbits(&s->pb, 5, sub->shift);
1155 for (i = 0; i < sub->order; i++)
1156 put_sbits(&s->pb, cbits, sub->coefs[i]);
1159 /* rice-encoded block */
1160 put_bits(&s->pb, 2, 0);
1162 /* partition order */
1163 porder = sub->rc.porder;
1164 psize = s->frame.blocksize >> porder;
1165 put_bits(&s->pb, 4, porder);
1168 part_end = &sub->residual[psize];
1169 for (p = 0; p < 1 << porder; p++) {
1170 int k = sub->rc.params[p];
1171 put_bits(&s->pb, 4, k);
1172 while (res < part_end)
1173 set_sr_golomb_flac(&s->pb, *res++, k, INT32_MAX, 0);
1174 part_end = FFMIN(frame_end, part_end + psize);
1181 static void write_frame_footer(FlacEncodeContext *s)
1184 flush_put_bits(&s->pb);
1185 crc = av_bswap16(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, s->pb.buf,
1186 put_bits_count(&s->pb)>>3));
1187 put_bits(&s->pb, 16, crc);
1188 flush_put_bits(&s->pb);
1192 static int write_frame(FlacEncodeContext *s, AVPacket *avpkt)
1194 init_put_bits(&s->pb, avpkt->data, avpkt->size);
1195 write_frame_header(s);
1197 write_frame_footer(s);
1198 return put_bits_count(&s->pb) >> 3;
1202 static void update_md5_sum(FlacEncodeContext *s, const int16_t *samples)
1206 for (i = 0; i < s->frame.blocksize * s->channels; i++) {
1207 int16_t smp = av_le2ne16(samples[i]);
1208 av_md5_update(s->md5ctx, (uint8_t *)&smp, 2);
1211 av_md5_update(s->md5ctx, (const uint8_t *)samples, s->frame.blocksize*s->channels*2);
1216 static int flac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
1217 const AVFrame *frame, int *got_packet_ptr)
1219 FlacEncodeContext *s;
1220 const int16_t *samples;
1221 int frame_bytes, out_bytes, ret;
1223 s = avctx->priv_data;
1225 /* when the last block is reached, update the header in extradata */
1227 s->max_framesize = s->max_encoded_framesize;
1228 av_md5_final(s->md5ctx, s->md5sum);
1229 write_streaminfo(s, avctx->extradata);
1232 samples = (const int16_t *)frame->data[0];
1234 /* change max_framesize for small final frame */
1235 if (frame->nb_samples < s->frame.blocksize) {
1236 s->max_framesize = ff_flac_get_max_frame_size(frame->nb_samples,
1240 init_frame(s, frame->nb_samples);
1242 copy_samples(s, samples);
1244 channel_decorrelation(s);
1246 frame_bytes = encode_frame(s);
1248 /* fallback to verbatim mode if the compressed frame is larger than it
1249 would be if encoded uncompressed. */
1250 if (frame_bytes > s->max_framesize) {
1251 s->frame.verbatim_only = 1;
1252 frame_bytes = encode_frame(s);
1255 if ((ret = ff_alloc_packet2(avctx, avpkt, frame_bytes)))
1258 out_bytes = write_frame(s, avpkt);
1261 s->sample_count += frame->nb_samples;
1262 update_md5_sum(s, samples);
1263 if (out_bytes > s->max_encoded_framesize)
1264 s->max_encoded_framesize = out_bytes;
1265 if (out_bytes < s->min_framesize)
1266 s->min_framesize = out_bytes;
1268 avpkt->pts = frame->pts;
1269 avpkt->duration = ff_samples_to_time_base(avctx, frame->nb_samples);
1270 avpkt->size = out_bytes;
1271 *got_packet_ptr = 1;
1276 static av_cold int flac_encode_close(AVCodecContext *avctx)
1278 if (avctx->priv_data) {
1279 FlacEncodeContext *s = avctx->priv_data;
1280 av_freep(&s->md5ctx);
1281 ff_lpc_end(&s->lpc_ctx);
1283 av_freep(&avctx->extradata);
1284 avctx->extradata_size = 0;
1285 #if FF_API_OLD_ENCODE_AUDIO
1286 av_freep(&avctx->coded_frame);
1291 #define FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
1292 static const AVOption options[] = {
1293 { "lpc_coeff_precision", "LPC coefficient precision", offsetof(FlacEncodeContext, options.lpc_coeff_precision), AV_OPT_TYPE_INT, {.i64 = 15 }, 0, MAX_LPC_PRECISION, FLAGS },
1294 { "lpc_type", "LPC algorithm", offsetof(FlacEncodeContext, options.lpc_type), AV_OPT_TYPE_INT, {.i64 = FF_LPC_TYPE_DEFAULT }, FF_LPC_TYPE_DEFAULT, FF_LPC_TYPE_NB-1, FLAGS, "lpc_type" },
1295 { "none", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = FF_LPC_TYPE_NONE }, INT_MIN, INT_MAX, FLAGS, "lpc_type" },
1296 { "fixed", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = FF_LPC_TYPE_FIXED }, INT_MIN, INT_MAX, FLAGS, "lpc_type" },
1297 { "levinson", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = FF_LPC_TYPE_LEVINSON }, INT_MIN, INT_MAX, FLAGS, "lpc_type" },
1298 { "cholesky", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = FF_LPC_TYPE_CHOLESKY }, INT_MIN, INT_MAX, FLAGS, "lpc_type" },
1299 { "lpc_passes", "Number of passes to use for Cholesky factorization during LPC analysis", offsetof(FlacEncodeContext, options.lpc_passes), AV_OPT_TYPE_INT, {.i64 = -1 }, INT_MIN, INT_MAX, FLAGS },
1300 { "min_partition_order", NULL, offsetof(FlacEncodeContext, options.min_partition_order), AV_OPT_TYPE_INT, {.i64 = -1 }, -1, MAX_PARTITION_ORDER, FLAGS },
1301 { "max_partition_order", NULL, offsetof(FlacEncodeContext, options.max_partition_order), AV_OPT_TYPE_INT, {.i64 = -1 }, -1, MAX_PARTITION_ORDER, FLAGS },
1302 { "prediction_order_method", "Search method for selecting prediction order", offsetof(FlacEncodeContext, options.prediction_order_method), AV_OPT_TYPE_INT, {.i64 = -1 }, -1, ORDER_METHOD_LOG, FLAGS, "predm" },
1303 { "estimation", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_EST }, INT_MIN, INT_MAX, FLAGS, "predm" },
1304 { "2level", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_2LEVEL }, INT_MIN, INT_MAX, FLAGS, "predm" },
1305 { "4level", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_4LEVEL }, INT_MIN, INT_MAX, FLAGS, "predm" },
1306 { "8level", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_8LEVEL }, INT_MIN, INT_MAX, FLAGS, "predm" },
1307 { "search", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_SEARCH }, INT_MIN, INT_MAX, FLAGS, "predm" },
1308 { "log", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_LOG }, INT_MIN, INT_MAX, FLAGS, "predm" },
1309 { "ch_mode", "Stereo decorrelation mode", offsetof(FlacEncodeContext, options.ch_mode), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, FLAC_CHMODE_MID_SIDE, FLAGS, "ch_mode" },
1310 { "auto", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = -1 }, INT_MIN, INT_MAX, FLAGS, "ch_mode" },
1311 { "indep", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FLAC_CHMODE_INDEPENDENT }, INT_MIN, INT_MAX, FLAGS, "ch_mode" },
1312 { "left_side", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FLAC_CHMODE_LEFT_SIDE }, INT_MIN, INT_MAX, FLAGS, "ch_mode" },
1313 { "right_side", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FLAC_CHMODE_RIGHT_SIDE }, INT_MIN, INT_MAX, FLAGS, "ch_mode" },
1314 { "mid_side", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FLAC_CHMODE_MID_SIDE }, INT_MIN, INT_MAX, FLAGS, "ch_mode" },
1318 static const AVClass flac_encoder_class = {
1320 av_default_item_name,
1322 LIBAVUTIL_VERSION_INT,
1325 AVCodec ff_flac_encoder = {
1327 .type = AVMEDIA_TYPE_AUDIO,
1328 .id = AV_CODEC_ID_FLAC,
1329 .priv_data_size = sizeof(FlacEncodeContext),
1330 .init = flac_encode_init,
1331 .encode2 = flac_encode_frame,
1332 .close = flac_encode_close,
1333 .capabilities = CODEC_CAP_SMALL_LAST_FRAME | CODEC_CAP_DELAY | CODEC_CAP_LOSSLESS,
1334 .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S16,
1335 AV_SAMPLE_FMT_NONE },
1336 .long_name = NULL_IF_CONFIG_SMALL("FLAC (Free Lossless Audio Codec)"),
1337 .priv_class = &flac_encoder_class,