1 /*
2 * FLAC audio encoder
3 * Copyright (c) 2006 Justin Ruggles <justin.ruggles@gmail.com>
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22 #include "libavutil/avassert.h"
23 #include "libavutil/channel_layout.h"
24 #include "libavutil/crc.h"
25 #include "libavutil/intmath.h"
26 #include "libavutil/md5.h"
27 #include "libavutil/opt.h"
28
29 #include "avcodec.h"
30 #include "bswapdsp.h"
31 #include "codec_internal.h"
32 #include "encode.h"
33 #include "put_bits.h"
34 #include "put_golomb.h"
35 #include "internal.h"
36 #include "lpc.h"
37 #include "flac.h"
38 #include "flacdata.h"
39 #include "flacdsp.h"
40
41 #define FLAC_SUBFRAME_CONSTANT 0
42 #define FLAC_SUBFRAME_VERBATIM 1
43 #define FLAC_SUBFRAME_FIXED 8
44 #define FLAC_SUBFRAME_LPC 32
45
46 #define MAX_FIXED_ORDER 4
47 #define MAX_PARTITION_ORDER 8
48 #define MAX_PARTITIONS (1 << MAX_PARTITION_ORDER)
49 #define MAX_LPC_PRECISION 15
50 #define MIN_LPC_SHIFT 0
51 #define MAX_LPC_SHIFT 15
52
53 enum CodingMode {
54 CODING_MODE_RICE = 4,
55 CODING_MODE_RICE2 = 5,
56 };
57
58 typedef struct CompressionOptions {
59 int compression_level;
60 int block_time_ms;
61 enum FFLPCType lpc_type;
62 int lpc_passes;
63 int lpc_coeff_precision;
64 int min_prediction_order;
65 int max_prediction_order;
66 int prediction_order_method;
67 int min_partition_order;
68 int max_partition_order;
69 int ch_mode;
70 int exact_rice_parameters;
71 int multi_dim_quant;
72 } CompressionOptions;
73
74 typedef struct RiceContext {
75 enum CodingMode coding_mode;
76 int porder;
77 int params[MAX_PARTITIONS];
78 } RiceContext;
79
80 typedef struct FlacSubframe {
81 int type;
82 int type_code;
83 int obits;
84 int wasted;
85 int order;
86 int32_t coefs[MAX_LPC_ORDER];
87 int shift;
88
89 RiceContext rc;
90 uint32_t rc_udata[FLAC_MAX_BLOCKSIZE];
91 uint64_t rc_sums[32][MAX_PARTITIONS];
92
93 int32_t samples[FLAC_MAX_BLOCKSIZE];
94 int32_t residual[FLAC_MAX_BLOCKSIZE+11];
95 } FlacSubframe;
96
97 typedef struct FlacFrame {
98 FlacSubframe subframes[FLAC_MAX_CHANNELS];
99 int blocksize;
100 int bs_code[2];
101 uint8_t crc8;
102 int ch_mode;
103 int verbatim_only;
104 } FlacFrame;
105
106 typedef struct FlacEncodeContext {
107 AVClass *class;
108 PutBitContext pb;
109 int channels;
110 int samplerate;
111 int sr_code[2];
112 int bps_code;
113 int max_blocksize;
114 int min_framesize;
115 int max_framesize;
116 int max_encoded_framesize;
117 uint32_t frame_count;
118 uint64_t sample_count;
119 uint8_t md5sum[16];
120 FlacFrame frame;
121 CompressionOptions options;
122 AVCodecContext *avctx;
123 LPCContext lpc_ctx;
124 struct AVMD5 *md5ctx;
125 uint8_t *md5_buffer;
126 unsigned int md5_buffer_size;
127 BswapDSPContext bdsp;
128 FLACDSPContext flac_dsp;
129
130 int flushed;
131 int64_t next_pts;
132 } FlacEncodeContext;
133
134
135 /**
136 * Write streaminfo metadata block to byte array.
137 */
write_streaminfo(FlacEncodeContext *s, uint8_t *header)138 static void write_streaminfo(FlacEncodeContext *s, uint8_t *header)
139 {
140 PutBitContext pb;
141
142 memset(header, 0, FLAC_STREAMINFO_SIZE);
143 init_put_bits(&pb, header, FLAC_STREAMINFO_SIZE);
144
145 /* streaminfo metadata block */
146 put_bits(&pb, 16, s->max_blocksize);
147 put_bits(&pb, 16, s->max_blocksize);
148 put_bits(&pb, 24, s->min_framesize);
149 put_bits(&pb, 24, s->max_framesize);
150 put_bits(&pb, 20, s->samplerate);
151 put_bits(&pb, 3, s->channels-1);
152 put_bits(&pb, 5, s->avctx->bits_per_raw_sample - 1);
153 /* write 36-bit sample count in 2 put_bits() calls */
154 put_bits(&pb, 24, (s->sample_count & 0xFFFFFF000LL) >> 12);
155 put_bits(&pb, 12, s->sample_count & 0x000000FFFLL);
156 flush_put_bits(&pb);
157 memcpy(&header[18], s->md5sum, 16);
158 }
159
160
161 /**
162 * Set blocksize based on samplerate.
163 * Choose the closest predefined blocksize >= BLOCK_TIME_MS milliseconds.
164 */
select_blocksize(int samplerate, int block_time_ms)165 static int select_blocksize(int samplerate, int block_time_ms)
166 {
167 int i;
168 int target;
169 int blocksize;
170
171 av_assert0(samplerate > 0);
172 blocksize = ff_flac_blocksize_table[1];
173 target = (samplerate * block_time_ms) / 1000;
174 for (i = 0; i < 16; i++) {
175 if (target >= ff_flac_blocksize_table[i] &&
176 ff_flac_blocksize_table[i] > blocksize) {
177 blocksize = ff_flac_blocksize_table[i];
178 }
179 }
180 return blocksize;
181 }
182
183
dprint_compression_options(FlacEncodeContext *s)184 static av_cold void dprint_compression_options(FlacEncodeContext *s)
185 {
186 AVCodecContext *avctx = s->avctx;
187 CompressionOptions *opt = &s->options;
188
189 av_log(avctx, AV_LOG_DEBUG, " compression: %d\n", opt->compression_level);
190
191 switch (opt->lpc_type) {
192 case FF_LPC_TYPE_NONE:
193 av_log(avctx, AV_LOG_DEBUG, " lpc type: None\n");
194 break;
195 case FF_LPC_TYPE_FIXED:
196 av_log(avctx, AV_LOG_DEBUG, " lpc type: Fixed pre-defined coefficients\n");
197 break;
198 case FF_LPC_TYPE_LEVINSON:
199 av_log(avctx, AV_LOG_DEBUG, " lpc type: Levinson-Durbin recursion with Welch window\n");
200 break;
201 case FF_LPC_TYPE_CHOLESKY:
202 av_log(avctx, AV_LOG_DEBUG, " lpc type: Cholesky factorization, %d pass%s\n",
203 opt->lpc_passes, opt->lpc_passes == 1 ? "" : "es");
204 break;
205 }
206
207 av_log(avctx, AV_LOG_DEBUG, " prediction order: %d, %d\n",
208 opt->min_prediction_order, opt->max_prediction_order);
209
210 switch (opt->prediction_order_method) {
211 case ORDER_METHOD_EST:
212 av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "estimate");
213 break;
214 case ORDER_METHOD_2LEVEL:
215 av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "2-level");
216 break;
217 case ORDER_METHOD_4LEVEL:
218 av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "4-level");
219 break;
220 case ORDER_METHOD_8LEVEL:
221 av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "8-level");
222 break;
223 case ORDER_METHOD_SEARCH:
224 av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "full search");
225 break;
226 case ORDER_METHOD_LOG:
227 av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "log search");
228 break;
229 }
230
231
232 av_log(avctx, AV_LOG_DEBUG, " partition order: %d, %d\n",
233 opt->min_partition_order, opt->max_partition_order);
234
235 av_log(avctx, AV_LOG_DEBUG, " block size: %d\n", avctx->frame_size);
236
237 av_log(avctx, AV_LOG_DEBUG, " lpc precision: %d\n",
238 opt->lpc_coeff_precision);
239 }
240
241
flac_encode_init(AVCodecContext *avctx)242 static av_cold int flac_encode_init(AVCodecContext *avctx)
243 {
244 int freq = avctx->sample_rate;
245 int channels = avctx->ch_layout.nb_channels;
246 FlacEncodeContext *s = avctx->priv_data;
247 int i, level, ret;
248 uint8_t *streaminfo;
249
250 s->avctx = avctx;
251
252 switch (avctx->sample_fmt) {
253 case AV_SAMPLE_FMT_S16:
254 avctx->bits_per_raw_sample = 16;
255 s->bps_code = 4;
256 break;
257 case AV_SAMPLE_FMT_S32:
258 if (avctx->bits_per_raw_sample != 24)
259 av_log(avctx, AV_LOG_WARNING, "encoding as 24 bits-per-sample\n");
260 avctx->bits_per_raw_sample = 24;
261 s->bps_code = 6;
262 break;
263 }
264
265 if (channels < 1 || channels > FLAC_MAX_CHANNELS) {
266 av_log(avctx, AV_LOG_ERROR, "%d channels not supported (max %d)\n",
267 channels, FLAC_MAX_CHANNELS);
268 return AVERROR(EINVAL);
269 }
270 s->channels = channels;
271
272 /* find samplerate in table */
273 if (freq < 1)
274 return AVERROR(EINVAL);
275 for (i = 4; i < 12; i++) {
276 if (freq == ff_flac_sample_rate_table[i]) {
277 s->samplerate = ff_flac_sample_rate_table[i];
278 s->sr_code[0] = i;
279 s->sr_code[1] = 0;
280 break;
281 }
282 }
283 /* if not in table, samplerate is non-standard */
284 if (i == 12) {
285 if (freq % 1000 == 0 && freq < 255000) {
286 s->sr_code[0] = 12;
287 s->sr_code[1] = freq / 1000;
288 } else if (freq % 10 == 0 && freq < 655350) {
289 s->sr_code[0] = 14;
290 s->sr_code[1] = freq / 10;
291 } else if (freq < 65535) {
292 s->sr_code[0] = 13;
293 s->sr_code[1] = freq;
294 } else {
295 av_log(avctx, AV_LOG_ERROR, "%d Hz not supported\n", freq);
296 return AVERROR(EINVAL);
297 }
298 s->samplerate = freq;
299 }
300
301 /* set compression option defaults based on avctx->compression_level */
302 if (avctx->compression_level < 0)
303 s->options.compression_level = 5;
304 else
305 s->options.compression_level = avctx->compression_level;
306
307 level = s->options.compression_level;
308 if (level > 12) {
309 av_log(avctx, AV_LOG_ERROR, "invalid compression level: %d\n",
310 s->options.compression_level);
311 return AVERROR(EINVAL);
312 }
313
314 s->options.block_time_ms = ((int[]){ 27, 27, 27,105,105,105,105,105,105,105,105,105,105})[level];
315
316 if (s->options.lpc_type == FF_LPC_TYPE_DEFAULT)
317 s->options.lpc_type = ((int[]){ FF_LPC_TYPE_FIXED, FF_LPC_TYPE_FIXED, FF_LPC_TYPE_FIXED,
318 FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON,
319 FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON,
320 FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON,
321 FF_LPC_TYPE_LEVINSON})[level];
322
323 if (s->options.min_prediction_order < 0)
324 s->options.min_prediction_order = ((int[]){ 2, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1})[level];
325 if (s->options.max_prediction_order < 0)
326 s->options.max_prediction_order = ((int[]){ 3, 4, 4, 6, 8, 8, 8, 8, 12, 12, 12, 32, 32})[level];
327
328 if (s->options.prediction_order_method < 0)
329 s->options.prediction_order_method = ((int[]){ ORDER_METHOD_EST, ORDER_METHOD_EST, ORDER_METHOD_EST,
330 ORDER_METHOD_EST, ORDER_METHOD_EST, ORDER_METHOD_EST,
331 ORDER_METHOD_4LEVEL, ORDER_METHOD_LOG, ORDER_METHOD_4LEVEL,
332 ORDER_METHOD_LOG, ORDER_METHOD_SEARCH, ORDER_METHOD_LOG,
333 ORDER_METHOD_SEARCH})[level];
334
335 if (s->options.min_partition_order > s->options.max_partition_order) {
336 av_log(avctx, AV_LOG_ERROR, "invalid partition orders: min=%d max=%d\n",
337 s->options.min_partition_order, s->options.max_partition_order);
338 return AVERROR(EINVAL);
339 }
340 if (s->options.min_partition_order < 0)
341 s->options.min_partition_order = ((int[]){ 2, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0})[level];
342 if (s->options.max_partition_order < 0)
343 s->options.max_partition_order = ((int[]){ 2, 2, 3, 3, 3, 8, 8, 8, 8, 8, 8, 8, 8})[level];
344
345 if (s->options.lpc_type == FF_LPC_TYPE_NONE) {
346 s->options.min_prediction_order = 0;
347 s->options.max_prediction_order = 0;
348 } else if (s->options.lpc_type == FF_LPC_TYPE_FIXED) {
349 if (s->options.min_prediction_order > MAX_FIXED_ORDER) {
350 av_log(avctx, AV_LOG_WARNING,
351 "invalid min prediction order %d, clamped to %d\n",
352 s->options.min_prediction_order, MAX_FIXED_ORDER);
353 s->options.min_prediction_order = MAX_FIXED_ORDER;
354 }
355 if (s->options.max_prediction_order > MAX_FIXED_ORDER) {
356 av_log(avctx, AV_LOG_WARNING,
357 "invalid max prediction order %d, clamped to %d\n",
358 s->options.max_prediction_order, MAX_FIXED_ORDER);
359 s->options.max_prediction_order = MAX_FIXED_ORDER;
360 }
361 }
362
363 if (s->options.max_prediction_order < s->options.min_prediction_order) {
364 av_log(avctx, AV_LOG_ERROR, "invalid prediction orders: min=%d max=%d\n",
365 s->options.min_prediction_order, s->options.max_prediction_order);
366 return AVERROR(EINVAL);
367 }
368
369 if (avctx->frame_size > 0) {
370 if (avctx->frame_size < FLAC_MIN_BLOCKSIZE ||
371 avctx->frame_size > FLAC_MAX_BLOCKSIZE) {
372 av_log(avctx, AV_LOG_ERROR, "invalid block size: %d\n",
373 avctx->frame_size);
374 return AVERROR(EINVAL);
375 }
376 } else {
377 s->avctx->frame_size = select_blocksize(s->samplerate, s->options.block_time_ms);
378 }
379 s->max_blocksize = s->avctx->frame_size;
380
381 /* set maximum encoded frame size in verbatim mode */
382 s->max_framesize = ff_flac_get_max_frame_size(s->avctx->frame_size,
383 s->channels,
384 s->avctx->bits_per_raw_sample);
385
386 /* initialize MD5 context */
387 s->md5ctx = av_md5_alloc();
388 if (!s->md5ctx)
389 return AVERROR(ENOMEM);
390 av_md5_init(s->md5ctx);
391
392 streaminfo = av_malloc(FLAC_STREAMINFO_SIZE);
393 if (!streaminfo)
394 return AVERROR(ENOMEM);
395 write_streaminfo(s, streaminfo);
396 avctx->extradata = streaminfo;
397 avctx->extradata_size = FLAC_STREAMINFO_SIZE;
398
399 s->frame_count = 0;
400 s->min_framesize = s->max_framesize;
401
402 if ((channels == 3 &&
403 av_channel_layout_compare(&avctx->ch_layout, &(AVChannelLayout)AV_CHANNEL_LAYOUT_SURROUND)) ||
404 (channels == 4 &&
405 av_channel_layout_compare(&avctx->ch_layout, &(AVChannelLayout)AV_CHANNEL_LAYOUT_2_2) &&
406 av_channel_layout_compare(&avctx->ch_layout, &(AVChannelLayout)AV_CHANNEL_LAYOUT_QUAD)) ||
407 (channels == 5 &&
408 av_channel_layout_compare(&avctx->ch_layout, &(AVChannelLayout)AV_CHANNEL_LAYOUT_5POINT0) &&
409 av_channel_layout_compare(&avctx->ch_layout, &(AVChannelLayout)AV_CHANNEL_LAYOUT_5POINT0_BACK)) ||
410 (channels == 6 &&
411 av_channel_layout_compare(&avctx->ch_layout, &(AVChannelLayout)AV_CHANNEL_LAYOUT_5POINT1) &&
412 av_channel_layout_compare(&avctx->ch_layout, &(AVChannelLayout)AV_CHANNEL_LAYOUT_5POINT1_BACK))) {
413 if (avctx->ch_layout.order != AV_CHANNEL_ORDER_UNSPEC) {
414 av_log(avctx, AV_LOG_ERROR, "Channel layout not supported by Flac, "
415 "output stream will have incorrect "
416 "channel layout.\n");
417 } else {
418 av_log(avctx, AV_LOG_WARNING, "No channel layout specified. The encoder "
419 "will use Flac channel layout for "
420 "%d channels.\n", channels);
421 }
422 }
423
424 ret = ff_lpc_init(&s->lpc_ctx, avctx->frame_size,
425 s->options.max_prediction_order, FF_LPC_TYPE_LEVINSON);
426
427 ff_bswapdsp_init(&s->bdsp);
428 ff_flacdsp_init(&s->flac_dsp, avctx->sample_fmt, channels,
429 avctx->bits_per_raw_sample);
430
431 dprint_compression_options(s);
432
433 return ret;
434 }
435
436
init_frame(FlacEncodeContext *s, int nb_samples)437 static void init_frame(FlacEncodeContext *s, int nb_samples)
438 {
439 int i, ch;
440 FlacFrame *frame;
441
442 frame = &s->frame;
443
444 for (i = 0; i < 16; i++) {
445 if (nb_samples == ff_flac_blocksize_table[i]) {
446 frame->blocksize = ff_flac_blocksize_table[i];
447 frame->bs_code[0] = i;
448 frame->bs_code[1] = 0;
449 break;
450 }
451 }
452 if (i == 16) {
453 frame->blocksize = nb_samples;
454 if (frame->blocksize <= 256) {
455 frame->bs_code[0] = 6;
456 frame->bs_code[1] = frame->blocksize-1;
457 } else {
458 frame->bs_code[0] = 7;
459 frame->bs_code[1] = frame->blocksize-1;
460 }
461 }
462
463 for (ch = 0; ch < s->channels; ch++) {
464 FlacSubframe *sub = &frame->subframes[ch];
465
466 sub->wasted = 0;
467 sub->obits = s->avctx->bits_per_raw_sample;
468
469 if (sub->obits > 16)
470 sub->rc.coding_mode = CODING_MODE_RICE2;
471 else
472 sub->rc.coding_mode = CODING_MODE_RICE;
473 }
474
475 frame->verbatim_only = 0;
476 }
477
478
479 /**
480 * Copy channel-interleaved input samples into separate subframes.
481 */
copy_samples(FlacEncodeContext *s, const void *samples)482 static void copy_samples(FlacEncodeContext *s, const void *samples)
483 {
484 int i, j, ch;
485 FlacFrame *frame;
486 int shift = av_get_bytes_per_sample(s->avctx->sample_fmt) * 8 -
487 s->avctx->bits_per_raw_sample;
488
489 #define COPY_SAMPLES(bits) do { \
490 const int ## bits ## _t *samples0 = samples; \
491 frame = &s->frame; \
492 for (i = 0, j = 0; i < frame->blocksize; i++) \
493 for (ch = 0; ch < s->channels; ch++, j++) \
494 frame->subframes[ch].samples[i] = samples0[j] >> shift; \
495 } while (0)
496
497 if (s->avctx->sample_fmt == AV_SAMPLE_FMT_S16)
498 COPY_SAMPLES(16);
499 else
500 COPY_SAMPLES(32);
501 }
502
503
rice_count_exact(const int32_t *res, int n, int k)504 static uint64_t rice_count_exact(const int32_t *res, int n, int k)
505 {
506 int i;
507 uint64_t count = 0;
508
509 for (i = 0; i < n; i++) {
510 int32_t v = -2 * res[i] - 1;
511 v ^= v >> 31;
512 count += (v >> k) + 1 + k;
513 }
514 return count;
515 }
516
517
subframe_count_exact(FlacEncodeContext *s, FlacSubframe *sub, int pred_order)518 static uint64_t subframe_count_exact(FlacEncodeContext *s, FlacSubframe *sub,
519 int pred_order)
520 {
521 int p, porder, psize;
522 int i, part_end;
523 uint64_t count = 0;
524
525 /* subframe header */
526 count += 8;
527
528 if (sub->wasted)
529 count += sub->wasted;
530
531 /* subframe */
532 if (sub->type == FLAC_SUBFRAME_CONSTANT) {
533 count += sub->obits;
534 } else if (sub->type == FLAC_SUBFRAME_VERBATIM) {
535 count += s->frame.blocksize * sub->obits;
536 } else {
537 /* warm-up samples */
538 count += pred_order * sub->obits;
539
540 /* LPC coefficients */
541 if (sub->type == FLAC_SUBFRAME_LPC)
542 count += 4 + 5 + pred_order * s->options.lpc_coeff_precision;
543
544 /* rice-encoded block */
545 count += 2;
546
547 /* partition order */
548 porder = sub->rc.porder;
549 psize = s->frame.blocksize >> porder;
550 count += 4;
551
552 /* residual */
553 i = pred_order;
554 part_end = psize;
555 for (p = 0; p < 1 << porder; p++) {
556 int k = sub->rc.params[p];
557 count += sub->rc.coding_mode;
558 count += rice_count_exact(&sub->residual[i], part_end - i, k);
559 i = part_end;
560 part_end = FFMIN(s->frame.blocksize, part_end + psize);
561 }
562 }
563
564 return count;
565 }
566
567
568 #define rice_encode_count(sum, n, k) (((n)*((k)+1))+((sum-(n>>1))>>(k)))
569
570 /**
571 * Solve for d/dk(rice_encode_count) = n-((sum-(n>>1))>>(k+1)) = 0.
572 */
find_optimal_param(uint64_t sum, int n, int max_param)573 static int find_optimal_param(uint64_t sum, int n, int max_param)
574 {
575 int k;
576 uint64_t sum2;
577
578 if (sum <= n >> 1)
579 return 0;
580 sum2 = sum - (n >> 1);
581 k = av_log2(av_clipl_int32(sum2 / n));
582 return FFMIN(k, max_param);
583 }
584
find_optimal_param_exact(uint64_t sums[32][MAX_PARTITIONS], int i, int max_param)585 static int find_optimal_param_exact(uint64_t sums[32][MAX_PARTITIONS], int i, int max_param)
586 {
587 int bestk = 0;
588 int64_t bestbits = INT64_MAX;
589 int k;
590
591 for (k = 0; k <= max_param; k++) {
592 int64_t bits = sums[k][i];
593 if (bits < bestbits) {
594 bestbits = bits;
595 bestk = k;
596 }
597 }
598
599 return bestk;
600 }
601
calc_optimal_rice_params(RiceContext *rc, int porder, uint64_t sums[32][MAX_PARTITIONS], int n, int pred_order, int max_param, int exact)602 static uint64_t calc_optimal_rice_params(RiceContext *rc, int porder,
603 uint64_t sums[32][MAX_PARTITIONS],
604 int n, int pred_order, int max_param, int exact)
605 {
606 int i;
607 int k, cnt, part;
608 uint64_t all_bits;
609
610 part = (1 << porder);
611 all_bits = 4 * part;
612
613 cnt = (n >> porder) - pred_order;
614 for (i = 0; i < part; i++) {
615 if (exact) {
616 k = find_optimal_param_exact(sums, i, max_param);
617 all_bits += sums[k][i];
618 } else {
619 k = find_optimal_param(sums[0][i], cnt, max_param);
620 all_bits += rice_encode_count(sums[0][i], cnt, k);
621 }
622 rc->params[i] = k;
623 cnt = n >> porder;
624 }
625
626 rc->porder = porder;
627
628 return all_bits;
629 }
630
631
calc_sum_top(int pmax, int kmax, const uint32_t *data, int n, int pred_order, uint64_t sums[32][MAX_PARTITIONS])632 static void calc_sum_top(int pmax, int kmax, const uint32_t *data, int n, int pred_order,
633 uint64_t sums[32][MAX_PARTITIONS])
634 {
635 int i, k;
636 int parts;
637 const uint32_t *res, *res_end;
638
639 /* sums for highest level */
640 parts = (1 << pmax);
641
642 for (k = 0; k <= kmax; k++) {
643 res = &data[pred_order];
644 res_end = &data[n >> pmax];
645 for (i = 0; i < parts; i++) {
646 if (kmax) {
647 uint64_t sum = (1LL + k) * (res_end - res);
648 while (res < res_end)
649 sum += *(res++) >> k;
650 sums[k][i] = sum;
651 } else {
652 uint64_t sum = 0;
653 while (res < res_end)
654 sum += *(res++);
655 sums[k][i] = sum;
656 }
657 res_end += n >> pmax;
658 }
659 }
660 }
661
calc_sum_next(int level, uint64_t sums[32][MAX_PARTITIONS], int kmax)662 static void calc_sum_next(int level, uint64_t sums[32][MAX_PARTITIONS], int kmax)
663 {
664 int i, k;
665 int parts = (1 << level);
666 for (i = 0; i < parts; i++) {
667 for (k=0; k<=kmax; k++)
668 sums[k][i] = sums[k][2*i] + sums[k][2*i+1];
669 }
670 }
671
calc_rice_params(RiceContext *rc, uint32_t udata[FLAC_MAX_BLOCKSIZE], uint64_t sums[32][MAX_PARTITIONS], int pmin, int pmax, const int32_t *data, int n, int pred_order, int exact)672 static uint64_t calc_rice_params(RiceContext *rc,
673 uint32_t udata[FLAC_MAX_BLOCKSIZE],
674 uint64_t sums[32][MAX_PARTITIONS],
675 int pmin, int pmax,
676 const int32_t *data, int n, int pred_order, int exact)
677 {
678 int i;
679 uint64_t bits[MAX_PARTITION_ORDER+1];
680 int opt_porder;
681 RiceContext tmp_rc;
682 int kmax = (1 << rc->coding_mode) - 2;
683
684 av_assert1(pmin >= 0 && pmin <= MAX_PARTITION_ORDER);
685 av_assert1(pmax >= 0 && pmax <= MAX_PARTITION_ORDER);
686 av_assert1(pmin <= pmax);
687
688 tmp_rc.coding_mode = rc->coding_mode;
689
690 for (i = 0; i < n; i++)
691 udata[i] = (2 * data[i]) ^ (data[i] >> 31);
692
693 calc_sum_top(pmax, exact ? kmax : 0, udata, n, pred_order, sums);
694
695 opt_porder = pmin;
696 bits[pmin] = UINT32_MAX;
697 for (i = pmax; ; ) {
698 bits[i] = calc_optimal_rice_params(&tmp_rc, i, sums, n, pred_order, kmax, exact);
699 if (bits[i] < bits[opt_porder] || pmax == pmin) {
700 opt_porder = i;
701 *rc = tmp_rc;
702 }
703 if (i == pmin)
704 break;
705 calc_sum_next(--i, sums, exact ? kmax : 0);
706 }
707
708 return bits[opt_porder];
709 }
710
711
get_max_p_order(int max_porder, int n, int order)712 static int get_max_p_order(int max_porder, int n, int order)
713 {
714 int porder = FFMIN(max_porder, av_log2(n^(n-1)));
715 if (order > 0)
716 porder = FFMIN(porder, av_log2(n/order));
717 return porder;
718 }
719
720
find_subframe_rice_params(FlacEncodeContext *s, FlacSubframe *sub, int pred_order)721 static uint64_t find_subframe_rice_params(FlacEncodeContext *s,
722 FlacSubframe *sub, int pred_order)
723 {
724 int pmin = get_max_p_order(s->options.min_partition_order,
725 s->frame.blocksize, pred_order);
726 int pmax = get_max_p_order(s->options.max_partition_order,
727 s->frame.blocksize, pred_order);
728
729 uint64_t bits = 8 + pred_order * sub->obits + 2 + sub->rc.coding_mode;
730 if (sub->type == FLAC_SUBFRAME_LPC)
731 bits += 4 + 5 + pred_order * s->options.lpc_coeff_precision;
732 bits += calc_rice_params(&sub->rc, sub->rc_udata, sub->rc_sums, pmin, pmax, sub->residual,
733 s->frame.blocksize, pred_order, s->options.exact_rice_parameters);
734 return bits;
735 }
736
737
encode_residual_fixed(int32_t *res, const int32_t *smp, int n, int order)738 static void encode_residual_fixed(int32_t *res, const int32_t *smp, int n,
739 int order)
740 {
741 int i;
742
743 for (i = 0; i < order; i++)
744 res[i] = smp[i];
745
746 if (order == 0) {
747 for (i = order; i < n; i++)
748 res[i] = smp[i];
749 } else if (order == 1) {
750 for (i = order; i < n; i++)
751 res[i] = smp[i] - smp[i-1];
752 } else if (order == 2) {
753 int a = smp[order-1] - smp[order-2];
754 for (i = order; i < n; i += 2) {
755 int b = smp[i ] - smp[i-1];
756 res[i] = b - a;
757 a = smp[i+1] - smp[i ];
758 res[i+1] = a - b;
759 }
760 } else if (order == 3) {
761 int a = smp[order-1] - smp[order-2];
762 int c = smp[order-1] - 2*smp[order-2] + smp[order-3];
763 for (i = order; i < n; i += 2) {
764 int b = smp[i ] - smp[i-1];
765 int d = b - a;
766 res[i] = d - c;
767 a = smp[i+1] - smp[i ];
768 c = a - b;
769 res[i+1] = c - d;
770 }
771 } else {
772 int a = smp[order-1] - smp[order-2];
773 int c = smp[order-1] - 2*smp[order-2] + smp[order-3];
774 int e = smp[order-1] - 3*smp[order-2] + 3*smp[order-3] - smp[order-4];
775 for (i = order; i < n; i += 2) {
776 int b = smp[i ] - smp[i-1];
777 int d = b - a;
778 int f = d - c;
779 res[i ] = f - e;
780 a = smp[i+1] - smp[i ];
781 c = a - b;
782 e = c - d;
783 res[i+1] = e - f;
784 }
785 }
786 }
787
788
encode_residual_ch(FlacEncodeContext *s, int ch)789 static int encode_residual_ch(FlacEncodeContext *s, int ch)
790 {
791 int i, n;
792 int min_order, max_order, opt_order, omethod;
793 FlacFrame *frame;
794 FlacSubframe *sub;
795 int32_t coefs[MAX_LPC_ORDER][MAX_LPC_ORDER];
796 int shift[MAX_LPC_ORDER];
797 int32_t *res, *smp;
798
799 frame = &s->frame;
800 sub = &frame->subframes[ch];
801 res = sub->residual;
802 smp = sub->samples;
803 n = frame->blocksize;
804
805 /* CONSTANT */
806 for (i = 1; i < n; i++)
807 if(smp[i] != smp[0])
808 break;
809 if (i == n) {
810 sub->type = sub->type_code = FLAC_SUBFRAME_CONSTANT;
811 res[0] = smp[0];
812 return subframe_count_exact(s, sub, 0);
813 }
814
815 /* VERBATIM */
816 if (frame->verbatim_only || n < 5) {
817 sub->type = sub->type_code = FLAC_SUBFRAME_VERBATIM;
818 memcpy(res, smp, n * sizeof(int32_t));
819 return subframe_count_exact(s, sub, 0);
820 }
821
822 min_order = s->options.min_prediction_order;
823 max_order = s->options.max_prediction_order;
824 omethod = s->options.prediction_order_method;
825
826 /* FIXED */
827 sub->type = FLAC_SUBFRAME_FIXED;
828 if (s->options.lpc_type == FF_LPC_TYPE_NONE ||
829 s->options.lpc_type == FF_LPC_TYPE_FIXED || n <= max_order) {
830 uint64_t bits[MAX_FIXED_ORDER+1];
831 if (max_order > MAX_FIXED_ORDER)
832 max_order = MAX_FIXED_ORDER;
833 opt_order = 0;
834 bits[0] = UINT32_MAX;
835 for (i = min_order; i <= max_order; i++) {
836 encode_residual_fixed(res, smp, n, i);
837 bits[i] = find_subframe_rice_params(s, sub, i);
838 if (bits[i] < bits[opt_order])
839 opt_order = i;
840 }
841 sub->order = opt_order;
842 sub->type_code = sub->type | sub->order;
843 if (sub->order != max_order) {
844 encode_residual_fixed(res, smp, n, sub->order);
845 find_subframe_rice_params(s, sub, sub->order);
846 }
847 return subframe_count_exact(s, sub, sub->order);
848 }
849
850 /* LPC */
851 sub->type = FLAC_SUBFRAME_LPC;
852 opt_order = ff_lpc_calc_coefs(&s->lpc_ctx, smp, n, min_order, max_order,
853 s->options.lpc_coeff_precision, coefs, shift, s->options.lpc_type,
854 s->options.lpc_passes, omethod,
855 MIN_LPC_SHIFT, MAX_LPC_SHIFT, 0);
856
857 if (omethod == ORDER_METHOD_2LEVEL ||
858 omethod == ORDER_METHOD_4LEVEL ||
859 omethod == ORDER_METHOD_8LEVEL) {
860 int levels = 1 << omethod;
861 uint64_t bits[1 << ORDER_METHOD_8LEVEL];
862 int order = -1;
863 int opt_index = levels-1;
864 opt_order = max_order-1;
865 bits[opt_index] = UINT32_MAX;
866 for (i = levels-1; i >= 0; i--) {
867 int last_order = order;
868 order = min_order + (((max_order-min_order+1) * (i+1)) / levels)-1;
869 order = av_clip(order, min_order - 1, max_order - 1);
870 if (order == last_order)
871 continue;
872 if (s->bps_code * 4 + s->options.lpc_coeff_precision + av_log2(order) <= 32) {
873 s->flac_dsp.lpc16_encode(res, smp, n, order+1, coefs[order],
874 shift[order]);
875 } else {
876 s->flac_dsp.lpc32_encode(res, smp, n, order+1, coefs[order],
877 shift[order]);
878 }
879 bits[i] = find_subframe_rice_params(s, sub, order+1);
880 if (bits[i] < bits[opt_index]) {
881 opt_index = i;
882 opt_order = order;
883 }
884 }
885 opt_order++;
886 } else if (omethod == ORDER_METHOD_SEARCH) {
887 // brute-force optimal order search
888 uint64_t bits[MAX_LPC_ORDER];
889 opt_order = 0;
890 bits[0] = UINT32_MAX;
891 for (i = min_order-1; i < max_order; i++) {
892 if (s->bps_code * 4 + s->options.lpc_coeff_precision + av_log2(i) <= 32) {
893 s->flac_dsp.lpc16_encode(res, smp, n, i+1, coefs[i], shift[i]);
894 } else {
895 s->flac_dsp.lpc32_encode(res, smp, n, i+1, coefs[i], shift[i]);
896 }
897 bits[i] = find_subframe_rice_params(s, sub, i+1);
898 if (bits[i] < bits[opt_order])
899 opt_order = i;
900 }
901 opt_order++;
902 } else if (omethod == ORDER_METHOD_LOG) {
903 uint64_t bits[MAX_LPC_ORDER];
904 int step;
905
906 opt_order = min_order - 1 + (max_order-min_order)/3;
907 memset(bits, -1, sizeof(bits));
908
909 for (step = 16; step; step >>= 1) {
910 int last = opt_order;
911 for (i = last-step; i <= last+step; i += step) {
912 if (i < min_order-1 || i >= max_order || bits[i] < UINT32_MAX)
913 continue;
914 if (s->bps_code * 4 + s->options.lpc_coeff_precision + av_log2(i) <= 32) {
915 s->flac_dsp.lpc32_encode(res, smp, n, i+1, coefs[i], shift[i]);
916 } else {
917 s->flac_dsp.lpc16_encode(res, smp, n, i+1, coefs[i], shift[i]);
918 }
919 bits[i] = find_subframe_rice_params(s, sub, i+1);
920 if (bits[i] < bits[opt_order])
921 opt_order = i;
922 }
923 }
924 opt_order++;
925 }
926
927 if (s->options.multi_dim_quant) {
928 int allsteps = 1;
929 int i, step, improved;
930 int64_t best_score = INT64_MAX;
931 int32_t qmax;
932
933 qmax = (1 << (s->options.lpc_coeff_precision - 1)) - 1;
934
935 for (i=0; i<opt_order; i++)
936 allsteps *= 3;
937
938 do {
939 improved = 0;
940 for (step = 0; step < allsteps; step++) {
941 int tmp = step;
942 int32_t lpc_try[MAX_LPC_ORDER];
943 int64_t score = 0;
944 int diffsum = 0;
945
946 for (i=0; i<opt_order; i++) {
947 int diff = ((tmp + 1) % 3) - 1;
948 lpc_try[i] = av_clip(coefs[opt_order - 1][i] + diff, -qmax, qmax);
949 tmp /= 3;
950 diffsum += !!diff;
951 }
952 if (diffsum >8)
953 continue;
954
955 if (s->bps_code * 4 + s->options.lpc_coeff_precision + av_log2(opt_order - 1) <= 32) {
956 s->flac_dsp.lpc16_encode(res, smp, n, opt_order, lpc_try, shift[opt_order-1]);
957 } else {
958 s->flac_dsp.lpc32_encode(res, smp, n, opt_order, lpc_try, shift[opt_order-1]);
959 }
960 score = find_subframe_rice_params(s, sub, opt_order);
961 if (score < best_score) {
962 best_score = score;
963 memcpy(coefs[opt_order-1], lpc_try, sizeof(*coefs));
964 improved=1;
965 }
966 }
967 } while(improved);
968 }
969
970 sub->order = opt_order;
971 sub->type_code = sub->type | (sub->order-1);
972 sub->shift = shift[sub->order-1];
973 for (i = 0; i < sub->order; i++)
974 sub->coefs[i] = coefs[sub->order-1][i];
975
976 if (s->bps_code * 4 + s->options.lpc_coeff_precision + av_log2(opt_order) <= 32) {
977 s->flac_dsp.lpc16_encode(res, smp, n, sub->order, sub->coefs, sub->shift);
978 } else {
979 s->flac_dsp.lpc32_encode(res, smp, n, sub->order, sub->coefs, sub->shift);
980 }
981
982 find_subframe_rice_params(s, sub, sub->order);
983
984 return subframe_count_exact(s, sub, sub->order);
985 }
986
987
count_frame_header(FlacEncodeContext *s)988 static int count_frame_header(FlacEncodeContext *s)
989 {
990 uint8_t av_unused tmp;
991 int count;
992
993 /*
994 <14> Sync code
995 <1> Reserved
996 <1> Blocking strategy
997 <4> Block size in inter-channel samples
998 <4> Sample rate
999 <4> Channel assignment
1000 <3> Sample size in bits
1001 <1> Reserved
1002 */
1003 count = 32;
1004
1005 /* coded frame number */
1006 PUT_UTF8(s->frame_count, tmp, count += 8;)
1007
1008 /* explicit block size */
1009 if (s->frame.bs_code[0] == 6)
1010 count += 8;
1011 else if (s->frame.bs_code[0] == 7)
1012 count += 16;
1013
1014 /* explicit sample rate */
1015 count += ((s->sr_code[0] == 12) + (s->sr_code[0] > 12) * 2) * 8;
1016
1017 /* frame header CRC-8 */
1018 count += 8;
1019
1020 return count;
1021 }
1022
1023
encode_frame(FlacEncodeContext *s)1024 static int encode_frame(FlacEncodeContext *s)
1025 {
1026 int ch;
1027 uint64_t count;
1028
1029 count = count_frame_header(s);
1030
1031 for (ch = 0; ch < s->channels; ch++)
1032 count += encode_residual_ch(s, ch);
1033
1034 count += (8 - (count & 7)) & 7; // byte alignment
1035 count += 16; // CRC-16
1036
1037 count >>= 3;
1038 if (count > INT_MAX)
1039 return AVERROR_BUG;
1040 return count;
1041 }
1042
1043
remove_wasted_bits(FlacEncodeContext *s)1044 static void remove_wasted_bits(FlacEncodeContext *s)
1045 {
1046 int ch, i;
1047
1048 for (ch = 0; ch < s->channels; ch++) {
1049 FlacSubframe *sub = &s->frame.subframes[ch];
1050 int32_t v = 0;
1051
1052 for (i = 0; i < s->frame.blocksize; i++) {
1053 v |= sub->samples[i];
1054 if (v & 1)
1055 break;
1056 }
1057
1058 if (v && !(v & 1)) {
1059 v = ff_ctz(v);
1060
1061 for (i = 0; i < s->frame.blocksize; i++)
1062 sub->samples[i] >>= v;
1063
1064 sub->wasted = v;
1065 sub->obits -= v;
1066
1067 /* for 24-bit, check if removing wasted bits makes the range better
1068 suited for using RICE instead of RICE2 for entropy coding */
1069 if (sub->obits <= 17)
1070 sub->rc.coding_mode = CODING_MODE_RICE;
1071 }
1072 }
1073 }
1074
1075
estimate_stereo_mode(const int32_t *left_ch, const int32_t *right_ch, int n, int max_rice_param)1076 static int estimate_stereo_mode(const int32_t *left_ch, const int32_t *right_ch, int n,
1077 int max_rice_param)
1078 {
1079 int i, best;
1080 int32_t lt, rt;
1081 uint64_t sum[4];
1082 uint64_t score[4];
1083 int k;
1084
1085 /* calculate sum of 2nd order residual for each channel */
1086 sum[0] = sum[1] = sum[2] = sum[3] = 0;
1087 for (i = 2; i < n; i++) {
1088 lt = left_ch[i] - 2*left_ch[i-1] + left_ch[i-2];
1089 rt = right_ch[i] - 2*right_ch[i-1] + right_ch[i-2];
1090 sum[2] += FFABS((lt + rt) >> 1);
1091 sum[3] += FFABS(lt - rt);
1092 sum[0] += FFABS(lt);
1093 sum[1] += FFABS(rt);
1094 }
1095 /* estimate bit counts */
1096 for (i = 0; i < 4; i++) {
1097 k = find_optimal_param(2 * sum[i], n, max_rice_param);
1098 sum[i] = rice_encode_count( 2 * sum[i], n, k);
1099 }
1100
1101 /* calculate score for each mode */
1102 score[0] = sum[0] + sum[1];
1103 score[1] = sum[0] + sum[3];
1104 score[2] = sum[1] + sum[3];
1105 score[3] = sum[2] + sum[3];
1106
1107 /* return mode with lowest score */
1108 best = 0;
1109 for (i = 1; i < 4; i++)
1110 if (score[i] < score[best])
1111 best = i;
1112
1113 return best;
1114 }
1115
1116
1117 /**
1118 * Perform stereo channel decorrelation.
1119 */
channel_decorrelation(FlacEncodeContext *s)1120 static void channel_decorrelation(FlacEncodeContext *s)
1121 {
1122 FlacFrame *frame;
1123 int32_t *left, *right;
1124 int i, n;
1125
1126 frame = &s->frame;
1127 n = frame->blocksize;
1128 left = frame->subframes[0].samples;
1129 right = frame->subframes[1].samples;
1130
1131 if (s->channels != 2) {
1132 frame->ch_mode = FLAC_CHMODE_INDEPENDENT;
1133 return;
1134 }
1135
1136 if (s->options.ch_mode < 0) {
1137 int max_rice_param = (1 << frame->subframes[0].rc.coding_mode) - 2;
1138 frame->ch_mode = estimate_stereo_mode(left, right, n, max_rice_param);
1139 } else
1140 frame->ch_mode = s->options.ch_mode;
1141
1142 /* perform decorrelation and adjust bits-per-sample */
1143 if (frame->ch_mode == FLAC_CHMODE_INDEPENDENT)
1144 return;
1145 if (frame->ch_mode == FLAC_CHMODE_MID_SIDE) {
1146 int32_t tmp;
1147 for (i = 0; i < n; i++) {
1148 tmp = left[i];
1149 left[i] = (tmp + right[i]) >> 1;
1150 right[i] = tmp - right[i];
1151 }
1152 frame->subframes[1].obits++;
1153 } else if (frame->ch_mode == FLAC_CHMODE_LEFT_SIDE) {
1154 for (i = 0; i < n; i++)
1155 right[i] = left[i] - right[i];
1156 frame->subframes[1].obits++;
1157 } else {
1158 for (i = 0; i < n; i++)
1159 left[i] -= right[i];
1160 frame->subframes[0].obits++;
1161 }
1162 }
1163
1164
write_utf8(PutBitContext *pb, uint32_t val)1165 static void write_utf8(PutBitContext *pb, uint32_t val)
1166 {
1167 uint8_t tmp;
1168 PUT_UTF8(val, tmp, put_bits(pb, 8, tmp);)
1169 }
1170
1171
write_frame_header(FlacEncodeContext *s)1172 static void write_frame_header(FlacEncodeContext *s)
1173 {
1174 FlacFrame *frame;
1175 int crc;
1176
1177 frame = &s->frame;
1178
1179 put_bits(&s->pb, 16, 0xFFF8);
1180 put_bits(&s->pb, 4, frame->bs_code[0]);
1181 put_bits(&s->pb, 4, s->sr_code[0]);
1182
1183 if (frame->ch_mode == FLAC_CHMODE_INDEPENDENT)
1184 put_bits(&s->pb, 4, s->channels-1);
1185 else
1186 put_bits(&s->pb, 4, frame->ch_mode + FLAC_MAX_CHANNELS - 1);
1187
1188 put_bits(&s->pb, 3, s->bps_code);
1189 put_bits(&s->pb, 1, 0);
1190 write_utf8(&s->pb, s->frame_count);
1191
1192 if (frame->bs_code[0] == 6)
1193 put_bits(&s->pb, 8, frame->bs_code[1]);
1194 else if (frame->bs_code[0] == 7)
1195 put_bits(&s->pb, 16, frame->bs_code[1]);
1196
1197 if (s->sr_code[0] == 12)
1198 put_bits(&s->pb, 8, s->sr_code[1]);
1199 else if (s->sr_code[0] > 12)
1200 put_bits(&s->pb, 16, s->sr_code[1]);
1201
1202 flush_put_bits(&s->pb);
1203 crc = av_crc(av_crc_get_table(AV_CRC_8_ATM), 0, s->pb.buf,
1204 put_bytes_output(&s->pb));
1205 put_bits(&s->pb, 8, crc);
1206 }
1207
1208
write_subframes(FlacEncodeContext *s)1209 static void write_subframes(FlacEncodeContext *s)
1210 {
1211 int ch;
1212
1213 for (ch = 0; ch < s->channels; ch++) {
1214 FlacSubframe *sub = &s->frame.subframes[ch];
1215 int i, p, porder, psize;
1216 int32_t *part_end;
1217 int32_t *res = sub->residual;
1218 int32_t *frame_end = &sub->residual[s->frame.blocksize];
1219
1220 /* subframe header */
1221 put_bits(&s->pb, 1, 0);
1222 put_bits(&s->pb, 6, sub->type_code);
1223 put_bits(&s->pb, 1, !!sub->wasted);
1224 if (sub->wasted)
1225 put_bits(&s->pb, sub->wasted, 1);
1226
1227 /* subframe */
1228 if (sub->type == FLAC_SUBFRAME_CONSTANT) {
1229 put_sbits(&s->pb, sub->obits, res[0]);
1230 } else if (sub->type == FLAC_SUBFRAME_VERBATIM) {
1231 while (res < frame_end)
1232 put_sbits(&s->pb, sub->obits, *res++);
1233 } else {
1234 /* warm-up samples */
1235 for (i = 0; i < sub->order; i++)
1236 put_sbits(&s->pb, sub->obits, *res++);
1237
1238 /* LPC coefficients */
1239 if (sub->type == FLAC_SUBFRAME_LPC) {
1240 int cbits = s->options.lpc_coeff_precision;
1241 put_bits( &s->pb, 4, cbits-1);
1242 put_sbits(&s->pb, 5, sub->shift);
1243 for (i = 0; i < sub->order; i++)
1244 put_sbits(&s->pb, cbits, sub->coefs[i]);
1245 }
1246
1247 /* rice-encoded block */
1248 put_bits(&s->pb, 2, sub->rc.coding_mode - 4);
1249
1250 /* partition order */
1251 porder = sub->rc.porder;
1252 psize = s->frame.blocksize >> porder;
1253 put_bits(&s->pb, 4, porder);
1254
1255 /* residual */
1256 part_end = &sub->residual[psize];
1257 for (p = 0; p < 1 << porder; p++) {
1258 int k = sub->rc.params[p];
1259 put_bits(&s->pb, sub->rc.coding_mode, k);
1260 while (res < part_end)
1261 set_sr_golomb_flac(&s->pb, *res++, k, INT32_MAX, 0);
1262 part_end = FFMIN(frame_end, part_end + psize);
1263 }
1264 }
1265 }
1266 }
1267
1268
write_frame_footer(FlacEncodeContext *s)1269 static void write_frame_footer(FlacEncodeContext *s)
1270 {
1271 int crc;
1272 flush_put_bits(&s->pb);
1273 crc = av_bswap16(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, s->pb.buf,
1274 put_bytes_output(&s->pb)));
1275 put_bits(&s->pb, 16, crc);
1276 flush_put_bits(&s->pb);
1277 }
1278
1279
write_frame(FlacEncodeContext *s, AVPacket *avpkt)1280 static int write_frame(FlacEncodeContext *s, AVPacket *avpkt)
1281 {
1282 init_put_bits(&s->pb, avpkt->data, avpkt->size);
1283 write_frame_header(s);
1284 write_subframes(s);
1285 write_frame_footer(s);
1286 return put_bytes_output(&s->pb);
1287 }
1288
1289
update_md5_sum(FlacEncodeContext *s, const void *samples)1290 static int update_md5_sum(FlacEncodeContext *s, const void *samples)
1291 {
1292 const uint8_t *buf;
1293 int buf_size = s->frame.blocksize * s->channels *
1294 ((s->avctx->bits_per_raw_sample + 7) / 8);
1295
1296 if (s->avctx->bits_per_raw_sample > 16 || HAVE_BIGENDIAN) {
1297 av_fast_malloc(&s->md5_buffer, &s->md5_buffer_size, buf_size);
1298 if (!s->md5_buffer)
1299 return AVERROR(ENOMEM);
1300 }
1301
1302 if (s->avctx->bits_per_raw_sample <= 16) {
1303 buf = (const uint8_t *)samples;
1304 #if HAVE_BIGENDIAN
1305 s->bdsp.bswap16_buf((uint16_t *) s->md5_buffer,
1306 (const uint16_t *) samples, buf_size / 2);
1307 buf = s->md5_buffer;
1308 #endif
1309 } else {
1310 int i;
1311 const int32_t *samples0 = samples;
1312 uint8_t *tmp = s->md5_buffer;
1313
1314 for (i = 0; i < s->frame.blocksize * s->channels; i++) {
1315 int32_t v = samples0[i] >> 8;
1316 AV_WL24(tmp + 3*i, v);
1317 }
1318 buf = s->md5_buffer;
1319 }
1320 av_md5_update(s->md5ctx, buf, buf_size);
1321
1322 return 0;
1323 }
1324
1325
flac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt, const AVFrame *frame, int *got_packet_ptr)1326 static int flac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
1327 const AVFrame *frame, int *got_packet_ptr)
1328 {
1329 FlacEncodeContext *s;
1330 int frame_bytes, out_bytes, ret;
1331
1332 s = avctx->priv_data;
1333
1334 /* when the last block is reached, update the header in extradata */
1335 if (!frame) {
1336 s->max_framesize = s->max_encoded_framesize;
1337 av_md5_final(s->md5ctx, s->md5sum);
1338 write_streaminfo(s, avctx->extradata);
1339
1340 if (!s->flushed) {
1341 uint8_t *side_data = av_packet_new_side_data(avpkt, AV_PKT_DATA_NEW_EXTRADATA,
1342 avctx->extradata_size);
1343 if (!side_data)
1344 return AVERROR(ENOMEM);
1345 memcpy(side_data, avctx->extradata, avctx->extradata_size);
1346
1347 avpkt->pts = s->next_pts;
1348
1349 *got_packet_ptr = 1;
1350 s->flushed = 1;
1351 }
1352
1353 return 0;
1354 }
1355
1356 /* change max_framesize for small final frame */
1357 if (frame->nb_samples < s->frame.blocksize) {
1358 s->max_framesize = ff_flac_get_max_frame_size(frame->nb_samples,
1359 s->channels,
1360 avctx->bits_per_raw_sample);
1361 }
1362
1363 init_frame(s, frame->nb_samples);
1364
1365 copy_samples(s, frame->data[0]);
1366
1367 channel_decorrelation(s);
1368
1369 remove_wasted_bits(s);
1370
1371 frame_bytes = encode_frame(s);
1372
1373 /* Fall back on verbatim mode if the compressed frame is larger than it
1374 would be if encoded uncompressed. */
1375 if (frame_bytes < 0 || frame_bytes > s->max_framesize) {
1376 s->frame.verbatim_only = 1;
1377 frame_bytes = encode_frame(s);
1378 if (frame_bytes < 0) {
1379 av_log(avctx, AV_LOG_ERROR, "Bad frame count\n");
1380 return frame_bytes;
1381 }
1382 }
1383
1384 if ((ret = ff_get_encode_buffer(avctx, avpkt, frame_bytes, 0)) < 0)
1385 return ret;
1386
1387 out_bytes = write_frame(s, avpkt);
1388
1389 s->frame_count++;
1390 s->sample_count += frame->nb_samples;
1391 if ((ret = update_md5_sum(s, frame->data[0])) < 0) {
1392 av_log(avctx, AV_LOG_ERROR, "Error updating MD5 checksum\n");
1393 return ret;
1394 }
1395 if (out_bytes > s->max_encoded_framesize)
1396 s->max_encoded_framesize = out_bytes;
1397 if (out_bytes < s->min_framesize)
1398 s->min_framesize = out_bytes;
1399
1400 avpkt->pts = frame->pts;
1401 avpkt->duration = ff_samples_to_time_base(avctx, frame->nb_samples);
1402
1403 s->next_pts = avpkt->pts + avpkt->duration;
1404
1405 av_shrink_packet(avpkt, out_bytes);
1406
1407 *got_packet_ptr = 1;
1408 return 0;
1409 }
1410
1411
flac_encode_close(AVCodecContext *avctx)1412 static av_cold int flac_encode_close(AVCodecContext *avctx)
1413 {
1414 FlacEncodeContext *s = avctx->priv_data;
1415
1416 av_freep(&s->md5ctx);
1417 av_freep(&s->md5_buffer);
1418 ff_lpc_end(&s->lpc_ctx);
1419 return 0;
1420 }
1421
1422 #define FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
1423 static const AVOption options[] = {
1424 { "lpc_coeff_precision", "LPC coefficient precision", offsetof(FlacEncodeContext, options.lpc_coeff_precision), AV_OPT_TYPE_INT, {.i64 = 15 }, 0, MAX_LPC_PRECISION, FLAGS },
1425 { "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" },
1426 { "none", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = FF_LPC_TYPE_NONE }, INT_MIN, INT_MAX, FLAGS, "lpc_type" },
1427 { "fixed", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = FF_LPC_TYPE_FIXED }, INT_MIN, INT_MAX, FLAGS, "lpc_type" },
1428 { "levinson", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = FF_LPC_TYPE_LEVINSON }, INT_MIN, INT_MAX, FLAGS, "lpc_type" },
1429 { "cholesky", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = FF_LPC_TYPE_CHOLESKY }, INT_MIN, INT_MAX, FLAGS, "lpc_type" },
1430 { "lpc_passes", "Number of passes to use for Cholesky factorization during LPC analysis", offsetof(FlacEncodeContext, options.lpc_passes), AV_OPT_TYPE_INT, {.i64 = 2 }, 1, INT_MAX, FLAGS },
1431 { "min_partition_order", NULL, offsetof(FlacEncodeContext, options.min_partition_order), AV_OPT_TYPE_INT, {.i64 = -1 }, -1, MAX_PARTITION_ORDER, FLAGS },
1432 { "max_partition_order", NULL, offsetof(FlacEncodeContext, options.max_partition_order), AV_OPT_TYPE_INT, {.i64 = -1 }, -1, MAX_PARTITION_ORDER, FLAGS },
1433 { "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" },
1434 { "estimation", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_EST }, INT_MIN, INT_MAX, FLAGS, "predm" },
1435 { "2level", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_2LEVEL }, INT_MIN, INT_MAX, FLAGS, "predm" },
1436 { "4level", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_4LEVEL }, INT_MIN, INT_MAX, FLAGS, "predm" },
1437 { "8level", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_8LEVEL }, INT_MIN, INT_MAX, FLAGS, "predm" },
1438 { "search", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_SEARCH }, INT_MIN, INT_MAX, FLAGS, "predm" },
1439 { "log", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_LOG }, INT_MIN, INT_MAX, FLAGS, "predm" },
1440 { "ch_mode", "Stereo decorrelation mode", offsetof(FlacEncodeContext, options.ch_mode), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, FLAC_CHMODE_MID_SIDE, FLAGS, "ch_mode" },
1441 { "auto", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = -1 }, INT_MIN, INT_MAX, FLAGS, "ch_mode" },
1442 { "indep", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FLAC_CHMODE_INDEPENDENT }, INT_MIN, INT_MAX, FLAGS, "ch_mode" },
1443 { "left_side", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FLAC_CHMODE_LEFT_SIDE }, INT_MIN, INT_MAX, FLAGS, "ch_mode" },
1444 { "right_side", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FLAC_CHMODE_RIGHT_SIDE }, INT_MIN, INT_MAX, FLAGS, "ch_mode" },
1445 { "mid_side", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FLAC_CHMODE_MID_SIDE }, INT_MIN, INT_MAX, FLAGS, "ch_mode" },
1446 { "exact_rice_parameters", "Calculate rice parameters exactly", offsetof(FlacEncodeContext, options.exact_rice_parameters), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, FLAGS },
1447 { "multi_dim_quant", "Multi-dimensional quantization", offsetof(FlacEncodeContext, options.multi_dim_quant), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, FLAGS },
1448 { "min_prediction_order", NULL, offsetof(FlacEncodeContext, options.min_prediction_order), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, MAX_LPC_ORDER, FLAGS },
1449 { "max_prediction_order", NULL, offsetof(FlacEncodeContext, options.max_prediction_order), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, MAX_LPC_ORDER, FLAGS },
1450
1451 { NULL },
1452 };
1453
1454 static const AVClass flac_encoder_class = {
1455 .class_name = "FLAC encoder",
1456 .item_name = av_default_item_name,
1457 .option = options,
1458 .version = LIBAVUTIL_VERSION_INT,
1459 };
1460
1461 const FFCodec ff_flac_encoder = {
1462 .p.name = "flac",
1463 .p.long_name = NULL_IF_CONFIG_SMALL("FLAC (Free Lossless Audio Codec)"),
1464 .p.type = AVMEDIA_TYPE_AUDIO,
1465 .p.id = AV_CODEC_ID_FLAC,
1466 .p.capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DELAY |
1467 AV_CODEC_CAP_SMALL_LAST_FRAME,
1468 .priv_data_size = sizeof(FlacEncodeContext),
1469 .init = flac_encode_init,
1470 FF_CODEC_ENCODE_CB(flac_encode_frame),
1471 .close = flac_encode_close,
1472 .p.sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S16,
1473 AV_SAMPLE_FMT_S32,
1474 AV_SAMPLE_FMT_NONE },
1475 .p.priv_class = &flac_encoder_class,
1476 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_INIT_CLEANUP,
1477 };
1478