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/crc.h"
24 #include "libavutil/intmath.h"
25 #include "libavutil/md5.h"
26 #include "libavutil/opt.h"
27
28 #include "avcodec.h"
29 #include "bswapdsp.h"
30 #include "put_bits.h"
31 #include "golomb.h"
32 #include "internal.h"
33 #include "lpc.h"
34 #include "flac.h"
35 #include "flacdata.h"
36 #include "flacdsp.h"
37
38 #define FLAC_SUBFRAME_CONSTANT 0
39 #define FLAC_SUBFRAME_VERBATIM 1
40 #define FLAC_SUBFRAME_FIXED 8
41 #define FLAC_SUBFRAME_LPC 32
42
43 #define MAX_FIXED_ORDER 4
44 #define MAX_PARTITION_ORDER 8
45 #define MAX_PARTITIONS (1 << MAX_PARTITION_ORDER)
46 #define MAX_LPC_PRECISION 15
47 #define MIN_LPC_SHIFT 0
48 #define MAX_LPC_SHIFT 15
49
50 enum CodingMode {
51 CODING_MODE_RICE = 4,
52 CODING_MODE_RICE2 = 5,
53 };
54
55 typedef struct CompressionOptions {
56 int compression_level;
57 int block_time_ms;
58 enum FFLPCType lpc_type;
59 int lpc_passes;
60 int lpc_coeff_precision;
61 int min_prediction_order;
62 int max_prediction_order;
63 int prediction_order_method;
64 int min_partition_order;
65 int max_partition_order;
66 int ch_mode;
67 int exact_rice_parameters;
68 int multi_dim_quant;
69 } CompressionOptions;
70
71 typedef struct RiceContext {
72 enum CodingMode coding_mode;
73 int porder;
74 int params[MAX_PARTITIONS];
75 } RiceContext;
76
77 typedef struct FlacSubframe {
78 int type;
79 int type_code;
80 int obits;
81 int wasted;
82 int order;
83 int32_t coefs[MAX_LPC_ORDER];
84 int shift;
85
86 RiceContext rc;
87 uint32_t rc_udata[FLAC_MAX_BLOCKSIZE];
88 uint64_t rc_sums[32][MAX_PARTITIONS];
89
90 int32_t samples[FLAC_MAX_BLOCKSIZE];
91 int32_t residual[FLAC_MAX_BLOCKSIZE+11];
92 } FlacSubframe;
93
94 typedef struct FlacFrame {
95 FlacSubframe subframes[FLAC_MAX_CHANNELS];
96 int blocksize;
97 int bs_code[2];
98 uint8_t crc8;
99 int ch_mode;
100 int verbatim_only;
101 } FlacFrame;
102
103 typedef struct FlacEncodeContext {
104 AVClass *class;
105 PutBitContext pb;
106 int channels;
107 int samplerate;
108 int sr_code[2];
109 int bps_code;
110 int max_blocksize;
111 int min_framesize;
112 int max_framesize;
113 int max_encoded_framesize;
114 uint32_t frame_count;
115 uint64_t sample_count;
116 uint8_t md5sum[16];
117 FlacFrame frame;
118 CompressionOptions options;
119 AVCodecContext *avctx;
120 LPCContext lpc_ctx;
121 struct AVMD5 *md5ctx;
122 uint8_t *md5_buffer;
123 unsigned int md5_buffer_size;
124 BswapDSPContext bdsp;
125 FLACDSPContext flac_dsp;
126
127 int flushed;
128 int64_t next_pts;
129 } FlacEncodeContext;
130
131
132 /**
133 * Write streaminfo metadata block to byte array.
134 */
write_streaminfo(FlacEncodeContext * s,uint8_t * header)135 static void write_streaminfo(FlacEncodeContext *s, uint8_t *header)
136 {
137 PutBitContext pb;
138
139 memset(header, 0, FLAC_STREAMINFO_SIZE);
140 init_put_bits(&pb, header, FLAC_STREAMINFO_SIZE);
141
142 /* streaminfo metadata block */
143 put_bits(&pb, 16, s->max_blocksize);
144 put_bits(&pb, 16, s->max_blocksize);
145 put_bits(&pb, 24, s->min_framesize);
146 put_bits(&pb, 24, s->max_framesize);
147 put_bits(&pb, 20, s->samplerate);
148 put_bits(&pb, 3, s->channels-1);
149 put_bits(&pb, 5, s->avctx->bits_per_raw_sample - 1);
150 /* write 36-bit sample count in 2 put_bits() calls */
151 put_bits(&pb, 24, (s->sample_count & 0xFFFFFF000LL) >> 12);
152 put_bits(&pb, 12, s->sample_count & 0x000000FFFLL);
153 flush_put_bits(&pb);
154 memcpy(&header[18], s->md5sum, 16);
155 }
156
157
158 /**
159 * Set blocksize based on samplerate.
160 * Choose the closest predefined blocksize >= BLOCK_TIME_MS milliseconds.
161 */
select_blocksize(int samplerate,int block_time_ms)162 static int select_blocksize(int samplerate, int block_time_ms)
163 {
164 int i;
165 int target;
166 int blocksize;
167
168 av_assert0(samplerate > 0);
169 blocksize = ff_flac_blocksize_table[1];
170 target = (samplerate * block_time_ms) / 1000;
171 for (i = 0; i < 16; i++) {
172 if (target >= ff_flac_blocksize_table[i] &&
173 ff_flac_blocksize_table[i] > blocksize) {
174 blocksize = ff_flac_blocksize_table[i];
175 }
176 }
177 return blocksize;
178 }
179
180
dprint_compression_options(FlacEncodeContext * s)181 static av_cold void dprint_compression_options(FlacEncodeContext *s)
182 {
183 AVCodecContext *avctx = s->avctx;
184 CompressionOptions *opt = &s->options;
185
186 av_log(avctx, AV_LOG_DEBUG, " compression: %d\n", opt->compression_level);
187
188 switch (opt->lpc_type) {
189 case FF_LPC_TYPE_NONE:
190 av_log(avctx, AV_LOG_DEBUG, " lpc type: None\n");
191 break;
192 case FF_LPC_TYPE_FIXED:
193 av_log(avctx, AV_LOG_DEBUG, " lpc type: Fixed pre-defined coefficients\n");
194 break;
195 case FF_LPC_TYPE_LEVINSON:
196 av_log(avctx, AV_LOG_DEBUG, " lpc type: Levinson-Durbin recursion with Welch window\n");
197 break;
198 case FF_LPC_TYPE_CHOLESKY:
199 av_log(avctx, AV_LOG_DEBUG, " lpc type: Cholesky factorization, %d pass%s\n",
200 opt->lpc_passes, opt->lpc_passes == 1 ? "" : "es");
201 break;
202 }
203
204 av_log(avctx, AV_LOG_DEBUG, " prediction order: %d, %d\n",
205 opt->min_prediction_order, opt->max_prediction_order);
206
207 switch (opt->prediction_order_method) {
208 case ORDER_METHOD_EST:
209 av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "estimate");
210 break;
211 case ORDER_METHOD_2LEVEL:
212 av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "2-level");
213 break;
214 case ORDER_METHOD_4LEVEL:
215 av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "4-level");
216 break;
217 case ORDER_METHOD_8LEVEL:
218 av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "8-level");
219 break;
220 case ORDER_METHOD_SEARCH:
221 av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "full search");
222 break;
223 case ORDER_METHOD_LOG:
224 av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "log search");
225 break;
226 }
227
228
229 av_log(avctx, AV_LOG_DEBUG, " partition order: %d, %d\n",
230 opt->min_partition_order, opt->max_partition_order);
231
232 av_log(avctx, AV_LOG_DEBUG, " block size: %d\n", avctx->frame_size);
233
234 av_log(avctx, AV_LOG_DEBUG, " lpc precision: %d\n",
235 opt->lpc_coeff_precision);
236 }
237
238
flac_encode_init(AVCodecContext * avctx)239 static av_cold int flac_encode_init(AVCodecContext *avctx)
240 {
241 int freq = avctx->sample_rate;
242 int channels = avctx->channels;
243 FlacEncodeContext *s = avctx->priv_data;
244 int i, level, ret;
245 uint8_t *streaminfo;
246
247 s->avctx = avctx;
248
249 switch (avctx->sample_fmt) {
250 case AV_SAMPLE_FMT_S16:
251 avctx->bits_per_raw_sample = 16;
252 s->bps_code = 4;
253 break;
254 case AV_SAMPLE_FMT_S32:
255 if (avctx->bits_per_raw_sample != 24)
256 av_log(avctx, AV_LOG_WARNING, "encoding as 24 bits-per-sample\n");
257 avctx->bits_per_raw_sample = 24;
258 s->bps_code = 6;
259 break;
260 }
261
262 if (channels < 1 || channels > FLAC_MAX_CHANNELS) {
263 av_log(avctx, AV_LOG_ERROR, "%d channels not supported (max %d)\n",
264 channels, FLAC_MAX_CHANNELS);
265 return AVERROR(EINVAL);
266 }
267 s->channels = channels;
268
269 /* find samplerate in table */
270 if (freq < 1)
271 return AVERROR(EINVAL);
272 for (i = 4; i < 12; i++) {
273 if (freq == ff_flac_sample_rate_table[i]) {
274 s->samplerate = ff_flac_sample_rate_table[i];
275 s->sr_code[0] = i;
276 s->sr_code[1] = 0;
277 break;
278 }
279 }
280 /* if not in table, samplerate is non-standard */
281 if (i == 12) {
282 if (freq % 1000 == 0 && freq < 255000) {
283 s->sr_code[0] = 12;
284 s->sr_code[1] = freq / 1000;
285 } else if (freq % 10 == 0 && freq < 655350) {
286 s->sr_code[0] = 14;
287 s->sr_code[1] = freq / 10;
288 } else if (freq < 65535) {
289 s->sr_code[0] = 13;
290 s->sr_code[1] = freq;
291 } else {
292 av_log(avctx, AV_LOG_ERROR, "%d Hz not supported\n", freq);
293 return AVERROR(EINVAL);
294 }
295 s->samplerate = freq;
296 }
297
298 /* set compression option defaults based on avctx->compression_level */
299 if (avctx->compression_level < 0)
300 s->options.compression_level = 5;
301 else
302 s->options.compression_level = avctx->compression_level;
303
304 level = s->options.compression_level;
305 if (level > 12) {
306 av_log(avctx, AV_LOG_ERROR, "invalid compression level: %d\n",
307 s->options.compression_level);
308 return AVERROR(EINVAL);
309 }
310
311 s->options.block_time_ms = ((int[]){ 27, 27, 27,105,105,105,105,105,105,105,105,105,105})[level];
312
313 if (s->options.lpc_type == FF_LPC_TYPE_DEFAULT)
314 s->options.lpc_type = ((int[]){ FF_LPC_TYPE_FIXED, FF_LPC_TYPE_FIXED, FF_LPC_TYPE_FIXED,
315 FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON,
316 FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON,
317 FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON,
318 FF_LPC_TYPE_LEVINSON})[level];
319
320 if (s->options.min_prediction_order < 0)
321 s->options.min_prediction_order = ((int[]){ 2, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1})[level];
322 if (s->options.max_prediction_order < 0)
323 s->options.max_prediction_order = ((int[]){ 3, 4, 4, 6, 8, 8, 8, 8, 12, 12, 12, 32, 32})[level];
324
325 if (s->options.prediction_order_method < 0)
326 s->options.prediction_order_method = ((int[]){ ORDER_METHOD_EST, ORDER_METHOD_EST, ORDER_METHOD_EST,
327 ORDER_METHOD_EST, ORDER_METHOD_EST, ORDER_METHOD_EST,
328 ORDER_METHOD_4LEVEL, ORDER_METHOD_LOG, ORDER_METHOD_4LEVEL,
329 ORDER_METHOD_LOG, ORDER_METHOD_SEARCH, ORDER_METHOD_LOG,
330 ORDER_METHOD_SEARCH})[level];
331
332 if (s->options.min_partition_order > s->options.max_partition_order) {
333 av_log(avctx, AV_LOG_ERROR, "invalid partition orders: min=%d max=%d\n",
334 s->options.min_partition_order, s->options.max_partition_order);
335 return AVERROR(EINVAL);
336 }
337 if (s->options.min_partition_order < 0)
338 s->options.min_partition_order = ((int[]){ 2, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0})[level];
339 if (s->options.max_partition_order < 0)
340 s->options.max_partition_order = ((int[]){ 2, 2, 3, 3, 3, 8, 8, 8, 8, 8, 8, 8, 8})[level];
341
342 #if FF_API_PRIVATE_OPT
343 FF_DISABLE_DEPRECATION_WARNINGS
344 if (avctx->min_prediction_order >= 0) {
345 if (s->options.lpc_type == FF_LPC_TYPE_FIXED) {
346 if (avctx->min_prediction_order > MAX_FIXED_ORDER) {
347 av_log(avctx, AV_LOG_WARNING,
348 "invalid min prediction order %d, clamped to %d\n",
349 avctx->min_prediction_order, MAX_FIXED_ORDER);
350 avctx->min_prediction_order = MAX_FIXED_ORDER;
351 }
352 } else if (avctx->min_prediction_order < MIN_LPC_ORDER ||
353 avctx->min_prediction_order > MAX_LPC_ORDER) {
354 av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n",
355 avctx->min_prediction_order);
356 return AVERROR(EINVAL);
357 }
358 s->options.min_prediction_order = avctx->min_prediction_order;
359 }
360 if (avctx->max_prediction_order >= 0) {
361 if (s->options.lpc_type == FF_LPC_TYPE_FIXED) {
362 if (avctx->max_prediction_order > MAX_FIXED_ORDER) {
363 av_log(avctx, AV_LOG_WARNING,
364 "invalid max prediction order %d, clamped to %d\n",
365 avctx->max_prediction_order, MAX_FIXED_ORDER);
366 avctx->max_prediction_order = MAX_FIXED_ORDER;
367 }
368 } else if (avctx->max_prediction_order < MIN_LPC_ORDER ||
369 avctx->max_prediction_order > MAX_LPC_ORDER) {
370 av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n",
371 avctx->max_prediction_order);
372 return AVERROR(EINVAL);
373 }
374 s->options.max_prediction_order = avctx->max_prediction_order;
375 }
376 FF_ENABLE_DEPRECATION_WARNINGS
377 #endif
378 if (s->options.lpc_type == FF_LPC_TYPE_NONE) {
379 s->options.min_prediction_order = 0;
380 s->options.max_prediction_order = 0;
381 } else if (s->options.lpc_type == FF_LPC_TYPE_FIXED) {
382 if (s->options.min_prediction_order > MAX_FIXED_ORDER) {
383 av_log(avctx, AV_LOG_WARNING,
384 "invalid min prediction order %d, clamped to %d\n",
385 s->options.min_prediction_order, MAX_FIXED_ORDER);
386 s->options.min_prediction_order = MAX_FIXED_ORDER;
387 }
388 if (s->options.max_prediction_order > MAX_FIXED_ORDER) {
389 av_log(avctx, AV_LOG_WARNING,
390 "invalid max prediction order %d, clamped to %d\n",
391 s->options.max_prediction_order, MAX_FIXED_ORDER);
392 s->options.max_prediction_order = MAX_FIXED_ORDER;
393 }
394 }
395
396 if (s->options.max_prediction_order < s->options.min_prediction_order) {
397 av_log(avctx, AV_LOG_ERROR, "invalid prediction orders: min=%d max=%d\n",
398 s->options.min_prediction_order, s->options.max_prediction_order);
399 return AVERROR(EINVAL);
400 }
401
402 if (avctx->frame_size > 0) {
403 if (avctx->frame_size < FLAC_MIN_BLOCKSIZE ||
404 avctx->frame_size > FLAC_MAX_BLOCKSIZE) {
405 av_log(avctx, AV_LOG_ERROR, "invalid block size: %d\n",
406 avctx->frame_size);
407 return AVERROR(EINVAL);
408 }
409 } else {
410 s->avctx->frame_size = select_blocksize(s->samplerate, s->options.block_time_ms);
411 }
412 s->max_blocksize = s->avctx->frame_size;
413
414 /* set maximum encoded frame size in verbatim mode */
415 s->max_framesize = ff_flac_get_max_frame_size(s->avctx->frame_size,
416 s->channels,
417 s->avctx->bits_per_raw_sample);
418
419 /* initialize MD5 context */
420 s->md5ctx = av_md5_alloc();
421 if (!s->md5ctx)
422 return AVERROR(ENOMEM);
423 av_md5_init(s->md5ctx);
424
425 streaminfo = av_malloc(FLAC_STREAMINFO_SIZE);
426 if (!streaminfo)
427 return AVERROR(ENOMEM);
428 write_streaminfo(s, streaminfo);
429 avctx->extradata = streaminfo;
430 avctx->extradata_size = FLAC_STREAMINFO_SIZE;
431
432 s->frame_count = 0;
433 s->min_framesize = s->max_framesize;
434
435 if (channels == 3 &&
436 avctx->channel_layout != (AV_CH_LAYOUT_STEREO|AV_CH_FRONT_CENTER) ||
437 channels == 4 &&
438 avctx->channel_layout != AV_CH_LAYOUT_2_2 &&
439 avctx->channel_layout != AV_CH_LAYOUT_QUAD ||
440 channels == 5 &&
441 avctx->channel_layout != AV_CH_LAYOUT_5POINT0 &&
442 avctx->channel_layout != AV_CH_LAYOUT_5POINT0_BACK ||
443 channels == 6 &&
444 avctx->channel_layout != AV_CH_LAYOUT_5POINT1 &&
445 avctx->channel_layout != AV_CH_LAYOUT_5POINT1_BACK) {
446 if (avctx->channel_layout) {
447 av_log(avctx, AV_LOG_ERROR, "Channel layout not supported by Flac, "
448 "output stream will have incorrect "
449 "channel layout.\n");
450 } else {
451 av_log(avctx, AV_LOG_WARNING, "No channel layout specified. The encoder "
452 "will use Flac channel layout for "
453 "%d channels.\n", channels);
454 }
455 }
456
457 ret = ff_lpc_init(&s->lpc_ctx, avctx->frame_size,
458 s->options.max_prediction_order, FF_LPC_TYPE_LEVINSON);
459
460 ff_bswapdsp_init(&s->bdsp);
461 ff_flacdsp_init(&s->flac_dsp, avctx->sample_fmt, channels,
462 avctx->bits_per_raw_sample);
463
464 dprint_compression_options(s);
465
466 return ret;
467 }
468
469
init_frame(FlacEncodeContext * s,int nb_samples)470 static void init_frame(FlacEncodeContext *s, int nb_samples)
471 {
472 int i, ch;
473 FlacFrame *frame;
474
475 frame = &s->frame;
476
477 for (i = 0; i < 16; i++) {
478 if (nb_samples == ff_flac_blocksize_table[i]) {
479 frame->blocksize = ff_flac_blocksize_table[i];
480 frame->bs_code[0] = i;
481 frame->bs_code[1] = 0;
482 break;
483 }
484 }
485 if (i == 16) {
486 frame->blocksize = nb_samples;
487 if (frame->blocksize <= 256) {
488 frame->bs_code[0] = 6;
489 frame->bs_code[1] = frame->blocksize-1;
490 } else {
491 frame->bs_code[0] = 7;
492 frame->bs_code[1] = frame->blocksize-1;
493 }
494 }
495
496 for (ch = 0; ch < s->channels; ch++) {
497 FlacSubframe *sub = &frame->subframes[ch];
498
499 sub->wasted = 0;
500 sub->obits = s->avctx->bits_per_raw_sample;
501
502 if (sub->obits > 16)
503 sub->rc.coding_mode = CODING_MODE_RICE2;
504 else
505 sub->rc.coding_mode = CODING_MODE_RICE;
506 }
507
508 frame->verbatim_only = 0;
509 }
510
511
512 /**
513 * Copy channel-interleaved input samples into separate subframes.
514 */
copy_samples(FlacEncodeContext * s,const void * samples)515 static void copy_samples(FlacEncodeContext *s, const void *samples)
516 {
517 int i, j, ch;
518 FlacFrame *frame;
519 int shift = av_get_bytes_per_sample(s->avctx->sample_fmt) * 8 -
520 s->avctx->bits_per_raw_sample;
521
522 #define COPY_SAMPLES(bits) do { \
523 const int ## bits ## _t *samples0 = samples; \
524 frame = &s->frame; \
525 for (i = 0, j = 0; i < frame->blocksize; i++) \
526 for (ch = 0; ch < s->channels; ch++, j++) \
527 frame->subframes[ch].samples[i] = samples0[j] >> shift; \
528 } while (0)
529
530 if (s->avctx->sample_fmt == AV_SAMPLE_FMT_S16)
531 COPY_SAMPLES(16);
532 else
533 COPY_SAMPLES(32);
534 }
535
536
rice_count_exact(const int32_t * res,int n,int k)537 static uint64_t rice_count_exact(const int32_t *res, int n, int k)
538 {
539 int i;
540 uint64_t count = 0;
541
542 for (i = 0; i < n; i++) {
543 int32_t v = -2 * res[i] - 1;
544 v ^= v >> 31;
545 count += (v >> k) + 1 + k;
546 }
547 return count;
548 }
549
550
subframe_count_exact(FlacEncodeContext * s,FlacSubframe * sub,int pred_order)551 static uint64_t subframe_count_exact(FlacEncodeContext *s, FlacSubframe *sub,
552 int pred_order)
553 {
554 int p, porder, psize;
555 int i, part_end;
556 uint64_t count = 0;
557
558 /* subframe header */
559 count += 8;
560
561 if (sub->wasted)
562 count += sub->wasted;
563
564 /* subframe */
565 if (sub->type == FLAC_SUBFRAME_CONSTANT) {
566 count += sub->obits;
567 } else if (sub->type == FLAC_SUBFRAME_VERBATIM) {
568 count += s->frame.blocksize * sub->obits;
569 } else {
570 /* warm-up samples */
571 count += pred_order * sub->obits;
572
573 /* LPC coefficients */
574 if (sub->type == FLAC_SUBFRAME_LPC)
575 count += 4 + 5 + pred_order * s->options.lpc_coeff_precision;
576
577 /* rice-encoded block */
578 count += 2;
579
580 /* partition order */
581 porder = sub->rc.porder;
582 psize = s->frame.blocksize >> porder;
583 count += 4;
584
585 /* residual */
586 i = pred_order;
587 part_end = psize;
588 for (p = 0; p < 1 << porder; p++) {
589 int k = sub->rc.params[p];
590 count += sub->rc.coding_mode;
591 count += rice_count_exact(&sub->residual[i], part_end - i, k);
592 i = part_end;
593 part_end = FFMIN(s->frame.blocksize, part_end + psize);
594 }
595 }
596
597 return count;
598 }
599
600
601 #define rice_encode_count(sum, n, k) (((n)*((k)+1))+((sum-(n>>1))>>(k)))
602
603 /**
604 * Solve for d/dk(rice_encode_count) = n-((sum-(n>>1))>>(k+1)) = 0.
605 */
find_optimal_param(uint64_t sum,int n,int max_param)606 static int find_optimal_param(uint64_t sum, int n, int max_param)
607 {
608 int k;
609 uint64_t sum2;
610
611 if (sum <= n >> 1)
612 return 0;
613 sum2 = sum - (n >> 1);
614 k = av_log2(av_clipl_int32(sum2 / n));
615 return FFMIN(k, max_param);
616 }
617
find_optimal_param_exact(uint64_t sums[32][MAX_PARTITIONS],int i,int max_param)618 static int find_optimal_param_exact(uint64_t sums[32][MAX_PARTITIONS], int i, int max_param)
619 {
620 int bestk = 0;
621 int64_t bestbits = INT64_MAX;
622 int k;
623
624 for (k = 0; k <= max_param; k++) {
625 int64_t bits = sums[k][i];
626 if (bits < bestbits) {
627 bestbits = bits;
628 bestk = k;
629 }
630 }
631
632 return bestk;
633 }
634
calc_optimal_rice_params(RiceContext * rc,int porder,uint64_t sums[32][MAX_PARTITIONS],int n,int pred_order,int max_param,int exact)635 static uint64_t calc_optimal_rice_params(RiceContext *rc, int porder,
636 uint64_t sums[32][MAX_PARTITIONS],
637 int n, int pred_order, int max_param, int exact)
638 {
639 int i;
640 int k, cnt, part;
641 uint64_t all_bits;
642
643 part = (1 << porder);
644 all_bits = 4 * part;
645
646 cnt = (n >> porder) - pred_order;
647 for (i = 0; i < part; i++) {
648 if (exact) {
649 k = find_optimal_param_exact(sums, i, max_param);
650 all_bits += sums[k][i];
651 } else {
652 k = find_optimal_param(sums[0][i], cnt, max_param);
653 all_bits += rice_encode_count(sums[0][i], cnt, k);
654 }
655 rc->params[i] = k;
656 cnt = n >> porder;
657 }
658
659 rc->porder = porder;
660
661 return all_bits;
662 }
663
664
calc_sum_top(int pmax,int kmax,const uint32_t * data,int n,int pred_order,uint64_t sums[32][MAX_PARTITIONS])665 static void calc_sum_top(int pmax, int kmax, const uint32_t *data, int n, int pred_order,
666 uint64_t sums[32][MAX_PARTITIONS])
667 {
668 int i, k;
669 int parts;
670 const uint32_t *res, *res_end;
671
672 /* sums for highest level */
673 parts = (1 << pmax);
674
675 for (k = 0; k <= kmax; k++) {
676 res = &data[pred_order];
677 res_end = &data[n >> pmax];
678 for (i = 0; i < parts; i++) {
679 if (kmax) {
680 uint64_t sum = (1LL + k) * (res_end - res);
681 while (res < res_end)
682 sum += *(res++) >> k;
683 sums[k][i] = sum;
684 } else {
685 uint64_t sum = 0;
686 while (res < res_end)
687 sum += *(res++);
688 sums[k][i] = sum;
689 }
690 res_end += n >> pmax;
691 }
692 }
693 }
694
calc_sum_next(int level,uint64_t sums[32][MAX_PARTITIONS],int kmax)695 static void calc_sum_next(int level, uint64_t sums[32][MAX_PARTITIONS], int kmax)
696 {
697 int i, k;
698 int parts = (1 << level);
699 for (i = 0; i < parts; i++) {
700 for (k=0; k<=kmax; k++)
701 sums[k][i] = sums[k][2*i] + sums[k][2*i+1];
702 }
703 }
704
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)705 static uint64_t calc_rice_params(RiceContext *rc,
706 uint32_t udata[FLAC_MAX_BLOCKSIZE],
707 uint64_t sums[32][MAX_PARTITIONS],
708 int pmin, int pmax,
709 const int32_t *data, int n, int pred_order, int exact)
710 {
711 int i;
712 uint64_t bits[MAX_PARTITION_ORDER+1];
713 int opt_porder;
714 RiceContext tmp_rc;
715 int kmax = (1 << rc->coding_mode) - 2;
716
717 av_assert1(pmin >= 0 && pmin <= MAX_PARTITION_ORDER);
718 av_assert1(pmax >= 0 && pmax <= MAX_PARTITION_ORDER);
719 av_assert1(pmin <= pmax);
720
721 tmp_rc.coding_mode = rc->coding_mode;
722
723 for (i = 0; i < n; i++)
724 udata[i] = (2 * data[i]) ^ (data[i] >> 31);
725
726 calc_sum_top(pmax, exact ? kmax : 0, udata, n, pred_order, sums);
727
728 opt_porder = pmin;
729 bits[pmin] = UINT32_MAX;
730 for (i = pmax; ; ) {
731 bits[i] = calc_optimal_rice_params(&tmp_rc, i, sums, n, pred_order, kmax, exact);
732 if (bits[i] < bits[opt_porder] || pmax == pmin) {
733 opt_porder = i;
734 *rc = tmp_rc;
735 }
736 if (i == pmin)
737 break;
738 calc_sum_next(--i, sums, exact ? kmax : 0);
739 }
740
741 return bits[opt_porder];
742 }
743
744
get_max_p_order(int max_porder,int n,int order)745 static int get_max_p_order(int max_porder, int n, int order)
746 {
747 int porder = FFMIN(max_porder, av_log2(n^(n-1)));
748 if (order > 0)
749 porder = FFMIN(porder, av_log2(n/order));
750 return porder;
751 }
752
753
find_subframe_rice_params(FlacEncodeContext * s,FlacSubframe * sub,int pred_order)754 static uint64_t find_subframe_rice_params(FlacEncodeContext *s,
755 FlacSubframe *sub, int pred_order)
756 {
757 int pmin = get_max_p_order(s->options.min_partition_order,
758 s->frame.blocksize, pred_order);
759 int pmax = get_max_p_order(s->options.max_partition_order,
760 s->frame.blocksize, pred_order);
761
762 uint64_t bits = 8 + pred_order * sub->obits + 2 + sub->rc.coding_mode;
763 if (sub->type == FLAC_SUBFRAME_LPC)
764 bits += 4 + 5 + pred_order * s->options.lpc_coeff_precision;
765 bits += calc_rice_params(&sub->rc, sub->rc_udata, sub->rc_sums, pmin, pmax, sub->residual,
766 s->frame.blocksize, pred_order, s->options.exact_rice_parameters);
767 return bits;
768 }
769
770
encode_residual_fixed(int32_t * res,const int32_t * smp,int n,int order)771 static void encode_residual_fixed(int32_t *res, const int32_t *smp, int n,
772 int order)
773 {
774 int i;
775
776 for (i = 0; i < order; i++)
777 res[i] = smp[i];
778
779 if (order == 0) {
780 for (i = order; i < n; i++)
781 res[i] = smp[i];
782 } else if (order == 1) {
783 for (i = order; i < n; i++)
784 res[i] = smp[i] - smp[i-1];
785 } else if (order == 2) {
786 int a = smp[order-1] - smp[order-2];
787 for (i = order; i < n; i += 2) {
788 int b = smp[i ] - smp[i-1];
789 res[i] = b - a;
790 a = smp[i+1] - smp[i ];
791 res[i+1] = a - b;
792 }
793 } else if (order == 3) {
794 int a = smp[order-1] - smp[order-2];
795 int c = smp[order-1] - 2*smp[order-2] + smp[order-3];
796 for (i = order; i < n; i += 2) {
797 int b = smp[i ] - smp[i-1];
798 int d = b - a;
799 res[i] = d - c;
800 a = smp[i+1] - smp[i ];
801 c = a - b;
802 res[i+1] = c - d;
803 }
804 } else {
805 int a = smp[order-1] - smp[order-2];
806 int c = smp[order-1] - 2*smp[order-2] + smp[order-3];
807 int e = smp[order-1] - 3*smp[order-2] + 3*smp[order-3] - smp[order-4];
808 for (i = order; i < n; i += 2) {
809 int b = smp[i ] - smp[i-1];
810 int d = b - a;
811 int f = d - c;
812 res[i ] = f - e;
813 a = smp[i+1] - smp[i ];
814 c = a - b;
815 e = c - d;
816 res[i+1] = e - f;
817 }
818 }
819 }
820
821
encode_residual_ch(FlacEncodeContext * s,int ch)822 static int encode_residual_ch(FlacEncodeContext *s, int ch)
823 {
824 int i, n;
825 int min_order, max_order, opt_order, omethod;
826 FlacFrame *frame;
827 FlacSubframe *sub;
828 int32_t coefs[MAX_LPC_ORDER][MAX_LPC_ORDER];
829 int shift[MAX_LPC_ORDER];
830 int32_t *res, *smp;
831
832 frame = &s->frame;
833 sub = &frame->subframes[ch];
834 res = sub->residual;
835 smp = sub->samples;
836 n = frame->blocksize;
837
838 /* CONSTANT */
839 for (i = 1; i < n; i++)
840 if(smp[i] != smp[0])
841 break;
842 if (i == n) {
843 sub->type = sub->type_code = FLAC_SUBFRAME_CONSTANT;
844 res[0] = smp[0];
845 return subframe_count_exact(s, sub, 0);
846 }
847
848 /* VERBATIM */
849 if (frame->verbatim_only || n < 5) {
850 sub->type = sub->type_code = FLAC_SUBFRAME_VERBATIM;
851 memcpy(res, smp, n * sizeof(int32_t));
852 return subframe_count_exact(s, sub, 0);
853 }
854
855 min_order = s->options.min_prediction_order;
856 max_order = s->options.max_prediction_order;
857 omethod = s->options.prediction_order_method;
858
859 /* FIXED */
860 sub->type = FLAC_SUBFRAME_FIXED;
861 if (s->options.lpc_type == FF_LPC_TYPE_NONE ||
862 s->options.lpc_type == FF_LPC_TYPE_FIXED || n <= max_order) {
863 uint64_t bits[MAX_FIXED_ORDER+1];
864 if (max_order > MAX_FIXED_ORDER)
865 max_order = MAX_FIXED_ORDER;
866 opt_order = 0;
867 bits[0] = UINT32_MAX;
868 for (i = min_order; i <= max_order; i++) {
869 encode_residual_fixed(res, smp, n, i);
870 bits[i] = find_subframe_rice_params(s, sub, i);
871 if (bits[i] < bits[opt_order])
872 opt_order = i;
873 }
874 sub->order = opt_order;
875 sub->type_code = sub->type | sub->order;
876 if (sub->order != max_order) {
877 encode_residual_fixed(res, smp, n, sub->order);
878 find_subframe_rice_params(s, sub, sub->order);
879 }
880 return subframe_count_exact(s, sub, sub->order);
881 }
882
883 /* LPC */
884 sub->type = FLAC_SUBFRAME_LPC;
885 opt_order = ff_lpc_calc_coefs(&s->lpc_ctx, smp, n, min_order, max_order,
886 s->options.lpc_coeff_precision, coefs, shift, s->options.lpc_type,
887 s->options.lpc_passes, omethod,
888 MIN_LPC_SHIFT, MAX_LPC_SHIFT, 0);
889
890 if (omethod == ORDER_METHOD_2LEVEL ||
891 omethod == ORDER_METHOD_4LEVEL ||
892 omethod == ORDER_METHOD_8LEVEL) {
893 int levels = 1 << omethod;
894 uint64_t bits[1 << ORDER_METHOD_8LEVEL];
895 int order = -1;
896 int opt_index = levels-1;
897 opt_order = max_order-1;
898 bits[opt_index] = UINT32_MAX;
899 for (i = levels-1; i >= 0; i--) {
900 int last_order = order;
901 order = min_order + (((max_order-min_order+1) * (i+1)) / levels)-1;
902 order = av_clip(order, min_order - 1, max_order - 1);
903 if (order == last_order)
904 continue;
905 if (s->bps_code * 4 + s->options.lpc_coeff_precision + av_log2(order) <= 32) {
906 s->flac_dsp.lpc16_encode(res, smp, n, order+1, coefs[order],
907 shift[order]);
908 } else {
909 s->flac_dsp.lpc32_encode(res, smp, n, order+1, coefs[order],
910 shift[order]);
911 }
912 bits[i] = find_subframe_rice_params(s, sub, order+1);
913 if (bits[i] < bits[opt_index]) {
914 opt_index = i;
915 opt_order = order;
916 }
917 }
918 opt_order++;
919 } else if (omethod == ORDER_METHOD_SEARCH) {
920 // brute-force optimal order search
921 uint64_t bits[MAX_LPC_ORDER];
922 opt_order = 0;
923 bits[0] = UINT32_MAX;
924 for (i = min_order-1; i < max_order; i++) {
925 if (s->bps_code * 4 + s->options.lpc_coeff_precision + av_log2(i) <= 32) {
926 s->flac_dsp.lpc16_encode(res, smp, n, i+1, coefs[i], shift[i]);
927 } else {
928 s->flac_dsp.lpc32_encode(res, smp, n, i+1, coefs[i], shift[i]);
929 }
930 bits[i] = find_subframe_rice_params(s, sub, i+1);
931 if (bits[i] < bits[opt_order])
932 opt_order = i;
933 }
934 opt_order++;
935 } else if (omethod == ORDER_METHOD_LOG) {
936 uint64_t bits[MAX_LPC_ORDER];
937 int step;
938
939 opt_order = min_order - 1 + (max_order-min_order)/3;
940 memset(bits, -1, sizeof(bits));
941
942 for (step = 16; step; step >>= 1) {
943 int last = opt_order;
944 for (i = last-step; i <= last+step; i += step) {
945 if (i < min_order-1 || i >= max_order || bits[i] < UINT32_MAX)
946 continue;
947 if (s->bps_code * 4 + s->options.lpc_coeff_precision + av_log2(i) <= 32) {
948 s->flac_dsp.lpc32_encode(res, smp, n, i+1, coefs[i], shift[i]);
949 } else {
950 s->flac_dsp.lpc16_encode(res, smp, n, i+1, coefs[i], shift[i]);
951 }
952 bits[i] = find_subframe_rice_params(s, sub, i+1);
953 if (bits[i] < bits[opt_order])
954 opt_order = i;
955 }
956 }
957 opt_order++;
958 }
959
960 if (s->options.multi_dim_quant) {
961 int allsteps = 1;
962 int i, step, improved;
963 int64_t best_score = INT64_MAX;
964 int32_t qmax;
965
966 qmax = (1 << (s->options.lpc_coeff_precision - 1)) - 1;
967
968 for (i=0; i<opt_order; i++)
969 allsteps *= 3;
970
971 do {
972 improved = 0;
973 for (step = 0; step < allsteps; step++) {
974 int tmp = step;
975 int32_t lpc_try[MAX_LPC_ORDER];
976 int64_t score = 0;
977 int diffsum = 0;
978
979 for (i=0; i<opt_order; i++) {
980 int diff = ((tmp + 1) % 3) - 1;
981 lpc_try[i] = av_clip(coefs[opt_order - 1][i] + diff, -qmax, qmax);
982 tmp /= 3;
983 diffsum += !!diff;
984 }
985 if (diffsum >8)
986 continue;
987
988 if (s->bps_code * 4 + s->options.lpc_coeff_precision + av_log2(opt_order - 1) <= 32) {
989 s->flac_dsp.lpc16_encode(res, smp, n, opt_order, lpc_try, shift[opt_order-1]);
990 } else {
991 s->flac_dsp.lpc32_encode(res, smp, n, opt_order, lpc_try, shift[opt_order-1]);
992 }
993 score = find_subframe_rice_params(s, sub, opt_order);
994 if (score < best_score) {
995 best_score = score;
996 memcpy(coefs[opt_order-1], lpc_try, sizeof(*coefs));
997 improved=1;
998 }
999 }
1000 } while(improved);
1001 }
1002
1003 sub->order = opt_order;
1004 sub->type_code = sub->type | (sub->order-1);
1005 sub->shift = shift[sub->order-1];
1006 for (i = 0; i < sub->order; i++)
1007 sub->coefs[i] = coefs[sub->order-1][i];
1008
1009 if (s->bps_code * 4 + s->options.lpc_coeff_precision + av_log2(opt_order) <= 32) {
1010 s->flac_dsp.lpc16_encode(res, smp, n, sub->order, sub->coefs, sub->shift);
1011 } else {
1012 s->flac_dsp.lpc32_encode(res, smp, n, sub->order, sub->coefs, sub->shift);
1013 }
1014
1015 find_subframe_rice_params(s, sub, sub->order);
1016
1017 return subframe_count_exact(s, sub, sub->order);
1018 }
1019
1020
count_frame_header(FlacEncodeContext * s)1021 static int count_frame_header(FlacEncodeContext *s)
1022 {
1023 uint8_t av_unused tmp;
1024 int count;
1025
1026 /*
1027 <14> Sync code
1028 <1> Reserved
1029 <1> Blocking strategy
1030 <4> Block size in inter-channel samples
1031 <4> Sample rate
1032 <4> Channel assignment
1033 <3> Sample size in bits
1034 <1> Reserved
1035 */
1036 count = 32;
1037
1038 /* coded frame number */
1039 PUT_UTF8(s->frame_count, tmp, count += 8;)
1040
1041 /* explicit block size */
1042 if (s->frame.bs_code[0] == 6)
1043 count += 8;
1044 else if (s->frame.bs_code[0] == 7)
1045 count += 16;
1046
1047 /* explicit sample rate */
1048 count += ((s->sr_code[0] == 12) + (s->sr_code[0] > 12) * 2) * 8;
1049
1050 /* frame header CRC-8 */
1051 count += 8;
1052
1053 return count;
1054 }
1055
1056
encode_frame(FlacEncodeContext * s)1057 static int encode_frame(FlacEncodeContext *s)
1058 {
1059 int ch;
1060 uint64_t count;
1061
1062 count = count_frame_header(s);
1063
1064 for (ch = 0; ch < s->channels; ch++)
1065 count += encode_residual_ch(s, ch);
1066
1067 count += (8 - (count & 7)) & 7; // byte alignment
1068 count += 16; // CRC-16
1069
1070 count >>= 3;
1071 if (count > INT_MAX)
1072 return AVERROR_BUG;
1073 return count;
1074 }
1075
1076
remove_wasted_bits(FlacEncodeContext * s)1077 static void remove_wasted_bits(FlacEncodeContext *s)
1078 {
1079 int ch, i;
1080
1081 for (ch = 0; ch < s->channels; ch++) {
1082 FlacSubframe *sub = &s->frame.subframes[ch];
1083 int32_t v = 0;
1084
1085 for (i = 0; i < s->frame.blocksize; i++) {
1086 v |= sub->samples[i];
1087 if (v & 1)
1088 break;
1089 }
1090
1091 if (v && !(v & 1)) {
1092 v = ff_ctz(v);
1093
1094 for (i = 0; i < s->frame.blocksize; i++)
1095 sub->samples[i] >>= v;
1096
1097 sub->wasted = v;
1098 sub->obits -= v;
1099
1100 /* for 24-bit, check if removing wasted bits makes the range better
1101 suited for using RICE instead of RICE2 for entropy coding */
1102 if (sub->obits <= 17)
1103 sub->rc.coding_mode = CODING_MODE_RICE;
1104 }
1105 }
1106 }
1107
1108
estimate_stereo_mode(const int32_t * left_ch,const int32_t * right_ch,int n,int max_rice_param)1109 static int estimate_stereo_mode(const int32_t *left_ch, const int32_t *right_ch, int n,
1110 int max_rice_param)
1111 {
1112 int i, best;
1113 int32_t lt, rt;
1114 uint64_t sum[4];
1115 uint64_t score[4];
1116 int k;
1117
1118 /* calculate sum of 2nd order residual for each channel */
1119 sum[0] = sum[1] = sum[2] = sum[3] = 0;
1120 for (i = 2; i < n; i++) {
1121 lt = left_ch[i] - 2*left_ch[i-1] + left_ch[i-2];
1122 rt = right_ch[i] - 2*right_ch[i-1] + right_ch[i-2];
1123 sum[2] += FFABS((lt + rt) >> 1);
1124 sum[3] += FFABS(lt - rt);
1125 sum[0] += FFABS(lt);
1126 sum[1] += FFABS(rt);
1127 }
1128 /* estimate bit counts */
1129 for (i = 0; i < 4; i++) {
1130 k = find_optimal_param(2 * sum[i], n, max_rice_param);
1131 sum[i] = rice_encode_count( 2 * sum[i], n, k);
1132 }
1133
1134 /* calculate score for each mode */
1135 score[0] = sum[0] + sum[1];
1136 score[1] = sum[0] + sum[3];
1137 score[2] = sum[1] + sum[3];
1138 score[3] = sum[2] + sum[3];
1139
1140 /* return mode with lowest score */
1141 best = 0;
1142 for (i = 1; i < 4; i++)
1143 if (score[i] < score[best])
1144 best = i;
1145
1146 return best;
1147 }
1148
1149
1150 /**
1151 * Perform stereo channel decorrelation.
1152 */
channel_decorrelation(FlacEncodeContext * s)1153 static void channel_decorrelation(FlacEncodeContext *s)
1154 {
1155 FlacFrame *frame;
1156 int32_t *left, *right;
1157 int i, n;
1158
1159 frame = &s->frame;
1160 n = frame->blocksize;
1161 left = frame->subframes[0].samples;
1162 right = frame->subframes[1].samples;
1163
1164 if (s->channels != 2) {
1165 frame->ch_mode = FLAC_CHMODE_INDEPENDENT;
1166 return;
1167 }
1168
1169 if (s->options.ch_mode < 0) {
1170 int max_rice_param = (1 << frame->subframes[0].rc.coding_mode) - 2;
1171 frame->ch_mode = estimate_stereo_mode(left, right, n, max_rice_param);
1172 } else
1173 frame->ch_mode = s->options.ch_mode;
1174
1175 /* perform decorrelation and adjust bits-per-sample */
1176 if (frame->ch_mode == FLAC_CHMODE_INDEPENDENT)
1177 return;
1178 if (frame->ch_mode == FLAC_CHMODE_MID_SIDE) {
1179 int32_t tmp;
1180 for (i = 0; i < n; i++) {
1181 tmp = left[i];
1182 left[i] = (tmp + right[i]) >> 1;
1183 right[i] = tmp - right[i];
1184 }
1185 frame->subframes[1].obits++;
1186 } else if (frame->ch_mode == FLAC_CHMODE_LEFT_SIDE) {
1187 for (i = 0; i < n; i++)
1188 right[i] = left[i] - right[i];
1189 frame->subframes[1].obits++;
1190 } else {
1191 for (i = 0; i < n; i++)
1192 left[i] -= right[i];
1193 frame->subframes[0].obits++;
1194 }
1195 }
1196
1197
write_utf8(PutBitContext * pb,uint32_t val)1198 static void write_utf8(PutBitContext *pb, uint32_t val)
1199 {
1200 uint8_t tmp;
1201 PUT_UTF8(val, tmp, put_bits(pb, 8, tmp);)
1202 }
1203
1204
write_frame_header(FlacEncodeContext * s)1205 static void write_frame_header(FlacEncodeContext *s)
1206 {
1207 FlacFrame *frame;
1208 int crc;
1209
1210 frame = &s->frame;
1211
1212 put_bits(&s->pb, 16, 0xFFF8);
1213 put_bits(&s->pb, 4, frame->bs_code[0]);
1214 put_bits(&s->pb, 4, s->sr_code[0]);
1215
1216 if (frame->ch_mode == FLAC_CHMODE_INDEPENDENT)
1217 put_bits(&s->pb, 4, s->channels-1);
1218 else
1219 put_bits(&s->pb, 4, frame->ch_mode + FLAC_MAX_CHANNELS - 1);
1220
1221 put_bits(&s->pb, 3, s->bps_code);
1222 put_bits(&s->pb, 1, 0);
1223 write_utf8(&s->pb, s->frame_count);
1224
1225 if (frame->bs_code[0] == 6)
1226 put_bits(&s->pb, 8, frame->bs_code[1]);
1227 else if (frame->bs_code[0] == 7)
1228 put_bits(&s->pb, 16, frame->bs_code[1]);
1229
1230 if (s->sr_code[0] == 12)
1231 put_bits(&s->pb, 8, s->sr_code[1]);
1232 else if (s->sr_code[0] > 12)
1233 put_bits(&s->pb, 16, s->sr_code[1]);
1234
1235 flush_put_bits(&s->pb);
1236 crc = av_crc(av_crc_get_table(AV_CRC_8_ATM), 0, s->pb.buf,
1237 put_bits_count(&s->pb) >> 3);
1238 put_bits(&s->pb, 8, crc);
1239 }
1240
1241
write_subframes(FlacEncodeContext * s)1242 static void write_subframes(FlacEncodeContext *s)
1243 {
1244 int ch;
1245
1246 for (ch = 0; ch < s->channels; ch++) {
1247 FlacSubframe *sub = &s->frame.subframes[ch];
1248 int i, p, porder, psize;
1249 int32_t *part_end;
1250 int32_t *res = sub->residual;
1251 int32_t *frame_end = &sub->residual[s->frame.blocksize];
1252
1253 /* subframe header */
1254 put_bits(&s->pb, 1, 0);
1255 put_bits(&s->pb, 6, sub->type_code);
1256 put_bits(&s->pb, 1, !!sub->wasted);
1257 if (sub->wasted)
1258 put_bits(&s->pb, sub->wasted, 1);
1259
1260 /* subframe */
1261 if (sub->type == FLAC_SUBFRAME_CONSTANT) {
1262 put_sbits(&s->pb, sub->obits, res[0]);
1263 } else if (sub->type == FLAC_SUBFRAME_VERBATIM) {
1264 while (res < frame_end)
1265 put_sbits(&s->pb, sub->obits, *res++);
1266 } else {
1267 /* warm-up samples */
1268 for (i = 0; i < sub->order; i++)
1269 put_sbits(&s->pb, sub->obits, *res++);
1270
1271 /* LPC coefficients */
1272 if (sub->type == FLAC_SUBFRAME_LPC) {
1273 int cbits = s->options.lpc_coeff_precision;
1274 put_bits( &s->pb, 4, cbits-1);
1275 put_sbits(&s->pb, 5, sub->shift);
1276 for (i = 0; i < sub->order; i++)
1277 put_sbits(&s->pb, cbits, sub->coefs[i]);
1278 }
1279
1280 /* rice-encoded block */
1281 put_bits(&s->pb, 2, sub->rc.coding_mode - 4);
1282
1283 /* partition order */
1284 porder = sub->rc.porder;
1285 psize = s->frame.blocksize >> porder;
1286 put_bits(&s->pb, 4, porder);
1287
1288 /* residual */
1289 part_end = &sub->residual[psize];
1290 for (p = 0; p < 1 << porder; p++) {
1291 int k = sub->rc.params[p];
1292 put_bits(&s->pb, sub->rc.coding_mode, k);
1293 while (res < part_end)
1294 set_sr_golomb_flac(&s->pb, *res++, k, INT32_MAX, 0);
1295 part_end = FFMIN(frame_end, part_end + psize);
1296 }
1297 }
1298 }
1299 }
1300
1301
write_frame_footer(FlacEncodeContext * s)1302 static void write_frame_footer(FlacEncodeContext *s)
1303 {
1304 int crc;
1305 flush_put_bits(&s->pb);
1306 crc = av_bswap16(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, s->pb.buf,
1307 put_bits_count(&s->pb)>>3));
1308 put_bits(&s->pb, 16, crc);
1309 flush_put_bits(&s->pb);
1310 }
1311
1312
write_frame(FlacEncodeContext * s,AVPacket * avpkt)1313 static int write_frame(FlacEncodeContext *s, AVPacket *avpkt)
1314 {
1315 init_put_bits(&s->pb, avpkt->data, avpkt->size);
1316 write_frame_header(s);
1317 write_subframes(s);
1318 write_frame_footer(s);
1319 return put_bits_count(&s->pb) >> 3;
1320 }
1321
1322
update_md5_sum(FlacEncodeContext * s,const void * samples)1323 static int update_md5_sum(FlacEncodeContext *s, const void *samples)
1324 {
1325 const uint8_t *buf;
1326 int buf_size = s->frame.blocksize * s->channels *
1327 ((s->avctx->bits_per_raw_sample + 7) / 8);
1328
1329 if (s->avctx->bits_per_raw_sample > 16 || HAVE_BIGENDIAN) {
1330 av_fast_malloc(&s->md5_buffer, &s->md5_buffer_size, buf_size);
1331 if (!s->md5_buffer)
1332 return AVERROR(ENOMEM);
1333 }
1334
1335 if (s->avctx->bits_per_raw_sample <= 16) {
1336 buf = (const uint8_t *)samples;
1337 #if HAVE_BIGENDIAN
1338 s->bdsp.bswap16_buf((uint16_t *) s->md5_buffer,
1339 (const uint16_t *) samples, buf_size / 2);
1340 buf = s->md5_buffer;
1341 #endif
1342 } else {
1343 int i;
1344 const int32_t *samples0 = samples;
1345 uint8_t *tmp = s->md5_buffer;
1346
1347 for (i = 0; i < s->frame.blocksize * s->channels; i++) {
1348 int32_t v = samples0[i] >> 8;
1349 AV_WL24(tmp + 3*i, v);
1350 }
1351 buf = s->md5_buffer;
1352 }
1353 av_md5_update(s->md5ctx, buf, buf_size);
1354
1355 return 0;
1356 }
1357
1358
flac_encode_frame(AVCodecContext * avctx,AVPacket * avpkt,const AVFrame * frame,int * got_packet_ptr)1359 static int flac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
1360 const AVFrame *frame, int *got_packet_ptr)
1361 {
1362 FlacEncodeContext *s;
1363 int frame_bytes, out_bytes, ret;
1364
1365 s = avctx->priv_data;
1366
1367 /* when the last block is reached, update the header in extradata */
1368 if (!frame) {
1369 s->max_framesize = s->max_encoded_framesize;
1370 av_md5_final(s->md5ctx, s->md5sum);
1371 write_streaminfo(s, avctx->extradata);
1372
1373 #if FF_API_SIDEDATA_ONLY_PKT
1374 FF_DISABLE_DEPRECATION_WARNINGS
1375 if (avctx->side_data_only_packets && !s->flushed) {
1376 FF_ENABLE_DEPRECATION_WARNINGS
1377 #else
1378 if (!s->flushed) {
1379 #endif
1380 uint8_t *side_data = av_packet_new_side_data(avpkt, AV_PKT_DATA_NEW_EXTRADATA,
1381 avctx->extradata_size);
1382 if (!side_data)
1383 return AVERROR(ENOMEM);
1384 memcpy(side_data, avctx->extradata, avctx->extradata_size);
1385
1386 avpkt->pts = s->next_pts;
1387
1388 *got_packet_ptr = 1;
1389 s->flushed = 1;
1390 }
1391
1392 return 0;
1393 }
1394
1395 /* change max_framesize for small final frame */
1396 if (frame->nb_samples < s->frame.blocksize) {
1397 s->max_framesize = ff_flac_get_max_frame_size(frame->nb_samples,
1398 s->channels,
1399 avctx->bits_per_raw_sample);
1400 }
1401
1402 init_frame(s, frame->nb_samples);
1403
1404 copy_samples(s, frame->data[0]);
1405
1406 channel_decorrelation(s);
1407
1408 remove_wasted_bits(s);
1409
1410 frame_bytes = encode_frame(s);
1411
1412 /* Fall back on verbatim mode if the compressed frame is larger than it
1413 would be if encoded uncompressed. */
1414 if (frame_bytes < 0 || frame_bytes > s->max_framesize) {
1415 s->frame.verbatim_only = 1;
1416 frame_bytes = encode_frame(s);
1417 if (frame_bytes < 0) {
1418 av_log(avctx, AV_LOG_ERROR, "Bad frame count\n");
1419 return frame_bytes;
1420 }
1421 }
1422
1423 if ((ret = ff_alloc_packet2(avctx, avpkt, frame_bytes, 0)) < 0)
1424 return ret;
1425
1426 out_bytes = write_frame(s, avpkt);
1427
1428 s->frame_count++;
1429 s->sample_count += frame->nb_samples;
1430 if ((ret = update_md5_sum(s, frame->data[0])) < 0) {
1431 av_log(avctx, AV_LOG_ERROR, "Error updating MD5 checksum\n");
1432 return ret;
1433 }
1434 if (out_bytes > s->max_encoded_framesize)
1435 s->max_encoded_framesize = out_bytes;
1436 if (out_bytes < s->min_framesize)
1437 s->min_framesize = out_bytes;
1438
1439 avpkt->pts = frame->pts;
1440 avpkt->duration = ff_samples_to_time_base(avctx, frame->nb_samples);
1441 avpkt->size = out_bytes;
1442
1443 s->next_pts = avpkt->pts + avpkt->duration;
1444
1445 *got_packet_ptr = 1;
1446 return 0;
1447 }
1448
1449
1450 static av_cold int flac_encode_close(AVCodecContext *avctx)
1451 {
1452 if (avctx->priv_data) {
1453 FlacEncodeContext *s = avctx->priv_data;
1454 av_freep(&s->md5ctx);
1455 av_freep(&s->md5_buffer);
1456 ff_lpc_end(&s->lpc_ctx);
1457 }
1458 av_freep(&avctx->extradata);
1459 avctx->extradata_size = 0;
1460 return 0;
1461 }
1462
1463 #define FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
1464 static const AVOption options[] = {
1465 { "lpc_coeff_precision", "LPC coefficient precision", offsetof(FlacEncodeContext, options.lpc_coeff_precision), AV_OPT_TYPE_INT, {.i64 = 15 }, 0, MAX_LPC_PRECISION, FLAGS },
1466 { "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" },
1467 { "none", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = FF_LPC_TYPE_NONE }, INT_MIN, INT_MAX, FLAGS, "lpc_type" },
1468 { "fixed", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = FF_LPC_TYPE_FIXED }, INT_MIN, INT_MAX, FLAGS, "lpc_type" },
1469 { "levinson", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = FF_LPC_TYPE_LEVINSON }, INT_MIN, INT_MAX, FLAGS, "lpc_type" },
1470 { "cholesky", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = FF_LPC_TYPE_CHOLESKY }, INT_MIN, INT_MAX, FLAGS, "lpc_type" },
1471 { "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 },
1472 { "min_partition_order", NULL, offsetof(FlacEncodeContext, options.min_partition_order), AV_OPT_TYPE_INT, {.i64 = -1 }, -1, MAX_PARTITION_ORDER, FLAGS },
1473 { "max_partition_order", NULL, offsetof(FlacEncodeContext, options.max_partition_order), AV_OPT_TYPE_INT, {.i64 = -1 }, -1, MAX_PARTITION_ORDER, FLAGS },
1474 { "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" },
1475 { "estimation", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_EST }, INT_MIN, INT_MAX, FLAGS, "predm" },
1476 { "2level", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_2LEVEL }, INT_MIN, INT_MAX, FLAGS, "predm" },
1477 { "4level", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_4LEVEL }, INT_MIN, INT_MAX, FLAGS, "predm" },
1478 { "8level", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_8LEVEL }, INT_MIN, INT_MAX, FLAGS, "predm" },
1479 { "search", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_SEARCH }, INT_MIN, INT_MAX, FLAGS, "predm" },
1480 { "log", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_LOG }, INT_MIN, INT_MAX, FLAGS, "predm" },
1481 { "ch_mode", "Stereo decorrelation mode", offsetof(FlacEncodeContext, options.ch_mode), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, FLAC_CHMODE_MID_SIDE, FLAGS, "ch_mode" },
1482 { "auto", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = -1 }, INT_MIN, INT_MAX, FLAGS, "ch_mode" },
1483 { "indep", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FLAC_CHMODE_INDEPENDENT }, INT_MIN, INT_MAX, FLAGS, "ch_mode" },
1484 { "left_side", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FLAC_CHMODE_LEFT_SIDE }, INT_MIN, INT_MAX, FLAGS, "ch_mode" },
1485 { "right_side", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FLAC_CHMODE_RIGHT_SIDE }, INT_MIN, INT_MAX, FLAGS, "ch_mode" },
1486 { "mid_side", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FLAC_CHMODE_MID_SIDE }, INT_MIN, INT_MAX, FLAGS, "ch_mode" },
1487 { "exact_rice_parameters", "Calculate rice parameters exactly", offsetof(FlacEncodeContext, options.exact_rice_parameters), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, FLAGS },
1488 { "multi_dim_quant", "Multi-dimensional quantization", offsetof(FlacEncodeContext, options.multi_dim_quant), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, FLAGS },
1489 { "min_prediction_order", NULL, offsetof(FlacEncodeContext, options.min_prediction_order), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, MAX_LPC_ORDER, FLAGS },
1490 { "max_prediction_order", NULL, offsetof(FlacEncodeContext, options.max_prediction_order), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, MAX_LPC_ORDER, FLAGS },
1491
1492 { NULL },
1493 };
1494
1495 static const AVClass flac_encoder_class = {
1496 .class_name = "FLAC encoder",
1497 .item_name = av_default_item_name,
1498 .option = options,
1499 .version = LIBAVUTIL_VERSION_INT,
1500 };
1501
1502 AVCodec ff_flac_encoder = {
1503 .name = "flac",
1504 .long_name = NULL_IF_CONFIG_SMALL("FLAC (Free Lossless Audio Codec)"),
1505 .type = AVMEDIA_TYPE_AUDIO,
1506 .id = AV_CODEC_ID_FLAC,
1507 .priv_data_size = sizeof(FlacEncodeContext),
1508 .init = flac_encode_init,
1509 .encode2 = flac_encode_frame,
1510 .close = flac_encode_close,
1511 .capabilities = AV_CODEC_CAP_SMALL_LAST_FRAME | AV_CODEC_CAP_DELAY,
1512 .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S16,
1513 AV_SAMPLE_FMT_S32,
1514 AV_SAMPLE_FMT_NONE },
1515 .priv_class = &flac_encoder_class,
1516 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
1517 };
1518