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 #include "avcodec.h"
28 #include "bswapdsp.h"
29 #include "put_bits.h"
30 #include "golomb.h"
31 #include "internal.h"
32 #include "lpc.h"
33 #include "flac.h"
34 #include "flacdata.h"
35 #include "flacdsp.h"
36 
37 #define FLAC_SUBFRAME_CONSTANT  0
38 #define FLAC_SUBFRAME_VERBATIM  1
39 #define FLAC_SUBFRAME_FIXED     8
40 #define FLAC_SUBFRAME_LPC      32
41 
42 #define MAX_FIXED_ORDER     4
43 #define MAX_PARTITION_ORDER 8
44 #define MAX_PARTITIONS     (1 << MAX_PARTITION_ORDER)
45 #define MAX_LPC_PRECISION  15
46 #define MAX_LPC_SHIFT      15
47 
48 enum CodingMode {
49     CODING_MODE_RICE  = 4,
50     CODING_MODE_RICE2 = 5,
51 };
52 
53 typedef struct CompressionOptions {
54     int compression_level;
55     int block_time_ms;
56     enum FFLPCType lpc_type;
57     int lpc_passes;
58     int lpc_coeff_precision;
59     int min_prediction_order;
60     int max_prediction_order;
61     int prediction_order_method;
62     int min_partition_order;
63     int max_partition_order;
64     int ch_mode;
65 } CompressionOptions;
66 
67 typedef struct RiceContext {
68     enum CodingMode coding_mode;
69     int porder;
70     int params[MAX_PARTITIONS];
71 } RiceContext;
72 
73 typedef struct FlacSubframe {
74     int type;
75     int type_code;
76     int obits;
77     int wasted;
78     int order;
79     int32_t coefs[MAX_LPC_ORDER];
80     int shift;
81     RiceContext rc;
82     int32_t samples[FLAC_MAX_BLOCKSIZE];
83     int32_t residual[FLAC_MAX_BLOCKSIZE+11];
84 } FlacSubframe;
85 
86 typedef struct FlacFrame {
87     FlacSubframe subframes[FLAC_MAX_CHANNELS];
88     int blocksize;
89     int bs_code[2];
90     uint8_t crc8;
91     int ch_mode;
92     int verbatim_only;
93 } FlacFrame;
94 
95 typedef struct FlacEncodeContext {
96     AVClass *class;
97     PutBitContext pb;
98     int channels;
99     int samplerate;
100     int sr_code[2];
101     int bps_code;
102     int max_blocksize;
103     int min_framesize;
104     int max_framesize;
105     int max_encoded_framesize;
106     uint32_t frame_count;
107     uint64_t sample_count;
108     uint8_t md5sum[16];
109     FlacFrame frame;
110     CompressionOptions options;
111     AVCodecContext *avctx;
112     LPCContext lpc_ctx;
113     struct AVMD5 *md5ctx;
114     uint8_t *md5_buffer;
115     unsigned int md5_buffer_size;
116     BswapDSPContext bdsp;
117     FLACDSPContext flac_dsp;
118 
119     int flushed;
120     int64_t next_pts;
121 } FlacEncodeContext;
122 
123 
124 /**
125  * Write streaminfo metadata block to byte array.
126  */
write_streaminfo(FlacEncodeContext * s,uint8_t * header)127 static void write_streaminfo(FlacEncodeContext *s, uint8_t *header)
128 {
129     PutBitContext pb;
130 
131     memset(header, 0, FLAC_STREAMINFO_SIZE);
132     init_put_bits(&pb, header, FLAC_STREAMINFO_SIZE);
133 
134     /* streaminfo metadata block */
135     put_bits(&pb, 16, s->max_blocksize);
136     put_bits(&pb, 16, s->max_blocksize);
137     put_bits(&pb, 24, s->min_framesize);
138     put_bits(&pb, 24, s->max_framesize);
139     put_bits(&pb, 20, s->samplerate);
140     put_bits(&pb, 3, s->channels-1);
141     put_bits(&pb,  5, s->avctx->bits_per_raw_sample - 1);
142     /* write 36-bit sample count in 2 put_bits() calls */
143     put_bits(&pb, 24, (s->sample_count & LLN(0xFFFFFF000)) >> 12);
144     put_bits(&pb, 12,  s->sample_count & LLN(0x000000FFF));
145     flush_put_bits(&pb);
146     memcpy(&header[18], s->md5sum, 16);
147 }
148 
149 
150 /**
151  * Set blocksize based on samplerate.
152  * Choose the closest predefined blocksize >= BLOCK_TIME_MS milliseconds.
153  */
select_blocksize(int samplerate,int block_time_ms)154 static int select_blocksize(int samplerate, int block_time_ms)
155 {
156     int i;
157     int target;
158     int blocksize;
159 
160     av_assert0(samplerate > 0);
161     blocksize = ff_flac_blocksize_table[1];
162     target    = (samplerate * block_time_ms) / 1000;
163     for (i = 0; i < 16; i++) {
164         if (target >= ff_flac_blocksize_table[i] &&
165             ff_flac_blocksize_table[i] > blocksize) {
166             blocksize = ff_flac_blocksize_table[i];
167         }
168     }
169     return blocksize;
170 }
171 
172 
dprint_compression_options(FlacEncodeContext * s)173 static av_cold void dprint_compression_options(FlacEncodeContext *s)
174 {
175     AVCodecContext     *avctx = s->avctx;
176     CompressionOptions *opt   = &s->options;
177 
178     av_log(avctx, AV_LOG_DEBUG, " compression: %d\n", opt->compression_level);
179 
180     switch (opt->lpc_type) {
181     case FF_LPC_TYPE_NONE:
182         av_log(avctx, AV_LOG_DEBUG, " lpc type: None\n");
183         break;
184     case FF_LPC_TYPE_FIXED:
185         av_log(avctx, AV_LOG_DEBUG, " lpc type: Fixed pre-defined coefficients\n");
186         break;
187     case FF_LPC_TYPE_LEVINSON:
188         av_log(avctx, AV_LOG_DEBUG, " lpc type: Levinson-Durbin recursion with Welch window\n");
189         break;
190     case FF_LPC_TYPE_CHOLESKY:
191         av_log(avctx, AV_LOG_DEBUG, " lpc type: Cholesky factorization, %d pass%s\n",
192                opt->lpc_passes, opt->lpc_passes == 1 ? "" : "es");
193         break;
194     }
195 
196     av_log(avctx, AV_LOG_DEBUG, " prediction order: %d, %d\n",
197            opt->min_prediction_order, opt->max_prediction_order);
198 
199     switch (opt->prediction_order_method) {
200     case ORDER_METHOD_EST:
201         av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "estimate");
202         break;
203     case ORDER_METHOD_2LEVEL:
204         av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "2-level");
205         break;
206     case ORDER_METHOD_4LEVEL:
207         av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "4-level");
208         break;
209     case ORDER_METHOD_8LEVEL:
210         av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "8-level");
211         break;
212     case ORDER_METHOD_SEARCH:
213         av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "full search");
214         break;
215     case ORDER_METHOD_LOG:
216         av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "log search");
217         break;
218     }
219 
220 
221     av_log(avctx, AV_LOG_DEBUG, " partition order: %d, %d\n",
222            opt->min_partition_order, opt->max_partition_order);
223 
224     av_log(avctx, AV_LOG_DEBUG, " block size: %d\n", avctx->frame_size);
225 
226     av_log(avctx, AV_LOG_DEBUG, " lpc precision: %d\n",
227            opt->lpc_coeff_precision);
228 }
229 
230 
flac_encode_init(AVCodecContext * avctx)231 static av_cold int flac_encode_init(AVCodecContext *avctx)
232 {
233     int freq = avctx->sample_rate;
234     int channels = avctx->channels;
235     FlacEncodeContext *s = avctx->priv_data;
236     int i, level, ret;
237     uint8_t *streaminfo;
238 
239     s->avctx = avctx;
240 
241     switch (avctx->sample_fmt) {
242     case AV_SAMPLE_FMT_S16:
243         avctx->bits_per_raw_sample = 16;
244         s->bps_code                = 4;
245         break;
246     case AV_SAMPLE_FMT_S32:
247         if (avctx->bits_per_raw_sample != 24)
248             av_log(avctx, AV_LOG_WARNING, "encoding as 24 bits-per-sample\n");
249         avctx->bits_per_raw_sample = 24;
250         s->bps_code                = 6;
251         break;
252     }
253 
254     if (channels < 1 || channels > FLAC_MAX_CHANNELS) {
255         av_log(avctx, AV_LOG_ERROR, "%d channels not supported (max %d)\n",
256                channels, FLAC_MAX_CHANNELS);
257         return AVERROR(EINVAL);
258     }
259     s->channels = channels;
260 
261     /* find samplerate in table */
262     if (freq < 1)
263         return -1;
264     for (i = 4; i < 12; i++) {
265         if (freq == ff_flac_sample_rate_table[i]) {
266             s->samplerate = ff_flac_sample_rate_table[i];
267             s->sr_code[0] = i;
268             s->sr_code[1] = 0;
269             break;
270         }
271     }
272     /* if not in table, samplerate is non-standard */
273     if (i == 12) {
274         if (freq % 1000 == 0 && freq < 255000) {
275             s->sr_code[0] = 12;
276             s->sr_code[1] = freq / 1000;
277         } else if (freq % 10 == 0 && freq < 655350) {
278             s->sr_code[0] = 14;
279             s->sr_code[1] = freq / 10;
280         } else if (freq < 65535) {
281             s->sr_code[0] = 13;
282             s->sr_code[1] = freq;
283         } else {
284             av_log(avctx, AV_LOG_ERROR, "%d Hz not supported\n", freq);
285             return AVERROR(EINVAL);
286         }
287         s->samplerate = freq;
288     }
289 
290     /* set compression option defaults based on avctx->compression_level */
291     if (avctx->compression_level < 0)
292         s->options.compression_level = 5;
293     else
294         s->options.compression_level = avctx->compression_level;
295 
296     level = s->options.compression_level;
297     if (level > 12) {
298         av_log(avctx, AV_LOG_ERROR, "invalid compression level: %d\n",
299                s->options.compression_level);
300         return AVERROR(EINVAL);
301     }
302 
303 	s->options.block_time_ms = ((int[]){ 27, 27, 27,105,105,105,105,105,105,105,105,105,105})[level];
304 
305     if (s->options.lpc_type == FF_LPC_TYPE_DEFAULT)
306 		s->options.lpc_type  = ((int[]){ FF_LPC_TYPE_FIXED,    FF_LPC_TYPE_FIXED,    FF_LPC_TYPE_FIXED,
307                                          FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON,
308                                          FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON,
309                                          FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON,
310                                          FF_LPC_TYPE_LEVINSON})[level];
311     s->options.min_prediction_order = ((int[]){  2,  0,  0,  1,  1,  1,  1,  1,  1,  1,  1,  1,  1})[level];
312     s->options.max_prediction_order = ((int[]){  3,  4,  4,  6,  8,  8,  8,  8, 12, 12, 12, 32, 32})[level];
313 
314     if (s->options.prediction_order_method < 0)
315 		s->options.prediction_order_method = ((int[]){ ORDER_METHOD_EST,    ORDER_METHOD_EST,    ORDER_METHOD_EST,
316                                                        ORDER_METHOD_EST,    ORDER_METHOD_EST,    ORDER_METHOD_EST,
317                                                        ORDER_METHOD_4LEVEL, ORDER_METHOD_LOG,    ORDER_METHOD_4LEVEL,
318                                                        ORDER_METHOD_LOG,    ORDER_METHOD_SEARCH, ORDER_METHOD_LOG,
319                                                        ORDER_METHOD_SEARCH})[level];
320 
321     if (s->options.min_partition_order > s->options.max_partition_order) {
322         av_log(avctx, AV_LOG_ERROR, "invalid partition orders: min=%d max=%d\n",
323                s->options.min_partition_order, s->options.max_partition_order);
324         return AVERROR(EINVAL);
325     }
326     if (s->options.min_partition_order < 0)
327 		s->options.min_partition_order = ((int[]){  2,  2,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0})[level];
328 	if (s->options.max_partition_order < 0)
329 		s->options.max_partition_order = ((int[]){  2,  2,  3,  3,  3,  8,  8,  8,  8,  8,  8,  8,  8})[level];
330 
331 	if (s->options.lpc_type == FF_LPC_TYPE_NONE) {
332         s->options.min_prediction_order = 0;
333     } else if (avctx->min_prediction_order >= 0) {
334         if (s->options.lpc_type == FF_LPC_TYPE_FIXED) {
335             if (avctx->min_prediction_order > MAX_FIXED_ORDER) {
336                 av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n",
337                        avctx->min_prediction_order);
338                 return AVERROR(EINVAL);
339             }
340         } else if (avctx->min_prediction_order < MIN_LPC_ORDER ||
341                    avctx->min_prediction_order > MAX_LPC_ORDER) {
342             av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n",
343                    avctx->min_prediction_order);
344             return AVERROR(EINVAL);
345         }
346         s->options.min_prediction_order = avctx->min_prediction_order;
347     }
348     if (s->options.lpc_type == FF_LPC_TYPE_NONE) {
349         s->options.max_prediction_order = 0;
350     } else if (avctx->max_prediction_order >= 0) {
351         if (s->options.lpc_type == FF_LPC_TYPE_FIXED) {
352             if (avctx->max_prediction_order > MAX_FIXED_ORDER) {
353                 av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n",
354                        avctx->max_prediction_order);
355                 return AVERROR(EINVAL);
356             }
357         } else if (avctx->max_prediction_order < MIN_LPC_ORDER ||
358                    avctx->max_prediction_order > MAX_LPC_ORDER) {
359             av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n",
360                    avctx->max_prediction_order);
361             return AVERROR(EINVAL);
362         }
363         s->options.max_prediction_order = avctx->max_prediction_order;
364     }
365     if (s->options.max_prediction_order < s->options.min_prediction_order) {
366         av_log(avctx, AV_LOG_ERROR, "invalid prediction orders: min=%d max=%d\n",
367                s->options.min_prediction_order, s->options.max_prediction_order);
368         return AVERROR(EINVAL);
369     }
370 
371     if (avctx->frame_size > 0) {
372         if (avctx->frame_size < FLAC_MIN_BLOCKSIZE ||
373                 avctx->frame_size > FLAC_MAX_BLOCKSIZE) {
374             av_log(avctx, AV_LOG_ERROR, "invalid block size: %d\n",
375                    avctx->frame_size);
376             return AVERROR(EINVAL);
377         }
378     } else {
379         s->avctx->frame_size = select_blocksize(s->samplerate, s->options.block_time_ms);
380     }
381     s->max_blocksize = s->avctx->frame_size;
382 
383     /* set maximum encoded frame size in verbatim mode */
384     s->max_framesize = ff_flac_get_max_frame_size(s->avctx->frame_size,
385                                                   s->channels,
386                                                   s->avctx->bits_per_raw_sample);
387 
388     /* initialize MD5 context */
389     s->md5ctx = av_md5_alloc();
390     if (!s->md5ctx)
391         return AVERROR(ENOMEM);
392     av_md5_init(s->md5ctx);
393 
394     streaminfo = av_malloc(FLAC_STREAMINFO_SIZE);
395     if (!streaminfo)
396         return AVERROR(ENOMEM);
397     write_streaminfo(s, streaminfo);
398     avctx->extradata = streaminfo;
399     avctx->extradata_size = FLAC_STREAMINFO_SIZE;
400 
401     s->frame_count   = 0;
402     s->min_framesize = s->max_framesize;
403 
404     if (channels == 3 &&
405             avctx->channel_layout != (AV_CH_LAYOUT_STEREO|AV_CH_FRONT_CENTER) ||
406         channels == 4 &&
407             avctx->channel_layout != AV_CH_LAYOUT_2_2 &&
408             avctx->channel_layout != AV_CH_LAYOUT_QUAD ||
409         channels == 5 &&
410             avctx->channel_layout != AV_CH_LAYOUT_5POINT0 &&
411             avctx->channel_layout != AV_CH_LAYOUT_5POINT0_BACK ||
412         channels == 6 &&
413             avctx->channel_layout != AV_CH_LAYOUT_5POINT1 &&
414             avctx->channel_layout != AV_CH_LAYOUT_5POINT1_BACK) {
415         if (avctx->channel_layout) {
416             av_log(avctx, AV_LOG_ERROR, "Channel layout not supported by Flac, "
417                                              "output stream will have incorrect "
418                                              "channel layout.\n");
419         } else {
420             av_log(avctx, AV_LOG_WARNING, "No channel layout specified. The encoder "
421                                                "will use Flac channel layout for "
422                                                "%d channels.\n", channels);
423         }
424     }
425 
426     ret = ff_lpc_init(&s->lpc_ctx, avctx->frame_size,
427                       s->options.max_prediction_order, FF_LPC_TYPE_LEVINSON);
428 
429     ff_bswapdsp_init(&s->bdsp);
430     ff_flacdsp_init(&s->flac_dsp, avctx->sample_fmt,
431                     avctx->bits_per_raw_sample);
432 
433     dprint_compression_options(s);
434 
435     return ret;
436 }
437 
438 
init_frame(FlacEncodeContext * s,int nb_samples)439 static void init_frame(FlacEncodeContext *s, int nb_samples)
440 {
441     int i, ch;
442     FlacFrame *frame;
443 
444     frame = &s->frame;
445 
446     for (i = 0; i < 16; i++) {
447         if (nb_samples == ff_flac_blocksize_table[i]) {
448             frame->blocksize  = ff_flac_blocksize_table[i];
449             frame->bs_code[0] = i;
450             frame->bs_code[1] = 0;
451             break;
452         }
453     }
454     if (i == 16) {
455         frame->blocksize = nb_samples;
456         if (frame->blocksize <= 256) {
457             frame->bs_code[0] = 6;
458             frame->bs_code[1] = frame->blocksize-1;
459         } else {
460             frame->bs_code[0] = 7;
461             frame->bs_code[1] = frame->blocksize-1;
462         }
463     }
464 
465     for (ch = 0; ch < s->channels; ch++) {
466         FlacSubframe *sub = &frame->subframes[ch];
467 
468         sub->wasted = 0;
469         sub->obits  = s->avctx->bits_per_raw_sample;
470 
471         if (sub->obits > 16)
472             sub->rc.coding_mode = CODING_MODE_RICE2;
473         else
474             sub->rc.coding_mode = CODING_MODE_RICE;
475     }
476 
477     frame->verbatim_only = 0;
478 }
479 
480 
481 /**
482  * Copy channel-interleaved input samples into separate subframes.
483  */
copy_samples(FlacEncodeContext * s,const void * samples)484 static void copy_samples(FlacEncodeContext *s, const void *samples)
485 {
486     int i, j, ch;
487     FlacFrame *frame;
488     int shift = av_get_bytes_per_sample(s->avctx->sample_fmt) * 8 -
489                 s->avctx->bits_per_raw_sample;
490 
491 #define COPY_SAMPLES(bits) do {                                     \
492     const int ## bits ## _t *samples0 = samples;                    \
493     frame = &s->frame;                                              \
494     for (i = 0, j = 0; i < frame->blocksize; i++)                   \
495         for (ch = 0; ch < s->channels; ch++, j++)                   \
496             frame->subframes[ch].samples[i] = samples0[j] >> shift; \
497 } while (0)
498 
499     if (s->avctx->sample_fmt == AV_SAMPLE_FMT_S16)
500         COPY_SAMPLES(16);
501     else
502         COPY_SAMPLES(32);
503 }
504 
505 
rice_count_exact(int32_t * res,int n,int k)506 static uint64_t rice_count_exact(int32_t *res, int n, int k)
507 {
508     int i;
509     uint64_t count = 0;
510 
511     for (i = 0; i < n; i++) {
512         int32_t v = -2 * res[i] - 1;
513         v ^= v >> 31;
514         count += (v >> k) + 1 + k;
515     }
516     return count;
517 }
518 
519 
subframe_count_exact(FlacEncodeContext * s,FlacSubframe * sub,int pred_order)520 static uint64_t subframe_count_exact(FlacEncodeContext *s, FlacSubframe *sub,
521                                      int pred_order)
522 {
523     int p, porder, psize;
524     int i, part_end;
525     uint64_t count = 0;
526 
527     /* subframe header */
528     count += 8;
529 
530     /* subframe */
531     if (sub->type == FLAC_SUBFRAME_CONSTANT) {
532         count += sub->obits;
533     } else if (sub->type == FLAC_SUBFRAME_VERBATIM) {
534         count += s->frame.blocksize * sub->obits;
535     } else {
536         /* warm-up samples */
537         count += pred_order * sub->obits;
538 
539         /* LPC coefficients */
540         if (sub->type == FLAC_SUBFRAME_LPC)
541             count += 4 + 5 + pred_order * s->options.lpc_coeff_precision;
542 
543         /* rice-encoded block */
544         count += 2;
545 
546         /* partition order */
547         porder = sub->rc.porder;
548         psize  = s->frame.blocksize >> porder;
549         count += 4;
550 
551         /* residual */
552         i        = pred_order;
553         part_end = psize;
554         for (p = 0; p < 1 << porder; p++) {
555             int k = sub->rc.params[p];
556             count += sub->rc.coding_mode;
557             count += rice_count_exact(&sub->residual[i], part_end - i, k);
558             i = part_end;
559             part_end = FFMIN(s->frame.blocksize, part_end + psize);
560         }
561     }
562 
563     return count;
564 }
565 
566 
567 #define rice_encode_count(sum, n, k) (((n)*((k)+1))+((sum-(n>>1))>>(k)))
568 
569 /**
570  * Solve for d/dk(rice_encode_count) = n-((sum-(n>>1))>>(k+1)) = 0.
571  */
find_optimal_param(uint64_t sum,int n,int max_param)572 static int find_optimal_param(uint64_t sum, int n, int max_param)
573 {
574     int k;
575     uint64_t sum2;
576 
577     if (sum <= n >> 1)
578         return 0;
579     sum2 = sum - (n >> 1);
580     k    = av_log2(av_clipl_int32(sum2 / n));
581     return FFMIN(k, max_param);
582 }
583 
584 
calc_optimal_rice_params(RiceContext * rc,int porder,uint64_t * sums,int n,int pred_order)585 static uint64_t calc_optimal_rice_params(RiceContext *rc, int porder,
586                                          uint64_t *sums, int n, int pred_order)
587 {
588     int i;
589     int k, cnt, part, max_param;
590     uint64_t all_bits;
591 
592     max_param = (1 << rc->coding_mode) - 2;
593 
594     part     = (1 << porder);
595     all_bits = 4 * part;
596 
597     cnt = (n >> porder) - pred_order;
598     for (i = 0; i < part; i++) {
599         k = find_optimal_param(sums[i], cnt, max_param);
600         rc->params[i] = k;
601         all_bits += rice_encode_count(sums[i], cnt, k);
602         cnt = n >> porder;
603     }
604 
605     rc->porder = porder;
606 
607     return all_bits;
608 }
609 
610 
calc_sums(int pmin,int pmax,uint32_t * data,int n,int pred_order,uint64_t sums[][MAX_PARTITIONS])611 static void calc_sums(int pmin, int pmax, uint32_t *data, int n, int pred_order,
612                       uint64_t sums[][MAX_PARTITIONS])
613 {
614     int i, j;
615     int parts;
616     uint32_t *res, *res_end;
617 
618     /* sums for highest level */
619     parts   = (1 << pmax);
620     res     = &data[pred_order];
621     res_end = &data[n >> pmax];
622     for (i = 0; i < parts; i++) {
623         uint64_t sum = 0;
624         while (res < res_end)
625             sum += *(res++);
626         sums[pmax][i] = sum;
627         res_end += n >> pmax;
628     }
629     /* sums for lower levels */
630     for (i = pmax - 1; i >= pmin; i--) {
631         parts = (1 << i);
632         for (j = 0; j < parts; j++)
633             sums[i][j] = sums[i+1][2*j] + sums[i+1][2*j+1];
634     }
635 }
636 
637 
calc_rice_params(RiceContext * rc,int pmin,int pmax,int32_t * data,int n,int pred_order)638 static uint64_t calc_rice_params(RiceContext *rc, int pmin, int pmax,
639                                  int32_t *data, int n, int pred_order)
640 {
641     int i;
642     uint64_t bits[MAX_PARTITION_ORDER+1];
643     int opt_porder;
644     RiceContext tmp_rc;
645     uint32_t *udata;
646     uint64_t sums[MAX_PARTITION_ORDER+1][MAX_PARTITIONS];
647 
648     av_assert1(pmin >= 0 && pmin <= MAX_PARTITION_ORDER);
649     av_assert1(pmax >= 0 && pmax <= MAX_PARTITION_ORDER);
650     av_assert1(pmin <= pmax);
651 
652     tmp_rc.coding_mode = rc->coding_mode;
653 
654     udata = av_malloc_array(n,  sizeof(uint32_t));
655     for (i = 0; i < n; i++)
656         udata[i] = (2*data[i]) ^ (data[i]>>31);
657 
658     calc_sums(pmin, pmax, udata, n, pred_order, sums);
659 
660     opt_porder = pmin;
661     bits[pmin] = UINT32_MAX;
662     for (i = pmin; i <= pmax; i++) {
663         bits[i] = calc_optimal_rice_params(&tmp_rc, i, sums[i], n, pred_order);
664         if (bits[i] <= bits[opt_porder]) {
665             opt_porder = i;
666             *rc = tmp_rc;
667         }
668     }
669 
670     av_freep(&udata);
671     return bits[opt_porder];
672 }
673 
674 
get_max_p_order(int max_porder,int n,int order)675 static int get_max_p_order(int max_porder, int n, int order)
676 {
677     int porder = FFMIN(max_porder, av_log2(n^(n-1)));
678     if (order > 0)
679         porder = FFMIN(porder, av_log2(n/order));
680     return porder;
681 }
682 
683 
find_subframe_rice_params(FlacEncodeContext * s,FlacSubframe * sub,int pred_order)684 static uint64_t find_subframe_rice_params(FlacEncodeContext *s,
685                                           FlacSubframe *sub, int pred_order)
686 {
687     int pmin = get_max_p_order(s->options.min_partition_order,
688                                s->frame.blocksize, pred_order);
689     int pmax = get_max_p_order(s->options.max_partition_order,
690                                s->frame.blocksize, pred_order);
691 
692     uint64_t bits = 8 + pred_order * sub->obits + 2 + sub->rc.coding_mode;
693     if (sub->type == FLAC_SUBFRAME_LPC)
694         bits += 4 + 5 + pred_order * s->options.lpc_coeff_precision;
695     bits += calc_rice_params(&sub->rc, pmin, pmax, sub->residual,
696                              s->frame.blocksize, pred_order);
697     return bits;
698 }
699 
700 
encode_residual_fixed(int32_t * res,const int32_t * smp,int n,int order)701 static void encode_residual_fixed(int32_t *res, const int32_t *smp, int n,
702                                   int order)
703 {
704     int i;
705 
706     for (i = 0; i < order; i++)
707         res[i] = smp[i];
708 
709     if (order == 0) {
710         for (i = order; i < n; i++)
711             res[i] = smp[i];
712     } else if (order == 1) {
713         for (i = order; i < n; i++)
714             res[i] = smp[i] - smp[i-1];
715     } else if (order == 2) {
716         int a = smp[order-1] - smp[order-2];
717         for (i = order; i < n; i += 2) {
718             int b    = smp[i  ] - smp[i-1];
719             res[i]   = b - a;
720             a        = smp[i+1] - smp[i  ];
721             res[i+1] = a - b;
722         }
723     } else if (order == 3) {
724         int a = smp[order-1] -   smp[order-2];
725         int c = smp[order-1] - 2*smp[order-2] + smp[order-3];
726         for (i = order; i < n; i += 2) {
727             int b    = smp[i  ] - smp[i-1];
728             int d    = b - a;
729             res[i]   = d - c;
730             a        = smp[i+1] - smp[i  ];
731             c        = a - b;
732             res[i+1] = c - d;
733         }
734     } else {
735         int a = smp[order-1] -   smp[order-2];
736         int c = smp[order-1] - 2*smp[order-2] +   smp[order-3];
737         int e = smp[order-1] - 3*smp[order-2] + 3*smp[order-3] - smp[order-4];
738         for (i = order; i < n; i += 2) {
739             int b    = smp[i  ] - smp[i-1];
740             int d    = b - a;
741             int f    = d - c;
742             res[i  ] = f - e;
743             a        = smp[i+1] - smp[i  ];
744             c        = a - b;
745             e        = c - d;
746             res[i+1] = e - f;
747         }
748     }
749 }
750 
751 
encode_residual_ch(FlacEncodeContext * s,int ch)752 static int encode_residual_ch(FlacEncodeContext *s, int ch)
753 {
754     int i, n;
755     int min_order, max_order, opt_order, omethod;
756     FlacFrame *frame;
757     FlacSubframe *sub;
758     int32_t coefs[MAX_LPC_ORDER][MAX_LPC_ORDER];
759     int shift[MAX_LPC_ORDER];
760     int32_t *res, *smp;
761 
762     frame = &s->frame;
763     sub   = &frame->subframes[ch];
764     res   = sub->residual;
765     smp   = sub->samples;
766     n     = frame->blocksize;
767 
768     /* CONSTANT */
769     for (i = 1; i < n; i++)
770         if(smp[i] != smp[0])
771             break;
772     if (i == n) {
773         sub->type = sub->type_code = FLAC_SUBFRAME_CONSTANT;
774         res[0] = smp[0];
775         return subframe_count_exact(s, sub, 0);
776     }
777 
778     /* VERBATIM */
779     if (frame->verbatim_only || n < 5) {
780         sub->type = sub->type_code = FLAC_SUBFRAME_VERBATIM;
781         memcpy(res, smp, n * sizeof(int32_t));
782         return subframe_count_exact(s, sub, 0);
783     }
784 
785     min_order  = s->options.min_prediction_order;
786     max_order  = s->options.max_prediction_order;
787     omethod    = s->options.prediction_order_method;
788 
789     /* FIXED */
790     sub->type = FLAC_SUBFRAME_FIXED;
791     if (s->options.lpc_type == FF_LPC_TYPE_NONE  ||
792         s->options.lpc_type == FF_LPC_TYPE_FIXED || n <= max_order) {
793         uint64_t bits[MAX_FIXED_ORDER+1];
794         if (max_order > MAX_FIXED_ORDER)
795             max_order = MAX_FIXED_ORDER;
796         opt_order = 0;
797         bits[0]   = UINT32_MAX;
798         for (i = min_order; i <= max_order; i++) {
799             encode_residual_fixed(res, smp, n, i);
800             bits[i] = find_subframe_rice_params(s, sub, i);
801             if (bits[i] < bits[opt_order])
802                 opt_order = i;
803         }
804         sub->order     = opt_order;
805         sub->type_code = sub->type | sub->order;
806         if (sub->order != max_order) {
807             encode_residual_fixed(res, smp, n, sub->order);
808             find_subframe_rice_params(s, sub, sub->order);
809         }
810         return subframe_count_exact(s, sub, sub->order);
811     }
812 
813     /* LPC */
814     sub->type = FLAC_SUBFRAME_LPC;
815     opt_order = ff_lpc_calc_coefs(&s->lpc_ctx, smp, n, min_order, max_order,
816                                   s->options.lpc_coeff_precision, coefs, shift, s->options.lpc_type,
817                                   s->options.lpc_passes, omethod,
818                                   MAX_LPC_SHIFT, 0);
819 
820     if (omethod == ORDER_METHOD_2LEVEL ||
821         omethod == ORDER_METHOD_4LEVEL ||
822         omethod == ORDER_METHOD_8LEVEL) {
823         int levels = 1 << omethod;
824         uint64_t bits[1 << ORDER_METHOD_8LEVEL];
825         int order       = -1;
826         int opt_index   = levels-1;
827         opt_order       = max_order-1;
828         bits[opt_index] = UINT32_MAX;
829         for (i = levels-1; i >= 0; i--) {
830             int last_order = order;
831             order = min_order + (((max_order-min_order+1) * (i+1)) / levels)-1;
832             order = av_clip(order, min_order - 1, max_order - 1);
833             if (order == last_order)
834                 continue;
835             s->flac_dsp.lpc_encode(res, smp, n, order+1, coefs[order],
836                                    shift[order]);
837             bits[i] = find_subframe_rice_params(s, sub, order+1);
838             if (bits[i] < bits[opt_index]) {
839                 opt_index = i;
840                 opt_order = order;
841             }
842         }
843         opt_order++;
844     } else if (omethod == ORDER_METHOD_SEARCH) {
845         // brute-force optimal order search
846         uint64_t bits[MAX_LPC_ORDER];
847         opt_order = 0;
848         bits[0]   = UINT32_MAX;
849         for (i = min_order-1; i < max_order; i++) {
850             s->flac_dsp.lpc_encode(res, smp, n, i+1, coefs[i], shift[i]);
851             bits[i] = find_subframe_rice_params(s, sub, i+1);
852             if (bits[i] < bits[opt_order])
853                 opt_order = i;
854         }
855         opt_order++;
856     } else if (omethod == ORDER_METHOD_LOG) {
857         uint64_t bits[MAX_LPC_ORDER];
858         int step;
859 
860         opt_order = min_order - 1 + (max_order-min_order)/3;
861         memset(bits, -1, sizeof(bits));
862 
863         for (step = 16; step; step >>= 1) {
864             int last = opt_order;
865             for (i = last-step; i <= last+step; i += step) {
866                 if (i < min_order-1 || i >= max_order || bits[i] < UINT32_MAX)
867                     continue;
868                 s->flac_dsp.lpc_encode(res, smp, n, i+1, coefs[i], shift[i]);
869                 bits[i] = find_subframe_rice_params(s, sub, i+1);
870                 if (bits[i] < bits[opt_order])
871                     opt_order = i;
872             }
873         }
874         opt_order++;
875     }
876 
877     sub->order     = opt_order;
878     sub->type_code = sub->type | (sub->order-1);
879     sub->shift     = shift[sub->order-1];
880     for (i = 0; i < sub->order; i++)
881         sub->coefs[i] = coefs[sub->order-1][i];
882 
883     s->flac_dsp.lpc_encode(res, smp, n, sub->order, sub->coefs, sub->shift);
884 
885     find_subframe_rice_params(s, sub, sub->order);
886 
887     return subframe_count_exact(s, sub, sub->order);
888 }
889 
890 
count_frame_header(FlacEncodeContext * s)891 static int count_frame_header(FlacEncodeContext *s)
892 {
893     uint8_t av_unused tmp;
894     int count;
895 
896     /*
897     <14> Sync code
898     <1>  Reserved
899     <1>  Blocking strategy
900     <4>  Block size in inter-channel samples
901     <4>  Sample rate
902     <4>  Channel assignment
903     <3>  Sample size in bits
904     <1>  Reserved
905     */
906     count = 32;
907 
908     /* coded frame number */
909     PUT_UTF8(s->frame_count, tmp, count += 8;)
910 
911     /* explicit block size */
912     if (s->frame.bs_code[0] == 6)
913         count += 8;
914     else if (s->frame.bs_code[0] == 7)
915         count += 16;
916 
917     /* explicit sample rate */
918     count += ((s->sr_code[0] == 12) + (s->sr_code[0] > 12)) * 8;
919 
920     /* frame header CRC-8 */
921     count += 8;
922 
923     return count;
924 }
925 
926 
encode_frame(FlacEncodeContext * s)927 static int encode_frame(FlacEncodeContext *s)
928 {
929     int ch;
930     uint64_t count;
931 
932     count = count_frame_header(s);
933 
934     for (ch = 0; ch < s->channels; ch++)
935         count += encode_residual_ch(s, ch);
936 
937     count += (8 - (count & 7)) & 7; // byte alignment
938     count += 16;                    // CRC-16
939 
940     count >>= 3;
941     if (count > INT_MAX)
942         return AVERROR_BUG;
943     return count;
944 }
945 
946 
remove_wasted_bits(FlacEncodeContext * s)947 static void remove_wasted_bits(FlacEncodeContext *s)
948 {
949     int ch, i;
950 
951     for (ch = 0; ch < s->channels; ch++) {
952         FlacSubframe *sub = &s->frame.subframes[ch];
953         int32_t v         = 0;
954 
955         for (i = 0; i < s->frame.blocksize; i++) {
956             v |= sub->samples[i];
957             if (v & 1)
958                 break;
959         }
960 
961         if (v && !(v & 1)) {
962             v = av_ctz(v);
963 
964             for (i = 0; i < s->frame.blocksize; i++)
965                 sub->samples[i] >>= v;
966 
967             sub->wasted = v;
968             sub->obits -= v;
969 
970             /* for 24-bit, check if removing wasted bits makes the range better
971                suited for using RICE instead of RICE2 for entropy coding */
972             if (sub->obits <= 17)
973                 sub->rc.coding_mode = CODING_MODE_RICE;
974         }
975     }
976 }
977 
978 
estimate_stereo_mode(int32_t * left_ch,int32_t * right_ch,int n,int max_rice_param)979 static int estimate_stereo_mode(int32_t *left_ch, int32_t *right_ch, int n,
980                                 int max_rice_param)
981 {
982     int i, best;
983     int32_t lt, rt;
984     uint64_t sum[4];
985     uint64_t score[4];
986     int k;
987 
988     /* calculate sum of 2nd order residual for each channel */
989     sum[0] = sum[1] = sum[2] = sum[3] = 0;
990     for (i = 2; i < n; i++) {
991         lt = left_ch[i]  - 2*left_ch[i-1]  + left_ch[i-2];
992         rt = right_ch[i] - 2*right_ch[i-1] + right_ch[i-2];
993         sum[2] += FFABS((lt + rt) >> 1);
994         sum[3] += FFABS(lt - rt);
995         sum[0] += FFABS(lt);
996         sum[1] += FFABS(rt);
997     }
998     /* estimate bit counts */
999     for (i = 0; i < 4; i++) {
1000         k      = find_optimal_param(2 * sum[i], n, max_rice_param);
1001         sum[i] = rice_encode_count( 2 * sum[i], n, k);
1002     }
1003 
1004     /* calculate score for each mode */
1005     score[0] = sum[0] + sum[1];
1006     score[1] = sum[0] + sum[3];
1007     score[2] = sum[1] + sum[3];
1008     score[3] = sum[2] + sum[3];
1009 
1010     /* return mode with lowest score */
1011     best = 0;
1012     for (i = 1; i < 4; i++)
1013         if (score[i] < score[best])
1014             best = i;
1015 
1016     return best;
1017 }
1018 
1019 
1020 /**
1021  * Perform stereo channel decorrelation.
1022  */
channel_decorrelation(FlacEncodeContext * s)1023 static void channel_decorrelation(FlacEncodeContext *s)
1024 {
1025     FlacFrame *frame;
1026     int32_t *left, *right;
1027     int i, n;
1028 
1029     frame = &s->frame;
1030     n     = frame->blocksize;
1031     left  = frame->subframes[0].samples;
1032     right = frame->subframes[1].samples;
1033 
1034     if (s->channels != 2) {
1035         frame->ch_mode = FLAC_CHMODE_INDEPENDENT;
1036         return;
1037     }
1038 
1039     if (s->options.ch_mode < 0) {
1040         int max_rice_param = (1 << frame->subframes[0].rc.coding_mode) - 2;
1041         frame->ch_mode = estimate_stereo_mode(left, right, n, max_rice_param);
1042     } else
1043         frame->ch_mode = s->options.ch_mode;
1044 
1045     /* perform decorrelation and adjust bits-per-sample */
1046     if (frame->ch_mode == FLAC_CHMODE_INDEPENDENT)
1047         return;
1048     if (frame->ch_mode == FLAC_CHMODE_MID_SIDE) {
1049         int32_t tmp;
1050         for (i = 0; i < n; i++) {
1051             tmp      = left[i];
1052             left[i]  = (tmp + right[i]) >> 1;
1053             right[i] =  tmp - right[i];
1054         }
1055         frame->subframes[1].obits++;
1056     } else if (frame->ch_mode == FLAC_CHMODE_LEFT_SIDE) {
1057         for (i = 0; i < n; i++)
1058             right[i] = left[i] - right[i];
1059         frame->subframes[1].obits++;
1060     } else {
1061         for (i = 0; i < n; i++)
1062             left[i] -= right[i];
1063         frame->subframes[0].obits++;
1064     }
1065 }
1066 
1067 
write_utf8(PutBitContext * pb,uint32_t val)1068 static void write_utf8(PutBitContext *pb, uint32_t val)
1069 {
1070     uint8_t tmp;
1071     PUT_UTF8(val, tmp, put_bits(pb, 8, tmp);)
1072 }
1073 
1074 
write_frame_header(FlacEncodeContext * s)1075 static void write_frame_header(FlacEncodeContext *s)
1076 {
1077     FlacFrame *frame;
1078     int crc;
1079 
1080     frame = &s->frame;
1081 
1082     put_bits(&s->pb, 16, 0xFFF8);
1083     put_bits(&s->pb, 4, frame->bs_code[0]);
1084     put_bits(&s->pb, 4, s->sr_code[0]);
1085 
1086     if (frame->ch_mode == FLAC_CHMODE_INDEPENDENT)
1087         put_bits(&s->pb, 4, s->channels-1);
1088     else
1089         put_bits(&s->pb, 4, frame->ch_mode + FLAC_MAX_CHANNELS - 1);
1090 
1091     put_bits(&s->pb, 3, s->bps_code);
1092     put_bits(&s->pb, 1, 0);
1093     write_utf8(&s->pb, s->frame_count);
1094 
1095     if (frame->bs_code[0] == 6)
1096         put_bits(&s->pb, 8, frame->bs_code[1]);
1097     else if (frame->bs_code[0] == 7)
1098         put_bits(&s->pb, 16, frame->bs_code[1]);
1099 
1100     if (s->sr_code[0] == 12)
1101         put_bits(&s->pb, 8, s->sr_code[1]);
1102     else if (s->sr_code[0] > 12)
1103         put_bits(&s->pb, 16, s->sr_code[1]);
1104 
1105     flush_put_bits(&s->pb);
1106     crc = av_crc(av_crc_get_table(AV_CRC_8_ATM), 0, s->pb.buf,
1107                  put_bits_count(&s->pb) >> 3);
1108     put_bits(&s->pb, 8, crc);
1109 }
1110 
1111 
write_subframes(FlacEncodeContext * s)1112 static void write_subframes(FlacEncodeContext *s)
1113 {
1114     int ch;
1115 
1116     for (ch = 0; ch < s->channels; ch++) {
1117         FlacSubframe *sub = &s->frame.subframes[ch];
1118         int i, p, porder, psize;
1119         int32_t *part_end;
1120         int32_t *res       =  sub->residual;
1121         int32_t *frame_end = &sub->residual[s->frame.blocksize];
1122 
1123         /* subframe header */
1124         put_bits(&s->pb, 1, 0);
1125         put_bits(&s->pb, 6, sub->type_code);
1126         put_bits(&s->pb, 1, !!sub->wasted);
1127         if (sub->wasted)
1128             put_bits(&s->pb, sub->wasted, 1);
1129 
1130         /* subframe */
1131         if (sub->type == FLAC_SUBFRAME_CONSTANT) {
1132             put_sbits(&s->pb, sub->obits, res[0]);
1133         } else if (sub->type == FLAC_SUBFRAME_VERBATIM) {
1134             while (res < frame_end)
1135                 put_sbits(&s->pb, sub->obits, *res++);
1136         } else {
1137             /* warm-up samples */
1138             for (i = 0; i < sub->order; i++)
1139                 put_sbits(&s->pb, sub->obits, *res++);
1140 
1141             /* LPC coefficients */
1142             if (sub->type == FLAC_SUBFRAME_LPC) {
1143                 int cbits = s->options.lpc_coeff_precision;
1144                 put_bits( &s->pb, 4, cbits-1);
1145                 put_sbits(&s->pb, 5, sub->shift);
1146                 for (i = 0; i < sub->order; i++)
1147                     put_sbits(&s->pb, cbits, sub->coefs[i]);
1148             }
1149 
1150             /* rice-encoded block */
1151             put_bits(&s->pb, 2, sub->rc.coding_mode - 4);
1152 
1153             /* partition order */
1154             porder  = sub->rc.porder;
1155             psize   = s->frame.blocksize >> porder;
1156             put_bits(&s->pb, 4, porder);
1157 
1158             /* residual */
1159             part_end  = &sub->residual[psize];
1160             for (p = 0; p < 1 << porder; p++) {
1161                 int k = sub->rc.params[p];
1162                 put_bits(&s->pb, sub->rc.coding_mode, k);
1163                 while (res < part_end)
1164                     set_sr_golomb_flac(&s->pb, *res++, k, INT32_MAX, 0);
1165                 part_end = FFMIN(frame_end, part_end + psize);
1166             }
1167         }
1168     }
1169 }
1170 
1171 
write_frame_footer(FlacEncodeContext * s)1172 static void write_frame_footer(FlacEncodeContext *s)
1173 {
1174     int crc;
1175     flush_put_bits(&s->pb);
1176     crc = av_bswap16(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, s->pb.buf,
1177                             put_bits_count(&s->pb)>>3));
1178     put_bits(&s->pb, 16, crc);
1179     flush_put_bits(&s->pb);
1180 }
1181 
1182 
write_frame(FlacEncodeContext * s,AVPacket * avpkt)1183 static int write_frame(FlacEncodeContext *s, AVPacket *avpkt)
1184 {
1185     init_put_bits(&s->pb, avpkt->data, avpkt->size);
1186     write_frame_header(s);
1187     write_subframes(s);
1188     write_frame_footer(s);
1189     return put_bits_count(&s->pb) >> 3;
1190 }
1191 
1192 
update_md5_sum(FlacEncodeContext * s,const void * samples)1193 static int update_md5_sum(FlacEncodeContext *s, const void *samples)
1194 {
1195     const uint8_t *buf;
1196     int buf_size = s->frame.blocksize * s->channels *
1197                    ((s->avctx->bits_per_raw_sample + 7) / 8);
1198 
1199     if (s->avctx->bits_per_raw_sample > 16 || HAVE_BIGENDIAN) {
1200         av_fast_malloc(&s->md5_buffer, &s->md5_buffer_size, buf_size);
1201         if (!s->md5_buffer)
1202             return AVERROR(ENOMEM);
1203     }
1204 
1205     if (s->avctx->bits_per_raw_sample <= 16) {
1206         buf = (const uint8_t *)samples;
1207 #if HAVE_BIGENDIAN
1208         s->bdsp.bswap16_buf((uint16_t *) s->md5_buffer,
1209                             (const uint16_t *) samples, buf_size / 2);
1210         buf = s->md5_buffer;
1211 #endif
1212     } else {
1213         int i;
1214         const int32_t *samples0 = samples;
1215         uint8_t *tmp            = s->md5_buffer;
1216 
1217         for (i = 0; i < s->frame.blocksize * s->channels; i++) {
1218             int32_t v = samples0[i] >> 8;
1219             *tmp++    = (v      ) & 0xFF;
1220             *tmp++    = (v >>  8) & 0xFF;
1221             *tmp++    = (v >> 16) & 0xFF;
1222         }
1223         buf = s->md5_buffer;
1224     }
1225     av_md5_update(s->md5ctx, buf, buf_size);
1226 
1227     return 0;
1228 }
1229 
1230 
flac_encode_frame(AVCodecContext * avctx,AVPacket * avpkt,const AVFrame * frame,int * got_packet_ptr)1231 static int flac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
1232                              const AVFrame *frame, int *got_packet_ptr)
1233 {
1234     FlacEncodeContext *s;
1235     int frame_bytes, out_bytes, ret;
1236 
1237     s = avctx->priv_data;
1238 
1239     /* when the last block is reached, update the header in extradata */
1240     if (!frame) {
1241         s->max_framesize = s->max_encoded_framesize;
1242         av_md5_final(s->md5ctx, s->md5sum);
1243         write_streaminfo(s, avctx->extradata);
1244 
1245         if (avctx->side_data_only_packets && !s->flushed) {
1246             uint8_t *side_data = av_packet_new_side_data(avpkt, AV_PKT_DATA_NEW_EXTRADATA,
1247                                                          avctx->extradata_size);
1248             if (!side_data)
1249                 return AVERROR(ENOMEM);
1250             memcpy(side_data, avctx->extradata, avctx->extradata_size);
1251 
1252             avpkt->pts = s->next_pts;
1253 
1254             *got_packet_ptr = 1;
1255             s->flushed = 1;
1256         }
1257 
1258         return 0;
1259     }
1260 
1261     /* change max_framesize for small final frame */
1262     if (frame->nb_samples < s->frame.blocksize) {
1263         s->max_framesize = ff_flac_get_max_frame_size(frame->nb_samples,
1264                                                       s->channels,
1265                                                       avctx->bits_per_raw_sample);
1266     }
1267 
1268     init_frame(s, frame->nb_samples);
1269 
1270     copy_samples(s, frame->data[0]);
1271 
1272     channel_decorrelation(s);
1273 
1274     remove_wasted_bits(s);
1275 
1276     frame_bytes = encode_frame(s);
1277 
1278     /* Fall back on verbatim mode if the compressed frame is larger than it
1279        would be if encoded uncompressed. */
1280     if (frame_bytes < 0 || frame_bytes > s->max_framesize) {
1281         s->frame.verbatim_only = 1;
1282         frame_bytes = encode_frame(s);
1283         if (frame_bytes < 0) {
1284             av_log(avctx, AV_LOG_ERROR, "Bad frame count\n");
1285             return frame_bytes;
1286         }
1287     }
1288 
1289     if ((ret = ff_alloc_packet2(avctx, avpkt, frame_bytes)) < 0)
1290         return ret;
1291 
1292     out_bytes = write_frame(s, avpkt);
1293 
1294     s->frame_count++;
1295     s->sample_count += frame->nb_samples;
1296     if ((ret = update_md5_sum(s, frame->data[0])) < 0) {
1297         av_log(avctx, AV_LOG_ERROR, "Error updating MD5 checksum\n");
1298         return ret;
1299     }
1300     if (out_bytes > s->max_encoded_framesize)
1301         s->max_encoded_framesize = out_bytes;
1302     if (out_bytes < s->min_framesize)
1303         s->min_framesize = out_bytes;
1304 
1305     avpkt->pts      = frame->pts;
1306     avpkt->duration = ff_samples_to_time_base(avctx, frame->nb_samples);
1307     avpkt->size     = out_bytes;
1308 
1309     s->next_pts = avpkt->pts + avpkt->duration;
1310 
1311     *got_packet_ptr = 1;
1312     return 0;
1313 }
1314 
1315 
flac_encode_close(AVCodecContext * avctx)1316 static av_cold int flac_encode_close(AVCodecContext *avctx)
1317 {
1318     if (avctx->priv_data) {
1319         FlacEncodeContext *s = avctx->priv_data;
1320         av_freep(&s->md5ctx);
1321         av_freep(&s->md5_buffer);
1322         ff_lpc_end(&s->lpc_ctx);
1323     }
1324     av_freep(&avctx->extradata);
1325     avctx->extradata_size = 0;
1326     return 0;
1327 }
1328 
1329 #define FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
1330 static const AVOption options[] = {
1331 	{ "lpc_coeff_precision", "LPC coefficient precision", offsetof(FlacEncodeContext, options.lpc_coeff_precision), AV_OPT_TYPE_INT, {.i64 = 15 }, 0, MAX_LPC_PRECISION, FLAGS },
1332     { "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" },
1333     { "none",     NULL, 0, AV_OPT_TYPE_CONST, {.i64 = FF_LPC_TYPE_NONE },     INT_MIN, INT_MAX, FLAGS, "lpc_type" },
1334     { "fixed",    NULL, 0, AV_OPT_TYPE_CONST, {.i64 = FF_LPC_TYPE_FIXED },    INT_MIN, INT_MAX, FLAGS, "lpc_type" },
1335     { "levinson", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = FF_LPC_TYPE_LEVINSON }, INT_MIN, INT_MAX, FLAGS, "lpc_type" },
1336     { "cholesky", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = FF_LPC_TYPE_CHOLESKY }, INT_MIN, INT_MAX, FLAGS, "lpc_type" },
1337     { "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 },
1338     { "min_partition_order",  NULL, offsetof(FlacEncodeContext, options.min_partition_order),  AV_OPT_TYPE_INT, {.i64 = -1 },      -1, MAX_PARTITION_ORDER, FLAGS },
1339     { "max_partition_order",  NULL, offsetof(FlacEncodeContext, options.max_partition_order),  AV_OPT_TYPE_INT, {.i64 = -1 },      -1, MAX_PARTITION_ORDER, FLAGS },
1340     { "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" },
1341     { "estimation", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_EST },    INT_MIN, INT_MAX, FLAGS, "predm" },
1342     { "2level",     NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_2LEVEL }, INT_MIN, INT_MAX, FLAGS, "predm" },
1343     { "4level",     NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_4LEVEL }, INT_MIN, INT_MAX, FLAGS, "predm" },
1344     { "8level",     NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_8LEVEL }, INT_MIN, INT_MAX, FLAGS, "predm" },
1345     { "search",     NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_SEARCH }, INT_MIN, INT_MAX, FLAGS, "predm" },
1346     { "log",        NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_LOG },    INT_MIN, INT_MAX, FLAGS, "predm" },
1347     { "ch_mode", "Stereo decorrelation mode", offsetof(FlacEncodeContext, options.ch_mode), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, FLAC_CHMODE_MID_SIDE, FLAGS, "ch_mode" },
1348     { "auto",       NULL, 0, AV_OPT_TYPE_CONST, { .i64 = -1                      }, INT_MIN, INT_MAX, FLAGS, "ch_mode" },
1349     { "indep",      NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FLAC_CHMODE_INDEPENDENT }, INT_MIN, INT_MAX, FLAGS, "ch_mode" },
1350     { "left_side",  NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FLAC_CHMODE_LEFT_SIDE   }, INT_MIN, INT_MAX, FLAGS, "ch_mode" },
1351     { "right_side", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FLAC_CHMODE_RIGHT_SIDE  }, INT_MIN, INT_MAX, FLAGS, "ch_mode" },
1352     { "mid_side",   NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FLAC_CHMODE_MID_SIDE    }, INT_MIN, INT_MAX, FLAGS, "ch_mode" },
1353 	{ NULL },
1354 };
1355 
1356 static const AVClass flac_encoder_class = {
1357     "FLAC encoder",
1358     av_default_item_name,
1359     options,
1360     LIBAVUTIL_VERSION_INT,
1361 };
1362 
1363 AVCodec ff_flac_encoder = {
1364 	.name           = "flac",
1365     .long_name      = NULL_IF_CONFIG_SMALL("FLAC (Free Lossless Audio Codec)"),
1366     .type           = AVMEDIA_TYPE_AUDIO,
1367     .id             = AV_CODEC_ID_FLAC,
1368     .priv_data_size = sizeof(FlacEncodeContext),
1369     .init           = flac_encode_init,
1370     .encode2        = flac_encode_frame,
1371     .close          = flac_encode_close,
1372     .capabilities   = CODEC_CAP_SMALL_LAST_FRAME | CODEC_CAP_DELAY | CODEC_CAP_LOSSLESS,
1373     .sample_fmts    = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S16,
1374                                                      AV_SAMPLE_FMT_S32,
1375                                                      AV_SAMPLE_FMT_NONE },
1376     .priv_class     = &flac_encoder_class,
1377 };
1378