Statistics
| Branch: | Revision:

ffmpeg / libavcodec / flacenc.c @ 2912e87a

History | View | Annotate | Download (40.4 KB)

1
/**
2
 * FLAC audio encoder
3
 * Copyright (c) 2006  Justin Ruggles <justin.ruggles@gmail.com>
4
 *
5
 * This file is part of Libav.
6
 *
7
 * Libav 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
 * Libav 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 Libav; 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/crc.h"
23
#include "libavutil/md5.h"
24
#include "avcodec.h"
25
#include "get_bits.h"
26
#include "golomb.h"
27
#include "lpc.h"
28
#include "flac.h"
29
#include "flacdata.h"
30

    
31
#define FLAC_SUBFRAME_CONSTANT  0
32
#define FLAC_SUBFRAME_VERBATIM  1
33
#define FLAC_SUBFRAME_FIXED     8
34
#define FLAC_SUBFRAME_LPC      32
35

    
36
#define MAX_FIXED_ORDER     4
37
#define MAX_PARTITION_ORDER 8
38
#define MAX_PARTITIONS     (1 << MAX_PARTITION_ORDER)
39
#define MAX_LPC_PRECISION  15
40
#define MAX_LPC_SHIFT      15
41
#define MAX_RICE_PARAM     14
42

    
43
typedef struct CompressionOptions {
44
    int compression_level;
45
    int block_time_ms;
46
    enum AVLPCType lpc_type;
47
    int lpc_passes;
48
    int lpc_coeff_precision;
49
    int min_prediction_order;
50
    int max_prediction_order;
51
    int prediction_order_method;
52
    int min_partition_order;
53
    int max_partition_order;
54
} CompressionOptions;
55

    
56
typedef struct RiceContext {
57
    int porder;
58
    int params[MAX_PARTITIONS];
59
} RiceContext;
60

    
61
typedef struct FlacSubframe {
62
    int type;
63
    int type_code;
64
    int obits;
65
    int order;
66
    int32_t coefs[MAX_LPC_ORDER];
67
    int shift;
68
    RiceContext rc;
69
    int32_t samples[FLAC_MAX_BLOCKSIZE];
70
    int32_t residual[FLAC_MAX_BLOCKSIZE+1];
71
} FlacSubframe;
72

    
73
typedef struct FlacFrame {
74
    FlacSubframe subframes[FLAC_MAX_CHANNELS];
75
    int blocksize;
76
    int bs_code[2];
77
    uint8_t crc8;
78
    int ch_mode;
79
    int verbatim_only;
80
} FlacFrame;
81

    
82
typedef struct FlacEncodeContext {
83
    PutBitContext pb;
84
    int channels;
85
    int samplerate;
86
    int sr_code[2];
87
    int max_blocksize;
88
    int min_framesize;
89
    int max_framesize;
90
    int max_encoded_framesize;
91
    uint32_t frame_count;
92
    uint64_t sample_count;
93
    uint8_t md5sum[16];
94
    FlacFrame frame;
95
    CompressionOptions options;
96
    AVCodecContext *avctx;
97
    LPCContext lpc_ctx;
98
    struct AVMD5 *md5ctx;
99
} FlacEncodeContext;
100

    
101

    
102
/**
103
 * Write streaminfo metadata block to byte array.
104
 */
105
static void write_streaminfo(FlacEncodeContext *s, uint8_t *header)
106
{
107
    PutBitContext pb;
108

    
109
    memset(header, 0, FLAC_STREAMINFO_SIZE);
110
    init_put_bits(&pb, header, FLAC_STREAMINFO_SIZE);
111

    
112
    /* streaminfo metadata block */
113
    put_bits(&pb, 16, s->max_blocksize);
114
    put_bits(&pb, 16, s->max_blocksize);
115
    put_bits(&pb, 24, s->min_framesize);
116
    put_bits(&pb, 24, s->max_framesize);
117
    put_bits(&pb, 20, s->samplerate);
118
    put_bits(&pb, 3, s->channels-1);
119
    put_bits(&pb, 5, 15);       /* bits per sample - 1 */
120
    /* write 36-bit sample count in 2 put_bits() calls */
121
    put_bits(&pb, 24, (s->sample_count & 0xFFFFFF000LL) >> 12);
122
    put_bits(&pb, 12,  s->sample_count & 0x000000FFFLL);
123
    flush_put_bits(&pb);
124
    memcpy(&header[18], s->md5sum, 16);
125
}
126

    
127

    
128
/**
129
 * Set blocksize based on samplerate.
130
 * Choose the closest predefined blocksize >= BLOCK_TIME_MS milliseconds.
131
 */
132
static int select_blocksize(int samplerate, int block_time_ms)
133
{
134
    int i;
135
    int target;
136
    int blocksize;
137

    
138
    assert(samplerate > 0);
139
    blocksize = ff_flac_blocksize_table[1];
140
    target    = (samplerate * block_time_ms) / 1000;
141
    for (i = 0; i < 16; i++) {
142
        if (target >= ff_flac_blocksize_table[i] &&
143
            ff_flac_blocksize_table[i] > blocksize) {
144
            blocksize = ff_flac_blocksize_table[i];
145
        }
146
    }
147
    return blocksize;
148
}
149

    
150

    
151
static av_cold void dprint_compression_options(FlacEncodeContext *s)
152
{
153
    AVCodecContext     *avctx = s->avctx;
154
    CompressionOptions *opt   = &s->options;
155

    
156
    av_log(avctx, AV_LOG_DEBUG, " compression: %d\n", opt->compression_level);
157

    
158
    switch (opt->lpc_type) {
159
    case AV_LPC_TYPE_NONE:
160
        av_log(avctx, AV_LOG_DEBUG, " lpc type: None\n");
161
        break;
162
    case AV_LPC_TYPE_FIXED:
163
        av_log(avctx, AV_LOG_DEBUG, " lpc type: Fixed pre-defined coefficients\n");
164
        break;
165
    case AV_LPC_TYPE_LEVINSON:
166
        av_log(avctx, AV_LOG_DEBUG, " lpc type: Levinson-Durbin recursion with Welch window\n");
167
        break;
168
    case AV_LPC_TYPE_CHOLESKY:
169
        av_log(avctx, AV_LOG_DEBUG, " lpc type: Cholesky factorization, %d pass%s\n",
170
               opt->lpc_passes, opt->lpc_passes == 1 ? "" : "es");
171
        break;
172
    }
173

    
174
    av_log(avctx, AV_LOG_DEBUG, " prediction order: %d, %d\n",
175
           opt->min_prediction_order, opt->max_prediction_order);
176

    
177
    switch (opt->prediction_order_method) {
178
    case ORDER_METHOD_EST:
179
        av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "estimate");
180
        break;
181
    case ORDER_METHOD_2LEVEL:
182
        av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "2-level");
183
        break;
184
    case ORDER_METHOD_4LEVEL:
185
        av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "4-level");
186
        break;
187
    case ORDER_METHOD_8LEVEL:
188
        av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "8-level");
189
        break;
190
    case ORDER_METHOD_SEARCH:
191
        av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "full search");
192
        break;
193
    case ORDER_METHOD_LOG:
194
        av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "log search");
195
        break;
196
    }
197

    
198

    
199
    av_log(avctx, AV_LOG_DEBUG, " partition order: %d, %d\n",
200
           opt->min_partition_order, opt->max_partition_order);
201

    
202
    av_log(avctx, AV_LOG_DEBUG, " block size: %d\n", avctx->frame_size);
203

    
204
    av_log(avctx, AV_LOG_DEBUG, " lpc precision: %d\n",
205
           opt->lpc_coeff_precision);
206
}
207

    
208

    
209
static av_cold int flac_encode_init(AVCodecContext *avctx)
210
{
211
    int freq = avctx->sample_rate;
212
    int channels = avctx->channels;
213
    FlacEncodeContext *s = avctx->priv_data;
214
    int i, level, ret;
215
    uint8_t *streaminfo;
216

    
217
    s->avctx = avctx;
218

    
219
    if (avctx->sample_fmt != AV_SAMPLE_FMT_S16)
220
        return -1;
221

    
222
    if (channels < 1 || channels > FLAC_MAX_CHANNELS)
223
        return -1;
224
    s->channels = channels;
225

    
226
    /* find samplerate in table */
227
    if (freq < 1)
228
        return -1;
229
    for (i = 4; i < 12; i++) {
230
        if (freq == ff_flac_sample_rate_table[i]) {
231
            s->samplerate = ff_flac_sample_rate_table[i];
232
            s->sr_code[0] = i;
233
            s->sr_code[1] = 0;
234
            break;
235
        }
236
    }
237
    /* if not in table, samplerate is non-standard */
238
    if (i == 12) {
239
        if (freq % 1000 == 0 && freq < 255000) {
240
            s->sr_code[0] = 12;
241
            s->sr_code[1] = freq / 1000;
242
        } else if (freq % 10 == 0 && freq < 655350) {
243
            s->sr_code[0] = 14;
244
            s->sr_code[1] = freq / 10;
245
        } else if (freq < 65535) {
246
            s->sr_code[0] = 13;
247
            s->sr_code[1] = freq;
248
        } else {
249
            return -1;
250
        }
251
        s->samplerate = freq;
252
    }
253

    
254
    /* set compression option defaults based on avctx->compression_level */
255
    if (avctx->compression_level < 0)
256
        s->options.compression_level = 5;
257
    else
258
        s->options.compression_level = avctx->compression_level;
259

    
260
    level = s->options.compression_level;
261
    if (level > 12) {
262
        av_log(avctx, AV_LOG_ERROR, "invalid compression level: %d\n",
263
               s->options.compression_level);
264
        return -1;
265
    }
266

    
267
    s->options.block_time_ms = ((int[]){ 27, 27, 27,105,105,105,105,105,105,105,105,105,105})[level];
268

    
269
    s->options.lpc_type      = ((int[]){ AV_LPC_TYPE_FIXED,    AV_LPC_TYPE_FIXED,    AV_LPC_TYPE_FIXED,
270
                                         AV_LPC_TYPE_LEVINSON, AV_LPC_TYPE_LEVINSON, AV_LPC_TYPE_LEVINSON,
271
                                         AV_LPC_TYPE_LEVINSON, AV_LPC_TYPE_LEVINSON, AV_LPC_TYPE_LEVINSON,
272
                                         AV_LPC_TYPE_LEVINSON, AV_LPC_TYPE_LEVINSON, AV_LPC_TYPE_LEVINSON,
273
                                         AV_LPC_TYPE_LEVINSON})[level];
274

    
275
    s->options.min_prediction_order = ((int[]){  2,  0,  0,  1,  1,  1,  1,  1,  1,  1,  1,  1,  1})[level];
276
    s->options.max_prediction_order = ((int[]){  3,  4,  4,  6,  8,  8,  8,  8, 12, 12, 12, 32, 32})[level];
277

    
278
    s->options.prediction_order_method = ((int[]){ ORDER_METHOD_EST,    ORDER_METHOD_EST,    ORDER_METHOD_EST,
279
                                                   ORDER_METHOD_EST,    ORDER_METHOD_EST,    ORDER_METHOD_EST,
280
                                                   ORDER_METHOD_4LEVEL, ORDER_METHOD_LOG,    ORDER_METHOD_4LEVEL,
281
                                                   ORDER_METHOD_LOG,    ORDER_METHOD_SEARCH, ORDER_METHOD_LOG,
282
                                                   ORDER_METHOD_SEARCH})[level];
283

    
284
    s->options.min_partition_order = ((int[]){  2,  2,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0})[level];
285
    s->options.max_partition_order = ((int[]){  2,  2,  3,  3,  3,  8,  8,  8,  8,  8,  8,  8,  8})[level];
286

    
287
    /* set compression option overrides from AVCodecContext */
288
#if FF_API_USE_LPC
289
    /* for compatibility with deprecated AVCodecContext.use_lpc */
290
    if (avctx->use_lpc == 0) {
291
        s->options.lpc_type = AV_LPC_TYPE_FIXED;
292
    } else if (avctx->use_lpc == 1) {
293
        s->options.lpc_type = AV_LPC_TYPE_LEVINSON;
294
    } else if (avctx->use_lpc > 1) {
295
        s->options.lpc_type   = AV_LPC_TYPE_CHOLESKY;
296
        s->options.lpc_passes = avctx->use_lpc - 1;
297
    }
298
#endif
299
    if (avctx->lpc_type > AV_LPC_TYPE_DEFAULT) {
300
        if (avctx->lpc_type > AV_LPC_TYPE_CHOLESKY) {
301
            av_log(avctx, AV_LOG_ERROR, "unknown lpc type: %d\n", avctx->lpc_type);
302
            return -1;
303
        }
304
        s->options.lpc_type = avctx->lpc_type;
305
        if (s->options.lpc_type == AV_LPC_TYPE_CHOLESKY) {
306
            if (avctx->lpc_passes < 0) {
307
                // default number of passes for Cholesky
308
                s->options.lpc_passes = 2;
309
            } else if (avctx->lpc_passes == 0) {
310
                av_log(avctx, AV_LOG_ERROR, "invalid number of lpc passes: %d\n",
311
                       avctx->lpc_passes);
312
                return -1;
313
            } else {
314
                s->options.lpc_passes = avctx->lpc_passes;
315
            }
316
        }
317
    }
318

    
319
    if (s->options.lpc_type == AV_LPC_TYPE_NONE) {
320
        s->options.min_prediction_order = 0;
321
    } else if (avctx->min_prediction_order >= 0) {
322
        if (s->options.lpc_type == AV_LPC_TYPE_FIXED) {
323
            if (avctx->min_prediction_order > MAX_FIXED_ORDER) {
324
                av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n",
325
                       avctx->min_prediction_order);
326
                return -1;
327
            }
328
        } else if (avctx->min_prediction_order < MIN_LPC_ORDER ||
329
                   avctx->min_prediction_order > MAX_LPC_ORDER) {
330
            av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n",
331
                   avctx->min_prediction_order);
332
            return -1;
333
        }
334
        s->options.min_prediction_order = avctx->min_prediction_order;
335
    }
336
    if (s->options.lpc_type == AV_LPC_TYPE_NONE) {
337
        s->options.max_prediction_order = 0;
338
    } else if (avctx->max_prediction_order >= 0) {
339
        if (s->options.lpc_type == AV_LPC_TYPE_FIXED) {
340
            if (avctx->max_prediction_order > MAX_FIXED_ORDER) {
341
                av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n",
342
                       avctx->max_prediction_order);
343
                return -1;
344
            }
345
        } else if (avctx->max_prediction_order < MIN_LPC_ORDER ||
346
                   avctx->max_prediction_order > MAX_LPC_ORDER) {
347
            av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n",
348
                   avctx->max_prediction_order);
349
            return -1;
350
        }
351
        s->options.max_prediction_order = avctx->max_prediction_order;
352
    }
353
    if (s->options.max_prediction_order < s->options.min_prediction_order) {
354
        av_log(avctx, AV_LOG_ERROR, "invalid prediction orders: min=%d max=%d\n",
355
               s->options.min_prediction_order, s->options.max_prediction_order);
356
        return -1;
357
    }
358

    
359
    if (avctx->prediction_order_method >= 0) {
360
        if (avctx->prediction_order_method > ORDER_METHOD_LOG) {
361
            av_log(avctx, AV_LOG_ERROR, "invalid prediction order method: %d\n",
362
                   avctx->prediction_order_method);
363
            return -1;
364
        }
365
        s->options.prediction_order_method = avctx->prediction_order_method;
366
    }
367

    
368
    if (avctx->min_partition_order >= 0) {
369
        if (avctx->min_partition_order > MAX_PARTITION_ORDER) {
370
            av_log(avctx, AV_LOG_ERROR, "invalid min partition order: %d\n",
371
                   avctx->min_partition_order);
372
            return -1;
373
        }
374
        s->options.min_partition_order = avctx->min_partition_order;
375
    }
376
    if (avctx->max_partition_order >= 0) {
377
        if (avctx->max_partition_order > MAX_PARTITION_ORDER) {
378
            av_log(avctx, AV_LOG_ERROR, "invalid max partition order: %d\n",
379
                   avctx->max_partition_order);
380
            return -1;
381
        }
382
        s->options.max_partition_order = avctx->max_partition_order;
383
    }
384
    if (s->options.max_partition_order < s->options.min_partition_order) {
385
        av_log(avctx, AV_LOG_ERROR, "invalid partition orders: min=%d max=%d\n",
386
               s->options.min_partition_order, s->options.max_partition_order);
387
        return -1;
388
    }
389

    
390
    if (avctx->frame_size > 0) {
391
        if (avctx->frame_size < FLAC_MIN_BLOCKSIZE ||
392
                avctx->frame_size > FLAC_MAX_BLOCKSIZE) {
393
            av_log(avctx, AV_LOG_ERROR, "invalid block size: %d\n",
394
                   avctx->frame_size);
395
            return -1;
396
        }
397
    } else {
398
        s->avctx->frame_size = select_blocksize(s->samplerate, s->options.block_time_ms);
399
    }
400
    s->max_blocksize = s->avctx->frame_size;
401

    
402
    /* set LPC precision */
403
    if (avctx->lpc_coeff_precision > 0) {
404
        if (avctx->lpc_coeff_precision > MAX_LPC_PRECISION) {
405
            av_log(avctx, AV_LOG_ERROR, "invalid lpc coeff precision: %d\n",
406
                   avctx->lpc_coeff_precision);
407
            return -1;
408
        }
409
        s->options.lpc_coeff_precision = avctx->lpc_coeff_precision;
410
    } else {
411
        /* default LPC precision */
412
        s->options.lpc_coeff_precision = 15;
413
    }
414

    
415
    /* set maximum encoded frame size in verbatim mode */
416
    s->max_framesize = ff_flac_get_max_frame_size(s->avctx->frame_size,
417
                                                  s->channels, 16);
418

    
419
    /* initialize MD5 context */
420
    s->md5ctx = av_malloc(av_md5_size);
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
    avctx->coded_frame = avcodec_alloc_frame();
436
    if (!avctx->coded_frame)
437
        return AVERROR(ENOMEM);
438

    
439
    ret = ff_lpc_init(&s->lpc_ctx, avctx->frame_size,
440
                      s->options.max_prediction_order, AV_LPC_TYPE_LEVINSON);
441

    
442
    dprint_compression_options(s);
443

    
444
    return ret;
445
}
446

    
447

    
448
static void init_frame(FlacEncodeContext *s)
449
{
450
    int i, ch;
451
    FlacFrame *frame;
452

    
453
    frame = &s->frame;
454

    
455
    for (i = 0; i < 16; i++) {
456
        if (s->avctx->frame_size == ff_flac_blocksize_table[i]) {
457
            frame->blocksize  = ff_flac_blocksize_table[i];
458
            frame->bs_code[0] = i;
459
            frame->bs_code[1] = 0;
460
            break;
461
        }
462
    }
463
    if (i == 16) {
464
        frame->blocksize = s->avctx->frame_size;
465
        if (frame->blocksize <= 256) {
466
            frame->bs_code[0] = 6;
467
            frame->bs_code[1] = frame->blocksize-1;
468
        } else {
469
            frame->bs_code[0] = 7;
470
            frame->bs_code[1] = frame->blocksize-1;
471
        }
472
    }
473

    
474
    for (ch = 0; ch < s->channels; ch++)
475
        frame->subframes[ch].obits = 16;
476

    
477
    frame->verbatim_only = 0;
478
}
479

    
480

    
481
/**
482
 * Copy channel-interleaved input samples into separate subframes.
483
 */
484
static void copy_samples(FlacEncodeContext *s, const int16_t *samples)
485
{
486
    int i, j, ch;
487
    FlacFrame *frame;
488

    
489
    frame = &s->frame;
490
    for (i = 0, j = 0; i < frame->blocksize; i++)
491
        for (ch = 0; ch < s->channels; ch++, j++)
492
            frame->subframes[ch].samples[i] = samples[j];
493
}
494

    
495

    
496
static int rice_count_exact(int32_t *res, int n, int k)
497
{
498
    int i;
499
    int count = 0;
500

    
501
    for (i = 0; i < n; i++) {
502
        int32_t v = -2 * res[i] - 1;
503
        v ^= v >> 31;
504
        count += (v >> k) + 1 + k;
505
    }
506
    return count;
507
}
508

    
509

    
510
static int subframe_count_exact(FlacEncodeContext *s, FlacSubframe *sub,
511
                                int pred_order)
512
{
513
    int p, porder, psize;
514
    int i, part_end;
515
    int count = 0;
516

    
517
    /* subframe header */
518
    count += 8;
519

    
520
    /* subframe */
521
    if (sub->type == FLAC_SUBFRAME_CONSTANT) {
522
        count += sub->obits;
523
    } else if (sub->type == FLAC_SUBFRAME_VERBATIM) {
524
        count += s->frame.blocksize * sub->obits;
525
    } else {
526
        /* warm-up samples */
527
        count += pred_order * sub->obits;
528

    
529
        /* LPC coefficients */
530
        if (sub->type == FLAC_SUBFRAME_LPC)
531
            count += 4 + 5 + pred_order * s->options.lpc_coeff_precision;
532

    
533
        /* rice-encoded block */
534
        count += 2;
535

    
536
        /* partition order */
537
        porder = sub->rc.porder;
538
        psize  = s->frame.blocksize >> porder;
539
        count += 4;
540

    
541
        /* residual */
542
        i        = pred_order;
543
        part_end = psize;
544
        for (p = 0; p < 1 << porder; p++) {
545
            int k = sub->rc.params[p];
546
            count += 4;
547
            count += rice_count_exact(&sub->residual[i], part_end - i, k);
548
            i = part_end;
549
            part_end = FFMIN(s->frame.blocksize, part_end + psize);
550
        }
551
    }
552

    
553
    return count;
554
}
555

    
556

    
557
#define rice_encode_count(sum, n, k) (((n)*((k)+1))+((sum-(n>>1))>>(k)))
558

    
559
/**
560
 * Solve for d/dk(rice_encode_count) = n-((sum-(n>>1))>>(k+1)) = 0.
561
 */
562
static int find_optimal_param(uint32_t sum, int n)
563
{
564
    int k;
565
    uint32_t sum2;
566

    
567
    if (sum <= n >> 1)
568
        return 0;
569
    sum2 = sum - (n >> 1);
570
    k    = av_log2(n < 256 ? FASTDIV(sum2, n) : sum2 / n);
571
    return FFMIN(k, MAX_RICE_PARAM);
572
}
573

    
574

    
575
static uint32_t calc_optimal_rice_params(RiceContext *rc, int porder,
576
                                         uint32_t *sums, int n, int pred_order)
577
{
578
    int i;
579
    int k, cnt, part;
580
    uint32_t all_bits;
581

    
582
    part     = (1 << porder);
583
    all_bits = 4 * part;
584

    
585
    cnt = (n >> porder) - pred_order;
586
    for (i = 0; i < part; i++) {
587
        k = find_optimal_param(sums[i], cnt);
588
        rc->params[i] = k;
589
        all_bits += rice_encode_count(sums[i], cnt, k);
590
        cnt = n >> porder;
591
    }
592

    
593
    rc->porder = porder;
594

    
595
    return all_bits;
596
}
597

    
598

    
599
static void calc_sums(int pmin, int pmax, uint32_t *data, int n, int pred_order,
600
                      uint32_t sums[][MAX_PARTITIONS])
601
{
602
    int i, j;
603
    int parts;
604
    uint32_t *res, *res_end;
605

    
606
    /* sums for highest level */
607
    parts   = (1 << pmax);
608
    res     = &data[pred_order];
609
    res_end = &data[n >> pmax];
610
    for (i = 0; i < parts; i++) {
611
        uint32_t sum = 0;
612
        while (res < res_end)
613
            sum += *(res++);
614
        sums[pmax][i] = sum;
615
        res_end += n >> pmax;
616
    }
617
    /* sums for lower levels */
618
    for (i = pmax - 1; i >= pmin; i--) {
619
        parts = (1 << i);
620
        for (j = 0; j < parts; j++)
621
            sums[i][j] = sums[i+1][2*j] + sums[i+1][2*j+1];
622
    }
623
}
624

    
625

    
626
static uint32_t calc_rice_params(RiceContext *rc, int pmin, int pmax,
627
                                 int32_t *data, int n, int pred_order)
628
{
629
    int i;
630
    uint32_t bits[MAX_PARTITION_ORDER+1];
631
    int opt_porder;
632
    RiceContext tmp_rc;
633
    uint32_t *udata;
634
    uint32_t sums[MAX_PARTITION_ORDER+1][MAX_PARTITIONS];
635

    
636
    assert(pmin >= 0 && pmin <= MAX_PARTITION_ORDER);
637
    assert(pmax >= 0 && pmax <= MAX_PARTITION_ORDER);
638
    assert(pmin <= pmax);
639

    
640
    udata = av_malloc(n * sizeof(uint32_t));
641
    for (i = 0; i < n; i++)
642
        udata[i] = (2*data[i]) ^ (data[i]>>31);
643

    
644
    calc_sums(pmin, pmax, udata, n, pred_order, sums);
645

    
646
    opt_porder = pmin;
647
    bits[pmin] = UINT32_MAX;
648
    for (i = pmin; i <= pmax; i++) {
649
        bits[i] = calc_optimal_rice_params(&tmp_rc, i, sums[i], n, pred_order);
650
        if (bits[i] <= bits[opt_porder]) {
651
            opt_porder = i;
652
            *rc = tmp_rc;
653
        }
654
    }
655

    
656
    av_freep(&udata);
657
    return bits[opt_porder];
658
}
659

    
660

    
661
static int get_max_p_order(int max_porder, int n, int order)
662
{
663
    int porder = FFMIN(max_porder, av_log2(n^(n-1)));
664
    if (order > 0)
665
        porder = FFMIN(porder, av_log2(n/order));
666
    return porder;
667
}
668

    
669

    
670
static uint32_t find_subframe_rice_params(FlacEncodeContext *s,
671
                                          FlacSubframe *sub, int pred_order)
672
{
673
    int pmin = get_max_p_order(s->options.min_partition_order,
674
                               s->frame.blocksize, pred_order);
675
    int pmax = get_max_p_order(s->options.max_partition_order,
676
                               s->frame.blocksize, pred_order);
677

    
678
    uint32_t bits = 8 + pred_order * sub->obits + 2 + 4;
679
    if (sub->type == FLAC_SUBFRAME_LPC)
680
        bits += 4 + 5 + pred_order * s->options.lpc_coeff_precision;
681
    bits += calc_rice_params(&sub->rc, pmin, pmax, sub->residual,
682
                             s->frame.blocksize, pred_order);
683
    return bits;
684
}
685

    
686

    
687
static void encode_residual_fixed(int32_t *res, const int32_t *smp, int n,
688
                                  int order)
689
{
690
    int i;
691

    
692
    for (i = 0; i < order; i++)
693
        res[i] = smp[i];
694

    
695
    if (order == 0) {
696
        for (i = order; i < n; i++)
697
            res[i] = smp[i];
698
    } else if (order == 1) {
699
        for (i = order; i < n; i++)
700
            res[i] = smp[i] - smp[i-1];
701
    } else if (order == 2) {
702
        int a = smp[order-1] - smp[order-2];
703
        for (i = order; i < n; i += 2) {
704
            int b    = smp[i  ] - smp[i-1];
705
            res[i]   = b - a;
706
            a        = smp[i+1] - smp[i  ];
707
            res[i+1] = a - b;
708
        }
709
    } else if (order == 3) {
710
        int a = smp[order-1] -   smp[order-2];
711
        int c = smp[order-1] - 2*smp[order-2] + smp[order-3];
712
        for (i = order; i < n; i += 2) {
713
            int b    = smp[i  ] - smp[i-1];
714
            int d    = b - a;
715
            res[i]   = d - c;
716
            a        = smp[i+1] - smp[i  ];
717
            c        = a - b;
718
            res[i+1] = c - d;
719
        }
720
    } else {
721
        int a = smp[order-1] -   smp[order-2];
722
        int c = smp[order-1] - 2*smp[order-2] +   smp[order-3];
723
        int e = smp[order-1] - 3*smp[order-2] + 3*smp[order-3] - smp[order-4];
724
        for (i = order; i < n; i += 2) {
725
            int b    = smp[i  ] - smp[i-1];
726
            int d    = b - a;
727
            int f    = d - c;
728
            res[i  ] = f - e;
729
            a        = smp[i+1] - smp[i  ];
730
            c        = a - b;
731
            e        = c - d;
732
            res[i+1] = e - f;
733
        }
734
    }
735
}
736

    
737

    
738
#define LPC1(x) {\
739
    int c = coefs[(x)-1];\
740
    p0   += c * s;\
741
    s     = smp[i-(x)+1];\
742
    p1   += c * s;\
743
}
744

    
745
static av_always_inline void encode_residual_lpc_unrolled(int32_t *res,
746
                                    const int32_t *smp, int n, int order,
747
                                    const int32_t *coefs, int shift, int big)
748
{
749
    int i;
750
    for (i = order; i < n; i += 2) {
751
        int s  = smp[i-order];
752
        int p0 = 0, p1 = 0;
753
        if (big) {
754
            switch (order) {
755
            case 32: LPC1(32)
756
            case 31: LPC1(31)
757
            case 30: LPC1(30)
758
            case 29: LPC1(29)
759
            case 28: LPC1(28)
760
            case 27: LPC1(27)
761
            case 26: LPC1(26)
762
            case 25: LPC1(25)
763
            case 24: LPC1(24)
764
            case 23: LPC1(23)
765
            case 22: LPC1(22)
766
            case 21: LPC1(21)
767
            case 20: LPC1(20)
768
            case 19: LPC1(19)
769
            case 18: LPC1(18)
770
            case 17: LPC1(17)
771
            case 16: LPC1(16)
772
            case 15: LPC1(15)
773
            case 14: LPC1(14)
774
            case 13: LPC1(13)
775
            case 12: LPC1(12)
776
            case 11: LPC1(11)
777
            case 10: LPC1(10)
778
            case  9: LPC1( 9)
779
                     LPC1( 8)
780
                     LPC1( 7)
781
                     LPC1( 6)
782
                     LPC1( 5)
783
                     LPC1( 4)
784
                     LPC1( 3)
785
                     LPC1( 2)
786
                     LPC1( 1)
787
            }
788
        } else {
789
            switch (order) {
790
            case  8: LPC1( 8)
791
            case  7: LPC1( 7)
792
            case  6: LPC1( 6)
793
            case  5: LPC1( 5)
794
            case  4: LPC1( 4)
795
            case  3: LPC1( 3)
796
            case  2: LPC1( 2)
797
            case  1: LPC1( 1)
798
            }
799
        }
800
        res[i  ] = smp[i  ] - (p0 >> shift);
801
        res[i+1] = smp[i+1] - (p1 >> shift);
802
    }
803
}
804

    
805

    
806
static void encode_residual_lpc(int32_t *res, const int32_t *smp, int n,
807
                                int order, const int32_t *coefs, int shift)
808
{
809
    int i;
810
    for (i = 0; i < order; i++)
811
        res[i] = smp[i];
812
#if CONFIG_SMALL
813
    for (i = order; i < n; i += 2) {
814
        int j;
815
        int s  = smp[i];
816
        int p0 = 0, p1 = 0;
817
        for (j = 0; j < order; j++) {
818
            int c = coefs[j];
819
            p1   += c * s;
820
            s     = smp[i-j-1];
821
            p0   += c * s;
822
        }
823
        res[i  ] = smp[i  ] - (p0 >> shift);
824
        res[i+1] = smp[i+1] - (p1 >> shift);
825
    }
826
#else
827
    switch (order) {
828
    case  1: encode_residual_lpc_unrolled(res, smp, n, 1, coefs, shift, 0); break;
829
    case  2: encode_residual_lpc_unrolled(res, smp, n, 2, coefs, shift, 0); break;
830
    case  3: encode_residual_lpc_unrolled(res, smp, n, 3, coefs, shift, 0); break;
831
    case  4: encode_residual_lpc_unrolled(res, smp, n, 4, coefs, shift, 0); break;
832
    case  5: encode_residual_lpc_unrolled(res, smp, n, 5, coefs, shift, 0); break;
833
    case  6: encode_residual_lpc_unrolled(res, smp, n, 6, coefs, shift, 0); break;
834
    case  7: encode_residual_lpc_unrolled(res, smp, n, 7, coefs, shift, 0); break;
835
    case  8: encode_residual_lpc_unrolled(res, smp, n, 8, coefs, shift, 0); break;
836
    default: encode_residual_lpc_unrolled(res, smp, n, order, coefs, shift, 1); break;
837
    }
838
#endif
839
}
840

    
841

    
842
static int encode_residual_ch(FlacEncodeContext *s, int ch)
843
{
844
    int i, n;
845
    int min_order, max_order, opt_order, omethod;
846
    FlacFrame *frame;
847
    FlacSubframe *sub;
848
    int32_t coefs[MAX_LPC_ORDER][MAX_LPC_ORDER];
849
    int shift[MAX_LPC_ORDER];
850
    int32_t *res, *smp;
851

    
852
    frame = &s->frame;
853
    sub   = &frame->subframes[ch];
854
    res   = sub->residual;
855
    smp   = sub->samples;
856
    n     = frame->blocksize;
857

    
858
    /* CONSTANT */
859
    for (i = 1; i < n; i++)
860
        if(smp[i] != smp[0])
861
            break;
862
    if (i == n) {
863
        sub->type = sub->type_code = FLAC_SUBFRAME_CONSTANT;
864
        res[0] = smp[0];
865
        return subframe_count_exact(s, sub, 0);
866
    }
867

    
868
    /* VERBATIM */
869
    if (frame->verbatim_only || n < 5) {
870
        sub->type = sub->type_code = FLAC_SUBFRAME_VERBATIM;
871
        memcpy(res, smp, n * sizeof(int32_t));
872
        return subframe_count_exact(s, sub, 0);
873
    }
874

    
875
    min_order  = s->options.min_prediction_order;
876
    max_order  = s->options.max_prediction_order;
877
    omethod    = s->options.prediction_order_method;
878

    
879
    /* FIXED */
880
    sub->type = FLAC_SUBFRAME_FIXED;
881
    if (s->options.lpc_type == AV_LPC_TYPE_NONE  ||
882
        s->options.lpc_type == AV_LPC_TYPE_FIXED || n <= max_order) {
883
        uint32_t bits[MAX_FIXED_ORDER+1];
884
        if (max_order > MAX_FIXED_ORDER)
885
            max_order = MAX_FIXED_ORDER;
886
        opt_order = 0;
887
        bits[0]   = UINT32_MAX;
888
        for (i = min_order; i <= max_order; i++) {
889
            encode_residual_fixed(res, smp, n, i);
890
            bits[i] = find_subframe_rice_params(s, sub, i);
891
            if (bits[i] < bits[opt_order])
892
                opt_order = i;
893
        }
894
        sub->order     = opt_order;
895
        sub->type_code = sub->type | sub->order;
896
        if (sub->order != max_order) {
897
            encode_residual_fixed(res, smp, n, sub->order);
898
            find_subframe_rice_params(s, sub, sub->order);
899
        }
900
        return subframe_count_exact(s, sub, sub->order);
901
    }
902

    
903
    /* LPC */
904
    sub->type = FLAC_SUBFRAME_LPC;
905
    opt_order = ff_lpc_calc_coefs(&s->lpc_ctx, smp, n, min_order, max_order,
906
                                  s->options.lpc_coeff_precision, coefs, shift, s->options.lpc_type,
907
                                  s->options.lpc_passes, omethod,
908
                                  MAX_LPC_SHIFT, 0);
909

    
910
    if (omethod == ORDER_METHOD_2LEVEL ||
911
        omethod == ORDER_METHOD_4LEVEL ||
912
        omethod == ORDER_METHOD_8LEVEL) {
913
        int levels = 1 << omethod;
914
        uint32_t bits[1 << ORDER_METHOD_8LEVEL];
915
        int order;
916
        int opt_index   = levels-1;
917
        opt_order       = max_order-1;
918
        bits[opt_index] = UINT32_MAX;
919
        for (i = levels-1; i >= 0; i--) {
920
            order = min_order + (((max_order-min_order+1) * (i+1)) / levels)-1;
921
            if (order < 0)
922
                order = 0;
923
            encode_residual_lpc(res, smp, n, order+1, coefs[order], shift[order]);
924
            bits[i] = find_subframe_rice_params(s, sub, order+1);
925
            if (bits[i] < bits[opt_index]) {
926
                opt_index = i;
927
                opt_order = order;
928
            }
929
        }
930
        opt_order++;
931
    } else if (omethod == ORDER_METHOD_SEARCH) {
932
        // brute-force optimal order search
933
        uint32_t bits[MAX_LPC_ORDER];
934
        opt_order = 0;
935
        bits[0]   = UINT32_MAX;
936
        for (i = min_order-1; i < max_order; i++) {
937
            encode_residual_lpc(res, smp, n, i+1, coefs[i], shift[i]);
938
            bits[i] = find_subframe_rice_params(s, sub, i+1);
939
            if (bits[i] < bits[opt_order])
940
                opt_order = i;
941
        }
942
        opt_order++;
943
    } else if (omethod == ORDER_METHOD_LOG) {
944
        uint32_t bits[MAX_LPC_ORDER];
945
        int step;
946

    
947
        opt_order = min_order - 1 + (max_order-min_order)/3;
948
        memset(bits, -1, sizeof(bits));
949

    
950
        for (step = 16; step; step >>= 1) {
951
            int last = opt_order;
952
            for (i = last-step; i <= last+step; i += step) {
953
                if (i < min_order-1 || i >= max_order || bits[i] < UINT32_MAX)
954
                    continue;
955
                encode_residual_lpc(res, smp, n, i+1, coefs[i], shift[i]);
956
                bits[i] = find_subframe_rice_params(s, sub, i+1);
957
                if (bits[i] < bits[opt_order])
958
                    opt_order = i;
959
            }
960
        }
961
        opt_order++;
962
    }
963

    
964
    sub->order     = opt_order;
965
    sub->type_code = sub->type | (sub->order-1);
966
    sub->shift     = shift[sub->order-1];
967
    for (i = 0; i < sub->order; i++)
968
        sub->coefs[i] = coefs[sub->order-1][i];
969

    
970
    encode_residual_lpc(res, smp, n, sub->order, sub->coefs, sub->shift);
971

    
972
    find_subframe_rice_params(s, sub, sub->order);
973

    
974
    return subframe_count_exact(s, sub, sub->order);
975
}
976

    
977

    
978
static int count_frame_header(FlacEncodeContext *s)
979
{
980
    uint8_t tmp;
981
    int count;
982

    
983
    /*
984
    <14> Sync code
985
    <1>  Reserved
986
    <1>  Blocking strategy
987
    <4>  Block size in inter-channel samples
988
    <4>  Sample rate
989
    <4>  Channel assignment
990
    <3>  Sample size in bits
991
    <1>  Reserved
992
    */
993
    count = 32;
994

    
995
    /* coded frame number */
996
    PUT_UTF8(s->frame_count, tmp, count += 8;)
997

    
998
    /* explicit block size */
999
    if (s->frame.bs_code[0] == 6)
1000
        count += 8;
1001
    else if (s->frame.bs_code[0] == 7)
1002
        count += 16;
1003

    
1004
    /* explicit sample rate */
1005
    count += ((s->sr_code[0] == 12) + (s->sr_code[0] > 12)) * 8;
1006

    
1007
    /* frame header CRC-8 */
1008
    count += 8;
1009

    
1010
    return count;
1011
}
1012

    
1013

    
1014
static int encode_frame(FlacEncodeContext *s)
1015
{
1016
    int ch, count;
1017

    
1018
    count = count_frame_header(s);
1019

    
1020
    for (ch = 0; ch < s->channels; ch++)
1021
        count += encode_residual_ch(s, ch);
1022

    
1023
    count += (8 - (count & 7)) & 7; // byte alignment
1024
    count += 16;                    // CRC-16
1025

    
1026
    return count >> 3;
1027
}
1028

    
1029

    
1030
static int estimate_stereo_mode(int32_t *left_ch, int32_t *right_ch, int n)
1031
{
1032
    int i, best;
1033
    int32_t lt, rt;
1034
    uint64_t sum[4];
1035
    uint64_t score[4];
1036
    int k;
1037

    
1038
    /* calculate sum of 2nd order residual for each channel */
1039
    sum[0] = sum[1] = sum[2] = sum[3] = 0;
1040
    for (i = 2; i < n; i++) {
1041
        lt = left_ch[i]  - 2*left_ch[i-1]  + left_ch[i-2];
1042
        rt = right_ch[i] - 2*right_ch[i-1] + right_ch[i-2];
1043
        sum[2] += FFABS((lt + rt) >> 1);
1044
        sum[3] += FFABS(lt - rt);
1045
        sum[0] += FFABS(lt);
1046
        sum[1] += FFABS(rt);
1047
    }
1048
    /* estimate bit counts */
1049
    for (i = 0; i < 4; i++) {
1050
        k      = find_optimal_param(2 * sum[i], n);
1051
        sum[i] = rice_encode_count( 2 * sum[i], n, k);
1052
    }
1053

    
1054
    /* calculate score for each mode */
1055
    score[0] = sum[0] + sum[1];
1056
    score[1] = sum[0] + sum[3];
1057
    score[2] = sum[1] + sum[3];
1058
    score[3] = sum[2] + sum[3];
1059

    
1060
    /* return mode with lowest score */
1061
    best = 0;
1062
    for (i = 1; i < 4; i++)
1063
        if (score[i] < score[best])
1064
            best = i;
1065
    if (best == 0) {
1066
        return FLAC_CHMODE_INDEPENDENT;
1067
    } else if (best == 1) {
1068
        return FLAC_CHMODE_LEFT_SIDE;
1069
    } else if (best == 2) {
1070
        return FLAC_CHMODE_RIGHT_SIDE;
1071
    } else {
1072
        return FLAC_CHMODE_MID_SIDE;
1073
    }
1074
}
1075

    
1076

    
1077
/**
1078
 * Perform stereo channel decorrelation.
1079
 */
1080
static void channel_decorrelation(FlacEncodeContext *s)
1081
{
1082
    FlacFrame *frame;
1083
    int32_t *left, *right;
1084
    int i, n;
1085

    
1086
    frame = &s->frame;
1087
    n     = frame->blocksize;
1088
    left  = frame->subframes[0].samples;
1089
    right = frame->subframes[1].samples;
1090

    
1091
    if (s->channels != 2) {
1092
        frame->ch_mode = FLAC_CHMODE_INDEPENDENT;
1093
        return;
1094
    }
1095

    
1096
    frame->ch_mode = estimate_stereo_mode(left, right, n);
1097

    
1098
    /* perform decorrelation and adjust bits-per-sample */
1099
    if (frame->ch_mode == FLAC_CHMODE_INDEPENDENT)
1100
        return;
1101
    if (frame->ch_mode == FLAC_CHMODE_MID_SIDE) {
1102
        int32_t tmp;
1103
        for (i = 0; i < n; i++) {
1104
            tmp      = left[i];
1105
            left[i]  = (tmp + right[i]) >> 1;
1106
            right[i] =  tmp - right[i];
1107
        }
1108
        frame->subframes[1].obits++;
1109
    } else if (frame->ch_mode == FLAC_CHMODE_LEFT_SIDE) {
1110
        for (i = 0; i < n; i++)
1111
            right[i] = left[i] - right[i];
1112
        frame->subframes[1].obits++;
1113
    } else {
1114
        for (i = 0; i < n; i++)
1115
            left[i] -= right[i];
1116
        frame->subframes[0].obits++;
1117
    }
1118
}
1119

    
1120

    
1121
static void write_utf8(PutBitContext *pb, uint32_t val)
1122
{
1123
    uint8_t tmp;
1124
    PUT_UTF8(val, tmp, put_bits(pb, 8, tmp);)
1125
}
1126

    
1127

    
1128
static void write_frame_header(FlacEncodeContext *s)
1129
{
1130
    FlacFrame *frame;
1131
    int crc;
1132

    
1133
    frame = &s->frame;
1134

    
1135
    put_bits(&s->pb, 16, 0xFFF8);
1136
    put_bits(&s->pb, 4, frame->bs_code[0]);
1137
    put_bits(&s->pb, 4, s->sr_code[0]);
1138

    
1139
    if (frame->ch_mode == FLAC_CHMODE_INDEPENDENT)
1140
        put_bits(&s->pb, 4, s->channels-1);
1141
    else
1142
        put_bits(&s->pb, 4, frame->ch_mode);
1143

    
1144
    put_bits(&s->pb, 3, 4); /* bits-per-sample code */
1145
    put_bits(&s->pb, 1, 0);
1146
    write_utf8(&s->pb, s->frame_count);
1147

    
1148
    if (frame->bs_code[0] == 6)
1149
        put_bits(&s->pb, 8, frame->bs_code[1]);
1150
    else if (frame->bs_code[0] == 7)
1151
        put_bits(&s->pb, 16, frame->bs_code[1]);
1152

    
1153
    if (s->sr_code[0] == 12)
1154
        put_bits(&s->pb, 8, s->sr_code[1]);
1155
    else if (s->sr_code[0] > 12)
1156
        put_bits(&s->pb, 16, s->sr_code[1]);
1157

    
1158
    flush_put_bits(&s->pb);
1159
    crc = av_crc(av_crc_get_table(AV_CRC_8_ATM), 0, s->pb.buf,
1160
                 put_bits_count(&s->pb) >> 3);
1161
    put_bits(&s->pb, 8, crc);
1162
}
1163

    
1164

    
1165
static void write_subframes(FlacEncodeContext *s)
1166
{
1167
    int ch;
1168

    
1169
    for (ch = 0; ch < s->channels; ch++) {
1170
        FlacSubframe *sub = &s->frame.subframes[ch];
1171
        int i, p, porder, psize;
1172
        int32_t *part_end;
1173
        int32_t *res       =  sub->residual;
1174
        int32_t *frame_end = &sub->residual[s->frame.blocksize];
1175

    
1176
        /* subframe header */
1177
        put_bits(&s->pb, 1, 0);
1178
        put_bits(&s->pb, 6, sub->type_code);
1179
        put_bits(&s->pb, 1, 0); /* no wasted bits */
1180

    
1181
        /* subframe */
1182
        if (sub->type == FLAC_SUBFRAME_CONSTANT) {
1183
            put_sbits(&s->pb, sub->obits, res[0]);
1184
        } else if (sub->type == FLAC_SUBFRAME_VERBATIM) {
1185
            while (res < frame_end)
1186
                put_sbits(&s->pb, sub->obits, *res++);
1187
        } else {
1188
            /* warm-up samples */
1189
            for (i = 0; i < sub->order; i++)
1190
                put_sbits(&s->pb, sub->obits, *res++);
1191

    
1192
            /* LPC coefficients */
1193
            if (sub->type == FLAC_SUBFRAME_LPC) {
1194
                int cbits = s->options.lpc_coeff_precision;
1195
                put_bits( &s->pb, 4, cbits-1);
1196
                put_sbits(&s->pb, 5, sub->shift);
1197
                for (i = 0; i < sub->order; i++)
1198
                    put_sbits(&s->pb, cbits, sub->coefs[i]);
1199
            }
1200

    
1201
            /* rice-encoded block */
1202
            put_bits(&s->pb, 2, 0);
1203

    
1204
            /* partition order */
1205
            porder  = sub->rc.porder;
1206
            psize   = s->frame.blocksize >> porder;
1207
            put_bits(&s->pb, 4, porder);
1208

    
1209
            /* residual */
1210
            part_end  = &sub->residual[psize];
1211
            for (p = 0; p < 1 << porder; p++) {
1212
                int k = sub->rc.params[p];
1213
                put_bits(&s->pb, 4, k);
1214
                while (res < part_end)
1215
                    set_sr_golomb_flac(&s->pb, *res++, k, INT32_MAX, 0);
1216
                part_end = FFMIN(frame_end, part_end + psize);
1217
            }
1218
        }
1219
    }
1220
}
1221

    
1222

    
1223
static void write_frame_footer(FlacEncodeContext *s)
1224
{
1225
    int crc;
1226
    flush_put_bits(&s->pb);
1227
    crc = av_bswap16(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, s->pb.buf,
1228
                            put_bits_count(&s->pb)>>3));
1229
    put_bits(&s->pb, 16, crc);
1230
    flush_put_bits(&s->pb);
1231
}
1232

    
1233

    
1234
static int write_frame(FlacEncodeContext *s, uint8_t *frame, int buf_size)
1235
{
1236
    init_put_bits(&s->pb, frame, buf_size);
1237
    write_frame_header(s);
1238
    write_subframes(s);
1239
    write_frame_footer(s);
1240
    return put_bits_count(&s->pb) >> 3;
1241
}
1242

    
1243

    
1244
static void update_md5_sum(FlacEncodeContext *s, const int16_t *samples)
1245
{
1246
#if HAVE_BIGENDIAN
1247
    int i;
1248
    for (i = 0; i < s->frame.blocksize * s->channels; i++) {
1249
        int16_t smp = av_le2ne16(samples[i]);
1250
        av_md5_update(s->md5ctx, (uint8_t *)&smp, 2);
1251
    }
1252
#else
1253
    av_md5_update(s->md5ctx, (const uint8_t *)samples, s->frame.blocksize*s->channels*2);
1254
#endif
1255
}
1256

    
1257

    
1258
static int flac_encode_frame(AVCodecContext *avctx, uint8_t *frame,
1259
                             int buf_size, void *data)
1260
{
1261
    FlacEncodeContext *s;
1262
    const int16_t *samples = data;
1263
    int frame_bytes, out_bytes;
1264

    
1265
    s = avctx->priv_data;
1266

    
1267
    /* when the last block is reached, update the header in extradata */
1268
    if (!data) {
1269
        s->max_framesize = s->max_encoded_framesize;
1270
        av_md5_final(s->md5ctx, s->md5sum);
1271
        write_streaminfo(s, avctx->extradata);
1272
        return 0;
1273
    }
1274

    
1275
    /* change max_framesize for small final frame */
1276
    if (avctx->frame_size < s->frame.blocksize) {
1277
        s->max_framesize = ff_flac_get_max_frame_size(avctx->frame_size,
1278
                                                      s->channels, 16);
1279
    }
1280

    
1281
    init_frame(s);
1282

    
1283
    copy_samples(s, samples);
1284

    
1285
    channel_decorrelation(s);
1286

    
1287
    frame_bytes = encode_frame(s);
1288

    
1289
    /* fallback to verbatim mode if the compressed frame is larger than it
1290
       would be if encoded uncompressed. */
1291
    if (frame_bytes > s->max_framesize) {
1292
        s->frame.verbatim_only = 1;
1293
        frame_bytes = encode_frame(s);
1294
    }
1295

    
1296
    if (buf_size < frame_bytes) {
1297
        av_log(avctx, AV_LOG_ERROR, "output buffer too small\n");
1298
        return 0;
1299
    }
1300
    out_bytes = write_frame(s, frame, buf_size);
1301

    
1302
    s->frame_count++;
1303
    avctx->coded_frame->pts = s->sample_count;
1304
    s->sample_count += avctx->frame_size;
1305
    update_md5_sum(s, samples);
1306
    if (out_bytes > s->max_encoded_framesize)
1307
        s->max_encoded_framesize = out_bytes;
1308
    if (out_bytes < s->min_framesize)
1309
        s->min_framesize = out_bytes;
1310

    
1311
    return out_bytes;
1312
}
1313

    
1314

    
1315
static av_cold int flac_encode_close(AVCodecContext *avctx)
1316
{
1317
    if (avctx->priv_data) {
1318
        FlacEncodeContext *s = avctx->priv_data;
1319
        av_freep(&s->md5ctx);
1320
        ff_lpc_end(&s->lpc_ctx);
1321
    }
1322
    av_freep(&avctx->extradata);
1323
    avctx->extradata_size = 0;
1324
    av_freep(&avctx->coded_frame);
1325
    return 0;
1326
}
1327

    
1328

    
1329
AVCodec ff_flac_encoder = {
1330
    "flac",
1331
    AVMEDIA_TYPE_AUDIO,
1332
    CODEC_ID_FLAC,
1333
    sizeof(FlacEncodeContext),
1334
    flac_encode_init,
1335
    flac_encode_frame,
1336
    flac_encode_close,
1337
    NULL,
1338
    .capabilities = CODEC_CAP_SMALL_LAST_FRAME | CODEC_CAP_DELAY,
1339
    .sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE},
1340
    .long_name = NULL_IF_CONFIG_SMALL("FLAC (Free Lossless Audio Codec)"),
1341
};